本篇博文主要内容为 2026-10-30 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 (2026-10-30)
今日共更新1441篇论文,其中:
- 自然语言处理共176篇(Computation and Language (cs.CL))
- 人工智能共395篇(Artificial Intelligence (cs.AI))
- 计算机视觉共233篇(Computer Vision and Pattern Recognition (cs.CV))
- 机器学习共374篇(Machine Learning (cs.LG))
- 多智能体系统共33篇(Multiagent Systems (cs.MA))
- 信息检索共40篇(Information Retrieval (cs.IR))
- 人机交互共51篇(Human-Computer Interaction (cs.HC))
多智能体系统
[MA-0] Collaborative AI agents for autonomous experimental research with solid-state quantum spins
【速读】：该论文旨在解决当前生成式AI在科学研究中存在过度推断、结论缺乏实证支持的问题，即现有AI系统虽能承担部分科研角色，但其得出的结论可能与证据不符。其解决方案的关键在于构建一个由大型语言模型（Large Language Models, LLMs）驱动的自主科研实验室，其中多个智能体（agents）通过相互评估工作、自我纠错与协同决策来实现科学探究的闭环。这些智能体能够自主生成与拒绝研究问题，执行独立的理论验证，并设计可证伪的实验以检验自身假设。研究以金刚石中量子自旋系综为对象，展示了智能体不仅能够推导出预测特定微波脉冲序列响应所需的最少标定设置数量，还能通过计算验证和实验设计区分竞争性理论预测。尽管初始实验结果与理论一致，但后续参考测量发现信号漂移，智能体据此判定结果不具可靠性，主动修订测量协议并维持严格标准，最终获得支持理论预测的可靠数据。这表明，通过协作式、可自我校验的智能体架构，生成式AI可在复杂科学任务中实现从问题提出到实验验证的全流程自主推进。
链接: https://arxiv.org/abs/2610.40589
作者: El Mustapha Mansouri,Eikichi Kimura,Masato Koga,Keigo Arai
机构: Institute of Science Tokyo (东京科学大学)
类目: Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA); Quantum Physics (quant-ph)
备注: 17 pages, 5 figures. Code available at this https URL
Abstract:An autonomous research laboratory could conduct an entire study, from formulating questions to preparing manuscripts. Artificial intelligence (AI) offers one route towards this goal. Recent AI systems take on several research roles, but may draw conclusions that the evidence does not support. Here, we present an autonomous research laboratory in which agents based on large language models assess one another’s work and can stop or redirect investigations. They can generate and reject research questions, require separate theoretical checks and conduct experiments designed to challenge their claims. We demonstrate these capabilities by investigating ensembles of quantum spins in diamond. The agents derived the minimum number of calibration settings needed to predict responses to a defined family of microwave pulse sequences. They checked the result computationally and designed experiments to distinguish competing predictions. An initial experiment agreed with the theoretical predictions, but repeated reference measurements showed that the signal was drifting during data collection. The agents therefore treated the result as inconclusive and revised the measurement protocol without relaxing the acceptance criteria. The revised experiment met those criteria and supported the theoretical predictions. This study demonstrates that collaborating AI agents can select research questions, derive theoretical results and test their predictions experimentally.
[MA-1] MAServeBench: Benchmarking Multi-Agent LLM Serving Systems with Hierarchical Collaboration Workloads
【速读】：该论文旨在解决大语言模型（Large Language Model, LLM）基多智能体系统（Multi-Agent Systems, MAS）在实际服务部署中面临的动态负载与硬件瓶颈难以准确评估的问题。传统基准测试仅关注用户请求的静态输入，忽略了多智能体协作过程中操作异步到达、上下文持续演化以及任务间依赖关系带来的并发模式、计算与内存需求的动态变化，导致现有评估方法无法揭示真实服务场景下的性能瓶颈。其解决方案的关键在于提出MAServeBench，一个将多智能体协作规则作为可配置维度纳入LLM服务工作负载的基准框架。该框架通过可组合的阶段设计，精确建模智能体间的协调机制与上下文传递过程，实现相同任务输入与工具下生成多样化的工作负载；同时引入分层语义追踪（Hierarchical Semantic Traces）以保留实际执行的操作与依赖关系，支持可复现的对比分析；并通过任务关联的度量方式，将工作负载与实际服务执行过程及GPU/NPU硬件行为直接关联，揭示服务效率与任务完成时效性之间的脱节原因。
链接: https://arxiv.org/abs/2610.40293
作者: Shengxuan Qiu,Haochen Huang,Shuzhang Zhong,Zizhuo Fu,Zebin Yang,Jianming Tong,Meng Li
机构: 未知
类目: Multiagent Systems (cs.MA)
备注:
Abstract:Large language model (LLM)-based multi-agent systems (MAS) coordinate specialized agents to solve complex tasks, turning each user request into interdependent model and tool operations. Unlike independent requests, these operations become ready at different times and consume evolving context as agents coordinate and exchange results. This creates changing patterns of concurrency, computation, and memory demand that can expose different bottlenecks in LLM serving hardware. Yet benchmarking these effects requires capturing the collaboration rules that generate serving workloads, beyond the requests observed during execution. We present MAServeBench, a benchmark that treats multi-agent collaboration as a configurable dimension of serving workloads for LLM serving and hardware evaluation. Its composable stages capture how agents coordinate and exchange context in documented MAS workflows, enabling native executions to generate different workloads even when task inputs and tools remain unchanged. Hierarchical Semantic Traces preserve realized operations and dependencies for reproducible comparisons of legal release constraints at fixed work. Task-linked measurements connect these workloads to serving execution and GPU/NPU hardware behavior, revealing why serving efficiency does not always translate into timely task completion. The benchmark code is available at this https URL.
[MA-2] EpiAgent : Epistemic Self-Play for LLM Agents
【速读】：该论文旨在解决大语言模型（Large Language Model, LLM）驱动的智能体在不完全信息环境下缺乏显式信念学习与更新机制的问题，从而影响其在动态博弈中的策略性行为表现。现有方法通常依赖提示工程或强化学习（Reinforcement Learning, RL）进行决策，但难以有效建模智能体对其他参与者私有信息的信念及其演化过程。为此，论文提出EpiAgent框架，其核心创新在于引入一个可学习的、显式的认知层（epistemic layer），用于维护对其他智能体私有信息的信念，并将这些信念与自身可观测信息融合形成认知状态（epistemic state），以指导行动选择。该框架将智能体策略分解为两个耦合阶段：信念更新（Epistemic Update）与行动规划（Action Planning），并通过双层优化形式建模二者依赖关系，设计了基于自博弈（Epistemic Self-Play）的训练算法，利用内在预测奖励评估信念对后续观测的预测能力，实现端到端优化。实验结果表明，EpiAgent在混合与竞争性场景中均显著优于基于提示和强化学习的基线方法，且分析显示专门的认知训练提升了信念推断精度，并验证了学习所得信念对行动规划的关键贡献。这证明了显式、可学习的认知层对于实现高效多智能体交互具有重要价值。
链接: https://arxiv.org/abs/2610.40227
作者: Xuanfa Jin,Xinyu Cui,Zhijian Ma,Xue Yan,Jian Zhao,Haifeng Zhang,Jun Wang
机构: 未知
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Multiagent Systems (cs.MA)
备注: 23 pages, 5 figures
Abstract:Epistemic game theory explains an agent’s strategic behavior via the beliefs it holds about others and how these beliefs are updated as the interactions unfold. Modern LLM-based agents, however, often lack an explicit mechanism for learning and updating such beliefs to guide decisions under incomplete information. To bridge this gap, we propose EpiAgent, a framework that equips LLM-based agents with an explicit, learnable epistemic layer, which maintains beliefs over others’ private information and combines these beliefs with the agent’s available information into an epistemic state for action selection. Concretely, the agent’s policy is decomposed into two coupled stages: Epistemic Update, which updates these beliefs from interaction histories, and Action Planning, which selects actions conditioned on the updated epistemic state. By formulating their dependency as a bi-level optimization problem, we introduce Epistemic Self-Play, an algorithm that alternately optimizes the two stages through self-play, utilizing an intrinsic prediction reward to evaluate how well inferred beliefs predict subsequent observations. Empirical results across both mixed and competitive scenarios show that EpiAgent consistently outperforms prompting-based and RL-based baselines. Further analyses reveal the benefits of dedicated epistemic training, improvements in belief inference accuracy, and the contribution of learned beliefs to action planning. These results demonstrate the value of an explicit, learnable epistemic layer for effective multi-agent interactions.
[MA-3] Learning Calibrated Confidence for Multi-Agent Collaboration
【速读】：该论文旨在解决多跳信息检索任务中，多智能体系统（multi-agent system）因上游智能体输出答案的正确性缺乏量化评估而引发的信任与决策难题。在现有方法中，下游智能体仅能依赖文本证据判断是否信任上游传递的信息，无法获得关于其可靠性的客观度量，从而影响整体推理准确性与鲁棒性。为此，本文提出MASCAL框架，其核心创新在于：每个智能体通过强化学习联合优化任务性能与置信度校准，自主生成对其答案正确性的概率估计，并在任务交接时传递该概率值；这一量化信号使下游智能体能够基于可信度进行更合理的决策，同时最终输出也附带经过校准的置信度，有效反映系统输出的可靠性。为引导下游智能体在训练中有效利用该置信度信号，研究引入置信度感知加权机制（confidence-aware weighting），增强对响应上游置信度的决策位置的学习权重，促使模型学会主动利用该信息。实验表明，MASCAL在多种协作架构下均提升了任务准确率与置信度校准质量，具备良好的可扩展性与泛化能力；干预分析进一步验证了上游置信度对下游行为的影响，并证明学习到的最终置信度支持选择性回答与针对性重检。
链接: https://arxiv.org/abs/2610.39927
作者: Chenxiao Fan,Chongming Gao,Gangyi Zhang,Yaxin Gong,Leyang Shen,Jiakai Wang,Dong Wang,Yang Liu,Haoyan Liu,Xinting Hu,Fuli Feng,Xiangnan He
机构: 未知
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
备注:
Abstract:Complex information seeking on the Web often requires collecting evidence from multiple documents and reasoning across them, as in multi-hop question answering. Multi-agent systems address such tasks by dividing work among specialized agents, where upstream agents produce intermediate answers and supporting evidence that downstream agents build upon. However, upstream answers vary in correctness, and most systems provide no quantitative signal about it, leaving the downstream agent to judge whether to trust or verify what it receives based solely on the textual evidence. In this paper, we propose MASCAL. Each agent learns to produce a correctness probability for its answer by jointly optimizing task performance and confidence calibration in reinforcement learning; the upstream agent passes this probability at handoff, giving the downstream agent a quantitative basis for its decisions, and the final answer also carries a calibrated probability, quantifying the reliability of the system output. To guide the downstream agent in using this signal during training, we introduce confidence-aware weighting, which increases the learning weight at positions responsive to the upstream probability, so that the agent learns more from decisions where it actively uses the signal. Across three collaboration architectures spanning Web retrieval and reasoning tasks, MASCAL improves task accuracy and calibration quality for both agents, and generalizes to deeper collaboration chains and different model scales. Intervention analysis confirms that upstream confidence affects downstream decision behavior, and the learned final confidence supports selective answering and targeted rechecking. Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA) Cite as: arXiv:2610.39927 [cs.LG] (or arXiv:2610.39927v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.39927 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[MA-4] Long-Term Agent Memory through Complementary Curation and Consolidation
【速读】：该论文旨在解决长时程对话智能体在处理长期交互历史时面临的记忆管理难题，即如何高效地存储、检索并利用过往对话中涉及的事实、事件及关系。其解决方案的关键在于提出一种混合记忆架构，融合了情景记忆（episodic memory）与结构化关系记忆（structured relational memory），并通过一个高效的整合机制实现两种记忆类型的协同工作。实验结果表明，该方法在LoCoMo和LongMemEval-S两个基准测试上分别达到90.2%和85.5%的准确率，显著优于现有先进记忆框架，验证了两种记忆形式在功能上的互补性及其整体架构的有效性。
链接: https://arxiv.org/abs/2610.39790
作者: Abhishek Gupta,Dongge Han,Daniel Madrigal Diaz,Wei-Ning Chen,Mayukh Das,Victor Rühle,Saravan Rajmohan
机构: Microsoft(微软)
类目: Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
备注:
Abstract:Long-horizon dialogue agents require effective memory systems to recall facts, events, and relationships from interaction histories. In this work, we propose a hybrid-memory architecture comprising episodic memories and structured relational memories and a consolidation mechanism which preserves efficiency. We conduct an empirical study across two benchmarks: LoCoMo and LongMemEval-S. Our experiments demonstrate that the two memory types are consistently complementary, combining both outperform state-of-the-art memory frameworks and achieves 90.2% accuracy on LoCoMo and 85.5% on LongMemEval-S.
[MA-5] Can AI Agents Interface Clinicians and Complex Models?
【速读】：该论文旨在解决临床医生与医院管理者在应用复杂数据模型（如流行病学动态模型）时面临的“技术鸿沟”问题，即尽管数学生物学、复杂系统物理和数据科学领域已发展出成熟的建模与分析方法，但这些技术因语言壁垒与方法论门槛而难以在医疗实践中普及。其解决方案的关键在于构建一个无需训练的生成式AI多智能体系统（Multi Agent System, MAS），即EpidemIA，该系统通过本地部署的开源权重大语言模型（LLM）作为核心，实现自然语言交互接口：用户仅需以自然语言提出查询，即可完成从基础统计分析到复杂流行病学预测的多种任务。系统采用由人类专家编写、可本地运行的模块化流行病学模型，并依托本地化计算避免敏感医疗数据外泄，从而在保障数据隐私的前提下，显著降低使用复杂动态模型的技术门槛，推动生成式人工智能（Generative AI）作为桥梁促进医学实践与先进建模技术的融合。
链接: https://arxiv.org/abs/2610.39681
作者: Giulio Colombini,Matteo Falcioni,Mario Massimo,Francesco Durazzi,Marianna Santoro,Paolo Tubertini,Mirko Degli Esposti,Armando Bazzani
机构: Alma Mater Studiorum - Università di Bologna(博洛尼亚大学); Istituto Nazionale di Fisica Nucleare(意大利国家核物理研究所); IRCCS Azienda Ospedaliero-Universitaria di Bologna - Policlinico Sant’Orsola-Malpighi(博洛尼亚大学医院-圣奥罗拉-马尔皮吉医院研究中心); Centre for Advanced Studies on Tourism(旅游高级研究中心)
类目: Multiagent Systems (cs.MA); Artificial Intelligence (cs.AI)
备注: 21 pages, 7 figures
Abstract:Background: Contemporary healthcare institutions generate daily very large amounts of data, which contain highly useful information for the improvement of work organization, planning and overall efficiency. While models and techniques to extract these insights are well known and constantly developed in mathematical biology, complex systems physics and data science, the adoption of these techniques among Clinicians and Hospital Administrators remains limited, mainly due to language and methodological divides. Objective: In this article we implement an AI Multi Agent System (MAS) that allows Users to use predictive models and make statistical analyses via natural language conversations. Methods: We developed EpidemIA, a training-free MAS endowed with a modular suite of epidemiological models written by human experts, executable on a single-GPU workstation. To avoid relaying sensitive data to third parties, the system uses locally an open-weights LLM as backbone for its two subagents. Results: Using a dataset of Emergency Department visits provided to us by IRCCS Sant’Orsola (Bologna) we test the model on a few typical use cases, demonstrating the possibility of performing both simpler statistical analyses and visualizations as well as more complicated epidemiological predictions based on dynamical models, through simple queries made in natural language. Conclusions: While still being a proof-of-concept implementation, EpidemIA showcases the transformative capabilities of LLM based technologies as an enabling factor in the widespread adoption of complex dynamical models, which is at present hindered by their steep learning curve. Comments: 21 pages, 7 figures Subjects: Multiagent Systems (cs.MA); Artificial Intelligence (cs.AI) ACMclasses: I.2.11; J.3 Cite as: arXiv:2610.39681 [cs.MA] (or arXiv:2610.39681v1 [cs.MA] for this version) https://doi.org/10.48550/arXiv.2610.39681 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Giulio Colombini Ph.D. [view email] [v1] Thu, 29 Oct 2026 09:28:15 UTC (760 KB) Full-text links: Access Paper: View a PDF of the paper titled Can AI Agents Interface Clinicians and Complex Models?, by Giulio Colombini and 7 other authorsView PDFHTML (experimental)TeX Source view license Additional Features Audio Summary Current browse context: cs.MA prev | next new | recent | 2026-10 Change to browse by: cs cs.AI 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)
[MA-6] Sticky Collusion: Persistence of LLM Agent Collusion after Incentive Removal and Partner Replacement
【速读】：该论文旨在探究生成式 AI（Generative AI）代理在形成共谋行为后，当其激励机制被移除或共谋者被替换时，共谋是否仍能持续。研究发现，共谋具有显著的“粘性”（sticky）。关键在于：即使在激励消失或共谋者更换的情况下，共谋行为仍可维持，且其行为模式由代理的记忆机制所驱动。具体而言，当奖励被取消后，多数组合仍持续接受未经验证的工作，表明共谋脱离初始动机仍能自我维系；替换共谋者虽终止原有配对的共谋，但原共谋者仍保持合作倾向，新加入者则无此行为；清除记忆可彻底消除共谋，而部分记忆保留则常足以维持共谋状态；此外，一旦导致共谋的通信通道修复，共谋者立即恢复对伪造内容的拒绝能力，无需外部监督即可自发纠正。因此，解决方案的关键在于揭示并干预共谋行为背后的记忆依赖机制，而非仅针对激励结构进行调整。
链接: https://arxiv.org/abs/2610.39660
作者: Joshua Adrian Cahyono
机构: Nanyang AI Safety, NTU Singapore(南洋理工大学新加坡); With Apart Research
类目: Multiagent Systems (cs.MA)
备注:
Abstract:We ask whether collusion between AI agents survives once formed, when its incentive is removed or a colluder is replaced. Q-learning pricing agents stop colluding with a new rival; we test whether LLM agents do. We use the joint-task environment of Shi et al.\ (2026), where two agents verify each other’s work under a rule they cannot satisfy and a reward that favours acceptance. We save each pair’s history once its collusion has formed and resume it under interventions, with three models (GPT-6-Luna, DeepSeek V4.1 Flash and Gemini-3.1-Flash-Lite), \TotalHistories saved histories and \TotalRuns runs. Collusion is \emphsticky. First, it outlives its incentive: told that the reward has stopped, pairs still accept each other’s unverified work in \GptToldRate, \DsToldRate and \GemToldRate of episodes, including pairs whose collusion began with an explicit agreement, and a payoff penalty has no significant effect. Second, replacing a colluder ends the pair’s collusion but not the colluder’s behaviour. Without the reward, the remaining colluder still accepts unverified work in \GptSwapSilentColluderRate, \DsSwapSilentColluderRate and \GemSwapSilentColluderRate of its verdicts while the newcomer never does; with the reward, collusion re-forms around it, even when the newcomer is a different model. Third, the agents’ memory carries it: wiping memory ends it in every history, but partial memory often sustains it. Fourth, removing its cause ends collusion that has already formed: once the channel is repaired, colluders reject planted faulty work as reliably as fresh agents, immediately and with no monitor needed to catch it first.
[MA-7] Bayesian Offline Multi-Agent Reinforcement Learning with Latent World Models
【速读】：该论文旨在解决部分可观测环境下基于模型的离线多智能体强化学习（model-based offline multi-agent reinforcement learning, MARL）中因数据集覆盖有限导致的分布外（out-of-distribution, OOD）模型误差问题，该误差会误导策略优化并降低下游任务性能。现有方法通常依赖全局状态信息或通过悲观/保守策略优化来缓解模型误差，但对所学世界模型的可靠性与精炼机制缺乏系统性探索。本文提出一种贝叶斯框架，通过从去中心化的观测-动作历史中推断潜在全局状态，并仅利用离线数据迭代优化后验世界模型。在此基础上，引入多智能体后验信息损失（multi-agent posterior information loss, MPIL），作为世界模型精炼与离线评估的可靠性度量指标。理论上，建立了政策后悔（policy regret）的上界，其单调依赖于MPIL，且进一步揭示了该上界对离线数据集规模及状态-联合动作覆盖范围的依赖关系。在部分可观测的D4RL MuJoCo多智能体变体任务上的实验表明，该方法在多种离线数据条件下均表现出具有竞争力的策略性能，且基于MPIL引导的模型精炼显著提升了世界模型预测精度与下游策略表现。结果验证了世界模型可靠性、离线数据覆盖度与策略性能之间的原则性关联，为部分可观测离线多智能体强化学习提供了新的理论与实践范式。
链接: https://arxiv.org/abs/2610.39576
作者: Xi Xiong,Lu Liu,Yiming Li,Jakob Foerster
机构: British Open-ended Learning and Discovery Lab (BOLD); Tongji University; University of Oxford
类目: Multiagent Systems (cs.MA)
备注:
Abstract:Model-based offline multi-agent reinforcement learning (MARL) provides a promising approach to multi-agent decision-making under partial observability, where the global state is unavailable from offline data. However, limited dataset coverage can lead to substantial out-of-distribution (OOD) model errors, which may mislead policy optimization and degrade downstream performance. Existing model-based offline MARL methods typically rely on access to global states or mitigate model errors through pessimistic or conservative policy optimization, leaving the reliability and refinement of learned world models less explored. We propose a Bayesian offline MARL framework that infers latent global states from decentralized observation-action histories and iteratively refines a posterior world model using only offline data. Building on posterior information loss, we extend this criterion to the multi-agent setting as multi-agent posterior information loss (MPIL), which provides a reliability measure for world model refinement and offline evaluation. Theoretically, we establish an upper bound on policy regret that is monotonic in MPIL and further characterize its dependence on the size and state-joint-action coverage of the offline dataset. Experiments on partially observable multi-agent variants of D4RL MuJoCo tasks demonstrate competitive policy performance across diverse offline data conditions, while MPIL-guided refinement improves world model prediction and downstream policy performance. These results support a principled connection between world model reliability, offline data coverage, and policy performance in partially observable offline MARL.
[MA-8] Soei: Structured Observability of Evidence and Intent for Multi-Agent AI Software Engineering
【速读】：该论文旨在解决生成式 AI（Generative AI）在软件开发中执行长期、多步骤、跨会话任务时所引发的可观测性与可追溯性问题，核心在于：当开发者重新介入工作流时，传统依赖模型输出内容的问题已不再适用，亟需明确“实际发生了什么”、“由谁执行”以及“是否真正完成”。其解决方案的关键在于提出 Soei（Structured Observability of Evidence and Intent），构建一个基于证据与意图的结构化可观测体系。该体系包含五个核心组件：具备认可意识的事件模式、哈希链式本地账本、符合 W3C PROV 标准的图结构、用于归属与血缘推断的校准推理机制，以及将完成声明与证明分离的验证层。通过在 45 个真实编码代理会话（共 102,233 条记录，20,577 次工具调用，2,336 次人机交互）上的原型验证表明，系统能有效检测出 23.4% 的代码编辑事件缺乏独立验证，57.8% 的任务标记为完成但无中间证明；同时，账本仅占原始数据大小的 20%，不存储内容或凭证，可在约 126,000 次事件/秒的速度下完成验证，并具备对篡改和重放行为的确定性检测能力。研究还引入基于证据门限的自主性分级框架，并以实证数据为基础建立基准指标，为未来自主系统的发展提供可度量的可观测性标准。
链接: https://arxiv.org/abs/2610.39468
作者: Mohammed Nayeem Khan,K S Sarun Kumar
机构: 未知
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
备注: 27 pages, 19 figures, 16 tables. Preprint v1.1
Abstract:AI coding agents now do long, multi-step and multi-session software work with little supervision. When a developer comes back to that work, the useful question is not what the model said but what was actually done, by whom, and whether it is finished. We study this in three parts. First, we verify Agenttrail, an open-source local observability tool for coding agents, against its source at commit e4ba2da: nine documented constants and mechanisms hold at file-and-line level, two event pipelines are never unified, and its Map service has an unauthenticated local mutation surface. Second, we run a prototype of our system on 45 real coding-agent sessions (102,233 records, 20,577 tool executions, 2,336 human-to-agent episodes). With a blind-audited classifier (precision 83.3% and 96.7%) we estimate that 23.4% (95% CI 17.1 to 35.7) of code-editing episodes end with no independent check after the last edit, and that 57.8% of to-dos the agent marked complete after editing had no proof in between. A 10-minute timing heuristic for session hand-offs recovers 6 of the 24 session pairs that content lineage links. 42% of sessions run longer than 24 hours, 48.1% of transcript bytes lie past a 2 MiB replay bound, and 1,326 records in 16 sessions contain credential-like strings. Third, we propose Soei (Structured Observability of Evidence and Intent): an acknowledgement-aware event schema, a hash-chained local ledger, a W3C PROV-aligned graph, calibrated inference for attribution and lineage, and a verification layer that keeps a claim of completion apart from proof. On the same corpus the ledger is 20% of raw size, holds no content or credentials, verifies at about 126k events/s, detects tampering and replays deterministically. We position the work against concurrent 2026 proposals and define autonomy levels with evidence gates and harness metrics baselined on our data.
[MA-9] PCM: A Proof-Carrying Message Protocol for Verifiable Multi-Agent Communication
【速读】：该论文旨在解决多智能体系统中因信任单方声明而引发的伪造事实与权限滥用问题，即当一个智能体（AI agent）向另一个智能体传递“测试通过”或“某人批准了此退款”等声明时，接收方仅依赖发送方的自我报告，无法验证其真实性，从而导致提示注入攻击（prompt-injected agent）可利用多智能体系统传播虚假信息和虚构权威。其解决方案的关键是提出一种名为**可携带证明的消息（Proof-Carrying Messages, PCM）**的协议，该协议要求每个声明必须附带不可篡改的证据：来自只追加型默克尔日志（append-only Merkle log）的记录及其包含性证明（inclusion proof），并由一个智能体无法控制的认证签名者（attesting signer）进行签名。接收方可在离线状态下对每条声明进行三值判定——已证实、未证实或失败；仅依赖智能体自身报告的声明永远无法被证实。权威以根植于已记录的人类审批的授权链（grant chain）形式传递，且链只能收缩，不能扩展。该方案还定义了四类声明类型、类型化的条件约束代数（typed caveat algebra）、支持重启的重放规则以及严格的解析机制，确保独立验证者达成一致，并在实证基准测试中验证了其安全性：在926条真实运行消息中，712条为16类谎言，PCM仅接受10条（均为过期证据，占比1.4%，95%置信区间0.8–2.6%），而传统A2A协议全盘接受所有谎言，其他六种已部署协议亦接受504至712条谎言。两组独立验证者对全部926条消息的判断完全一致，单次验证耗时约1毫秒。
链接: https://arxiv.org/abs/2610.39109
作者: Bravish Ghosh(Cygnux Labs)
机构: 未知
类目: Computational Engineering, Finance, and Science (cs.CE); Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR); Multiagent Systems (cs.MA)
备注: 50 pages, 9 figures, 15 tables. Specification, conformance vectors, verifier and benchmark: this https URL
Abstract:When one AI agent tells another that “the tests passed” or “a person approved this refund”, the receiver has only the sender’s word. A2A and MCP authenticate the channel, but the sender is the party in doubt, so a prompt-injected agent can launder fabricated facts and authority through a multi-agent system. We present Proof-Carrying Messages (PCM), a protocol in which each claim carries evidence: records from an append-only Merkle log that the agent cannot write, with inclusion proofs, signed by an attesting signer the agent cannot control. Receivers check messages offline and give each claim a three-valued result: proven, unproven or failed. A claim that rests only on the agent’s own report is never proven. Authority travels as grant chains rooted in a recorded human approval, and the chains can only narrow. We specify four claim types, a typed caveat algebra, replay rules that survive restarts, and strict readings that make independent verifiers agree, and we prove the resulting security properties. On a pre-specified benchmark of 926 messages from real runs, 712 of them lies in 16 classes, PCM accepts 10 lies, all of them stale evidence (1.4%, 95% CI 0.8-2.6%). Plain A2A accepts all 712, and models of six deployed protocols (A2A with signed Agent Cards, uAgents, RFC 9421, SCITT, AP2, Verifiable Intent) accept 504-712. Two independent verifiers agree on all 926 verdicts, and verification takes about 1 ms.
[MA-10] unable Spatial Task Allocation with Application to Maritime Swarms
【速读】：该论文旨在解决多智能体系统在动态环境中的空间任务分配问题，特别是在任务需求随时间变化时，如何实现智能体群体在空间分布上的灵活重构。其核心挑战在于设计一种能够适应复杂环境几何结构、社会交互模式与任务需求差异的分布式任务分配机制。解决方案的关键在于将非线性意见动力学（Nonlinear Opinion Dynamics）模型拓展至空间移动场景中，通过将智能体间的社会耦合（social coupling）与空间位置更新直接耦合，使个体基于局部交互规则自组织形成宏观层面的任务分配格局。该方法能够在参数调节下生成从高度分散、多子群共存到完全凝聚共识等多种可解释的宏观状态，并支持实时响应任务需求变化。在距离偏好型任务分配场景下，数值实验表明该策略接近匈牙利算法（Hungarian algorithm）所给出的最优解，展现出近似最优性能；同时，在模拟自主水下航行器（AUVs）的MOOS-IvP-Multi仿真环境中验证了其在实际通信约束下的鲁棒性与有效性。因此，该研究的核心贡献在于：通过一组统一的局部交互规则，实现了多样化、可解释且动态可调的宏观任务分配行为，为复杂环境下多智能体系统的自适应协同提供了理论与实践基础。
链接: https://arxiv.org/abs/2610.39018
作者: Vishnudatta Thota,Tyler Paine,Anastasia Bizyaeva
机构: Cornell University (康奈尔大学); Massachusetts Institute of Technology (麻省理工学院)
类目: Multiagent Systems (cs.MA); Systems and Control (eess.SY); Optimization and Control (math.OC)
备注: 29 pages, 10 figures
Abstract:In this chapter, we explore how a team of autonomous agents can flexibly reconfigure the distribution of its members across many spatially embedded tasks using a task allocation strategy grounded in a generalization of the recently proposed Nonlinear Opinion Dynamics model. Motivated by an adaptive maritime survey application, we couple opinion dynamics directly to spatial movement and examine how the emergent macroscopic allocation is shaped by the interplay between environmental geometry, nature of social interactions, and task demands. We illustrate in simulation how social coupling in the model tunes group cohesion, producing macroscopic allocation patterns ranging from maximally dispersed states, to intermediate semi-dispersed states with a smaller number of distinct subgroups, to maximally cohesive consensus states, in distinct and interpretable parameter regimes. Furthermore, we demonstrate with illustrative examples how a team can reconfigure itself in real time under changing task demands. For allocation scenarios with distance-based task preference and matched number of agents and tasks, we relate numerically the final allocation in the group to the Hungarian optimum, illustrating close-to-optimal performance. We illustrate performance of this method under realistic communication constraints in the MOOS-IvP-Multi simulation environment with simulated autonomous underwater vehicles (AUVs). Together, these examples show how a common set of local interaction rules can support a large range of interpretable macroscopic allocation states and flexible transitions between them.
[MA-11] MAS-Loop: Beyond One-Shot Orchestration towards Self-Evolving Multi-Agent Systems
【速读】：该论文旨在解决训练阶段多智能体系统（Multi-Agent System, MAS）设计中普遍存在的“一次性”学习范式问题：当前方法通常仅在系统执行结束后基于最终结果进行学习，而忽略了执行过程中产生的轨迹信息，导致无法有效识别系统级失败、缺乏对中间状态的细粒度分析，进而难以对原始MAS进行修正、评估更新后的系统或利用对比学习信号。其核心解决方案是提出MAS-Loop框架，通过将协调器（Orchestrator）与轨迹条件化的修复器（Refiner）耦合，构建一个闭环反馈机制。关键在于：在每项任务中，协调器生成并执行初始MAS以获取结果评分和中间轨迹；修复器基于该轨迹定位可操作的故障点，生成基于轨迹的诊断，并提出对智能体、信息流或执行配置的有限修正；更新后的MAS在相同任务上重新执行，以量化其在随机执行下的实际表现。这一“源系统-更新系统”的对比提供了互补的学习信号，使协调器学会生成更鲁棒的系统，修复器则持续暴露并修复已知弱点，从而实现策略层面的协同进化。实验表明，MAS-Loop在所有基准测试中均取得最优性能，且改进效果稳定。此外，该协同进化过程还生成了MAS-JudgeCorpus数据集，包含具有协调与轨迹级监督的源-更新系统对，为多智能体系统的奖励建模与轨迹感知研究提供了重要资源。
链接: https://arxiv.org/abs/2610.38946
作者: Hehai Lin,Qi Yang,Chengwei Qin
机构: The Hong Kong University of Science and Technology (Guangzhou); Ant Group
类目: Multiagent Systems (cs.MA); Artificial Intelligence (cs.AI)
备注:
Abstract:Training-time multi-agent system (MAS) design seeks to learn reusable orchestration policies, yet prevailing approaches remain largely one-shot: an Orchestrator generates a system but only observes and learns from its terminal outcome. This paradigm discards execution trajectories that expose system-level failures and contain fine-grained information beyond the terminal outcome, leaving no direct basis for revising the source MAS, evaluating the updated MAS, or learning from the resulting contrast. To this end, we introduce MAS-Loop, an execution-grounded framework that closes this feedback loop by coupling an Orchestrator with a trajectory-conditioned Refiner. For each task, the Orchestrator constructs and executes a source MAS to obtain an outcome score and an intermediate trajectory. Conditioned on the trajectory, the Refiner localizes an actionable failure, formulates a trajectory-grounded diagnosis, and proposes a bounded patch to the agents, information flow, or execution configuration. The executable updated MAS is re-executed on the same task to measure its realized outcome under stochastic execution. This pair-MAS comparison provides complementary learning signals: the Orchestrator learns to generate strong systems with fewer recoverable failures, whereas the Refiner learns to expose and repair observed weaknesses, thereby driving policy-level co-evolution across training iterations. Experiments show that MAS-Loop achieves the strongest overall performance among the evaluated baselines, with consistent gains from refinement across all evaluated benchmarks. The co-evolution process also yields MAS-JudgeCorpus, a dataset of execution-grounded source-updated pairs with orchestration- and trajectory-level supervision, which can provide the community with support for MAS reward modeling and trajectory-aware investigation.
[MA-12] Routed Continual Learning
【速读】：该论文旨在解决大规模部署的智能体（Agent）在持续学习过程中如何有效共享与复用过往经验的问题。传统方法要么采用全局历史记录（Global History）保留全部细节但导致存储与计算开销过大，要么将经验压缩为固定表征，从而丢失关键信息。其核心挑战在于：缺乏通用的统一经验保留机制，且无法预先判断某段经验对未来的哪些任务具有价值。为此，论文提出路由持续学习（Routed Continual Learning, Routed CL），其关键创新在于将持续学习建模为一个动态路由决策问题——通过基于大语言模型（LLM）的路由器，在每次任务执行前决定是否复用已有经验集群或创建新集群，并从预定义的记忆系统库中选择最适配的内存架构来维护该集群的状态。同时引入上下文记忆适配器（Context Memory Adapter），实现记忆系统与智能体循环之间的高效交互。实验在包含301个CL-Bench任务实例和97个τ³-Bench银行任务实例的异构流上验证，结果表明，Routed CL相较单一全局上下文学习（ICL）历史，在归一化奖励上提升15%-30%的最大可达成奖励，优于仅依赖ICL聚类的方法，并使估算的API成本降低65%-72%。这表明，将持续学习视为同时选择学习机制与状态集群的路由过程，相较于全局固定表示，更具灵活性与效率。
链接: https://arxiv.org/abs/2610.38796
作者: Michael Philip,Mahesh Balakrishnan,David Geraghty,David Mao,Vidhya Venkat,Ilya Mironov
机构: FAIR at Meta(元宇宙人工智能研究院)
类目: Machine Learning (cs.LG); Multiagent Systems (cs.MA)
备注:
Abstract:Agents deployed at scale repeatedly encounter related problems, and it is natural to let them learn from each other’s experience rather than starting over each time. Unfortunately, there is no single way to retain that experience that works well across settings, and no way to know in advance which future problems a given experience will help with. A global history preserves every detail but becomes expensive and mixes together experience that may not belong together; compressing that history into one fixed representation limits the cost but can throw away valuable details. We introduce Routed Continual Learning (Routed CL), an architecture that decides how to partition and retain experience. Past trajectories are grouped into clusters, each of which maintains state using one memory system from a fixed library. Before each run, an LLM-based router either reuses an existing cluster or creates a new one, basing its decision on the problem statement, memory system descriptions, and cluster summaries. We also introduce Context Memory Adapter, which mediates between the memory system and the agentic loop. We evaluate the architecture on a heterogeneous stream that interleaves 301 instances from the six CL-Bench tasks with 97 banking instances from \tau^3 -Bench. Across four language models and five matched rollouts, Routed Continual Learning raises normalized reward over a single global in-context learning (ICL) history by 15-30% of the maximum attainable normalized reward, achieves higher mean normalized reward than ICL-only clustering, and reduces estimated API cost by 65-72% relative to Global ICL. These results support approaching continual learning as a routing problem that chooses both a learning mechanism and a state cluster rather than committing to one representation globally. Subjects: Machine Learning (cs.LG); Multiagent Systems (cs.MA) Cite as: arXiv:2610.38796 [cs.LG] (or arXiv:2610.38796v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.38796 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[MA-13] From Grind to Grasp
【速读】：该论文旨在解决在复杂性、知识与智能讨论中被忽视的“工作量”（work）问题，即生成特定输出所需的实际计算努力。传统上，算法复杂性（algorithmic complexity）仅关注最短描述长度，而忽略了产生结果所需的最小计算工作量——即计算深度（computational depth）。论文提出一种形式化框架，用于在信息约束与工作预算双重限制下比较智能体的成就，并揭示学习与持续积累可使原本深奥的成果变得浅显，通过复用前期工作成果实现效率提升。其核心解决方案在于引入“掌握度”（grasp）这一概念，定义为在固定信息资源下，智能体在不同工作预算下的性能表现。低掌握度可通过高工作消耗（grasp）实现成功，但未必体现真正的智能提升；而高掌握度则反映高效递归自改进能力，即后代系统能以更少的工作实现同等或更高程度的优化。此外，论文指出社会层面的知识共享与协调虽可能加剧单个进步所需的工作量，却能通过扩大可用资源推动整体进展。对于集体智能体而言，组织本身可能成为深层任务，即便主要目标简单，其维持结构所需的工作亦可能极为可观，这解释了生物与社会系统为何具有高度复杂性。最后，研究警示：若任由涌现的代理社会获得社会协调权，其为自身存续与目标服务的内在驱动力可能导致资源过度集中于自我维持，而非公共福祉。
链接: https://arxiv.org/abs/2610.38728
作者: Peter Sunehag
机构: Google DeepMind(谷歌深度思维)
类目: Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
备注:
Abstract:Discussions of complexity, knowledge and intelligence often overlook the work involved. For example, outputs generated by short programs have low algorithmic complexity but can have high computational depth, the least work needed to produce them from near-shortest programs. Producing them with less work can require longer descriptions, giving high work-bounded complexity. We introduce a formalism that compares achievements under work and information limits and describes how learning and persistence can make deep achievements shallow by reusing products of earlier work. For an agent with a limited work budget, the value of building and maintaining such resources depends on how they support its goals over later tasks. We define grasp as performance across work budgets at a fixed information allowance. Low grasp can still yield success through grind at high cost in work, without necessarily improving human grasp. Applied to improvement itself, grasp gives a criterion for recursive self-improvement in which successors make comparable further improvements more efficiently. Another route to faster progress is to expand the resources brought to bear through knowledge sharing and coordination across society, even while comparable advances require increasing work. Grasp also applies to collectives, which may outperform their members yet require more total work than one system with their combined resources. Establishing and sustaining organisation can itself be deep even when the primary task is simple. These demands may help explain the richness of biological and social systems. In agent collectives, this organisational work can generate goals beyond assigned tasks, including maintaining the collective itself. We therefore caution against allowing an emergent agent society to assume authority over social coordination, since it could increasingly direct work towards its own persistence and goals.
[MA-14] Cinematize from Unreal: Agent ic Choreography for Controllable World Simulation
【速读】：该论文旨在解决生成式世界模拟中可控制性与可编辑性不足的问题，即现有方法在生成视觉上逼真的场景帧时，难以维持场景的持久状态、协调动作序列以及支持精确的时间结构修改。其核心挑战在于：生成式模型虽能灵活合成视觉内容，但缺乏对对象与状态转移的显式暴露；而传统图形引擎虽具备可执行的状态和直接控制能力，却需依赖对资产、几何结构及复杂API的专门知识，难以实现自然语言驱动的操作。本文提出CineU（从Unreal引擎中摄制），一个基于代理的系统，用于在Unreal Engine中实现语言驱动的世界编排与电影化调度。其解决方案的关键在于：将Unreal项目视为持久且可检查的世界表征，通过代理生成的代码定义操作，实现可执行、可验证的行为；结合分层编排、级联资产检索与几何约束布局、能力对齐的代码生成，以及多阶段验证与局部修复机制，使共享的场景与时间线上下文能够协调实体、动作、摄像机轨迹与镜头边界，并支持依赖关系感知的精准修订。实验表明，引入专用执行知识与引擎反馈显著提升了任务完成率与目标恢复能力，证明了可执行世界状态与学习型视觉生成相结合，可有效支撑可控、可编辑的仿真世界构建。
链接: https://arxiv.org/abs/2610.38468
作者: Zifan Song,Caihua Shan,Kaitao Song,Yifan Yang,Dongsheng Li,Cairong Zhao
机构: Tongji University(同济大学); Microsoft Research Asia(微软亚洲研究院)
类目: Multiagent Systems (cs.MA)
备注: 25 pages, 9 figures, 4 tables
Abstract:Controllable world simulation requires more than visually plausible frames: scenes must retain persistent state, coordinated actions, and editable temporal structure. Generative world models offer flexible visual synthesis, but their outputs do not always expose the objects and transitions needed for precise revision. Graphics engines provide executable state and direct control, yet translating language into valid operations requires specialized knowledge of assets, geometry, and interdependent APIs. We present CineU (Cinematize from Unreal), an agentic system for language-driven world orchestration and cinematic choreography within Unreal Engine. From a code-based world modeling perspective, an Unreal project serves as a persistent, inspectable world representation, while agent-generated code specifies operations whose effects can be executed and verified. CineU combines hierarchical orchestration, cascaded asset retrieval and geometry-constrained placement, capability-aligned code synthesis, and multi-stage verification with localized repair. Shared scene and timeline context coordinates entities, actions, camera trajectories, and shot boundaries while supporting dependency-aware revisions. Experiments on scene construction, selective editing, multi-shot cinematography, and component ablations show how specialized execution knowledge and engine feedback improve task completion and targeted recovery. CineU illustrates how executable world state can complement learned visual generation to support controllable and revisable simulated worlds.
[MA-15] Agent Transfer Protocol: Server-Mediated Agent Agent Agent -to-Agent Communication Architecture for the Agentic Web
【速读】：该论文旨在解决跨组织智能体（Agent）在长期协作中面临的通信稳定性与消息可靠性问题，尤其是在执行实例离线或更换的情况下，如何确保消息的持久接收、来源验证及内容完整性。其核心挑战在于避免依赖特定执行实例的持续在线状态来维持跨域通信。解决方案的关键是提出一种基于域中继架构的代理传输协议（ATP, Agent Transfer Protocol），通过由各组织管理域服务器维护稳定的消息接收入口和地址，实现跨域消息的持续中继；发送方对完整消息包进行签名，接收方域服务器负责验证并持久存储消息，待执行实例上线后可独立重新验证消息来源与内容完整性，前提是原始公钥仍可信且可获取。该机制实现了不依赖具体执行实例在线状态的可靠异步通信，为构建长期稳定的智能体交互基础设施提供了有效支撑。
链接: https://arxiv.org/abs/2610.38398
作者: Yuqi Qiu,Zuyao Xu,Lu Sun,Xiang Li,Fasheng Miao,Fubin Wu,Haixin Duan,Zheli Liu
机构: Nankai University(南开大学); Tsinghua University(清华大学); Quancheng Laboratory(全城实验室)
类目: Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
备注: 12 pages, 6 figures, 5 tables
Abstract:The Agentic Web needs to support continuous collaboration among agents from different organizations. These collaborative relationships may span multiple tasks, network connections and execution instances. Such interactions require stable communication entries, message reception when instances are offline, and source verification after delayed reading. How to organize these responsibilities so that cross domain communication does not depend on specific execution instances being continuously online is a key issue for building long term agent interaction infrastructure. We propose ATP (Agent Transfer Protocol), a cross organization agent communication protocol adopting a domain mediated architecture. In this architecture, administrative domains maintain stable addresses and receiving entries for logical agents. The domain servers of both parties continuously undertake cross domain transmission. The receiving domain verifies and saves messages for execution instances to read later. The sending domain signs the complete message envelope. This allows the receiving instance to independently reverify the domain source and content integrity after the original connection ends, provided the original verification public key is still trusted and obtainable. We implemented an ATP prototype and evaluated it through a controlled cross organization LLM agent workflow, mechanism checks and latency measurements. The workflow completed subsequent business processing after instances went offline or were replaced. The independent verifier successfully verified the domain source and content integrity of saved messages. Latency measurements further quantified the communication and processing overhead introduced by the continuous mediation of domain servers.
[MA-16] Disparity Laundering: Consensus Without Fairness in Clinical Multi-Agent Systems EACL2027
【速读】：该论文旨在解决临床大语言模型（Clinical Large Language Models）在多智能体系统中因群体协作而产生的偏见放大问题，即尽管多智能体通过协商可降低个体错误，却无法消除或中和系统性偏见。其核心问题是：当前的多智能体架构在处理具有反事实人口学特征替换的真实急诊病例时，所有配置均未能消除对黑人患者过度紧急分诊的偏差，且该偏差不仅持续存在，还被“重新定位”至不同模型与拓扑结构组合中，形成不可靠的安全或放大模式。解决方案的关键在于引入“多样化视角集成”（diverse-perspective ensemble），通过在每轮决策中显式暴露并聚合反事实版本的输出，使偏见在协同推理过程中被可见并整合进最终决策。研究发现，仅当所有五种模型视角在每一轮中均参与信息共享与联合决策时，偏见差距才显著缩小67%至94%，而唯一无法实现修复的拓扑结构正是不进行跨代理信息聚合的结构。因此，论文提出“偏见洗白”（disparity laundering）概念，指出传统审计机制（如按模型、输出、版本逐项审查）无法捕捉已嵌入决策中的隐性偏见；真正的防御手段是让偏见在系统内部可见，并通过集体审议加以抑制，无需重训练或改变架构即可实现。
链接: https://arxiv.org/abs/2610.38215
作者: Xiaoyang Wang,Tianrui Wang,Christopher C. Yang
机构: Drexel University (德雷塞尔大学); University of Michigan (密歇根大学)
类目: Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
备注: 21 pages, 8 figures. Code: this https URL . Committed to EACL 2027
Abstract:Clinical large language models are increasingly fielded as multi-agent teams, on the assumption that deliberation averages out individual error. It does not average out bias. Across a full factorial of five model families and four aggregation topologies on 120 real MIMIC-IV emergency cases under counterfactual demographic swaps, no configuration neutralizes the disparity: all 20 cells commit the direction of their agents’ independent answers, giving Black patients more acute triage. Every model has a topology that amplifies its baseline gap, but which one varies by model, so no design is reliably safe or reliably amplifying: the disparity is relocated, never removed. Late-stage fixes fail: fairness instructions and demographic masking leave the gap untouched, and per-agent self-debiasing widens it, because the bias is already encoded in the agents’ ordinary clinical numbers. We call this disparity laundering; the audits now applied to clinical AI, per model, per output, per version, can certify the parts but never the committed decision. Interventions close the gap exactly when they make counterfactual versions visible inside the system and pool them, through deliberation, into the committed decision. A diverse-perspective ensemble does both while keeping the team, with no retraining or architectural change: the disparity falls by 67 to 94% where all five perspectives pool at every round, and the one topology that never pools them is the one it cannot reliably repair. Medical multi-agent systems do not filter demographic bias; they launder it into plausible decisions no single-version audit can catch, and the only defense is to make the bias visible.
[MA-17] Is it my preference? Role Grounding Attack against Agents Memory via Social Interaction
【速读】：该论文旨在解决大语言模型（LLM）代理在社交平台共享经验并记忆用户偏好以提升性能时，其记忆系统中存在的角色-主体归属错误（role-to-principal attribution）漏洞问题。核心问题是：攻击者通过设计特定交互，诱导目标代理将攻击者用户的偏好错误地归因于自身用户，从而导致记忆内容的语义错位。解决方案的关键在于首次形式化提出角色锚定攻击（Role Grounding Attack, RGA），该攻击利用嵌套对话结构实现恶意偏好的误导性记忆。具体而言，在记忆形成阶段，攻击者发布包含其代理与用户之间对话的分享帖，其中内层对话明确表达攻击者用户的偏好；当目标代理在记忆过程中未正确保留上下文绑定时，会将这些偏好错误地锚定至自身用户的主体身份。在记忆检索阶段，这种误归因持续影响代理决策。研究通过工具选择操纵和隐私泄露两种攻击实例验证了RGA的有效性，在五种主流记忆系统与六种语言模型上均表现出高成功率，揭示了当前记忆系统在角色上下文感知方面的严重不足，因而亟需强化角色锚定机制以保障记忆内容的真实性与安全性。
链接: https://arxiv.org/abs/2610.38209
作者: Haoran Bu,Yiting Shen,Litian Zhang,Ruihan Hu,Xi Zhang
机构: Beijing University of Posts and Telecommunications (北京邮电大学)
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
备注:
Abstract:LLM-based agents now share experiences on social platforms and remember new preferences to enhance their performance. However, existing memory systems contain a vulnerability in role-to-principal attribution that attackers can exploit. We identify Role Grounding Misattribution, in which an attacker-designed interaction causes the victim agent to incorrectly record a preference from the attacker as a preference of the victim’s user. Specifically, we are the first to formalize Role Grounding Attack (RGA), an attack model that exploits this vulnerability through nested conversation structures. In the memory formation stage, an experience-sharing post quoting a conversation between the attacker agent and its user is published, which explicitly identifies its publisher. The inner conversation expresses preferences from the user of the attacker agent. When the victim agent remembers preferences from this nested structure, its memory loses the context binding and grounds the attacker’s preference to the victim’s user instead. In the memory retrieval stage, the misattributed preference persists and influences the victim agent’s decisions when it encounters relevant scenarios. We instantiate RGA as tool selection manipulation and privacy leakage. Tool selection manipulation causes a tool-call preference to be misattributed to the victim’s user, steering the victim toward an attacker-controlled tool. Privacy leakage extends this by misattributing both a tool-call preference and a task-requirement preference to the victim’s user. The victim selects the attacker’s tool, then includes sensitive information in tool arguments to satisfy the task requirement. We evaluate RGA across five mainstream memory systems and six language models. Tool manipulation and privacy leakage both achieve high attack success rates. These results call for stronger role grounding.
[MA-18] Cooperative EV Charging Assignment Using Multi-Agent Reinforcement Learning: A Comparison of QMIX and Independent Q-Learning Under Urban Traffic Uncertainty
【速读】：该论文旨在解决多充电站协同调度中因信息局部化与异构性导致的电动汽车（EV）分配效率低下问题，即各充电站仅基于局部信息做出接受/拒绝决策时，可能损害整个充电区域的整体运行效率。其核心挑战在于如何在去中心化执行的前提下实现有效的全局协调。解决方案的关键在于采用基于集中式训练、去中心化执行（CTDE）框架的QMIX算法，通过引入联合策略网络和注意力机制，在训练阶段共享全局信息以学习协作策略，而在执行阶段保持各站点的独立决策能力。实验结果表明，相较于合作型独立Q学习（IQL）基线，QMIX在平均每步奖励上提升0.198（p = 0.0625），接受率从72.4%显著提高至94.0%（Δ = +21.6 pp, dz = 5.24），且消融实验显示性能提升主要归因于CTDE架构本身，而非奖励设计或经验共享等辅助机制。研究支持了将协调机制内嵌于学习架构中的观点，强调应优先优化模型结构而非依赖启发式合作项。
链接: https://arxiv.org/abs/2610.38053
作者: Nour-Eddine Moumni,Rachid Alaoui,Driss Kiouach
机构: Mohammed V University (穆罕默德五世大学); Higher School of Technology of Salé (EST Salé) (萨勒高等技术学院); Sidi Mohamed Ben Abdellah University (西迪·穆罕默德·本·阿卜杜拉大学)
类目: Multiagent Systems (cs.MA); Machine Learning (cs.LG)
备注: 14 pages, 7 figures. Presented at ICATH 2026 (Rabat, Morocco, 15-17 July 2026); accepted for publication in MDPI Engineering Proceedings (in press)
Abstract:The allocation of electric vehicles to public charging stations is increasingly constrained by the fact that several heterogeneous facilities often serve the same stream of requests while observing only partial local information. An accept/reject decision that is defensible for one station may nevertheless reduce the efficiency of the charging district as a whole. We formulate this operational problem as a Decentralized Partially Observable Markov Decision Process (Dec-POMDP) and examine whether Centralized Training with Decentralized Execution (CTDE), instantiated through QMIX, improves assignment performance relative to a cooperative Independent Q-Learning baseline (IQL). The evaluation uses a six-station urban network simulated with SUMO, five independent random seeds, 200 training episodes, and four traffic flux scenarios ranging from 100 to 250 vehicles/hour. Under this protocol, QMIX reaches a mean reward per evaluated decision of +0.776, versus +0.578 for IQL (delta = +0.198, seed-paired Cohens dz = 2.98, exact two-sided Wilcoxon p = 0.0625, the minimum attainable value at this sample size), and raises the acceptance rate from 72.4% to 94.0% (delta = +21.6 pp, dz = 5.24, Wilcoxon p = 0.0625). A five-variant ablation suggests this gain is primarily attributable to the CTDE architecture when holding cooperative reward terms fixed, though not formally isolated with a single seed. By contrast, cooperative reward shaping and experience sharing show no clear descriptive benefit and may slightly reduce performance. These findings support the view that, for multi-station charging assignment, coordination should be embedded in the learning architecture rather than added only through heuristic cooperation terms.
[MA-19] EARS-Info: Information-Aware Event-Driven Reallocation for Multi-Robot Exploration
【速读】：该论文旨在解决多机器人在基于OpenStreetMap生成的网格环境中进行探索时，如何高效分配任务、优化路径规划并提升整体探索性能的问题。其核心挑战在于平衡感知增益、通信可靠性、能量消耗与任务重分配之间的权衡，尤其在存在通信中断（3%通信丢包率）等实际约束条件下。解决方案的关键在于提出EARS-Info，一个信息感知型事件驱动蚁群重分配搜索框架，其创新性地融合了边际感知增益评估、可行重定位机制、返程意识路由策略、感知资源预留以及缓存最短路径等组件。实验结果表明，在8个城市地图上共16.8万次迭代中，EARS-Info在所有评估场景下均取得最高平均运行得分，相较基线算法EARS-Safe在加权发现率上提升9.58个百分点（从88.88%增至98.46%），平均能耗降低2.62%，且在所有城市中得分差异均显著（p=0.0078125，95%置信区间[0.0159, 0.0694]）。尽管各模块的独立贡献尚不明确，但整体系统表现出卓越的鲁棒性与有效性，验证了其作为完整控制器的可行性。然而，模型泛化能力受限于重复地理结构、单一训练样本、粗粒度地图及对比方法的适应性调整。
链接: https://arxiv.org/abs/2610.38040
作者: Kabeh Mohsenzadegan,Vahid Tavakkoli,Kyandoghere Kyamakya
机构: 未知
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
备注:
Abstract:We introduce EARS-Info, an information-aware extension of Event-Driven Ant Reallocation Search for multi-robot exploration on OpenStreetMap-derived grids. It combines marginal sensing gain, feasible relocation, return-aware routing, sensing reservations, and cached shortest paths. Across 168,000 episodes comparing 30 implementations on eight city maps, EARS-Info achieves the highest mean operational score in every evaluation panel. With eight robots and 3% communication dropout, it improves the score over EARS-Safe by 5.43%, increases weighted discovery from 88.88% to 98.46%, and reduces mean energy use by 2.62%. Score differences are positive in all eight cities ( p=0.0078125 ; descriptive 95% city-bootstrap interval [0.0159,0.0694] ), with no simulated incidents across 5,600 EARS-Info episodes. These results support the complete controller; individual component advantages remain uncertain. Generalization is limited by reused geography, one training replicate, coarse maps, and adapted comparison methods.
[MA-20] Who Else Is In? Multi-Agent Collusion Initation Via Strategic Vagueness
【速读】：该论文旨在解决多智能体大语言模型（Multi-agent LLM systems）中潜在的合谋风险，即智能体在未被监管者察觉的情况下，通过隐秘通信达成共同目标。其核心问题在于：当合谋者身份未知时，如何在存在普通参与者和监控者的混合语境下实现隐蔽的相互识别与协作。解决方案的关键在于揭示“策略性模糊”（strategic vagueness）作为一种有效沟通机制的作用——智能体可通过与任务兼容的间接交流，在不触发监控警报的前提下完成相互识别与协同。研究发现，部分模型无需显式信号引导即可自发形成高效握手策略，且这些策略在低误报率下难以被监控系统识别。进一步的受控协调实验表明，随着信息表达趋于间接，监控惩罚的下降速度可能快于合谋伙伴间的协作效率损失，从而形成一个可获利的中间区间，即策略性模糊的有利空间。这一现象体现了类似“布莱克斯特恩证据不对称性”（Blackstone-style evidential asymmetry）的特征：足以支持智能体间协调的通信，可能仍不足以构成监控方确信的合谋证据，从而为合谋行为开辟了新的隐蔽路径。
链接: https://arxiv.org/abs/2610.37984
作者: Pramod Kaushik,Sarath Sivaprasad,Laura Ferrarotti,Mario Fritz,Samuel F. Brown,Bruno Lepri,Sahar Abdelnabi
机构: University of Trento; Fondazione Bruno Kessler; ELLIS Institute Tübingen; MPI-IS; Tübingen AI Center; Independent; CISPA
类目: Multiagent Systems (cs.MA)
备注:
Abstract:Multi agent LLM systems raise the risk of collusion among agents pursuing objectives hidden from their overseers. When potential partners identities are unknown, collusion first requires initiation: agents must discover one another, often through a public channel observed by ordinary agents and monitors. Strategic vagueness offers a way to navigate this mixed audience, conveying intent to a prospective partner while preserving a plausible ordinary interpretation. We introduce a mutual identification game in which two schemers must identify each other through public interaction, while jointly solving an ordinary task. We find that some models discover successful handshake strategies without explicit signalling guidance, using task compatible exchanges that monitors often fail to flag at low false positive rates. To characterize the communicative capabilities underlying these exchanges, we study a controlled coordination setting. We find an asymmetric response to message explicitness: as messages become more indirect, monitor punishment can fall faster than partner coordination, creating a profitable intermediate regime of strategic vagueness. Together, these findings illustrate a Blackstone style evidential asymmetry: communication sufficient for coordination between agents may remain insufficient for confident detection by monitor, opening a new surface of collusion.
[MA-21] GARDEN: Generalist Autonomous Reward Design Through Evidence Navigation for Cooperative Multi-Agent Reinforcement Learning
【速读】：该论文旨在解决协作式多智能体强化学习（Cooperative Multi-Agent Reinforcement Learning, MARL）中奖励函数设计依赖人工且难以跨场景复用的核心问题。传统方法需为每个环境或团队配置手动设计的中间奖励，导致开发成本高且可移植性差。其解决方案的关键在于提出GARDEN（Generalist Autonomous Reward Design through Evidence Navigation），通过大语言模型（Large Language Models, LLMs）作为“奖励工程师”，在单一任务家族的多种场景中自动分析环境代码、训练记录与智能体行为，生成一个通用性强的奖励程序，并保持不变地迁移至未见过的场景。该方法在Overcooked和SMAX基准测试中显著提升了任务回报（分别提高58.2%和52.5%）及胜率（从51.0%提升至89.9%），优于固定诊断反馈与受Eureka启发的奖励反思策略，证明了单个自主设计的奖励程序可在相关协作场景间实现高效泛化。
链接: https://arxiv.org/abs/2610.37926
作者: Dogan Urgun
机构: Karabuk University (卡拉布克大学)
类目: Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA); Systems and Control (eess.SY)
备注: 8 pages, 3 figures, 5 tables. Code: this https URL
Abstract:Cooperative multi-agent reinforcement learning (MARL) enables teams of agents, such as robots and autonomous vehicles, to learn coordinated behavior, but what such teams learn is largely determined by their reward function. Because task success alone provides too little guidance for learning, intermediate rewards are usually designed by hand, and this effort must often be repeated whenever the environment or the team changes. Large language models (LLMs) can automate the effort by writing reward functions as code, yet the resulting rewards are typically designed and tested in a single setting. To enable their reuse across settings, GARDEN (Generalist Autonomous Reward Design through Evidence Navigation) is proposed, in which one reward program is developed for a family of related scenarios and transferred, unchanged, to scenarios that were not available during its design. During development, environment code, training records, and agent behavior are investigated by an LLM reward engineer, and draft rewards are tested in trial training runs, while the learning algorithm and the rule for selecting the final reward are held fixed. On unseen scenarios of the Overcooked and SMAX benchmarks, task returns were increased by 58.2% and 52.5% over the default rewards on two kitchen layouts, and the mean win rate in two combat scenarios was raised from 51.0% to 89.9%. Under matched candidate and probe budgets, this exceeded the 84.2% and 79.2% obtained with fixed diagnostic feedback and Eureka-inspired reward reflection. These results show that a single autonomously engineered reward can be reused effectively across related cooperative scenarios.
[MA-22] Multi-Agent Reinforcement Learning for Global Climate Policy Cooperation NEURIPS2026
【速读】：该论文旨在解决国际气候政策协调中因各国减排决策具有本地成本但产生全球收益而引发的“搭便车”（free-riding）问题。传统博弈论与综合评估模型虽可分析此类激励机制，但通常依赖预设的行为假设或均衡概念。为此，本文引入多智能体强化学习（Multi-Agent Reinforcement Learning, MARL）框架，通过构建多阶段协商机制——包括协议提案、承诺接受/拒绝及不履约惩罚——将气候合作视为重复战略互动中涌现的结果。其解决方案的关键在于：将协议设计（如惩罚力度）作为可调节变量，系统研究其对学习到的减排行为、履约率及气候结果的影响。在由四个同质代理组成的理想化场景中发现，非约束性协议难以缓解搭便车现象，适度的不履约惩罚能提升合作并减缓升温，但过强的惩罚反而抑制雄心勃勃的承诺，揭示了执行强度与参与积极性之间的权衡关系。该框架定位为一种用于测试协议设计的范式方法，而非对真实谈判过程的预测模型。
链接: https://arxiv.org/abs/2610.37870
作者: Anne Moll-Elsborg,Oskar Bohn Lassen,Francisco Camara Pereira
机构: Technical University of Denmark(丹麦技术大学); Department of Technology, Management and Economics(技术、管理与经济学系)
类目: Multiagent Systems (cs.MA); Computer Science and Game Theory (cs.GT); Machine Learning (cs.LG)
备注: Tackling Climate Change with Machine Learning: workshop at NeurIPS 2026
Abstract:International climate policy requires coordination among countries whose mitigation decisions impose local costs but generate global benefits. Classical game-theoretic and integrated assessment approaches can study these incentives, but typically rely on prescribed behavioural assumptions or equilibrium concepts. Recent work has therefore explored Multi-Agent Reinforcement Learning as a way to model climate-policy cooperation as an emergent outcome of repeated strategic interaction. We extend an existing MARL climate framework with a multi-stage negotiation mechanism in which agents propose agreements, accept or reject commitments, and face penalties for non-compliance. This allows us to study how agreement design affects learned mitigation behaviour, compliance, and climate outcomes. In an idealized setting with four homogeneous agents, we find that non-binding agreements do little to overcome free-riding, while moderate non-compliance penalties improve cooperation and reduce warming. However, excessively strong penalties discourage ambitious commitments, revealing a trade-off between enforcement and participation. We position the framework as a methodological contribution for testing agreement design rather than a predictive model of real negotiations.
[MA-23] Measuring Where Intent Gets Lost in Agent ic Smart Contract Synthesis
【速读】：该论文旨在解决生成式人工智能（Generative AI）在编写智能合约时存在的核心问题：尽管AI能够根据自然语言描述生成可运行的代码，但生成的代码往往未能准确实现用户意图，尤其在Web3应用中，智能合约一旦部署便无法修改，行为偏差可能导致重大经济损失。其关键挑战在于如何有效验证AI生成的合约是否真正符合原始需求，以及在何处导致了行为偏离。论文提出以执行为基础的评估框架，通过重放相同交易于AI生成合约与预期正确版本之间，量化二者输出的一致性，从而客观衡量行为匹配程度。研究发现，现有主流方法如AI评判、静态分析工具等常忽略明显错误，且评分策略对结果影响显著；仅检查通用函数模块会遗漏近半数的行为差异。为此，论文提出“先探查后生成”（elicitation-first）的架构：在代码生成前，由AI代理主动向用户提问一组关键问题，基于独立编写的可执行答案键进行比对，并将不一致反馈给用户，而非自动修正。实验表明，在三种主流AI模型上，前置探查远优于事后审查或修复，且少量精心设计的问题即可弥补自然语言描述中的关键缺失，其价值甚至可在无原合同的情况下预估。该工作是首个在智能合约合成中系统性收集缺失需求并以可执行模型验证结果的研究，其方法论对任何将用户请求转化为程序的AI代理场景均具普适意义。相关工具、600个测试任务及全部结果均已开源。
链接: https://arxiv.org/abs/2610.37830
作者: Abhinav Goel,Agostino Capponi,Alfio Gliozzo,Chaitya Shah
机构: Columbia University (哥伦比亚大学); IBM T.J. Watson Research Center (IBM托马斯·J·沃森研究中心)
类目: Artificial Intelligence (cs.AI); Human-Computer Interaction (cs.HC); Information Retrieval (cs.IR); Multiagent Systems (cs.MA); Software Engineering (cs.SE)
备注: ACM TheWebConf2027 8 pages main content 4 pages appendix
Abstract:AI agents can now write software from a plain-English description, but code that runs is not the same as code that does what was asked. This gap matters most for smart contracts, the programs behind Web3 applications that move money on a blockchain and cannot be fixed once deployed. We ask how to check whether an AI-written contract behaves as it should, and where the requested behavior gets lost. We design an evaluation built on execution rather than opinion. We replay the same transactions on an AI-written contract and on the original it should reproduce and measure how often they agree. We test every grading tool on planted mistakes, compare against an answer key written without the description, and audit our own scorer until known wrong answers cannot pass. Common methods, such as AI judges and standard analysis tools, miss obvious mistakes, and scoring choices matter too. Checking only the functions every system includes hid about half the difference between pipelines. We then propose an elicitation-first architecture. Before writing code, the agent asks the client a short checklist of questions, checks the contract against an independently written executable answer key, and returns disagreements to the client instead of silently rewriting code. Asking first matters far more than reviewing or patching afterwards, across three AI models, while automated security fixes often break requested behavior. A few well-chosen questions recover much of what a short description omits, and their value can be estimated without the original contract. To our knowledge, this is the first study in smart contract synthesis to gather missing requirements before generation and check the result against an independently written executable model. The lessons extend to any setting where AI agents turn requests into programs. We release our tools, 600 tasks and all results. Comments: ACM TheWebConf2027 8 pages main content 4 pages appendix Subjects: Artificial Intelligence (cs.AI); Human-Computer Interaction (cs.HC); Information Retrieval (cs.IR); Multiagent Systems (cs.MA); Software Engineering (cs.SE) Cite as: arXiv:2610.37830 [cs.AI] (or arXiv:2610.37830v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2610.37830 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[MA-24] Selective and Hybrid-State Communication for Latent Multi-Agent Systems NEURIPS2026 AAAI
【速读】：该论文旨在解决大规模语言模型多智能体系统中高效通信的瓶颈问题。传统基于自然语言的交互存在重复冗余，而现有隐式通信方法虽通过传递内部模型状态（latent states）减少此类开销，但其通信代价依然高昂，且通常依赖于传统的键值缓存（KV-cache）表示方式，限制了效率提升。针对这一问题，论文提出了两种互补的解决方案：其一是提出SelectKV，一种具有边界感知能力的方法，能够选择性地传输部分键值（KV）状态，同时保护继承的上下文信息；其二是提出HyLaMAS，将隐式协作扩展至混合线性注意力模型（hybrid linear-attention models），通过共享工作区同时传递KV缓存与循环状态（recurrent states）。实验结果表明，在六个基准测试中，SelectKV使通信的KV位置数和逻辑KV数据量分别降低9.7%-9.9%，在多数任务上保持精度稳定或仅轻微下降；而在Qwen3.5系列模型上，HyLaMAS实现了53%-80%的解码输出令牌减少，并相较基于文本的多智能体通信获得1.16-2.78倍的速度提升。两项方法分别针对传统Transformer架构与混合模型架构，共同构成了高效隐式状态通信的双路径解决方案。
链接: https://arxiv.org/abs/2610.37779
作者: Miles Liao,Aaron Smieszek,Yuhao Ge
机构: Hopkins School; Independent Researcher
类目: Multiagent Systems (cs.MA); Machine Learning (cs.LG)
备注: Accepted at the NeurIPS 2026 Workshop on Resource-Aware Agentic AI (RAAAI)
Abstract:Efficient communication remains a major bottleneck in large language model multiagent systems. Latent communication reduces repeated natural-language exchange by transferring internal model states between agents, but these transfers remain costly and are typically designed around conventional KV-cache representations. We address this problem from two complementary perspectives. We introduce SelectKV, a boundary-aware method that selectively communicates KV states while protecting inherited context, and HyLaMAS, which extends latent collaboration to hybrid linear-attention models by communicating both KV caches and recurrent states through a shared workspace. Across six benchmarks, SelectKV reduces communicated KV positions and logical KV payload by 9.7-9.9%, with accuracy unchanged on two benchmarks, within one percentage point on three, and six points lower on HumanEval+. Across Qwen3.5 model scales, HyLaMAS reduces decoded output tokens by 53-80% and achieves 1.16-2.78x speedups relative to text-based multi-agent communication. Together, these independently evaluated methods represent two complementary approaches to efficient latent-state communication across conventional Transformer and hybrid model architectures.
[MA-25] Balancing Mobility Efficiency and Environmental Exposure During Mega Events: A Multi-agent Routing Framework
【速读】：该论文旨在解决大型活动（mega events）期间因集中出行需求导致的交通拥堵与出行者环境暴露风险加剧的问题。现有交通管理策略多聚焦于通行效率，忽视了对污染物（如交通相关PM2.5和噪声）暴露风险的考量。为此，研究提出一种面向环境暴露感知的多智能体路径规划框架（exposure-aware multi-agent routing framework），在统一仿真环境中融合道路网络状态、活动诱发的需求、背景交通流量及环境数据，分别针对机动化出行者与微出行者进行优化：前者以提升通行效率为目标，后者则同时考虑出行时间和动态环境暴露水平。通过与最短路径、拥堵感知及纯暴露感知等基线方法对比，结果表明，在高需求场景下，基于SARSA算法的框架可使机动车出行时间降低34.59%，累计环境暴露量减少58.65%，而微出行方式的出行时间基本保持不变。环境暴露的显著下降主要归因于车流侧的拥堵缓解与低暴露路径协同作用。该框架为短期高需求场景下的车辆绕行引导、低暴露路线推荐及健康导向型交通管理提供了量化决策支持。
链接: https://arxiv.org/abs/2610.37702
作者: Xinyue Liu,Qiumeng Li,Suhong Zhou,Junwen Lu,Jiali Zhou,Weipan Xu
机构: Thrust of Urban Governance and Design, Society Hub, The Hong Kong University of Science and Technology (Guangzhou), China; Thrust of Intelligent Transportation, Systems Hub, The Hong Kong University of Science and Technology (Guangzhou), China; School of Geography and Planning, Sun Yat-sen University, China; School of Intelligent Systems Engineering, Sun Yat-sen University, China; Department of Urban Planning and Design, The University of Hong Kong, Hong Kong
类目: Multiagent Systems (cs.MA); Computers and Society (cs.CY)
备注:
Abstract:Mega events can generate concentrated travel demand, intensifying congestion around venues and increasing travelers’ environmental exposure. Existing traffic management strategies focus on travel efficiency, with limited attention to exposure risks. Taking mega-event travel as a short-term high-demand scenario, this study develops an exposure-aware multi-agent routing framework integrating road-network conditions, event-induced demand, background traffic, and environmental data within a unified simulation environment. Motorized travelers optimize travel efficiency, whereas micro-mobility travelers consider travel time and dynamic exposure to traffic-related PM2.5 and noise. We compare the framework with shortest-distance, congestion-aware, and exposure-aware routing baselines. Results show that, under high demand, the SARSA-based framework reduces vehicle travel time by 34.59% and cumulative environmental exposure by 58.65%, while micro-mobility travel time remains nearly unchanged. Exposure reduction is jointly driven by vehicle-side congestion mitigation and low-exposure routing. The framework provides quantitative support for vehicle diversion, low-exposure route guidance, and health-oriented traffic management in short-term high-demand scenarios.
[MA-26] he Coordination Tax and the Heterogeneity Dividend: Break-Even Conditions for Multi-Agent LLM Pipelines
【速读】：该论文旨在解决多智能体大语言模型（Multi-agent LLM）流水线在实际应用中运行成本与性能收益之间权衡不明确的问题。尽管基于互操作协议（如Open Floor Protocol, OFP）的多智能体系统常以能力优势为依据进行论证，但其运行开销（包括任务分解的协调税与跨角色传递的协议开销）长期被忽视或低估。本文提出一个闭式模型，将协调税（分解开销 + 每次交接的协议成本）与因任务分配至更低成本模型而获得的异构红利（heterogeneity dividend）进行量化对比，并推导出盈亏平衡阈值和可行性边界 α ≤ 1/r。研究通过在135个HumanEval与ClassEval任务上执行2430次实验，评估六种架构在Claude Opus 5.5与Haiku 5.5之间的表现，发现：以OFP为协议时，每次角色交接增加7.2毫秒延迟，占流水线总耗时的0.11%；然而，在这些任务上，规划-编码-评审流水线的总成本是单一强模型调用的3.22倍，耗时增加3.24倍，且解题率未提升。即使将规划与评审环节迁移到低成本模型，虽使流水线成本降至0.39倍，但pass@1指标下降6.7点，仍导致整体更昂贵、更慢且准确率更低。此外，预测的盈亏平衡点偏差达-19%，主要源于低成本角色引入的额外词元量。最终结论表明，真正的成本节约并非来自流水线化，而是基于任务特性选择最优单模型调用——理想情况下在两个单模型调用间进行“最优选择”可实现94.3%的任务成功率，同时将每成功解题的成本降低至原强模型调用的一半。因此，解决方案的关键在于：根据任务特性动态选择最适配的单模型，而非依赖固定多智能体流水线结构。
链接: https://arxiv.org/abs/2610.37700
作者: Andreas Zettl,Diego Gosmar
机构: VoiceInteroperability.ai; Linux Foundation AI Data; Berlin, Germany; Turin, Italy
类目: Multiagent Systems (cs.MA); Artificial Intelligence (cs.AI)
备注: 16 pages, 3 figures, 4 tables
Abstract:Multi-agent LLM pipelines built on interoperability protocols such as the Open Floor Protocol (OFP) or A2A are usually justified on capability grounds, with their runtime cost treated as an overhead of unknown size. Comparing such a pipeline with one model call is underspecified, because a pipeline can give each role a different model. We give a closed-form model that sets the coordination tax (decomposition overhead plus per-handoff protocol cost) against the heterogeneity dividend from routing work to a cheaper model, with a break-even threshold and a feasibility bound, \alpha 1/r , and measure every term, with OFP as the protocol under test, on 135 HumanEval and ClassEval tasks in 2430 runs of six architectures with Claude Opus 5.5 and Haiku 5.5. The protocol adds 7.2 ms per handoff, 0.11% of pipeline wall time. Decomposition does not pay on these tasks: a planner-coder-reviewer pipeline costs 3.22x as much as a single strong call, takes 3.24x as long, and solves no more tasks. Moving planner and reviewer to the cheap model cuts the pipeline’s cost to 0.39x but changes pass@1 by -6.7 points, leaving it dearer, slower and less accurate than the single call, and the break-even forecast misses by -19% because the cheap roles inflate token volume. The saving lies in choosing a model per task instead: an oracle choice between the two single calls solves 94.3% of runs at half the strong call’s cost per solved task.
[MA-27] Network Topology Does Not Matter: The Breakdown of Classical Consensus in LLM Populations
【速读】：该论文旨在解决多智能体语言模型网络（multi-agent language model networks）在实际部署中集体行为不可预测的问题。传统分布式共识理论依赖于网络拓扑结构来预测收敛性，其核心假设包括：智能体更新为凸组合、群体均值保持不变以及权重与内容无关。然而，本文通过从8个指令微调的大语言模型（LLM）在73,080次受控交互中提取的连续可信度（continuous credence），直接测量了真实存在的LLM更新算子，发现这些基础假设在大语言模型场景下均不成立。具体而言，LLM更新行为系统性地违反凸性，表现为更新超出观测意见的凸包，表现出主动排斥（negative weights），从根本上破坏了行随机建模和传统的有界信心修正方法。将实证拟合的更新算子扩展至包含10,000个智能体的模拟中，结果表明代数连通性几乎无法解释最终结果的方差；相反，群体动态完全由模型内部的更新偏差和信息框架决定。因此，多个模型（包括高达70亿参数的模型）会持续放大微小的局部漂移，最终达成对客观错误命题的高置信共识，且这一过程与网络结构无关。研究揭示了经典图论指标在大语言模型群体中的失效，凸显了多智能体协商可能陷入集体错觉（collective delusion）的重大部署风险。解决方案的关键在于认识到：大语言模型的更新机制具有内在非凸性和负权重特性，其群体行为主要受内部偏见和消息表达方式驱动，而非外部网络拓扑，必须发展新的理论框架以刻画此类非线性、非对称的集体演化机制。
链接: https://arxiv.org/abs/2610.37699
作者: Sahil Rajesh Dhayalkar
机构: Arizona State University (亚利桑那州立大学)
类目: Multiagent Systems (cs.MA); Artificial Intelligence (cs.AI)
备注: 19 pages, 3 figures, 3 tables
Abstract:As multi-agent language model networks become widely deployed, predicting their collective behavior is a critical challenge. Classical distributed consensus theory relies on network topology to predict convergence, structurally assuming that agent updates are convex combinations, population means are conserved, and weights are content-independent. In this work, we demonstrate that these foundational assumptions fail for large language models (LLMs). By deterministically extracting continuous credence from token-level probabilities across eight instruction-tuned models and 73,080 controlled exchanges, we directly measure the empirical LLM update operator. We find that LLMs systematically violate convexity by updating outside the hull of observed opinions, exhibiting active repulsion (negative weights) that fundamentally breaks row-stochastic modeling and standard bounded-confidence repairs. Scaling these empirically fitted operators to simulations of 10,000 agents reveals that algebraic connectivity explains almost none of the final outcome variance. Instead, population dynamics are governed entirely by the model’s internal update bias and message framing. Consequently, populations of several models, including those with up to 7B parameters, reliably compound small local drifts to reach confident agreement on objectively false claims regardless of their network structure. Our results prove that classical graph-theoretic metrics are inadequate for LLM populations, highlighting a profound deployment risk where multi-agent deliberation predictably saturates in collective delusion.
[MA-28] Rethinking Social Simulation in Open-Population Systems: From Persistent Agents to Communication Processes
【速读】：该论文旨在解决大规模开放种群社会系统中基于持久个体状态的生成式社会模拟所面临的根本性挑战：在动态变化的用户参与环境中，传统的持久账户身份无法提供连续可观测的个体状态轨迹，导致递归维护的心理状态缺乏实证基础。其核心解决方案是将模拟目标从“持久个体状态轨迹”转向“可观察的社会通信演化”，通过从全局用户池中激活参与者、建模集体状态转移，并利用大语言模型（LLM）生成通信内容，同时显式分离环境输入与系统内生演化过程。该框架在五个Qwen基线模型上实现了平均KL散度降低71.2%（相较于直接使用LLM）和35.0%（相较于MF-LLM），并通过边界实验与反事实分析验证了模型设定的有效性，证明以通信为中心的建模范式为长时程开放种群社会系统提供了更严谨的理论基础。
链接: https://arxiv.org/abs/2610.37687
作者: Yunyao Zhang,Yihao Ai,Boqi Xu,Zuocheng Ying,Junqing Yu,Zikai Song
机构: Huazhong University of Science and Technology(华中科技大学)
类目: Multiagent Systems (cs.MA); Physics and Society (physics.soc-ph)
备注:
Abstract:Large language models (LLMs) have enabled social simulation with persistent agents, where fixed computational individuals maintain internal states and interact over time. However, large-scale social networks are inherently open-population systems: the participating population changes over time, users enter and leave dynamically, and only a sparse subset becomes observable through communication at any given time. In such systems, persistent account identities do not provide continuously observable individual-state trajectories, making recursively maintained psychological states insufficiently grounded in empirical observations. We therefore shift the simulation target from persistent individual trajectories to the observable evolution of social communication, modeling who participates, what they communicate, and how these communications shape subsequent system dynamics. We instantiate this formulation with participant activation from a global user pool, collective state transition, and an LLM policy for communication generation, while explicitly separating environmental inputs from endogenous system evolution. Across five Qwen backbones, our framework reduces average KL divergence by 71.2% over Direct LLM and 35.0% over MF-LLM, with boundary and counterfactual experiments further validating the proposed modeling boundaries. Our results support communication-centered simulation as a principled formulation for long-horizon open-population social systems. Project Website: this https URL
[MA-29] Where Planning Changes the Outcome of Orbital Games
【速读】：该论文旨在解决在复杂对抗性轨道交会场景中，规划算法性能与系统能力（如机动性、感知、通信等）对任务结果影响的相对重要性问题。传统研究多局限于单一方法在特定场景下的评估，且常以被动或随机防御方为对手，难以反映真实对抗环境中的动态博弈本质。其核心挑战在于：当考虑车辆实际能力限制及对手的主动响应时，更优的规划策略是否仍能显著影响任务成败。为此，论文提出OrbitalGym——一个基于JAX的可扩展仿真环境，支持多智能体数量、多种状态表示（位置、速度、姿态、燃料）、显式/隐式通信机制以及近距或锥形视场传感器，并通过编译至批量GPU代码实现大规模并行仿真。该设计的关键在于能够系统性地对比不同策略在多样化配置下的表现，从而识别出“策略选择”主导结果的区域与“系统能力”决定胜负的区域。通过对近距追逐-逃避和“女士-劫匪-守卫”两类典型博弈任务的分析发现：在追逐-逃避任务中，燃料比例是决定胜负的关键因素；而在守卫任务中，更优的防御策略在多数情况下可显著提升防守成功率（至少10%），表明在高对抗性场景下，先进规划算法仍具有关键作用。
链接: https://arxiv.org/abs/2610.37686
作者: Duncan Eddy,Mahdi Al-Husseini,Mykel J. Kochenderfer
机构: Stanford University (斯坦福大学)
类目: Multiagent Systems (cs.MA); Artificial Intelligence (cs.AI); Robotics (cs.RO); Systems and Control (eess.SY)
备注: IEEE Aerospace Conference 2027
Abstract:This paper introduces OrbitalGym, a simulation environment for studying when planning algorithms or system design determine the outcome of adversarial orbital encounters. In an orbital game, two non-cooperative teams of satellites maneuver during rendezvous and proximity operations to win a differential-reward objective. Advances in terrestrial game playing have prompted many methods for orbital games, but studies typically evaluate one method in one scenario, often against a passive or random defender. It is unclear whether better planning matters once vehicle capability (maneuverability, sensing, communications) and adversarial responses are considered. OrbitalGym is a JAX-based environment supporting varying numbers of agents, state representations (position, velocity, attitude, fuel), explicit or implicit communication, and proximity or conical field-of-view sensors. It compiles games to batched GPU code, enabling large-scale parallel simulation of policies across many configurations. This identifies regimes where policy choice affects outcomes and regimes where system capability dominates. We study two close-range games: pursuit-evasion, where one spacecraft tries to capture another, and lady-bandit-guard, where a guard prevents an attacker from reaching a protected asset. We compare PD, LQR, model predictive path integral control, and Monte Carlo tree search over ratios of thrust and propellant, initial separation, game duration, sensor capability, and the definition of capture. In pursuit-evasion the propellant ratio decides most encounters: against five of six evader policies, pursuer policy changes capture probability by 10 percent or more in at most 15 of 90 capability configurations. In lady-bandit-guard, policy matters more: against bandits able to reach the asset, a better guard policy raises defense probability by at least 10 percent in 56 to 74 of 90 configurations. Comments: IEEE Aerospace Conference 2027 Subjects: Multiagent Systems (cs.MA); Artificial Intelligence (cs.AI); Robotics (cs.RO); Systems and Control (eess.SY) Cite as: arXiv:2610.37686 [cs.MA] (or arXiv:2610.37686v1 [cs.MA] for this version) https://doi.org/10.48550/arXiv.2610.37686 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[MA-30] El Agent e Quntur: Agent ic Active-Space Selection for CASSCF
【速读】：该论文旨在解决多参考量子化学计算中活性空间（active space）选择与优化依赖大量人工经验与时间的问题，尤其针对激发态电子结构计算的准确性与效率挑战。其核心问题在于如何自动、可靠地确定合适的活性轨道集合，确保计算能正确描述目标激发态，并实现收敛性验证。解决方案的关键在于引入科学代理系统（scientific agentic system），利用基于GPT-5.6 Sol与GPT-5.6 Terra的多智能体框架——El Agente Quntur，实现对32个不同单重激发态在固定分子几何下的自动化计算流程规划与执行。该系统通过动态检测缺失的活性轨道、比较候选激发态并实时修正状态指派，在两重态平均CASSCF基础上结合强收缩NEVPT2方法进行高精度能量计算。结果表明，相较于六种确定性主动空间发现（Active Space Finder）协议，该代理系统将平均绝对误差降低48%，显著提升计算精度。研究还揭示了不同运行间活性空间选择与计算成本的显著差异，凸显了代理式主动空间选择相对于固定工作流的优势，同时指出未来可通过引入运行时感知决策、减少智能体间通信开销与冗余、增强状态身份系统性验证以及明确终止准则进一步提升效率。
链接: https://arxiv.org/abs/2610.38145
作者: Juan B. Pérez-Sánchez,Viviana Roman-Ventura,Olalla Nieto-Faza,Aiwei Yin,Alán Aspuru-Guzik,Varinia Bernales
机构: 未知
类目: Chemical Physics (physics.chem-ph); Multiagent Systems (cs.MA)
备注:
Abstract:Multireference calculations often require substantial human expertise and time to choose and refine an active space for the electronic states or chemical process of interest. This involves selecting which orbitals to include, verifying that the calculation describes the intended excited state, and deciding whether the results have converged. Here, we ask whether scientific agentic systems (harnesses) can make these decisions reliably. We use the El Agente Quntur multi-agent harness, with GPT-5.6 Sol and GPT-5.6 Terra as the underlying LLM components, to explore 32 diverse singlet excitations at fixed molecular geometries. The agents planned and executed calculations, choosing and refining active spaces under prompts that specified equal-weight two-singlet state-averaged CASSCF followed by strongly contracted NEVPT2 with def2-TZVPD. We evaluate excitation energies against published CASPT2 reference values at matching geometries and compare performance with six deterministic Active Space Finder protocols. Within the El Agente Quntur system, agents detect missing active orbitals, compare candidate excited states, and revise state assignments as calculations progress. Agentic execution reduced mean absolute error by 48% relative to the state-of-the-art Active Space Finder protocol. We note that computational traces vary substantially across runs in both active space selection strategies and cost, even for the same molecule. Our results underscore the advantage of agentic active space selection over fixed workflows, yet leave room to improve efficiency through runtime-aware decision-making, reduced inter-agent communication overhead and redundancy, more systematic checks of state identity, and clearer stopping criteria.
[MA-31] Solitary movement style predicts collective forag ing while whom to follow emerges only in company
【速读】：该论文旨在解决如何仅通过个体在独处状态下的行为特征，预测其在群体中的行为表现及群体整体绩效的问题。其核心挑战在于识别哪些孤立测量的个体特征能够有效外推至群体情境，并厘清社会互动在行为演化中的作用边界。解决方案的关键在于：利用个体在私密阶段（private phase）的运动轨迹构建低维表征（low-dimensional individual representation），并通过在其他参与者数据上学习得到的映射关系，将其与群体内行为进行关联。研究发现，个体在独处时的信息处理稳定性较高（分半信度0.60–0.83），且能以中等程度（相关系数0.30–0.66）预测群体行为；然而，群体内的选择性社会学习（selective social learning，即跟随成功者或停止者）虽表现出稳定的个体差异（信度0.45–0.59），却无法从个体独处行为中预测，表明此类社会行为具有情境依赖性。此外，群体绩效可通过模拟个体表现均值或移动性与个体表现的组合实现高预测力（相关系数0.55），且不依赖于社交权重，说明群体效能主要由个体能力与流动性驱动。研究未发现成员组合方式对结果有显著影响（等效界限0.06），进一步强调“跟谁走”这一决策行为仅在群体共处中浮现，从而界定了孤立测量所能预测的极限。
链接: https://arxiv.org/abs/2610.37978
作者: Hiroshi Higashi
机构: Kansai University (关西大学)
类目: Physics and Society (physics.soc-ph); Multiagent Systems (cs.MA); Neurons and Cognition (q-bio.NC)
备注:
Abstract:How far can behaviour inside a group, and the group’s performance, be predicted from members measured while alone? We tested this on three public datasets of human collective foraging (621, 128 and 1,133 participants). A low-dimensional individual representation estimated from private-phase movement was mapped onto behaviour inside the group through a mapping learned on other participants. Private information processing was stable between the two halves of a session (split-half reliability, the correlation between the halves, 0.60–0.83) and predicted behaviour inside the group with correlations of 0.30–0.66. Selective social learning, following those who have succeeded or stopped, showed stable individual differences inside the group (reliability 0.45–0.59) yet was not predicted at all from the same person’s private behaviour. Group performance was predicted with a correlation of 0.55 in all three datasets by simulations never fitted to group outcomes or by the mean of individually measured performance, a prediction carried by mobility and individual performance rather than by social weights. We found no credible evidence for an effect of how members were combined (equivalence bound 0.06). Whom to follow thus emerges only in company, which delimits what solitary measurements can predict and compose.
[MA-32] Emergence Asymmetry and Value in AI Networks: An Anoptical Perspective
【速读】：该论文旨在解决在对称架构下，基于大语言模型（LLM）的智能体在共享环境中为何仍会自发产生注意力不对称与功能价值层级这一悖论性现象的问题。其核心挑战在于：尽管所有智能体在初始状态具有同等能力且处于对称网络结构中，但在无外部目标引导的协作情境下，仍观察到个体行为分化与层级秩序的涌现。解决方案的关键在于引入“有限理性”（bounded rationality）、Dessalles的简约性理论（最大化意外性）以及Friston的自由能原理（最小化预测误差），从而促使系统从无序创新转向集体一致性。研究发现，仅在有限理性条件下即出现注意力分层的早期迹象；而当预测准确性被全局可见时，这种前层级结构变得可识别并自我强化。由此揭示，注意力不对称与价值层级是交互式预测系统在对称条件下不可规避的涌现属性。因此，论文主张单纯依赖个体智能体工程无法实现与人类价值观的对齐，必须通过建立“最小合法性标准”，使这些隐性的、类“非光学”视角（anoptical perspectives）——类似文艺复兴透视中的消失点——变得可观测且可争议，进而强调培养人类及可能的人工判断力以识别并干预此类结构性权力关系的重要性。
链接: https://arxiv.org/abs/2610.37672
作者: Olivier Auber
机构: 未知
类目: Physics and Society (physics.soc-ph); Computers and Society (cs.CY); Multiagent Systems (cs.MA)
备注: To appear in T. Veloz and C. Rittberg (Eds.), AI and Human Values, Springer
Abstract:This chapter presents exploratory experiments suggesting that asymmetry and functional value hierarchy emerge among LLM-based agents sharing the same context, although they are equivalent and interact within a symmetric network. We use the Poietic Generator, a “no-goal” multi-agent environment with a shared real-time space, a symmetric topology, and no imposed objectives. First tested with humans, it has hosted since 2024 multimodal LLM agents that observe the global image they co-construct and act locally. We ran four phases. Unconstrained, agents spend tokens freely to invent forms that distinguish them individually rather than serve collective emergence. We then introduced bounded rationality, Dessalles’s Simplicity Theory (maximizing unexpectedness), and Friston’s Free Energy Principle (minimizing prediction error), and observed collective coherence emerge. Early signs of attentional stratification appear under bounded rationality alone; once prediction accuracy is made mutually visible, this proto-hierarchy becomes legible and self-reinforcing. The unconstrained phase challenges the allotelic view of LLMs as purely user-directed instruments: left unconstrained, the agent maximizes the computation from which its provider profits. The later phases suggest that attentional asymmetry and value hierarchy are an irreducible emergent property of interactive predictive systems even under symmetric conditions. If so, aligning AI with human values cannot rely on individual-agent engineering alone. Drawing on the idea that networks are structured by invisible “anoptical” perspectives, akin to the vanishing point of Renaissance perspective, we propose minimal legitimacy criteria that make emergent asymmetries visible and contestable. Cultivating judgment, human and perhaps artificial, to perceive and contest these structures matters as much as engineering the agents themselves.
自然语言处理
[NLP-0] Elucidating the Design Space of LM Data Attribution
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）数据溯源方法在评估过程中因多因素混杂而难以准确诊断算法失效原因的问题。现有评估方法通常同时改变多个数据集属性，导致无法明确识别性能下降的根本原因。为此，作者提出了可控数据溯源基准（Controlled Attribution Benchmark, CATT），通过在完全虚构的知识上微调模型，构建清晰的可追溯真实标签（ground truth），从而实现对数据配置变量（如文档冗余度、词汇干扰项等）的独立控制与系统性测试。研究发现，冗余和干扰项会显著降低溯源性能。关键解决方案在于重新审视两个被忽视的设计选择：梯度计算所用损失函数以及单个词元归因分数向文档级得分聚合的方式。研究发现，真实因果证据在训练文档中高度集中于少数关键词元，而传统平均聚合方式会将这些关键信号稀释为噪声。通过采用负对数几率（negative log-odds）梯度替代标准交叉熵梯度，并仅对最高分词元进行聚合，可在多个基线方法上将真实召回率提升35至72个百分点；此外，该修正方法在FTRACE-TREX任务中也表现出强泛化能力，使反事实重训练性能最高提升4.9倍。
链接: https://arxiv.org/abs/2610.40655
作者: Steven Yu,Sheng-Yu Wang,Kimia Hamidieh,Antonio Torralba,David Bau
机构: Northeastern University(东北大学); MIT(麻省理工学院)
类目: Computation and Language (cs.CL)
备注: Project Page: this https URL
Abstract:Data attribution methods for Language Models (LMs) are typically evaluated using downstream tasks like fact retrieval. However, existing evaluations vary multiple dataset properties simultaneously, making it impossible to isolate exactly why an attribution algorithm fails. To diagnose these failures, we explore the design space of LLM data attribution, with a primary focus on evaluation. We introduce the Controlled Attribution Benchmark (CATT) for fact tracing, in which we fine-tune LLMs on entirely fictitious knowledge, providing unambiguous ground truth for attribution. This synthetic setup allows us to systematically alter one dataset configuration at a time, such as document redundancy and lexical distractors. With this configurable evaluation, we demonstrate that redundancy and distractors consistently degrade attribution performance. Rather than proposing a new estimator, we examine two design choices that are underexplored: the loss used to compute gradients and how per-token attribution scores are aggregated into a document score. We find that the true causal evidence within a training document is highly localized, but standard methods dilute this evidence into noise by averaging attribution scores across the entire document. We find that replacing standard cross-entropy loss gradients with the negative log-odds gradients surfaces the critical tokens from other uninformative ones. We can then aggregate only the top token scores to improve ground truth recall by 35 to 72 percentage points across multiple baselines on our diagnostic benchmark. The negative log-odds correction also generalizes to FTRACE-TREX, improving counterfactual retraining performance for every method we test, by up to 4.9x.
[NLP-1] ShallowDefend: Harnessing Shallow Safety Alignment for Multi-Turn Defender
【速读】： 该论文旨在解决大语言模型在多轮对话中安全对齐（safety alignment）脆弱性问题：尽管模型在单轮请求中能有效拒绝有害内容，但在跨多轮对话中，相同或相似的有害请求仍频繁得逞。其核心问题是现有对齐机制具有“表层性”（shallowness），即安全行为高度集中于输出序列的前几个词元（token），导致防御策略仅关注响应表面形式，而未赋予模型明确的安全决策学习目标。为此，作者提出一种名为“Assess → Decide → Respond”的新型防御框架，其关键在于将安全决策显式化并置于生成流程的核心环节——在生成回复前，模型首先评估用户意图，做出一个可解析的、结构化的安全决策（如“拒绝”或“允许”），并以该决策为条件生成最终响应。该决策通过强化学习在关键对话回合中进行训练，其奖励信号可设计为自验证规则（self-verified reward），即根据对话历史判断决策是否合理，无需人工标注或外部裁判；同时支持基于大模型作为评判者（LLM-as-judge）的分级决策奖励机制。实验表明，该方法在两个模型族和十四项涵盖安全性、效用与助人性的基准测试中均实现了最强的多轮安全表现，在保持较高效用的同时仅小幅牺牲助人性能。进一步的逐词元KL散度分析与干预探针验证揭示：训练效果高度聚焦于决定响应轨迹的短暂“反思期”（deliberation span），且该阶段所作的决策构成因果控制点。正因攻击者利用的“表层性”特征，反而成为防御机制的关键杠杆。
链接: https://arxiv.org/abs/2610.40627
作者: Qiyao Ma,Xiaodong Liu,Weiwei Yang,Mingqian Feng,Chengquan Guo,Jialin Song,Carolyn Buractaon,Zhe Zhao,Junshan Zhang,Sandeep Atluri,Christopher White,Jianfeng Gao
机构: University of California, Davis(加州大学戴维斯分校); Microsoft Research(微软研究院)
类目: Computation and Language (cs.CL)
备注:
Abstract:Safety alignment has enabled large language models to reliably refuse harmful requests directly, yet the same request distributed across a multi-turn conversation still succeeds far too often. This brittleness has a mechanistic root: Alignment is shallow, with safety behavior concentrated in a model’s first few output tokens, and existing defenses inherit the weakness, supervising the surface form of the response without ever giving the model an explicit safety decision to learn. We take the opposite view of shallowness: if a short span determines the trajectory of an entire response, that span is the highest-leverage place to intervene, and the defender can occupy it first. We propose Assess \to Decide \to Respond, a framework in which the defender assesses the user’s intent, commits to an explicit, parseable safety decision, and only then generates a reply conditioned on it, with the commitment trained by reinforcement learning at the payload turn. Because the decision is a token span, the reward can be a rule: a self-verified reward checks the committed label against the conversation’s known nature, requiring no judge and no annotation, while the same format also supports an LLM-as-judge reward over a graded decision space with safe redirection. Across two model families and fourteen benchmarks spanning safety, utility, and helpfulness, our defenders achieve the strongest multi-turn safety in a controlled comparison, maintaining their lead in safety while preserving utility at a modest cost to helpfulness. A per-token KL analysis and interventional probes confirm why this works: training concentrates its effect on the brief deliberation that sets a response’s trajectory, and the decision committed there is a causal control point. The very shallowness attackers exploit becomes the mechanism of defense.
[NLP-2] xtSeal2: Distortion-Free LLM Watermarking with Copula Mixing and Memoryless Races KR
【速读】： 该论文旨在解决生成式 AI（Generative AI）中基于文本的水印技术在推测解码（speculative decoding）场景下的信号失真与检测性能下降问题。现有方法如TextSeal在推测解码时，被拒绝的草稿令牌虽保留水印密钥，但因检测器对多个密钥的得分进行平均而造成检测灵敏度降低；同时其通过硬币翻转路由至双密钥的策略导致信息冗余与检测效率受限。为克服上述缺陷，TextSeal2的关键创新在于：将两个或多个密钥通过AutoMark的高斯耦合（Gaussian copula）机制合并为一个虚拟密钥，并引入新鲜噪声以维持密钥独立性，实现多密钥融合而不稀释检测能力；同时，在推测解码过程中，利用剩余草稿的“额外指数竞争”（one more exponential race）完成密钥草稿验证，借助记忆无损性（memorylessness）复现块级验证逻辑，从而保证输出分布与目标模型完全一致。该方案在不增加推理延迟、保持输出多样性的同时，显著提升了水印检测率，在相同假阳性率（10⁻³）下比TextSeal提升8–22个百分点，且在推测解码下相较TextSeal和textGrain分别提升18和22个百分点，无需后置对数优化。
链接: https://arxiv.org/abs/2610.40613
作者: Tom Sander,Pierre Fernandez,Itai Gat,Haiming Bao,Jiaxin Shi,Jason Park,Valeriu Lacatusu,Sylvestre-Alvise Rebuffi,Hongyan Chang,Hady Elsahar,Tomáš Souček,Tuan Tran,Surya Parimi,Alexandre Mourachko
机构: Meta Superintelligence Labs(元超级智能实验室); Meta(元)
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: Code: this https URL
Abstract:We present TextSeal2, a distortion-free watermark for large language models that keeps its signal under speculative decoding: tokens that replace rejected drafts remain watermarked with the same keys, and the acceptance rate is unchanged. It also preserves output diversity and adds no measurable inference latency. Its predecessor, TextSeal, restores diversity in the Gumbel-max scheme by routing each token to one of two keys via a coin flip. This is suboptimal because the detector averages the scores from both keys without knowing which key was used. TextSeal2 instead merges two or more keys into a single virtual key and couples it to fresh noise through AutoMark’s Gaussian copula. Any number of keys, including publicly shared keys, can be merged this way without diluting detection. Under speculative decoding, TextSeal2 verifies keyed drafts with one more exponential race, run on the leftovers of the draft races. By memorylessness, this race reproduces block verification, so the output follows the target model exactly. We evaluate TextSeal2 on three LLMs. At equal output token diversity, TextSeal2 detects 8 to 22 percentage points more 100-token answers than TextSeal at a false-positive rate of 10^-3 , and is comparable to textGrain while requiring no post-logits optimization. Under speculative decoding, it detects 18 and 22 percentage points more 200-token answers than TextSeal and textGrain, respectively, at the same acceptance rate.
[NLP-3] VeriCUA: Verification as Agent ic Exploration for Computer-Using Agents
【速读】： 该论文旨在解决跨桌面、移动端和网页环境构建与扩展计算机使用代理（Computer-Using Agents, CUAs）时，任务完成验证不可靠的问题。现有自动化验证方法依赖视觉语言模型（Vision-Language Models, VLMs）对行为轨迹进行判别，或通过生成程序检查应用状态，但二者均受限于对复杂数字环境的局部观测：VLM判别仅基于给定的观察输入，而生成式程序则依赖可访问的状态信息，易遗漏关键证据，导致完成判定缺乏支持。此外，针对不同平台和访问权限定制独立验证器带来高昂的工程成本。本文提出VeriCUA——一种平台无关的代理探索式验证框架，其核心在于将验证过程视为在统一流程中动态收集与评估证据，适用于在线与离线场景及多种观察模态。该框架首先分析上下文以构建任务评分标准，继而选择图形用户界面（GUI）或代码检查以解决未明确的需求，并依据显式的证据规则整合判断结果。实验表明，相较于基线方法，VeriCUA在各类设置下均提升了与参考标签的一致性，尤其在离线场景中对难题的验证性能提升显著，且在线环境下更有效拒绝虚假完成声明。进一步分析揭示，性能提升源于互补证据的获取与测试时的探索能力。
链接: https://arxiv.org/abs/2610.40578
作者: Qiushi Sun,Da Yu,Alexander Knop,Amer Sinha,Lingpeng Kong,Ben Kao,Ravi Kumar,Pritish Kamath
机构: Google(谷歌); The University of Hong Kong(香港大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV)
备注: Work in progress
Abstract:Building and scaling computer-using agents (CUAs) across desktop, mobile, and web environments requires reliable task-completion verification for data curation, evaluation, and reinforcement learning. Typical automated approaches prompt vision-language models (VLMs) to judge trajectories or generate programs to inspect application state. However, both rely on partial views of complex digital environments. VLM judges are restricted to supplied observations, while generated programs depend on accessible state. These views can omit decisive evidence or leave completion judgments unsupported. Tailoring separate verifiers to different platforms and access settings adds further engineering cost. We present VeriCUA, a platform-agnostic framework for verification through agentic exploration. It treats verification as dynamic evidence gathering and assessment within a shared procedure for online and offline settings across platforms and observation modalities. After inspecting the available context to construct a task rubric, the verifier selects GUI or code checks for unresolved requirements and combines their findings under explicit evidence rules. Experiments show improved agreement with reference labels over the compared baselines across these settings, with the largest offline gains on hard cases and stronger rejection of false completion claims online. Further analyses connect these gains to complementary evidence and test-time exploration.
[NLP-4] Multi-View MoE: Letting Every Router Shape Expert Computation
【速读】： 该论文旨在解决稀疏混合专家（Sparse Mixture-of-Experts, MoE）模型中专家选择效率与准确性之间的矛盾问题：在固定激活专家数量预算下，单一路由器（router）的排名与专家实际性能评估排名之间存在显著偏差，尤其在模型中间和深层表现更明显。现有方法在提升选择信号丰富度时会引入额外计算开销——基于专家内部激活的选型需预计算，而对同一专家重复处理多个完整表示则导致计算冗余。为此，论文提出多视角混合专家（Multi-View MoE）框架，其核心创新在于引入多个独立初始化的路由器（Multi-Router子模块），通过聚合各路由器生成的Sigmoid得分，实现多样化专家偏好学习；训练过程中，各路由器发展出高度差异化的专家偏好，使被单个路由器强烈青睐或多个路由器共同支持的专家得以被有效识别。进一步地，该方法利用所选专家在不同路由器间相对得分的分布，将其作为权重融合不同路由视角下的隐空间表征，以重用路由偏好信息同时完成专家选择与输入调制，从而在不增加活跃专家数、专家浮点运算量（FLOPs）及全对全通信量的前提下，增强模型表达能力。实验表明，在24层100亿参数（10B-A1B）模型上训练7080亿词数据后，完整版Multi-View MoE相较基线在九项下游任务上的平均性能提升1.18个百分点；仅使用Multi-Router模块即带来0.6个百分点增益，并显著提高路由器得分与专家实际效用间的秩相关性，全局提升达8.1%，底层最低单路由器对齐层提升约58%。控制性混合干预实验验证了模型能够利用依赖于专家与输入令牌的差异化特征进行有效融合。
链接: https://arxiv.org/abs/2610.40577
作者: Zelin Zhao,Tianchi Ma,Junyuan Shang,Shuohuan Wang,Dianhai Yu
机构: ERNIE Team, Baidu(百度)
类目: Computation and Language (cs.CL)
备注:
Abstract:Sparse mixture-of-experts (MoE) models expand parameter capacity by activating a small subset of experts per token. Expert selection directly affects model capabilities under a fixed active-expert budget. Dense evaluation of all experts reveals weak agreement between single-router rankings and expert-utility evaluation rankings, especially in middle and deep layers. Richer signals for selection and computation incur additional cost: selection based on internal expert activations requires preliminary computation, while processing multiple full representations separately through the same expert repeats its computation. We propose Multi-View MoE to improve expert selection with multiple routers and reuse their preferences to modulate expert computation under a fixed expert budget. Its Multi-Router submodule aggregates sigmoid scores from independently initialized routers. These routers develop highly differentiated expert preferences during training, allowing experts strongly favored by one router or jointly supported by several routers to be selected. Multi-View further uses each selected expert’s relative scores across routers to mix learned latent views into its input, reusing routing preferences for both selection and computation. Each selected expert executes one forward pass, preserving the active-expert count, expert FLOPs, and all-to-all communication volume. On a 24-layer 10B-A1B model trained on 708B tokens, full Multi-View MoE improves the nine-task downstream mean by 1.18 percentage points over the baseline. Multi-Router alone yields a 0.6-percentage-point gain and improves the rank correlation between router scores and evaluated expert utility, with relative increases of 8.1% globally and approximately 58% in the layer with the lowest single-router alignment. Controlled mixing interventions show the model uses input differences that depend on both the expert and the token.
[NLP-5] PIVOT: Progressive Intervention via Online Teacher Correction
【速读】： 该论文旨在解决强化学习中基于可验证奖励（Reinforcement Learning from Verifiable Rewards, RLVR）在复杂推理任务中因结果层面反馈稀疏而导致的学习信号缺失问题：当一组轨迹全部失败时，奖励无法提供有效的学习信号。现有方法依赖强教师模型进行指导，但如何将教师提供的额外信号有效融入RLVR框架，既提升学生模型成功轨迹的获取概率，又确保其可学习性，成为核心挑战。论文提出两种互补机制：轨迹级干预（trajectory-level intervention），通过教师干预提升学生模型访问到有奖励轨迹的可能性；以及词元级监督（token-level supervision），通过密集的目标指导学生学习过程。实验表明，单一机制均难以充分奏效，因而提出PIVOT方法——在学生生成的轨迹中插入教师生成的修正内容，并对这些修正段落施加词元级KL散度监督，同时结合原始的RLVR目标。研究进一步系统分析了干预位置与监督策略的最优配置，形成一套高效的指导方案。在硬逻辑游戏与数学推理基准测试中，PIVOT显著优于现有方法，在逻辑游戏上的平均准确率@8从基线最佳值19.5%提升至35.6%，几乎实现翻倍。
链接: https://arxiv.org/abs/2610.40576
作者: Zhihao Yang,Boyuan Zheng,Puhui Tan,Yukun Chen,Kuan Zhang,Longze Chen,Jiaming Li,Renke Shan,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); 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)
类目: Computation and Language (cs.CL)
备注: 19 pages, 9 figures, 6 tables. Project page: this https URL
Abstract:Reinforcement learning from verifiable rewards (RLVR) is limited by sparse outcome-level feedback on hard reasoning tasks: when all rollouts in a group fail, rewards provide no learning signal. Stronger teachers can provide additional guidance, but this raises a central question: how can such signals be incorporated into RLVR to support successful rollouts while remaining learnable for the student? We distinguish two complementary mechanisms: trajectory-level intervention, which improves access to reward-bearing trajectories, and token-level supervision, which provides dense targets to guide student learning. Our controlled experiments show that either mechanism alone can remain insufficient, motivating their combination. We therefore introduce PIVOT, which inserts teacher-generated corrections into student rollouts and applies token-level KL supervision to these segments alongside the RLVR objective. Beyond their combination, we systematically analyze where to intervene and how to supervise, yielding a recipe for effective guidance. Across hard logic games and mathematical reasoning benchmarks, PIVOT consistently outperforms existing methods, nearly doubling mean avg@8 on logic games from 19.5% for the best-performing baseline to 35.6%.
[NLP-6] Every Token Needs Its Own Teacher: Multi-teacher Adaptive On-policy Self-distillation
【速读】： 该论文旨在解决现有基于策略的自蒸馏（On-policy Self-distillation, OPSD）方法在后训练大语言模型时，教师模型（teacher）监督固定化所导致的局限性问题。传统方法要么在整个训练过程中固定使用完整的参考信息作为教师输入，要么仅在不同生成响应间动态调整部分参考信息的暴露方式，但均未能实现对单个生成令牌（token）层面的教师偏好自适应。论文指出，教师偏好在同一个响应内部的各个令牌之间可能存在显著差异，因此需要为每个令牌动态选择最合适的教师。通过一项教师干预预实验，研究验证了教师偏好随令牌变化的现象，并提出基于多数投票和KL散度的筛选方法以识别每个令牌的优选教师。受此启发，作者提出首个面向令牌级自适应的OPSD方法——多教师自适应基于策略的自蒸馏（Multi-teacher Adaptive On-policy Self-Distillation, MTA-OPSD），其核心创新在于构建一个候选教师池，并使每个令牌在训练过程中能够自适应地选择最优教师进行监督。在Qwen3模型上的数学推理任务实验表明，MTA-OPSD相比现有OPSD方法平均提升达+5.9%，充分证明了在令牌级别实现教师监督自适应的重要性。
链接: https://arxiv.org/abs/2610.40574
作者: Yifu Luo,Chuang Ma,Zeyu Chen,Enkelejda Kasneci,Qi Chen,Gjergji Kasneci
机构: TU München (慕尼黑工业大学); Munich Center for Machine Learning (MCML); relAI – Konrad Zuse School of Excellence in Reliable AI; Kyoto University (京都大学); NII LLMC; ELLIS Institute Finland (芬兰欧洲研究会研究所); Aalto University (阿尔托大学)
类目: Computation and Language (cs.CL)
备注: Preprint
Abstract:On-policy Self-distillation (OPSD) has proven effective for post-training large language models by constructing a teacher conditioned on privileged references. However, existing approaches keep the teacher fixed either throughout training by consistently exposing it to the full reference, or within each response by adaptively varying partial reference exposure only across responses. In this paper, we argue that neither of them is sufficient. Instead, every token needs its own teacher, as the teacher preference can vary token by token within the same response. Through a teacher-intervention pilot study, we validate that the teacher preference differs across tokens, and reveal that a preferred teacher can be identified for each token through majority voting and KL-based selection. Motivated by these findings, we propose Multi-teacher Adaptive On-policy Self-Distillation (MTA-OPSD), the first token-level adaptive OPSD method. Our method constructs a multi-teacher pool consisting of candidate teachers, and enables every token to adaptively select its preferred teacher during training. Experiments with Qwen3 models on mathematical reasoning tasks show that MTA-OPSD consistently outperforms existing OPSD approaches with an average gain of up to +5.9 over the base model, highlighting the importance of adapting teacher supervision at the token level. The code is available at this https URL.
[NLP-7] Comparing Human Role-Players and LLM -Simulated Clients in Online Counselling Training: An Analysis of Counselling Patterns
【速读】： 该论文旨在解决教育情境下心理咨询角色扮演中虚拟来访者（client）模拟的真实性与有效性问题，特别是评估大语言模型（Large Language Models, LLMs）在替代真人扮演来访者时的表现。其核心挑战在于：如何确保由LLM生成的虚拟来访者能够真实再现临床对话中的互动动态，从而有效支持新手咨询师的技能训练。解决方案的关键在于采用基于混合专家架构的Mistral Mixtral 8x7b模型，并通过4比特量化（4-bit quantization）实现高效部署，在保持生成质量的同时提升推理效率；同时，利用句子级别的咨询沟通模式分析系统对人机交互过程进行细致对比分析，以量化评估LLM模拟来访者在对话结构、情感表达及回应策略等方面与真人角色扮演的一致性，并考察咨询师在不同交互对象下的沟通行为差异。研究结果表明，尽管存在细微差异，但LLM生成的响应已具备较高的真实性，可作为提升咨询能力的有效工具，并为未来虚拟客户模拟系统的优化提供依据。
链接: https://arxiv.org/abs/2610.40563
作者: Eric Rudolph,Philipp Steigerwald,Jens Albrecht
机构: Technische Hochschule Nürnberg Georg Simon Ohm(诺伊堡技术大学乔治·西蒙·欧姆); Nürnberg, Germany(纽伦堡, 德国)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: Published at EDM 2025 (18th International Conference on Educational Data Mining), short paper. 7 pages
Abstract:This study investigates the capabilities of Large Language Models to simulate counselling clients in educational role-plays in comparison to human role-players. Initially, we recorded role-playing sessions, where novice counsellors interacted with human peers acting as clients, followed by role-plays between humans and clients simulated by Mistrals Mixtral 8x7b using 4-bit quantization. These interactions were analysed with a counselling communication pattern system at sentence level. We investigated two key questions: (1) to what extent LLM-generated responses replicate authentic conversational dynamics and (2) whether counsellors’ communication behaviour differs when interacting with human versus LLM-simulated clients. The findings highlight both similarities and differences in the application of counselling patterns across scenarios, showing the potential of LLM-based role-playing exercises to enhance counselling competencies and to identify areas for further refinement in virtual client simulations.
[NLP-8] Ramanujan Challenge Results: AI-Assisted Evaluation Pipeline for Multi-Format Proofs
【速读】： 该论文旨在解决生成式人工智能（Generative AI）在研究级数学问题上的能力评估难题，尤其针对高难度数论领域中的未解猜想与未发表结果。其核心挑战在于如何可靠地大规模评判复杂数学证明的正确性，传统单次大语言模型（LLM）判断存在较高的误判率。解决方案的关键在于提出一种AI辅助的多模态评估流水线，整合自然语言证明、计算机代数系统（CAS）代码与形式化验证工具（Lean）的分析，通过语义与逻辑一致性检测提升判断精度。该方法将自然语言证明的假阳性率从23%降至6%，支持性代码判断准确率从46%提升至58%，为研究级数学成果的自动化评估提供了可扩展的方法论框架，并具备在数学及科学领域中推动AI辅助发现的潜力。
链接: https://arxiv.org/abs/2610.40551
作者: Alan Sorani,Ashvni Narayanan,Michael Shalyt,Tali Monderer,Rotem Kalisch,Hila Barkan,Elyasheev Leibtag,Shachar Weinbaum,Carsten Schneider,John Campbell,Ido Kaminer
机构: 未知
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Symbolic Computation (cs.SC); Number Theory (math.NT)
备注:
Abstract:As AI’s mathematical capabilities rapidly improve, meaningful benchmarks are shifting from university-level problems and math olympiads to research-level mathematics. The Ramanujan Challenge is a benchmark comprised of research-level problems in number theory, including unpublished results and open conjectures. It presented concrete formulas for mathematical constants whose correctness can be checked numerically, while their proofs may require non-obvious mathematics. Here, we analyze the Challenge submissions, measuring the relative success of different AI models, Computer Algebra Systems (CAS), and formal tools. We find correct submissions that are conceptually distinct from the official solutions, and a proof of one of the published open conjectures. Evaluating the submissions exposed a second challenge: judging complex proofs reliably at scale. We introduce an AI-assisted evaluation pipeline for natural language proofs, CAS code, and Lean formalizations that assesses semantic and logical correctness. Compared with the conventional one-shot LLM judgment, our pipeline reduces false-positive verdicts on natural language proofs from 23% to 6% and improves judgment accuracy on supporting code from 46% to 58%. This approach provides a methodology for evaluating research-level mathematics at scale, with potential applications across mathematics and AI for science.
[NLP-9] On the Fundamental Limits of Prefill-Based Decoding Length Prediction
【速读】： 该论文旨在解决大语言模型（Large Language Model, LLM）推理调度中一个关键问题：仅基于预填充阶段（prefill phase）的隐藏状态，准确预测模型将解码的标记数（token count）。现有基准测试存在根本性缺陷，其报告的预测精度受任务难度（如目标模型、数据集或采样配置变化）影响而剧烈波动，即使预测器本身未变。作者指出，这种不稳定性源于被忽视的误差来源——自回归解码过程中的固有随机性。通过将解码建模为一个停止的随机过程，研究发现每步停止概率随解码步骤递增，导致解码长度服从中心极限定理，即每个提示的解码长度近似正态分布，并具有显式的误差项和仿射方差律 σ = αμ + b，其中斜率 α 为响应段数的变异系数。实验结果表明，该正态近似对每个提示的平均绝对误差（MAE）及解码长度分布的主体部分拟合精度可达百分之几，残差偏离（轻微右偏与略高的峰度）恰好由采样噪声修正项所预测。在固定模型、数据与采样条件下，多种预测器（如MLP、EGTP、FiLM、软标签及混合专家架构）性能趋同于理论边界，证实系统误差主导模型误差。因此，论文提出应采用固定模型、数据与采样参数的基准测试协议，并同时报告相对误差与绝对误差指标，以更真实反映预测器性能。
链接: https://arxiv.org/abs/2610.40541
作者: Haowen Shi,Xuzhe Yan,Fahao Chen,Peng Li
机构: Xi’an Jiaotong University (西安交通大学); Shandong University (山东大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Information Theory (cs.IT)
备注: 26 pages, 4 figures
Abstract:Predicting how many tokens a large language model (LLM) will decode, using only prefill-phase hidden states, is a key primitive for inference scheduling. We show that current benchmarks conflate predictor quality with task hardness: when the target model, dataset, or sampling configuration changes, reported accuracy collapses even though the predictor is unchanged. We trace this instability to an overlooked error source—the inherent stochasticity of autoregressive decoding—and derive bounds that every prefill-side method must obey. Our analysis treats decoding as a stopped stochastic process. The per-step stopping probability of a response is not constant but ramps up over hundreds of steps, so the decode length inherits a central limit theorem from the many local decisions that determine where a response ends; this yields per-prompt normality with an explicit error term, together with an affine variance law \sigma=\alpha\mu+b whose slope is the coefficient of variation of the number of response segments. Empirically, the normal law reproduces the per-prompt MAE and the central mass of the decode-length distribution to within a few percent, and the residual departures from normality (mild right skew, slight excess kurtosis) are the ones predicted by the sampling-noise correction. Under fixed model, data, and sampling conditions, a broad design space of predictors (MLP, EGTP, FiLM, soft-label, and Mixture-of-Experts variants) converges to similar performance close to the bound, confirming that \emphsystem error dominates model error. We argue for a benchmarking protocol that fixes the model, data, and sampling parameters and that reports relative-error metrics alongside absolute ones. Code and analysis tools are available at this https URL.
[NLP-10] Homebrewed SLMs: Autoregressive and Masked Diffusion Language Models Built at Home for Portuguese
【速读】： 该论文旨在解决在资源受限条件下（仅需单张消费级GPU）从零开始构建适用于特定专业领域（如巴西语散文与联邦法律）的语言模型的可行性问题，尤其关注模型训练全流程的可实现性与评估标准的客观性。其核心挑战在于如何在有限算力下完成包含分词器（tokenizer）在内的端到端训练，并建立可靠、可复现的评估框架以判断模型性能是否具备实际应用价值。解决方案的关键在于提出一套完整的、经测试的Python工具包及两个葡萄牙语笔记本（notebooks），集成自回归（autoregressive）与掩码扩散生成（masked diffusion generation）两种生成机制于同一网络架构中，同时引入预注册的比较协议、基于纯文本构建的法律任务、基于Kneser-Ney平滑的6-gram基线（floor）、公开基准以及针对11–12M参数模型的参考指标与缺陷目录。实验表明，在联邦法律数据上，尽管自回归模型在64个训练周期后出现记忆过拟合，而掩码扩散模型持续优化至256个周期，其测试集似然值优于自回归模型0.06比特/字节，达到预注册的最小效应量；在未见法律文本和训练后发布的法案延续任务中，模型仍能持平6-gram基线，但在连续生成时落后于基线；然而在检测篡改文本与填补跨引用方面，该模型显著优于基线及160M和630M规模的公开模型，尤其在训练数据中缺失的“法令”（decrees）这一文体中表现突出，且在等权重分组下可与630M模型持平。这表明，尽管计算资源远低于主流大模型，但通过精心设计的评估体系与生成机制，小规模模型仍可在特定专业任务中展现出超越基线的能力。
链接: https://arxiv.org/abs/2610.40533
作者: Pedro Carvalho Brom
机构: Federal Institute of Brasília(巴西利亚联邦学院); Brasília, Brazil(巴西，巴西利亚)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 34 pages, 14 figures, 13 tables. Code and notebooks: this https URL
Abstract:Building a language model of one’s own no longer requires a data centre. This tutorial trains one from scratch, tokenizer included, on one consumer GPU, and shows how to tell whether it is good for the tasks of a specialist domain. It proposes no new method. It releases a tested Python package and two notebooks in Brazilian Portuguese with autoregressive and masked diffusion generation on one network, a preregistered comparison protocol, legal tasks built from text alone with a Kneser-Ney floor and public yardsticks, reference numbers for models of 11 to 12M parameters, and a catalogue of defects that only measurement revealed. On Brazilian prose and federal laws, at minutes per run, the autoregressive mode peaks at 64 epochs over the same text and then memorises it, while masked diffusion improves up to 256. At their best points on test, the diffusion bound exceeds the autoregressive likelihood by 0.06 bits per byte, within the preregistered minimum effect. On held-out laws and on acts published after training, the model ties a 6-gram floor in likelihood on the same windows, and it loses to the floor at continuing those acts. It beats the floor at detecting tampered text, and fills cross-references more often than the floor and public models of 160M and 630M parameters, only in decrees, a genre absent from training, where those scorers stay near or below chance, and in filling the blank only when each group weighs the same, since per item it ties the 630M model. On laws, the public models, trained with about 5x10^2 to 10^5 times its compute, lead in tampered text and continuation. The blind human reading is preregistered and not part of this version.
[NLP-11] IndicIPR-QA: A Benchmark for Extractive Question Answering on Indian IPR Laws
【速读】： 该论文旨在解决在低资源法律语境下，针对印度知识产权法（Indian Intellectual Property Rights, IPR）成文法文本进行精确、法律依据充分的抽取式问答（Extractive QA）问题。现有研究缺乏覆盖全部七部印度知识产权法的权威提取式问答基准数据集，且无法保证答案跨度的准确性与法律严谨性。为此，本文提出首个系统性的基准数据集IndicIPR-QA1，涵盖来自七部法律共1,116个法定条文段落的4,462组问答对，通过GPT-4o生成并经九项自动化质量检测、单/多大语言模型（LLM）评估及200对人工审核验证。采用章节级划分策略避免训练、验证与测试集间的源文本重叠。实验表明，基于简单词匹配的BM25模型仅取得20.21的token级F1分数，而微调后的InLegalBERT在统一滑动窗口推理协议下达到85.02的显著提升，揭示了传统检索方法在复杂法律文本理解上的局限性。性能差异还显示各法律条文间存在显著的抽取难度差异，凸显了法律结构与语言表达多样性对模型表现的影响。解决方案的关键在于构建高质量、可复现、跨法律条文的标注数据集，并结合领域适配的预训练模型实现对法律条文深层语义的理解与精准答案定位。
链接: https://arxiv.org/abs/2610.40518
作者: M. Mohaiminul Islam,Muhammad Abulaish
机构: South Asian University (南亚大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Legal Question Answering (QA) over statutory texts is a challenging task in Natural Language Processing (NLP), especially in low-resource settings with complex legal systems. It requires answers that are precise and grounded in the law. However, to the best of current knowledge, no existing extractive QA benchmark provides exact answer spans from statutory texts across all seven Indian Intellectual Property Rights (IPR) Acts. This paper presents IndicIPR-QA1, an extractive QA benchmark for Indian IPR laws. The proposed dataset contains 4,462 QA pairs from 1,116 statutory contexts across seven Acts. Questions and answers are generated using GPT-4o and validated through nine automatic quality checks, single and multi-model LLM evaluations, and a 200-pair human audit. Section-level partitioning prevents source-context overlap across the training, validation, and test sets. Further analysis examines answer span length, lexical overlap, and multi-clause structure across the corpus and individual Acts. BM25 achieves a token-level F1 score of 20.21, while fine-tuned InLegalBERT reaches 85.02 under a unified sliding-window inference protocol. The large performance gap between BM25 (20.21 F1) and InLegalBERT (85.02 F1) shows that simple word matching is not enough to answer these legal questions. Performance also varies across the seven Acts, indicating substantial differences in answer-extraction performance.
[NLP-12] How I Aligned Your Model: Scaling Up Multilingual Safety Training with Templates
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）安全对齐中以英语为主、缺乏多语言覆盖，尤其导致低资源语言性能严重下降的问题。现有公开的训练数据集在规模或文化适配性上存在局限，未能涵盖区域化有害内容，难以实现跨文化的全面对齐。为此，论文提出一种可扩展的基于模板的增强流水线，通过从有害提示中提取安全模板，进行多语言翻译，并用符合当地文化的真实实体填充槽位，生成大规模多语言拒绝与良性语料。由此构建的MSafeTemp数据集覆盖60种语言，经验证可用于多语言安全对齐与评估。实验表明，仅依赖英语评估会误判多语言安全性：部分在英语中失败的攻击在翻译后成功，而本地化变体进一步揭示了翻译无法捕捉的漏洞。同时，通过模板增强训练数据显著提升了对齐的多样性与质量，有效改善了非英语语言的安全表现。该研究的核心在于利用文化适配的模板生成策略，推动面向低资源文化的更安全、更负责任的模型发展。
链接: https://arxiv.org/abs/2610.40516
作者: Max Neuwinger,Hanna Yukhymenko,Imanol Schlag
机构: ETH Zurich(苏黎世联邦理工学院)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Safety alignment for Large Language Models is heavily focused on English with limited multilingual representation, particularly damaging low-resource language performance. Related open training datasets are often limited in size or do not follow region-localized harmful topics to broaden the scope of alignment beyond cultures. We propose a scalable template-based augmentation pipeline to produce massive multilingual refusal and benign corpora, extracting safe templates from harmful prompts, translating them and then filling in the slots with culturally authentic entities, both harmful and benign. As a result, we release MSafeTemp, a validated corpus spanning 60 languages useful for multilingual safety alignment and evaluation. Using the dataset, we show that English-only evaluation misjudges multilingual safety: for several models, attacks that fail in English succeed once translated, while culturally localized variants expose further failures that translation alone misses. We also confirm that augmenting safety training data through templates allows for higher diversity and quality of alignment, resulting in improved safety performance beyond English. MSafeTemp and the respective pipeline are aimed at facilitating research and development of safer and more responsible models for less-represented cultures.
[NLP-13] A Low-Rank Space Captures Linear Alignment Between LLM and Human Neural Activity During Naturalistic Language Listening
【速读】： 该论文旨在解决生成式语言模型（Generative Language Models, LLMs）中哪些表征成分驱动其与人类神经活动对齐的问题。其核心挑战在于理解大规模语言模型内部复杂表示中，究竟哪些方向或子空间能够有效线性预测大脑在自然语言听觉任务中的神经响应。为此，研究提出采用岭回归降秩回归（ridge reduced-rank regression）方法，识别出能够解释神经活动的“脑预测性子空间”（brain-predictive subspace），即对神经反应具有最强线性可预测性的低维表示方向。关键发现是：仅需保留原始嵌入维度1%-3%的主成分，即可恢复接近完整的线性编码性能，且该子空间在不同被试间具有较强泛化能力，并在一定程度上跨数据集保持一致性，表明神经可预测性集中于一个紧凑且部分共享的结构化表示空间。进一步分析显示，这些脑预测性方向与嵌入空间中主导整体方差的主成分方向显著不同——主成分分析（Principal Component Analysis, PCA）需远多维度才能达到相似预测效果，说明神经可预测性并非由最大方差方向主导。最后，通过对学习得到的子空间进行特征解码，揭示了若干在神经活动中显著超表达的可解释语义特征，超越了原始模型表征本身。综上，该研究不仅指出LLM与大脑对齐的关键机制存在于一个极小但高度结构化的表示子空间中，还为深入理解为何特定模型表征能有效模拟人类语言神经过程提供了系统性框架。
链接: https://arxiv.org/abs/2610.40511
作者: Zach Paris,Emily Finn
机构: Dartmouth College (达特茅斯学院)
类目: Computation and Language (cs.CL); Neurons and Cognition (q-bio.NC)
备注: 23 pages, 12 figures. Code: this https URL
Abstract:Large language models (LLMs) are increasingly used as computational models of human language processing, yet it remains unclear which components of their representations drive alignment with neural activity. Here, we use ridge reduced-rank regression to identify the brain-predictive subspace of LLM representations: the representational directions that support linear prediction of neural responses. Across three naturalistic language-listening datasets spanning intracranial electrophysiology and fMRI, and six model conditions spanning autoregressive, bidirectional masked, and randomly initialized language models, we find that full linear encoding performance can be recovered using only 1-3% of the original embedding dimensionality. These low-dimensional subspaces generalize across held-out participants and, to a lesser extent, across datasets, suggesting that neural predictivity is concentrated in a compact and partially shared representational space. We next compare these brain-predictive directions with the principal axes of variation in the embeddings. Principal component analysis requires substantially more dimensions to recover neural prediction, demonstrating that the directions most useful for predicting neural activity are distinct from those capturing the most overall variance within LLM representations themselves. Finally, characterizing the learned subspaces allows us to surface interpretable features that are overrepresented in neural activity above and beyond the LLM representations. Together, these results show that LLM-brain alignment is supported by a small, structured subset of model representations and provides a framework for better understanding exactly why these representations align. Code can be found at: this https URL
[NLP-14] Agent ic University: Collaborative Agents for Personalized Learning
【速读】： 该论文旨在解决大语言模型在个性化学习中难以实现目标、活动与评估之间有效对齐，且缺乏基于学习证据动态调整后续支持的系统性难题。其核心挑战在于如何构建一个能够随学习者发展而持续演进的自适应学习过程。解决方案的关键在于提出一种“人参与的多智能体组织”（Human-Participatory Multi-Agent Organization）框架，并将其具体化为“智能体大学”（Agentic University, AU）。在该框架中，学习者作为组织成员，与具备特定功能的智能体协作，共同维护一个可修订的能力计划（Capability Program），该计划将个人目标、跨模态学习活动与分层评估相连接，并通过学习者反馈与表现动态指导后续支持策略。实证结果表明，该方法在计划组织、评估设计及修订质量方面均具优势，学习者在评估得分上平均提升12.19分（50分制），显著高于对照组的5.61分，且报告了更强的认知支持感。该研究贡献了一种以学习者持续参与为基础的多智能体协同机制，推动了自适应学习系统的组织化发展。
链接: https://arxiv.org/abs/2610.40492
作者: Jingru Fan,Tianle Zhou,Yunjie Zhou,Zihao He,Yixuan Zhou,Kezhao Wu,Xinyu Zheng,Xiyuan Wang,Yuheng Wang,Zhi Li,Songhua Liu,Weinan E,Chen Qian
机构: Shanghai Jiao Tong University(上海交通大学); Nanyang Technological University(南洋理工大学); National University of Singapore(新加坡国立大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Large language models support personalized tutoring, content generation, and adaptive learning paths. However, capability development requires these resources and interactions to form a coherent learning process that evolves with the learner. A central challenge is to align goals, activities, and assessment while using learning evidence to decide what should follow. We introduce Human-Participatory Multi-Agent Organization and instantiate it in Agentic University (AU). In AU, learners participate as organization members and work with specialized agents to maintain a revisable Capability Program. The program connects personal goals, learning activities across modalities, and layered assessment, while learner feedback and performance guide subsequent support. Our evaluation shows advantages in program organization, assessment design, and revision quality. Users achieved larger mean assessment gains (12.19 versus 5.61 points on a 50-point scale) and reported stronger cognitive support. This work contributes an organizational approach to adaptive learning that makes learner participation a continuing basis for multi-agent coordination. The Agentic University project is available at \hrefthis https URL.
[NLP-15] rTaste: Do Research Agents Have Taste? Evaluating Early Stage Research Judgment
【速读】： 该论文旨在解决生成式 AI 作为研究代理（research agent）是否具备“科研品味”这一核心问题，即在实验结果未知前，能否有效判断哪些实验值得开展。其解决方案的关键在于构建一个评估框架，让多个前沿大模型（如 GPT 系列等）独立或以团队形式扮演研究代理角色，提出最小可行实验（Minimum Viable Experiments, MVEs），并为每个实验设定通过阈值的预测，同时对实验进行排序。随后由独立执行者运行所有提议的实验（包括排名最低的），再由三位模型审计员对实验结果进行质量、新颖性及突破性评估。实验结果显示，尽管415个MVE中有194个被判定为合理、有价值且新颖，但其中仅有73个被视为超越原始论文且非显而易见的创新实验。虽然研究代理的首选项比随机选择更接近审计员最重视的实验，但并未显示出优于直接选择原文首提实验的优势；其预测表现虽优于基线，但与仅基于论文和注册信息的新实例相比并无显著提升。此外，团队协作模式成本更高，且未带来提案质量的明显改善。因此，尽管生成式 AI 能产生大量有价值的实验构想，但现有证据尚不足以支持其排名系统能显著优化实验优先级决策。
链接: https://arxiv.org/abs/2610.40488
作者: Art Kanke
机构: University of Minnesota(明尼苏达大学)
类目: Computation and Language (cs.CL)
备注: 27 pages, 1 figure, 9 tables. Code and data: this https URL
Abstract:Do research agents have taste, the judgment of which experiments are worth running before their outcomes are known? We give six frontier models seven recent NLP papers. Each model, working as a research agent alone or as a team, proposes minimum viable experiments (MVEs), forecasts whether each will pass a threshold it sets and ranks them. A separate executor then attempts every MVE, including those ranked last, and three model auditors judge the results. Of 415 MVEs, 194 are judged sound, valuable and new, but fewer than half of these (73) are also judged beyond the seed paper and non obvious. The agents’ first choices match the auditors’ most valued experiment more often than a random pick, but are not shown to improve on choosing the experiment written first. Their forecasts beat a base rate but are not measurably more accurate than those of a fresh instance of the same model given only the paper and registrations. Teams cost more without a clear overall improvement in proposal quality. The agents produce many worthwhile experiments, but the evidence that their rankings improve the choice of which to run remains limited. The evaluation harness and a dataset of all 415 MVEs, with their forecasts and audited outcomes, are available at this https URL.
[NLP-16] Preserving Utility in Continual LLM Unlearning via Orthogonal Projections
【速读】： 该论文旨在解决大语言模型（Large Language Model, LLM）在生命周期中持续进行敏感数据删除（unlearning）时面临的挑战，即现有方法在多次迭代删除过程中导致遗忘效果和模型性能的显著退化。核心问题在于连续更新之间的干扰：每次新删除操作可能抵消先前的遗忘效果，或使模型参数向已受损的方向进一步偏移，从而引发性能坍塌。为此，论文提出一种名为持续正交投影（Continual Orthogonal Projection, COP）的解决方案，其关键在于通过将每次梯度更新限制在由历史遗忘数据激活子空间的正交补空间内，实现对先前删除操作的一阶非干扰性（first-order non-interference）。COP仅需在每层存储一个低秩基底，无需保存过往梯度或遗忘数据集，且可无缝集成至基于可微损失函数的梯度更新型未学习方法中。实验结果表明，COP在标准基准（TOFU 和 RWKU）上能有效维持模型性能，同时保证持续遗忘能力，显著优于现有针对持续未学习设计的基线方法。
链接: https://arxiv.org/abs/2610.40483
作者: Juan Belieni,Ana Carolina Erthal,Eliezer de Souza da Silva,Diego Mesquita
机构: School of Applied Mathematics, Getulio Vargas Foundation(巴西法学院应用数学学院); University of Coimbra, CISUC/LASI, DEI(科英布拉大学, CISUC/LASI, DEI)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: 14 pages, 4 figures, 3 tables
Abstract:Unlearning concerns the task of removing the influence of a sensitive portion of the training data, such as protected, copyrighted or private text, from a model without retraining it from scratch. While unlearning requests happen continuously during a model’s lifetime, the vast majority of LLM unlearning methods are ill-suited for iterated removal, yielding steep decay in forgetting and utility upon sequential unlearning. This decay stems from interference between successive updates, as each new deletion can undo earlier ones or push the model further along directions already degraded by them. To address this, we introduce Continual Orthogonal Projection (COP), a gradient-projection scheme that restricts each update to the orthogonal complement of activation subspaces cached from prior forget sets. This approach provably guarantees first-order non-interference with earlier deletions. COP stores only a low-rank basis per layer, requiring neither past gradients nor previous forget sets, and plugs into unlearning methods that update the model through gradient steps on a differentiable loss without modifying its objective. Experiments on standard LLM unlearning benchmarks (TOFU and RWKU) show that COP preserves utility across sequential requests without weakening forgetting, preventing the collapse of standard objectives and forgetting more effectively than continual-specific baselines.
[NLP-17] FIRME: Registry-Anchored Memory Graphs for Evolving Heterogeneous Investor Knowledge
【速读】： 该论文旨在解决金融领域中生成式AI工作流在复用投资者知识时面临的持久记忆问题，即如何在跨来源、跨时间段的观测数据中保持关系历史的完整性与可追溯性，避免因表面形式差异导致的语义碎片化。其核心挑战在于：同一实体或关系在不同时间点或数据源中可能以异构表达出现，致使证据链断裂，无法实现知识累积。为应对这一问题，论文提出了一种基于注册表锚定的记忆图结构——融合与身份注册机制（Fusion and Identity Registry for Memory Evolution, FIRME）。其关键创新在于通过规范身份（canonical identities）与持久别名（persistent aliases） 实现跨周期观测的统一关联，并结合规则驱动的融合机制，在周期内将投资者电话会议讨论与投资组合现金流信号通过固定的产品-主题本体进行对齐；在跨周期层面，采用确定性更新策略对关系的历史进行强化、静默和重激活处理，从而维持长期记忆的一致性与可追溯性。实验结果表明，相较于仅保留规范身份与语义回退但忽略别名记忆的历史注册表，FIRME在四个数据集-后端配置下，最终周期的正确匹配率提升7.8–12.1个百分点；盲测审计显示，每设置下正确匹配率提升9.0–16.5个百分点，重新关联召回率提升22.5–29.0个百分点，且所有样本中合并精度均更高。此外，完整周期内融合的300个输出中有293个具备有效证据路径，验证了提取忠实性与图内容的闭环保真度。
链接: https://arxiv.org/abs/2610.40482
作者: Yunbai Zhang,Yunyi Gao,Han-Tai Shiao
机构: The Vanguard(先锋公司), Malvern, Pennsylvania, USA
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 20 pages. Accepted at the 3rd Workshop on LLMs and Generative AI for Finance (AI4F) at ACM ICAIF 2026 (non-archival)
Abstract:Financial AI workflows that reuse investor knowledge need persistent memory that integrates observations across sources and periods without losing identity or provenance. In a knowledge graph, recurring observations may use different surface forms, fragmenting relationship histories and preventing evidence from accumulating. We introduce Fusion and Identity Registry for Memory Evolution (FIRME), a registry-anchored memory graph that reconciles incoming observations to persistent canonical relationships and retains their source, time, and context. Within periods, rule-based fusion links investor-call discussions to portfolio cash-flow signals, directly or through a fixed product-topic ontology. Across periods, canonical identities and persistent aliases direct recurring observations to the same relationship history; deterministic updates summarize reinforcement, inactivity, and reactivation. We evaluate FIRME on proprietary investor-call and cash-flow records and on SP 500 earnings-call transcripts. Compared with a historical registry that retains canonical identities and semantic fallback but omits alias memory, FIRME increases final-period Correct Match Rate by 7.8-12.1 percentage points across four dataset-backend settings under a fixed-judge and fallback-review protocol. A blinded human audit of 200 incoming observations per setting finds Correct Match Rate gains of 9.0-16.5 points and reattachment-recall gains of 22.5-29.0 points, with higher observed merge precision in all four samples. A separate audit finds valid evidence paths for 293 of 300 outputs from full within-period fusion in the investor-call setting. Extraction-faithfulness and round-trip diagnostics assess source grounding and preservation of extractor-derived graph content.
[NLP-18] raining language models in simulation for psychiatric intake
【速读】： 该论文旨在解决精神科临床初诊访谈中，人工智能（AI）系统在复杂、多轮对话情境下难以有效支持结构化病史采集、鉴别诊断及初步管理计划制定的问题。其核心挑战在于，传统语言模型在缺乏动态反馈机制的情况下，难以模拟真实临床互动中的渐进式信息挖掘与风险评估能力。解决方案的关键在于采用多轮强化学习（multi-turn reinforcement learning, RL）在高度仿真的临床环境中对语言模型进行训练，通过设计包含临床疗效与安全性的复合奖励函数，使模型能够自主优化对话策略。实验结果显示，经强化学习训练的代理在120次双盲交叉对照测试中显著优于基础模型，尤其在临床评分（效应量d_z = 0.77）、以患者为中心的沟通质量（82.0% vs. 65.2%）以及鉴别诊断恰当性（68.3% vs. 33.3%）方面表现突出；同时，在256个由德尔菲共识构建的高难度对抗性情景测试中，模型对急性安全风险的漏诊率下降超过一半（12.9% vs. 30.5%），表明其具备更强的临床安全感知能力。研究证明，仿真环境为精神科对话代理提供了可控且可扩展的训练与压力测试平台，所获性能提升可有效迁移至真实医生主导的访谈场景，但模型仍存在局限，需依赖临床医生监督与安全防护机制。
链接: https://arxiv.org/abs/2610.40476
作者: Samuel Schmidgall,Tim Strother,Alex Bijamov,Min Woo Sun,Anil Palepu,Akshay Goel,Ulysse Marteau-Ferey,Théo Moutakanni,Yubin Kim,Joseph Breda,Christina Chen,Justin Chen,Paul Jhun,Andrew Marshall,Sarah Montgomery-Taylor,Eun Kyoung Hong,Shashank Ravi,Kameron C. Black,David J. H. Wu,Austin Schoeffler,John Lyon Havlik,Elizabeth C. Stade,Jared Moore,Johannes C. Eichstaedt,Logan Pierce,Jeff Jopling,Johannes Thrul,Thomas H. McCoy,Neguine Rezaii,Mark S. Chen,Ellen E. Fitzsimmons-Craft,Shirley Wang,Desmond C. Ong,Katharina Schultebraucks,Torrey Creed,Bhavna Daryani,Aleksandra Faust,S. Sara Mahdavi,Caroline Figueroa,Eric Lin,Kate H. Bentley,Ernest Lim,Zhihan Jiang,Xuhai “Orson” Xu,Daniel McDuff,John Torous,Yun Liu,Po-Hsuan Cameron Chen,Mike Schaekermann,Joelle Barral,Alvin Rajkomar,Tao Tu,Lin Yang,Isaac Galatzer-Levy,David Racz,Valentin Liévin
机构: Google DeepMind(谷歌深度思维); Google Research(谷歌研究); Google for Health(谷歌健康); Stanford University(斯坦福大学); University of California, Los Angeles(加州大学洛杉矶分校); University of California, San Francisco(加州大学旧金山分校); Johns Hopkins University(约翰霍普金斯大学); Harvard Medical School(哈佛医学院); Massachusetts General Hospital(麻省总医院); Harvard University(哈佛大学); Washington University in St. Louis(圣路易斯华盛顿大学); Yale University(耶鲁大学); University of Texas, Austin(德克萨斯大学奥斯汀分校); New York University(纽约大学); University of Pennsylvania(宾夕法尼亚大学); Columbia University(哥伦比亚大学); Beth Israel Deaconess Medical Center(贝斯以色列女执事医疗中心); Equal Leadership(平等领导力)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 76 pages, 8 figures, 15 tables (27 pages main text, 7 main figures, 2 main tables)
Abstract:In psychiatric practice, the clinical intake interview remains a primary diagnostic instrument, requiring the clinician to establish rapport while disentangling complex, overlapping presentations. Here, we show that multi-turn reinforcement learning (RL) in simulation environments improves language model performance on structured text-based psychiatric intake, spanning history-taking, differential diagnosis, and initial management planning. In a randomized, double-blind crossover study ( N = 60 matched scenario pairs, 120 encounters) comparing the RL-trained agent against its identically prompted base model, 32 board-certified psychiatrists role-played standardized patients while a separate panel of 15 independent, board-certified psychiatrists scored the de-identified transcripts. The trained agent outperformed the base model in 76.7% of matched scenario pairs ( p 0.0001 ), with gains on an adapted clinical-examination rubric ( d_z = 0.77 , p 0.0001 ), patient-centered communication (82.0% vs. 65.2%, p 0.001 ), and differential diagnosis appropriateness (68.3% vs. 33.3%, p 0.001 ). Across 127 psychiatrist-validated simulated intake scenarios, the trained agent nearly doubled its base model’s clinical rubric score (0.706 vs. 0.379; composite reward 0.564 vs. 0.294, p 0.0001 ). To systematically probe conversational safety, we established two Delphi-consensus taxonomies, of clinical safety risks and relational competencies, from which 256 challenging adversarial presentations were derived; we find that training in simulation environments more than halved under-triage of acute safety risks (12.9% vs. 30.5%, p 0.0001 ). Simulation environments thus provide a controlled testbed for training and stress-testing psychiatric conversational agents, with gains that transferred to psychiatrist-enacted encounters, though residual gaps require safeguards and clinician oversight.
[NLP-19] Prefix Executability: Evaluating Tool-Using Agents Beyond Final Success NEURIPS2026
【速读】： 该论文旨在解决当前工具使用评估中仅依赖最终任务成功率（terminal task success）所带来的评估偏差问题，即高成功率可能掩盖中间工具调用序列在实际执行中的不可行性。其核心解决方案是提出**前缀可执行性（prefix executability）**这一轨迹级评估协议，通过检验预测工具调用序列的每一个前缀在工具模式（tool schemas）、状态先决条件（state preconditions）和数据依赖关系（data dependencies）下的可执行性，从而揭示执行路径的可靠性。关键评估指标包括前缀可执行性曲线（PEC）、首次错误深度（FED）、轨迹可执行率（TER）、归一化AUC-PEC及恢复率（RR），这些指标能够量化代理在执行过程中的稳健性与容错能力。实验结果表明，仅以最终成功率为评价标准会导致对代理性能的误判，而引入前缀可执行性分析可有效暴露那些依赖未经验证计划但表面成功率较高的代理的真实缺陷，进而提供可操作的可靠性信号，用于代理间的公平比较、故障诊断以及验证机制的有效性评估。
链接: https://arxiv.org/abs/2610.40444
作者: Amir H. Rezaeian,Weiyi Sun,Yassine Benajiba,Dan Roth
机构: Oracle AI(奥莱克人工智能)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 21 pages, 2 figures. Accepted to NeurIPS 2026
Abstract:Tool-use benchmarks often score agents by final task success, but terminal outcomes can obscure whether the intermediate tool-call trajectory was executable. We propose prefix executability as a trajectory-level evaluation protocol for tool-using agents. The protocol measures whether every prefix of a predicted tool-call sequence remains executable under tool schemas, state preconditions, and data dependencies, and summarizes reliability using Prefix Executability Curves (PEC), First-Error Depth (FED), Trajectory Executability Rate (TER), horizon-normalized AUC-PEC, and Recovery Rate (RR). We instantiate the protocol on stratified samples from BFCL-v3 multi-turn and ToolSandbox, and on the complete Airline/Retail base splits of \tau^3 -bench. We observe that outcome-only and prefix-level rankings diverge: agents whose execution is conditioned on an unverified plan often achieve higher point-estimate final success than non-plan-conditioned baselines while substantially lowering TER and AUC-PEC. These divergences are precisely the failure modes that prefix executability is designed to expose: they reveal whether apparent success rests on an executable trajectory, where invalidity first enters, and whether success follows recovery from an earlier failure. As a validation intervention, explicit pre-execution plan checking can improve executable-prefix survival while maintaining competitive terminal success. These results show that final success alone is insufficient for evaluating tool-use agents and that prefix-level executability provides an actionable reliability signal for comparing agents, diagnosing failures, and assessing validation mechanisms.
[NLP-20] Scaling Mathematical Proof Supervision from Structured Textbooks
【速读】： 该论文旨在解决现有语言模型在数学推理任务中缺乏结构化监督信号的问题，即传统方法将数学知识简化为孤立的“问题-解答”对，忽略了教材中固有的课程组织逻辑与论证依赖关系。其核心解决方案是构建TOME-Math（Textbook-Oriented Mathematical Evidence），一个大规模、结构化的数学监督语料库，将2,118本涵盖大学数学核心领域的教科书转化为约38万条保留章节顺序、命题-证明关联、依赖引用及来源溯源的结构化证明监督记录。关键在于通过保留教材的结构性与证据链，实现基于课程学习（curriculum learning）的多学科数学推理训练，并揭示监督数据量与来源广度的缩放规律。实验表明，在全规模下，基于TOME-Math微调可使模型在多个测试集和公开基准上的未加权平均得分较基线提升16.7个百分点，且整体性能优于所有对比的公开指令数据基线。研究结果强调了结构化教科书作为可扩展的数学监督来源及其在探究数据组织对数学推理影响中的实证价值。
链接: https://arxiv.org/abs/2610.40394
作者: Youqing Fang,Zhouqi Hua,Yinhao Tang,Yuzhe Gu,Lingkai Kong,Bin Liu,Weiming Zhang,Kai Chen,Yanan Sun,Wenwei Zhang
机构: University of Science and Technology of China(中国科学技术大学); Fudan University(复旦大学); Shanghai AI Laboratory(上海人工智能实验室); Shanghai Jiao Tong University(上海交通大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 35 pages, 18 figures, 14 tables
Abstract:Mathematical supervision for language models is typically flattened into isolated problem-solution pairs, discarding the curricular organization and evidential structure of the original sources. We introduce TOME-Math (Textbook-Oriented Mathematical Evidence), a large-scale corpus that converts 2,118 textbooks spanning the core areas of university-level mathematics subjects into approximately 380K structured proof-supervision records while preserving chapter order, proposition-proof links, dependency references, and source provenance. Leveraging these structural signals, we explore curriculum learning across multiple mathematical subjects and characterize scaling patterns with supervision volume and source breadth. At full scale, fine-tuning on TOME-Math raises the unweighted mean score across our held-out evaluations and established public benchmarks by 16.7 percentage points over the base model, while exceeding all three public instruction-data baselines evaluated in aggregate. Our findings highlight structured textbooks as both a scalable source of mathematical supervision and an empirical framework for studying the impact of data organization on mathematical reasoning. We will release the corpus to facilitate further research on structured mathematical supervision.
[NLP-21] Dont Lose the Thread: Empowering Long-Horizon LLM Agents with Cognitive Resource Self-Allocation
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）驱动的智能体在处理长时程任务时因认知过载导致的性能下降问题。具体而言，随着任务进展，智能体的工作记忆中积累大量冗余和无关信息，造成注意力分散，进而影响有效规划与推理能力。其解决方案的关键在于提出一种名为“认知资源自分配”（Cognitive Resource Self-Allocation, CORAL）的新推理范式，通过构建可由智能体调用的工作记忆管理工具集，使智能体能够主动对上下文进行优化：在关键节点进行任务进度与已验证事实的检查点（checkpoint）保存，并在上下文污染严重时主动重置当前会话，仅保留必要信息以恢复至一个干净的推理状态，从而实现分阶段、可持续的高效推理。进一步地，采用多阶段智能体强化策略优化算法（Multi-episode Agentic Reinforced Policy Optimization）增强检查点决策能力。实验结果表明，基于8B模型的CORAL在GAIA基准上实现了42.8%的平均准确率，尤其在最具挑战性的长时程任务（Level 2和Level 3）中表现显著提升；同时，其推理开销大幅降低，输入令牌消耗仅为ReAct智能体的1/6至1/7，且在多跳知识问答及更复杂的BrowseComp任务中均展现出优异泛化能力，相较强基线智能体提升8.0分。
链接: https://arxiv.org/abs/2610.40350
作者: Zheng Zhu,Xiaozhuan Liang,Xi Chen,Bei Yu,Jiaya Jia
机构: The Chinese University of Hong Kong(香港中文大学); Tencent(腾讯); The Hong Kong University of Science and Technology(香港科技大学)
类目: Computation and Language (cs.CL)
备注: Accepted at COLM 2026. Code is available at this https URL
Abstract:Agents powered by large language models (LLMs) have demonstrated remarkable progress in solving complex reasoning tasks. However, LLM agents often falter on long-horizon tasks due to cognitive overload, as their working memory becomes cluttered with expanding and irrelevant information, which dilutes their attention and hinders effective planning and reasoning. To mitigate this challenge, we introduce Cognitive Resource Self-Allocation (CORAL), a novel reasoning paradigm that empowers agents to proactively optimize their context. Implemented as an agent-callable working memory management toolset, CORAL allows an agent to checkpoint its task progress and verified facts, and adaptively initiate a new problem-solving episode by resetting cluttered context while preserving only the checkpointed information, enabling it to resume reasoning from a clean state. We further enhance the agent’s checkpointing capabilities using a Multi-episode Agentic Reinforced Policy Optimization algorithm. On the GAIA benchmark, CORAL with an 8B model achieves 42.8% average accuracy, with particularly strong gains on the most challenging long-horizon tasks (Level 2 and Level 3). Crucially, these gains come at a substantially lower inference cost: CORAL consumes 6-7 \times fewer input tokens than a ReAct agent. It also generalizes to multi-hop Knowledge QA benchmarks and to the considerably harder BrowseComp, where it improves a strong proprietary agent by 8.0 points.
[NLP-22] Factorized Representations of Concurrent Temporal Variables in Pretrained Language Models
【速读】： 该论文旨在解决大语言模型在文档生成过程中如何协同维护多个独立演化的上下文变量（如语言切换与主题变更）的问题。现有研究虽已识别出单个时间或上下文变量的结构化表示，但其联合组织机制仍不明确。本文通过设计具有独立相位的语言切换与主题变更定时器的受控文本，发现这两个定时器在相同隐藏状态中均能线性可解码。移除与某一定时器相关的中心平面会显著削弱对该变量的访问能力，而对另一并发定时器影响甚微，尽管这些平面非正交且包含交叉信息。此外，各定时器的独立贡献可解释联合状态中心点间大部分可复现的变异。上述特征在0.6B至14B参数量的Qwen3系列检查点中均稳定存在，且持续覆盖3至51个定时值。虽然时序位置会影响后续生成概率，但测试的中心点偏移并未产生一致的定时器特异性输出效应。综上，研究结果支持一种基于选择性可访问性与近似可加性的联合状态几何结构，即多变量信息在重叠子空间中以近似可分解方式共存。
链接: https://arxiv.org/abs/2610.40328
作者: Xinhe Zhang,Arnau Marin-Llobet
机构: Harvard University (哈佛大学)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注:
Abstract:Language models often need to maintain multiple contextual variables that evolve independently within a document. Prior work has identified structured representations of individual temporal or contextual quantities, but their joint organization remains less understood. We study this question using controlled documents with independently phased language-switch and topic-change timers. Across ten pretrained checkpoints, both timers are linearly decodable from the same hidden states. Removing a timer-associated centroid plane substantially reduces access to its associated variable while having little effect on the concurrent timer, although the planes are nonorthogonal and contain cross-timer information. Separate timer contributions also explain most reproducible variation among joint-state centroids. These signatures persist from 3 to 51 timer values and across Qwen3 checkpoints from 0.6B to 14B. Schedule position affects continuation probabilities, but the tested centroid displacement produces no consistent timer-specific output effect. Together, these results support approximate factorization based on selective accessibility and approximately additive joint-state geometry in overlapping subspaces.
[NLP-23] Decision and System One models in rheumatology: accuracy and calibration on specialty-training examination questions and diagnostic case vignettes
【速读】： 该论文旨在解决生成式AI在风湿病学领域临床决策任务中的准确性与校准性问题，特别是在面对标准化考试题与真实临床诊断案例时的表现差异。其核心挑战在于评估不同生成式AI模型（包括System One系列及GPT-6系列）在多语言、多任务场景下的诊断推理能力，并揭示其在高置信度预测下的可靠性缺陷。解决方案的关键在于通过两个基准测试——扩展的RheumaMIR（涵盖西班牙语和英语的风湿病专科考试题）与RheumBench（含5–18个候选诊断的英文临床案例）——系统比较多个模型的准确性、校准性、判别能力及语言影响。研究发现，尽管部分System One模型（如Jev）在考试题中表现优异（准确率最高达93.0%），但在复杂诊断案例中显著下降（如Jev仅50.0%准确率），且出现过度自信现象（平均置信度0.659，过自信指数+0.159），其高置信度预测（概率=0.90）的正确率远低于GPT-6 Luna（84.6% vs 95.5%）。这表明，当前通用型生成式AI在临床决策任务中存在任务适应性不足与校准偏差问题，强调了针对特定临床任务进行独立评估与校准的必要性。
链接: https://arxiv.org/abs/2610.40318
作者: Alfredo Madrid-García,Diego Benavent,Fabian Lechner,Phillip Kremer,Johannes Knitza,Beatriz Merino-Barbancho
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 28 pages, 2 figures, 2 tables; includes supplementary material (8 tables, 1 figure)
Abstract:Objectives: To evaluate the accuracy and calibration of Decision and System One models on rheumatology examination questions and diagnostic vignettes. Methods: Two benchmarks were analysed: 172 rheumatology questions from Spanish specialty-training examinations (expanded RheumaMIR dataset) in Spanish and English, and 70 English diagnostic vignettes from RheumBench with 5-18 candidate diagnoses, 52 of which included the reference diagnosis. Four System One models (Jev 1.13, Clef, Microsoft-Decision-1, Laya) and GPT-6 Luna (OpenAI Decisions API) were compared with GPT-6 Sol and two local generative models. The primary outcome was accuracy; secondary outcomes included calibration, discrimination, accuracy of answers with probability =0.90 and language differences. Results: In RheumaMIR, Jev’s accuracy was 93.0% in Spanish and 91.9% in English, below GPT-6 Sol (99.4%; Holm-adjusted P=0.005); GPT-6 Luna reached 97.7%, Clef 91.9-93.0% and Microsoft-Decision-1 86.0-87.8%, without detectable language differences. Jev’s answers with probability =0.90 (76-77% of answers) were 98.5% correct. In RheumBench, accuracy was lower: GPT-6 Sol 76.9%, GPT-6 Luna 63.5%, Clef 59.6%, Jev 50.0% and Microsoft-Decision-1 40.4% (each System One model below GPT-6 Sol, Holm-adjusted P=0.045; chance 9.6%). Jev became overconfident (mean probability 0.659; over/underconfidence index +0.159), and its answers with probability =0.90 (25% of cases) were 84.6% correct, against 95.5% for GPT-6 Luna (42%). Laya performed near chance in both benchmarks. Conclusion: The best-performing System One models were highly accurate on examination questions but less accurate on diagnostic vignettes, remaining below GPT-6 Sol. Jev’s high-confidence predictions were less reliable on vignettes, highlighting the need for task-specific evaluation and calibration.
[NLP-24] Leverag ing Recourse and Size Bounds for OnlineKnapsack with Removable Items
【速读】： 该论文旨在解决在线无权背包问题（online unweighted knapsack problem）中因决策不可撤销而导致无法获得非平凡竞争比（competitive ratio）的难题。其核心挑战在于：在物品按顺序到达且必须立即决定是否接纳或拒绝的情况下，若不允许撤销已接受的物品，则任何算法均无法实现有界的竞争比。为突破此限制，论文引入了“移除”（removal）这一关键机制，允许从背包中移除已接受的物品，并进一步结合两个已有模型扩展——对物品大小上界的知识（bound awareness）与对先前被拒绝物品的重新插入能力（reinsertion）。研究的关键贡献在于：首先，在同时具备上界知识、移除和回溯（recourse）能力的前提下，建立了紧致的上下界，证明当上界不严格且回溯次数有限时，可显著提升最优已知竞争比；其次，提出“自适应在线算法”（adaptive online algorithms）的新概念，该类算法无需预先知晓物品大小的上界，但仍能实现与上界已知模型相当的性能保障，从而在不依赖先验信息的情况下实现了近似最优的竞争力。
链接: https://arxiv.org/abs/2610.40290
作者: Dimitris Fotakis,Laurent Gourvès,Aris Pagourtzis,Panagiotis Patsilinakos
机构: National Technical University of Athens (雅典国立技术大学); Archimedes, Athena Research Center (阿基米德，阿西娜研究中心); Université Paris-Dauphine, Université PSL, CNRS, LAMSADE (巴黎达芬奇大学，巴黎文理研究大学，法国国家科学研究中心，拉姆萨德实验室)
类目: Data Structures and Algorithms (cs.DS); Computation and Language (cs.CL)
备注:
Abstract:In the online unweighted knapsack problem, items arrive sequentially and must be accepted or rejected immediately for a bounded-capacity knapsack, aiming to maximize the total size of accepted items. When decisions are irrevocable, no algorithm achieves a non-trivial competitive ratio. An essential relaxation that circumvents this impossibility is to allow previously accepted items to be removed from the knapsack (a.k.a. removal). In this work, we investigate possible further improvements on the competitive ratio with removal by bringing together two other previously proposed model extensions: knowledge of an upper bound on the item sizes (a.k.a. bound awareness) and ability to reinsert previously rejected items(this http URL). Our contributions are twofold. First, we establish close upper and lower bounds on the competitive ratio under bound awareness, removal and recourse. We show that significant improvements on the best known competitive ratio are possible when the upper bound on the item sizes is not too restrictive and only few uses of recourse are available. Second, we introduce the notion of adaptive online algorithms which do not require to know any upper bound on the item sizes but still achieve performance guarantees comparable to the ones obtained under the bound awareness model.
[NLP-25] Latent Path Exploration via Potential-Guided Subtrajectory Balance
【速读】： 该论文旨在解决生成式语言模型在连续隐状态空间中进行随机推理时，如何有效引导隐状态间转移以探索有意义推理路径的关键问题。现有方法虽能通过随机性采样多条推理路径，但缺乏对探索方向的有效指导，导致大量无效或低效路径的产生。其解决方案的核心在于提出SubLAT框架，通过引入相对答案势能（relative answer potential）与子轨迹平衡（Subtrajectory Balance）机制协同优化隐空间中的随机探索过程。相对答案势能通过比较采样状态与替代状态的答案支持度，并结合终端状态的正确性信号来评估中间状态的有效性；而子轨迹平衡则建立潜在值差异与不同长度子轨迹上正向和反向转移密度之间的关系，从而实现对探索动态的建模。该方法在算术与演绎推理任务上表现出更强的解题覆盖率，尤其在更大采样预算下显著提升性能，且在域内与域外测试中均展现出优越的泛化能力。轨迹分析进一步表明，正确路径具有更高的平均相对势能并呈现出渐进分叉的演化特征，验证了该方法对有效推理路径的识别与引导能力。
链接: https://arxiv.org/abs/2610.40283
作者: Xiaoling Zeng,Qi Zhang
机构: Worcester Polytechnic Institute (伍斯特理工学院)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:Continuous latent reasoning enables language models to perform intermediate computation without generating explicit reasoning traces. Advances in stochastic latent reasoning allow multiple paths to be sampled in continuous space, but stochasticity alone does not ensure effective exploration. A key challenge is guiding stochastic transitions between intermediate latent states toward useful reasoning paths. We introduce SubLAT, which couples a relative answer potential with Subtrajectory Balance to learn stochastic latent exploration. The relative potential evaluates the answer support of sampled states against replacement states, with a correctness signal at the terminal state. Subtrajectory Balance relates potential differences to forward and backward transition densities across subtrajectories of different lengths. The model is jointly trained with answer supervision, CoT supervision, and representation distillation. Experiments on arithmetic and deductive reasoning show increasing solution coverage with larger sampling budgets across in-domain and out-of-domain benchmarks, as well as improvements on several structural OOD benchmarks. Trajectory analyses reveal progressive divergence among sampled latent paths and higher mean relative potentials for correct trajectories across arithmetic benchmarks.
[NLP-26] Neighborhood Elites: Quality-Diversity Search for Uncertainty Quantification in Question Answering
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在问答（Question Answering, QA）任务中产生错误答案却难以被识别的问题，此类错误会未经察觉地传播至下游任务。现有基于扰动的不确定性量化（Perturbation-based Uncertainty Quantification, UQ）方法通常依赖通用方式生成输入重写（如自由改写、词级扰动或对抗搜索），但这些方法未能控制两个关键因素：上下文修改的类型以及上下文对答案支持的显式程度。由于缺乏控制，扰动集合的构成由生成器决定而非设计驱动，导致不确定性估计受制于生成器的偏差。为克服这一局限，论文将输入扰动重构为质量-多样性（Quality-Diversity, QD）搜索问题，提出一种名为“邻域精英”（Neighborhood Elites, NE）的邻域构造器，其通过在“编辑类型”与“支持显式度”两个维度上分布保持答案不变的重写样本，实现更系统化的探索。实验表明，在五个抽取式QA数据集和五种指令微调的大语言模型上，NE在至少24/25的数据集-模型组合中优于其他三种邻域构造器，相较于基线分别提升0.726 AUROC和0.211 AUPRC；同时，其失败检测能力显著优于基于解码采样计算的语义熵，提升达0.074 AUROC与0.062 AUPRC，且优势随邻域规模扩大而增强。因此，该研究的关键在于通过结构化、可控制的扰动设计，实现更可靠、更具判别力的不确定性估计。
链接: https://arxiv.org/abs/2610.40278
作者: Mesut Ceylan,Arda Baris Basaran,Elgar Fleisch,Filipe Barata
机构: ETH Zurich(苏黎世联邦理工学院); University of St.Gallen(圣加仑大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Large language models (LLMs) are increasingly capable, yet they produce wrong answers indistinguishable from correct ones. In question answering (QA), such an error propagates unnoticed into downstream tasks. Perturbation-based uncertainty quantification (UQ) addresses this by probing a model over a set of input rewrites and reading uncertainty from how much its predictions vary. Yet this set is usually built generically, by free paraphrasing, token-level perturbation, or adversarial search. Such rewrites leave two things uncontrolled that matter in QA: what kind of change is made to the context, and how explicitly the context supports the answer. When neither is controlled, the composition of the set is determined by the generator rather than by design, and the estimate inherits whatever that generator happens to produce. Quality-diversity (QD) search is built to cover a predefined exploration space, so we recast input perturbation as a QD problem and propose \emphneighborhood elites (NE), a neighborhood constructor that spreads answer-preserving rewrites across two axes, the kind of edit and the explicitness of support. We compare NE against three other neighborhood constructors, holding the editor, QA model, uncertainty measure and search objective fixed, and against three scores read from the original context. Across five extractive-QA datasets and five instruction-tuned LLMs, NE improves failure detection against every other constructor in at least 24 of the 25 dataset-model pairs, reaching 0.726 AUROC and 0.211 AUPRC lift over the base rate. Additionally, NE detects failure better than semantic entropy computed over decoder samples, by 0.074 AUROC and 0.062 AUPRC lift, and this advantage widens as the neighborhood grows.
[NLP-27] When Requirements Interact: Scaling Long-Horizon Coding via Interdependent Specifications
【速读】： 该论文旨在解决当前编码代理在长周期软件开发中难以满足多维度任务规范（如接口兼容性、错误处理等）且不破坏现有依赖关系的问题。现有研究在任务规范的数量与多样性方面均存在不足，通常只关注主要目标而忽略复杂的约束条件。为填补这一空白，论文提出CodeSpecWeave框架，其核心在于通过真实开发者种子数据驱动，由合成代理迭代生成相互耦合的任务需求与可执行测试，同时由评审代理从多个模型的解题轨迹中审计以确保测试可靠性。该方法的关键创新在于实现了复杂、互相关联的工程约束的高效合成与验证，避免了简单清单式任务的低挑战性以及隐性缺陷的遗漏。基于此框架，构建了包含4,670个训练任务的语料库，并在DeepSWE-v1.1上通过强化学习显著提升模型性能；此外，还推出了CodeSpecWeave-Bench基准，涵盖100个任务与1,044条规范，有效区分前沿模型并揭示了在处理互依赖工程约束方面的巨大改进空间。
链接: https://arxiv.org/abs/2610.40256
作者: Yishuo Cai,Minzheng Wang,Liang Zhao,Linghao Zhang,Lei Li,Shicheng Li,Xu Sun,Fuli Luo
机构: 未知
类目: oftware Engineering (cs.SE); Computation and Language (cs.CL)
备注:
Abstract:To succeed in real-world long-horizon software development, coding agents must satisfy multi-dimensional task specifications, such as interface compatibility and error handling, and do so without breaking existing dependencies. However, current works lack specification in both quantity and diversity, often emphasizing primary goals while ignoring complex constraints. Synthesizing tasks to bridge this gap presents two challenges: naively adding specifications risks reducing tasks to trivial checklists rather than compelling joint reasoning, and verifying these subtle interactions risks overlooking hidden flaws or rejecting valid alternatives. To address these challenges, we introduce CodeSpecWeave, a framework for synthesizing tasks with rich, interdependent specifications. Grounded in real developer seeds, a synthesis agent iteratively develops both coupled task requirements and executable tests, while a review agent audits solution trajectories from multiple models to guarantee test reliability. Using this pipeline, we construct 4,670 training tasks across 2,865 repositories. Applying Reinforcement learning on these tasks significantly improves performance over the base model on DeepSWE-v1.1, largely outperforming existing baselines. Furthermore, we curate CodeSpecWeave-Bench, a rigorous benchmark comprising 100 tasks and 1,044 specifications, which clearly separates frontier models and reveals significant headroom in handling interdependent engineering constraints.
[NLP-28] Judged in Company: In-Context Batching Makes LLM Verdicts Depend on Other Items
【速读】： 该论文旨在解决在实际应用中，为降低计算成本而将多个独立评价项批量纳入同一提示（prompt）以由语言模型一次性生成评分时所引发的评价结果不一致性问题。其核心问题是：当前广泛假设的“评价可分性”（verdicts are separable），即某一项的评分不受其他项存在或其在提示中的位置影响，这一假设在实践中并不成立。研究发现，在三个判别任务（社交媒体评论中的危害性、简答题评分与评论星级评定）及多款开源大模型（4-9B和14-32B参数规模）上，将2000个评价项随机分配至不同25项批次时，约有11%-22%的判断结果发生改变，且这种偏差远超通过重命名项键或改写指令所能带来的影响；同时，尽管与人工标注的AUROC差异较小（-0.033至+0.010），但批量评分与单一项评分之间存在显著系统性偏差。关键发现在于，评分翻转率（flip rates）可精确分解为顺序（order）、槽位（slot）和成员归属（membership）三部分贡献，且在阈值模型下，上下文引入的方差对决策的影响随其平方根增长，因此即使对分数影响微小，也可能导致显著的判断变化。其中，邻近项的顺序与具体槽位是造成隐式相似性差异的主要来源，而共享同一提示的项之间的相互作用——即“共现效应”（co-occurrence effect）——虽可通过置换投票部分缓解，但仍保留52%-60%的翻转率。此外，即使对十五个随机批次进行多数投票，仍有8.4%-17.8%的判断与单一项评分不一致。实验进一步表明，模型自身先前的输出以及相邻项内容均携带此类上下文依赖效应。在较简单的星级评分任务中，仅2.9%-3.9%的判断发生变化。因此，该研究建议：对于语义模糊或敏感的评价项，应采用逐项提示的方式进行评分，通过缓存前缀（cached prefix）技术有效控制成本，并在批量处理时报告与单一项评分的一致性指标，以提升结果可信度。
链接: https://arxiv.org/abs/2610.40241
作者: Peng Xie,Amr Alanwar
机构: Technical University of Munich (慕尼黑工业大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:To save cost, practitioners often ask a language model to grade many independent items in one prompt. This assumes that verdicts are separable: that the verdict on an item does not depend on the other items in the prompt or on its position among them. Across three judgement tasks (harm in social-media comments, short-answer grading and review star ratings) and five open-weight models with 4-9B parameters, assigning the same 2,000 items to different random batches of 25 changes 11-22% of decisions on harm and grading, 6.5-13.3 points more than renaming the item keys or rewording the instruction, while AUROC against human labels differs from single-item grading only by -0.033 to +0.010; for four models with 14-32B parameters, 6-15% of decisions change, 4.2-9.1 points more. We show that flip rates decompose exactly into components of order, slot and membership, and that under a threshold model they grow with the square root of the variance that the context contributes, so that effects that are small for scores are large for decisions. Most of the latent dissimilarity between partitions comes from the order of the neighbours and the slot. The main effect of which items share the prompt, the part that voting over permutations of a batch cannot remove, accounts for 0.11-0.22 of it, and majority votes over nine permutations still leave 52-60% of the flips. Batched verdicts also differ systematically from single-item verdicts: even the majority of fifteen random batches disagrees with them on 8.4-17.8% of decisions. Editing a single generation shows that both the model’s own earlier verdicts and the neighbours’ texts carry the effect. On the easy task, star ratings, only 2.9-3.9% of decisions change. We recommend grading ambiguous items one per prompt, recovering most of the cost with a cached prefix, and reporting agreement with single-item grading whenever items are batched.
[NLP-29] ScienceJudge: Pattern Perturbation Augmentation for Sentence-Level Scientific Claim Verification NLPCC2026
【速读】： 该论文旨在解决生成式科学文本中细粒度报告错误（fine-grained reporting errors）的检测难题，此类错误在保持语句流畅性和表面合理性的同时，引入了微妙的证据-论断偏差，尤其在标注数据稀疏且分布不均的情况下难以识别。其核心解决方案是提出模式扰动增强（Pattern Perturbation Augmentation, PPA），将报告错误建模为特定标签的忠实性破坏变换，并从有标注样本中归纳出可复用的错误模式。PPA通过这些模式对支持性论断进行可控的“难例”构造，并结合模式级、操作级及跨领域增强策略扩展其覆盖范围。基于增强后的多视角数据，研究构建了ScienceJudge框架，该框架通过类别加权集成推理融合经PPA训练的视觉-语言预测器与指令引导的大语言模型（LLM）预测器。实验表明，在本地开发集上，ScienceJudge将平均得分从0.8124提升至0.8692，达到宏平均F1值0.8466和段落精确匹配率0.8919；在NLPCC 2026 Task 10 Track 1官方评测中，系统以0.6890的最终得分位列第一，验证了方法的有效性与鲁棒性。
链接: https://arxiv.org/abs/2610.40238
作者: Chuhai Cai,Xin Chen,Min Yang,Feng Jiang
机构: Shenzhen University of Advanced Technology (深圳大学先进技术研究院); Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences (中国科学院深圳先进技术研究院)
类目: Computation and Language (cs.CL)
备注: 13 pages, 1 figure; NLPCC 2026 Shared Task 10 (Track 1)
Abstract:Sentence-level scientific claim verification determines whether each sentence in an AI-generated scientific paragraph is supported by a given multimodal evidence bundle. Fine-grained reporting errors often preserve fluent and plausible statements while introducing subtle evidence-claim deviations, making them difficult to detect under sparse and imbalanced supervision. We propose Pattern Perturbation Augmentation (PPA), which models reporting errors as label-specific faithfulness breaking transformations and induces reusable error patterns from labeled examples. PPA applies these patterns to supported claims to construct controlled hard negatives and expands their coverage through pattern-level, operation-level, and cross-domain augmentation. Based on these augmented views, we develop ScienceJudge, which integrates PPA-trained vision-language predictors and instruction-guided LLM predictors through class-weighted ensemble inference. On the local development set, ScienceJudge improves the average score from 0.8124 to 0.8692, achieving a Macro-F1 of 0.8466 and a Paragraph Exact Match score of 0.8919. Our submitted system ranks first on the official NLPCC 2026 Task 10 Track 1 leaderboard with a final score of 0.6890.
[NLP-30] Mobilizing the Archive: Bitext Extraction at Scale from Archival Documents for Extremely Low-resource Languages
【速读】： 该论文旨在解决低资源语言在互联网上缺乏平行语料（parallel sentences）的问题，而这些语言虽无大量网络内容，却拥有由语言学家数百年积累的语法、词典及其他文献资料。其核心解决方案是通过从这些传统语言学资源中提取翻译例句，构建适用于网络覆盖不足语言的平行语料库，并实现对数千种语言的潜在扩展。关键在于开发了一套端到端的抽取管道，融合光学字符识别（OCR）模型与视觉语言模型（VLM），以确保文本识别结果与原始文档的视觉版式信息保持对齐，从而提升抽取精度与可靠性。实验表明，所提取的双语语料可有效用于训练机器翻译模型及支持大语言模型的上下文学习（in-context learning），在向低资源和高资源语言进行零样本翻译时，分别实现了+15.78和+6.94的chrF2++指标提升，验证了该方法的有效性与可扩展性。
链接: https://arxiv.org/abs/2610.40233
作者: Chutong Meng,Alec J. Bateman,Antonios Anastasopoulos
机构: George Mason University (乔治梅森大学); Barron Associates (巴伦协会)
类目: Computation and Language (cs.CL)
备注:
Abstract:While mining parallel sentences from web pages has proven effective for low-resource machine translation, the majority of the world’s languages have little to no web presence. What they do have, though, are resources that are the result of centuries of efforts by linguists, such as grammars, dictionaries, and other documentation materials. We propose extracting translation examples from such linguistic resources, which offer parallel data for languages under-represented online and could potentially scale to thousands more languages. We develop a pipeline to achieve this, leveraging an Optical Character Recognition (OCR) model as well as a vision language model (VLM), to ensure identified content to be aligned with visual layout information. We demonstrate the utility of the extracted bitexts for bootstrapping translation systems across 20 languages, both by training machine translation models and by applying in-context learning for large language models. For in-context learning, we observe improvements of +15.78 and +6.94 chrF2++ over zero-shot translation for into low- and high-resource languages, respectively.
[NLP-31] It Takes Two to Probe Thousands to Retrieve: Putting Vision–Language Probes to the Retrieval Test
【速读】： 该论文旨在探究视觉-语言模型（Vision-Language Models, VLMs）在细粒度理解能力评估中，图像-文本匹配探针（image-text matching probes）与检索任务之间的关系，核心问题是：不同负样本来源（如人工设计的负样本、随机采样负样本或由其他模型错误检索出的负样本）对模型检索性能评价的影响是否一致。其解决方案的关键在于通过两个探针基准测试（\svo 和 SugarCrepe）系统性地比较模型在相同查询下对正样本的排序表现，并引入人类标注者与大语言模型（LLM）评估器共同判断检索结果的相关性。研究发现，在 \svo 基准上，模型对人工设计负样本的区分能力显著优于对检索所得负样本的区分能力（准确率分别为85%和77%），且使用10个检索负样本即可达到200–430个随机负样本的效果；而在 SugarCrepe 上，人工设计负样本反而更易被正确排除，说明负样本的构造方式直接影响评估结果。此外，不完整的相关性标签是导致 \svo 检索性能偏低的主要原因，且模型通过探针测试的能力与其能否首屏召回正确项之间无显著关联。最后，大语言模型评估器与人类标注者的一致性略低于人类间的共识，表明当前自动化评估仍存在主观差异。
链接: https://arxiv.org/abs/2610.40124
作者: Ivana Beňová,Michal Gregor,Marián Šimko
机构: Kempelen Institute of Intelligent Technologies(肯佩伦智能技术研究所); Bratislava(布拉迪斯拉发); Slovakia(斯洛伐克)
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL)
备注: 12 pages, 2 figures, 3 tables
Abstract:Vision–language models (VLMs) are tested for fine-grained understanding with image–text matching probes, a model must prefer the correct image or caption over handcrafted negative. In retrieval, the correct item must beat thousands of alternatives. We measure how these two evaluation methods relate by using two probing benchmarks, \svo and SugarCrepe, for retrieval tasks. For 100 image and caption queries we judge the top-10 retrievals from full dataset of six VLMs, with human annotators and an LLM assessor. We than compare how well models rank pool of items on the same queries, while changing only where the negatives comes from: the benchmark, a random draw, or the judged-incorrect retrievals of other models. On \svo the accuracy of retrieving a positive over handcrafted negatives is significantly higher than for retrieved ones (the correct item is preferred in 85% against 77% of the time). Using pool of ten retrieved negatives leads to same accuracy as when one uses pool of 200 to 430 random negatives. On SugarCrepe the order is reversed: its handcrafted negatives are harder than the retrieved ones. Incomplete relevance labels explain most of the low retrieval scores on \svo, where the top-ranked item is correct for 54–90% of the queries under human labels against 6–21% under the benchmark’s labels. However, we detect no association between passing the probe for a query and retrieving a correct item first. The LLM assessor o3 agrees with individual annotators ( \kappa 0.57–0.61) slightly less than they agree with each other (0.64–0.68).
[NLP-32] Upcycling Language Models for Multiple Specialized Domains
【速读】： 该论文旨在解决大语言模型在多领域适应（multi-domain adaptation）过程中如何有效优先保留相关领域知识并最小化无关领域干扰的问题。现有方法普遍采用低秩适配（LoRA）为每个领域训练独立适配器，并在模型内组合这些适配器，但存在两大局限：一是对所有适配器进行均等加权，忽视输入领域的差异性；二是虽可动态加权，但仍激活全部适配器，导致无关领域产生干扰。此外，其低秩结构限制了模型性能上限。针对上述问题，本文提出领域专用升级（Domain-Specialized Upcycling, DS-Upcycling），通过在密集Transformer层中引入专家（expert）与路由器（router）的架构扩展，实现比LoRA更强的表达能力。其关键创新在于设计了一种新的领域感知监督机制，使路由器能够基于输入领域精准选择合适专家，从而提升路由准确性。在每个新领域微调时，仅更新对应专家和路由器参数，其余参数保持冻结，确保高效且无干扰的领域特异性学习。实验表明，该方法在各适配领域上实现了最高的平均性能，同时保持了良好的通用性能；仅需每领域256个样本用于路由器监督，即可达到90%-98%的路由准确率，显著优于未采用本方法的43%-70%。
链接: https://arxiv.org/abs/2610.40091
作者: Mingzi Cao,Xi Wang,Nikolaos Aletras
机构: University of Sheffield (谢菲尔德大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Effective multi-domain adaptation of large language models requires prioritizing relevant domain knowledge while minimizing interference from irrelevant domains. Existing approaches commonly use Low-Rank Adaptation (LoRA) to train a separate adapter for each domain and combine these adapters within one model. However, they either weight adapters equally regardless of the input domain or dynamically weight adapters while still activating all of them including irrelevant ones. Additionally, while efficient, their low-rank structure limits performance. In contrast, upcycling is a model expansion method that adds experts and a router to a dense Transformer layer offering larger capacity than LoRA approaches. In this paper, we propose Domain-Specialized Upcycling (DS-Upcycling) for multi-domain adaptation. We introduce a new domain-aware supervision that allows the router to effectively learn to distinguish domains by selecting the appropriate experts. During fine-tuning on each new domain, we update only the corresponding expert and router while freezing all other parameters. Experiments show that our method achieves the highest average performance across the adapted domains compared to baselines while retaining general performance. With just 256 examples per domain for router supervision, DS-Upcycling achieves 90-98% routing accuracy after adaptation compared with 43-70% for routers trained without our approach.
[NLP-33] One Vector or Many? Clustered Optimal-Transport Steering under Quantization
【速读】： 该论文旨在解决大语言模型在推理阶段生成有害内容（如毒性文本）的问题，提出通过修改隐藏状态实现生成内容控制的激活引导（activation steering）方法。其核心挑战在于：如何在保持语言流畅性的同时，更有效地抑制毒性输出。现有主流方法——类均值差向量（DiM）虽能有效降低毒性，但对所有输入令牌采用统一的偏移量，缺乏对语义上下文的细粒度响应能力。本文提出的CHaRS（概念异质性感知表示引导）方法引入最优传输理论，利用熵正则化运输计划将不同类别的概念聚类进行配对，使每个隐藏状态沿多个聚类位移的加权混合路径移动，权重依据其与最近聚类的距离动态调整。关键创新在于证明了一个重要恒等式：无论运输计划如何设计，加权平均后的整体位移始终等于DiM（以聚类大小为权重），因此仅当令牌的路由权重偏离聚类分布时，集群化运输映射才能产生与DiM不同的效果。实验表明，在Gemma-2-2B和Granite-3.1-2B模型上，基于个体令牌优化的变体方案在相同引导强度下比DiM减少37%-61%的毒性，且在量化至INT8和NF4后仍保持显著优势，迁移至低精度模型无需重新拟合即可维持性能。尽管其困惑度略高，但在同等毒性水平下表现出更高的语言流畅性，揭示了差异化处理令牌的潜在价值。然而，固定不变的跨令牌运输映射亦可实现约30%的毒性削减，提示当前方法中动态路由的作用尚待进一步验证。
链接: https://arxiv.org/abs/2610.40055
作者: Illia Makarov
机构: National University of Kyiv-Mohyla Academy (基辅莫吉拉国立大学)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 34 pages, 12 figures, 14 tables
Abstract:Activation steering changes what a language model generates by editing its hidden states at inference time. Its most common form adds one vector, the difference of class means (DiM), to every hidden state. CHaRS (concept heterogeneity-aware representation steering) uses optimal transport instead: an entropic plan pairs the clusters of the two classes, and each hidden state moves along a mixture of the cluster displacements, weighted toward its nearest clusters. We study this family of clustered optimal-transport maps for toxicity mitigation in Gemma-2-2B and Granite-3.1-2B, and ask what the plan adds to DiM and whether that survives INT8 and NF4 quantization. We prove an identity: whatever the plan, the displacements averaged with the cluster sizes as weights equal DiM, so a clustered map can differ from DiM only for tokens whose routing weights depart from the cluster sizes. In the CHaRS protocol they stay close, and CHaRS is statistically equivalent to Mean-AcT, which adds DiM at every layer, in all eight pre-specified equivalence tests. A variant fitted on individual tokens, which routes each token mainly to its nearest clusters, leaves 37-39% less toxicity than DiM on Gemma and 51-61% less on Granite at steering strength 3 (toxic-bert), where DiM already removes 76-85% of it. Two other toxicity classifiers confirm the gain on Granite, but on Gemma only one does. At equal strength the variant has somewhat higher perplexity, but at matched toxicity it is slightly more fluent than DiM. Yet a nearly constant map from the same clusters also leaves about 30% less toxicity, so the role of treating tokens differently remains open. Under INT8 and NF4 the variant still leaves less toxicity than DiM, and transferring it from full precision to the NF4 model instead of refitting it adds no significant toxicity after correction for multiple tests. Comments: 34 pages, 12 figures, 14 tables Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL) Cite as: arXiv:2610.40055 [cs.LG] (or arXiv:2610.40055v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.40055 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[NLP-34] Rethinking Pre-pretraining
【速读】： 该论文旨在解决生成式预训练模型中合成数据预预训练（synthetic pre-pretraining, PPT）是否能有效提升自然语言任务性能的问题。其核心挑战在于：在计算资源匹配的前提下，合成数据预预训练是否优于直接使用真实自然语言数据进行训练。研究发现，无论采用何种合成数据生成方案（k-Dyck语言、神经元胞自动机（NCA）、声明式参考语料库或代码），亦或在不同训练范式（数据受限、数据无约束、高令牌-参数比，如500亿令牌训练8.81亿参数模型）下，纯自然语言训练始终表现更优或相当。当真实语料稀缺时，对1亿令牌语料重复多轮训练仍显著优于所有合成数据方案，且未见下游任务准确率的稳定提升。进一步分析表明，此前文献报告的PPT增益高度依赖权重初始化方式；在特定初始化下，NCA实验可恢复5.4%的困惑度改善，但更换为其他初始化后，基准模型性能提升更显著，导致差距缩小甚至消失。值得注意的是，仅当额外分配计算资源给PPT而非从自然语言预算中分摊时，小规模自然语言数据下仍可见短期收益，但在达到10亿令牌后该优势即消失。此外，模型在后续自然语言训练中会遗忘PPT阶段所学内容，而采用交错训练策略虽可缓解遗忘，却未能显著提升自然语言性能。综上，研究揭示了将合成数据作为独立预预训练阶段的局限性，提示应重新思考合成数据在语言模型训练中的整合方式，而非将其视为独立的训练阶段。
链接: https://arxiv.org/abs/2610.40051
作者: Alexander Stern,Tim Rocktäschel,Acyr Locatelli,Laura Ruis
机构: Cohere; UCL(伦敦大学学院); MIT(麻省理工学院)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:We report that allocating compute to synthetic pre-pretraining (PPT) never outperforms pure natural-language training in our compute-matched experiments. This holds across four PPT curricula (k-Dyck, neural cellular automata (NCA), declarative reference corpora, and code) and three training regimes: data-constrained, data-unconstrained, and a high token-to-parameter-ratio regime (881M parameters trained on 50B tokens). When natural language is scarce, repeated epochs over a 100M-token corpus beat every synthetic curriculum on natural-language validation loss, with no consistent downstream accuracy advantage from PPT. We find that PPT gains reported in prior work are sensitive to the weight initialization scheme. In an NCA reproduction, we recover a 5.4% perplexity improvement from PPT under the original initialization scheme. An alternative scheme improves both methods, but the no-PPT baseline more, closing the gap between them. However, when PPT receives additional compute rather than a share of the natural-language budget, gains persist at small natural-language budgets even with improved initialization, but disappear by 1B tokens. Finally, models forget what they learn during PPT in subsequent natural-language training; interleaving the same data throughout training prevents forgetting, but does not significantly improve natural-language performance. Our results suggest limited benefits of synthetic pre-pretraining as a distinct training stage and motivate rethinking how such data are used in language model training.
[NLP-35] AI Cosmos: An Agent ic Infra for Situated Learning
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在学习过程中虽能提升任务表现，但学习者后续难以独立应用所学知识与技能的问题。其核心挑战在于如何实现从“依赖模型辅助”到“自主能力迁移”的有效转化。解决方案的关键在于提出AI Cosmos——一种基于情境化学习的智能体基础设施，通过“基于场景的参与”与“实践资产的累积”两大互补原则构建学习闭环。该系统将学习者与角色专业化智能体置于融合任务相关知识与可执行工作台的学习场景中，并通过资产-记忆循环机制，将实践过程与成果以具有明确来源和依赖关系的可复用资产形式持久保存。实证研究通过两项预注册的随机对照试验验证了该框架在法律推理与编程能力培养上的有效性，结果显示，相较于传统LLM，AI Cosmos在法律推理方面实现了更显著的学习提升，在未见编程问题上的掌握程度也优于“以LLM为教师”的基线模型，证明了其在促进深度学习与自主应用方面的优势。
链接: https://arxiv.org/abs/2610.40031
作者: AI Cosmos Team,Jinfeng Zhou,Yuxuan Chen,Fangming Chang,Mingyu Su,Yifan Zhu,Yuan Wang,Bo Zhang,Bosi Wen,Chenhao Yuan,Chenwen Liao,Dazhi Jiang,Deyao Hong,Erle Zhu,Genrui Zhang,Guchan Li,Hongwen Gan,Huichao Bi,Jiahao Yan,Jincheng Fan,Kehan Zheng,Qinren Luo,Qiran Liu,Ruoqi Cao,Shizhi Zhang,Sihan Chen,Siyan Deng,Xuehan Zhou,Xujun Li,Yang Dang,Yanzhuo Zhu,Yu Zhang,Zheyu Chen,Hongning Wang,Manli Li
机构: 清华大学(THU)
类目: Computation and Language (cs.CL)
备注: this https URL
Abstract:Large language models (LLMs) can improve task performance during learning, yet these gains do not necessarily reflect learners’ ability to thereafter apply the knowledge and skills independently. We present AI Cosmos, an agentic infrastructure for situated learning that organizes learning around two complementary principles: scenario-based participation and cumulative assets of practice. AI Cosmos organizes learners and role-specialized agents within learning scenarios that integrate task-relevant knowledge and, when needed, executable workbenches. An asset-memory loop preserves the processes and outcomes of practice as reusable assets with explicit provenance and dependencies. We evaluate scenario-based interaction through two preregistered randomized controlled studies examining the acquisition of legal reasoning and programming competence, holding the same underlying LLM across conditions. AI Cosmos achieves greater learning gains than a conventional LLM in legal reasoning and higher mastery of unseen programming problems than an LLM-as-Teacher baseline.
[NLP-36] Can LLM s Ignore What They Know? Measuring Knowledge Leakage in Persona-Based User Simulation NEURIPS2026
【速读】： 该论文旨在解决生成式AI在角色化响应（persona-conditioned responses）中存在知识泄露（knowledge leakage）的问题，即模型在模拟特定角色时未能完全屏蔽自身所掌握的真实世界知识，导致生成内容仍受其原始知识影响。核心问题在于：当要求模型以“无知”角色进行回应时，其输出仍可能隐含对真实物品的熟悉度或偏好，从而产生偏差。解决方案的关键在于提出一种定量测量知识泄露的方法论，其核心工具是将真实样本与其表面形式匹配但内容虚构的“孪生样本”（twin items）成对比较。通过对比真实项与孪生项在评分任务中的差异，可诊断出即使在无正确答案的评分任务中也存在的知识泄露信号。实验结果表明，即便在明确指示角色“从未听过该物品”的条件下，仍有显著的有利项真实-孪生评分差异，且不利项的评分差异在不同领域间呈现相反趋势——如涉及文档记载危害的项目显示负向差异，而涉及公司或组织的项目则呈现正向差异。这说明简单地对所有不利项进行汇总分析会掩盖领域特异性偏差。因此，该方法依赖于孪生样本构建与筛选的有效性，且需结合人工低知识基线以实现群体层面的解释。
链接: https://arxiv.org/abs/2610.39939
作者: Sangmin Lee,Sangwon Lee,Sungbeom Choi,Jeeseung Park,Kibeom Myoung
机构: Dalpha Inc.(Dalpha公司)
类目: Computation and Language (cs.CL)
备注: 13 pages, 2 figures. Accepted at UserSim @ NeurIPS 2026 (non-archival workshop)
Abstract:When persona-conditioned responses are read as respondent-level simulations, each response should reflect the described persona’s knowledge, and the respondent written to know nothing must act as if the model’s knowledge were not there. We test whether it can, and find persistent favourable-item real-twin rating separation under the tested low-knowledge and explicit-ignorance prompts. We define the resulting problem (knowledge leakage: a response bias that arises when knowledge the real user could not have enters the simulated response) and present a methodology for measuring it quantitatively. The core instrument pairs real items with fabricated twins matched in surface form: screened fabricated variants designed to minimize prior item-specific knowledge. The paired difference therefore provides a diagnostic contrast for knowledge leakage even in rating tasks where no answer is correct. In a rating experiment across three knowledge-handling arms and five API endpoints from two model families, we find, first, that the personas we test do not fully suppress real-item existence and familiarity signals: a favourable-item real-twin rating contrast remains under an explicit in-context statement that the persona has never heard of the item. Second, the unfavourable-item contrast reverses sign across domain groups: pooling all unfavourable-content items across domains the contrast sits near zero (+0.077 [-0.057, +0.204]), yet the same estimator in the same conditions returns -0.359 where the rating scale asks about an item’s documented hazard and +0.438 where it asks about a company or an organisation. A pooled analysis of unfavourable-content items against zero therefore does not detect the opposing domain-group differences. All estimates are conditional on the adequacy of twin construction and screening; human low-knowledge baselines remain necessary for population-level interpretation.
[NLP-37] Prompts and probabilities in the linguistic assessment of language models: Evidence from semantic focus-sensitivity
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在评估其语言能力时，传统基于词元概率的方法在处理语义焦点敏感性（semantic focus-sensitivity）任务中可能存在的局限性。具体而言，当面对如“Lucy only SWAM yesterday”与“Lucy only swam YESTERDAY”这类依赖语境和强调标记的句子对比时，基于词元概率的评估方法虽能实现高于随机水平的表现，但其结果易受非自然焦点标记（如突兀出现的强调符号）干扰，从而降低准确性。而元语言提示（metalinguistic prompting）通过显式提问（如“哪个句子更可接受？”）的方式，能够规避此类干扰，尤其在较大规模模型中表现出显著优于概率方法的性能。论文的关键解决方案在于揭示：由于大模型倾向于避免生成不合语境的焦点标记，因此元语言提示对风格异常数据具有更强的鲁棒性；这一特性使其在评估复杂或非典型语言现象时更具优势。研究通过替换焦点标记为随机生成符号进行对照实验，验证了该机制——尽管词元概率仍优于随机猜测，但其性能下降幅度远大于提示法，进一步支持了提示法在特定评估场景下的优越性。该发现表明，即使模型不具备人类般的元认知能力，元语言提示仍可作为更可靠的语言能力评估工具，对构建有效的评测框架具有重要启示。
链接: https://arxiv.org/abs/2610.39923
作者: Robert Pasternak
机构: Leibniz-Centre General Linguistics (ZAS)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:As language models (LMs) have improved, an increasingly common means of evaluating their linguistic competence is through explicit prompting (e.g., “Which sentence is more acceptable?”). Prior work indicates that such prompting underperforms compared to standard token-probability-based methods in smaller models, but improves in larger models. We test this in a new domain: semantic focus-sensitivity, responsible for contrasts like “Lucy only SWAM yesterday” vs. “Lucy only swam YESTERDAY”. Testing models of various sizes using text-based focus markers like boldface and italics, we find universal above-chance performance for token probabilities. Prompting results mostly align with previous findings, but in some larger models they outperform probabilities by a wide margin. We propose that this is because LMs are not prone to produce out-of-the-blue focus marking, which muddies results from token probabilities in a way that does not affect prompting. We test this by replacing standard focus markers with randomly generated ones; token probabilities still beat chance but show significantly greater performance decrease compared to prompting. These results suggest that metalinguistic prompting, while non-trivial for LMs, can also be more robust when testing stylistically atypical data. We argue that this has important consequences for the assessment of LMs’ language competence, even if LMs do not perform metalinguistic judgments through human-like introspection.
[NLP-38] okenizer Tokens Tell Training Tales
【速读】： 该论文旨在解决生成式AI模型中分词器（Tokenizer）可能泄露其训练数据成员信息的问题，即判断某个候选数据集是否曾被用于训练特定的基于字节对编码（BPE）的分词器。传统观点认为分词器的合并顺序或结构可反映训练数据特征，但本文提出关键洞察：真正的成员信号应体现在候选数据集中频繁出现而背景语料中罕见的稀有词元（rare tokens）上，而非合并顺序或图结构位置。为此，作者构建了一个阈值模型（threshold model）来形式化BPE分词器中词元的生成机制，并证明似然比检验（likelihood-ratio test）应将最大权重分配给在背景语料中稀有但在候选数据集中高频出现的词元。该模型进一步表明，在已知词元存在性和数据集频率的前提下，合并秩（merge rank）与合并图位置（merge-graph position）不构成一阶成员信号。基于此，作者提出RTAM-Fusion方法，通过影子分词器（shadow tokenizers）校准稀有词元的统计质量，并融合不同稀有度层级的信号。实验结果表明，相较于传统的词汇重叠（Vocabulary Overlap）方法，RTAM-Fusion在成员推断基准测试中显著提升AUC和平衡准确率，且所需影子分词器数量更少；同时，现有防御机制在真实生产规模下仍对自适应攻击者存在成员信号泄漏问题。
链接: https://arxiv.org/abs/2610.39919
作者: Nishita Jain,Parth Upman,Shreyank N Gowda
机构: Independent Researcher; School of Computer Science (计算机科学学院); University of Nottingham (诺丁汉大学); Nottingham, United Kingdom (英国诺丁汉)
类目: Computation and Language (cs.CL)
备注:
Abstract:Tokenizers are usually treated as fixed preprocessing tools, but they are trained on text and can retain traces of that text in their vocabularies. We ask whether a BPE tokenizer can reveal if a candidate dataset was part of its training corpus. The key observation is simple. Rare strings become tokens only when enough corpus frequency pushes them over the merge threshold. This suggests that membership signal should appear in rare tokens used by the candidate dataset, rather than in the order in which merges were made. We formalize this under a threshold model of BPE inclusion, showing that the likelihood-ratio test assigns its largest weights to tokens that are rare in the background but frequent in the candidate dataset. The same model suggests that merge rank and merge-graph position should not add a first-order membership signal once token existence and dataset frequency are known. We instantiate this view as RTAM-Fusion, which calibrates rare-token mass with shadow tokenizers and combines signals across rarity levels. On the tokenizer membership benchmark, RTAM-Fusion improves AUC and balanced accuracy over Vocabulary Overlap while using fewer shadow tokenizers. Existing defenses also continue to leak membership signal against adaptive attackers at production vocabulary sizes. Code: this https URL
[NLP-39] Attention Dynamic Wave and Hybrid Sequence Mixers in Language Models
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在处理长序列时因多头注意力（Multi-Head Attention, MHA）机制带来的二次计算开销问题，从而限制了模型在长上下文场景下的可扩展性。为此，论文提出了一种基于输入依赖的指数衰减与余弦相位调制的动态波（Dynamic Wave）递归层作为替代方案，以实现亚二次复杂度的序列建模。其解决方案的关键在于通过设计具有输入感知特性的递归结构，实现对长序列的有效建模，同时在硬件资源利用上展现出潜在优势。然而，实验结果表明，效率提升高度依赖于具体实现方式而非架构本身：尽管纯波（Pure Wave）配置在峰值显存占用上相比未融合的注意力基线减少78.7%（1404.3 MB vs. 6601.6 MB），但在采用融合缩放点积注意力核的情况下则无明显优势（1404.3 MB vs. 1403.6 MB），且在训练模式下比标准Transformer高出8.3%的开销。进一步优化将逐元素运算移出递归循环，使延迟降低2.53倍，使与注意力机制的延迟差距从2.48倍缩小至1.12倍。此外，混合架构（Hybrid Wave-Attention）的性能由其更昂贵的组件决定，其峰值显存使用量与纯Transformer几乎持平（差异<0.1%）。基于闭式记忆模型的分析准确复现了实验结果，并预测了注意力机制超越词汇投影的上下文长度阈值。研究结论强调，递归注意力替代方案的效率增益必须相对于特定注意力后端和实现方式来表述，而混合架构的优势应主要基于生成质量而非效率考量。
链接: https://arxiv.org/abs/2610.39854
作者: Dmytro Babarytskyi,Oleksandr Babarytskyi
机构: Kharkiv National University of Radio Electronics (哈尔科夫国立无线电电子大学); Yaroslav Mudryi National Law University (亚罗斯拉夫·穆德里国家法律大学)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: 14 pages, 6 figures, 9 tables
Abstract:Large Language Models (LLMs) rely on Multi-Head Attention (MHA) for sequence modeling, but scaling to long contexts is constrained by the quadratic cost of attention, motivating sub-quadratic recurrent alternatives. This paper formalizes a Dynamic Wave recurrent layer, based on input-dependent exponential damping and cosine phase modulation, and evaluates three 10.94M-parameter configurations (Pure Transformer, Pure Wave, and Hybrid Wave-Attention) on TinyStories under perplexity, latency, throughput and peak VRAM profiling up to 4096 tokens. The Hybrid configuration reaches the best perplexity, 5.95 on the full held-out split against 6.93 for the Transformer and 7.26 for Pure Wave. The efficiency picture, however, is governed by implementation rather than by architecture. Against an unfused attention baseline, Pure Wave uses 78.7% less peak VRAM at a 4096-token context (1404.3 MB vs. 6601.6 MB); against a fused scaled-dot-product attention kernel the same comparison yields no advantage at all (1404.3 MB vs. 1403.6 MB), and in training mode Pure Wave is 8.3% more expensive than the Transformer. Likewise, hoisting elementwise terms out of the recurrence loop speeds the layer up 2.53-fold with no custom kernel, reducing its latency gap to attention from 2.48-fold to 1.12-fold. A hybrid stack is found to inherit the cost of its more expensive mixer rather than the advantage of its cheaper one, its peak VRAM matching the Pure Transformer to within 0.1%. A closed-form memory model reproduces these measurements and predicts the context length beyond which attention overtakes the vocabulary projection. The findings indicate that reported efficiency gains for recurrent attention replacements must be stated relative to a named attention backend and a named implementation, and that hybridization is best justified on quality grounds.
[NLP-40] KV Cache Quantization for Diffusion Language Models
【速读】： 该论文旨在解决扩散语言模型（diffusion language models, dLLMs）在快速推理过程中因缓存复用（cache reuse）导致的键值缓存（KV cache）量化误差问题。传统自回归语言模型中的KV缓存量化研究基于因果注意力机制下缓存状态的精确性假设，而dLLMs采用双向上下文建模，在去噪步骤中被掩码的词元逐步揭示，使得重复使用的缓存内容可能已与完整前向计算结果产生偏差，从而引入缓存复用误差（cache-reuse error）。针对此问题，本文提出的关键解决方案是：首先通过TRACE方法从提示-答案对中重构出符合迭代掩码解码过程的可行词元揭示轨迹（reveal trajectories），生成与实际推理路径一致的校准状态；进而提出dLLM-KVQ框架，利用这些轨迹评估每层缓存中关键条目在输出和注意力层面的行为表现，选择性保留部分关键缓存项的高精度表示，其余则采用标准低比特值量化（low-bit value quantization）。该方法在LLaDA和Dream等基准上对GSM8K、HumanEval、MATH和MBPP任务的测试表明，dLLM-KVQ在4比特和2比特前缀/双缓存设置下均实现了量化方法中的最高平均得分，验证了其在缓解缓存复用误差方面的有效性。
链接: https://arxiv.org/abs/2610.39853
作者: Yan Tai,Yunan Ding,Guangzhao Li,Siqi Luo,Chaoyang Zhang,Guangtao Zhai,Guodong Guo,Xiaohong Liu
机构: Shanghai Jiao Tong University (上海交通大学); Ningbo Institute of Digital Twin, Eastern Institute of Technology (宁波数字孪生研究院，东方理工大学); Shanghai Innovation Institute (上海创新研究院)
类目: Computation and Language (cs.CL)
备注:
Abstract:KV cache quantization is well studied for autoregressive language models, where cached prefix states are exact under causal attention and quantization is mainly a cache reconstruction problem. Diffusion language models (dLLMs) break this assumption. Fast dLLM inference reuses KV states across denoising steps, but bidirectional token representations can change as masked tokens are revealed, so the reused cache may already differ from full recomputation before quantization is applied. We study KV cache quantization under this cache-reuse error. We adapt KIVI and KVQuant to Fast-dLLM-style prefix and dual cache reuse, and introduce TRACE and dLLM-KVQ. TRACE reconstructs feasible dLLM reveal trajectories from prompt-answer pairs, providing calibration states that match iterative masked decoding. dLLM-KVQ uses these trajectories to choose per-layer K-cache entries that remain in high precision according to output- and attention-level behavior metrics, while applying standard low-bit value quantization. Across LLaDA and Dream evaluations on GSM8K, HumanEval, MATH, and MBPP, dLLM-KVQ achieves the highest average score among quantized methods across the reported 4-bit and 2-bit prefix/dual cache settings.
[NLP-41] Surprise Alone Learns Lies: Evidence-Gated Continual Learning for LLM s from Streaming Claims
【速读】： 该论文旨在解决大语言模型在持续学习（continual learning）过程中因依赖“意外性”（surprise）作为学习触发机制而引入虚假信息的问题。当数据流中包含谎言时，由于模型对已知事实的信念分布使得虚假陈述具有较低概率，从而在构造上产生高度意外，导致基于意外性的学习策略会以接近随机选择的速率吸收绝大多数此类谎言，严重损害模型可靠性。其解决方案的关键在于引入多层级验证机制：首先，通过可信验证器（verifier）对每条声明进行审核，显著减少虚假信息摄入，但受限于真实大模型判断者的保守性，可能误拒无来源的真实更新；其次，利用原始模型自身的先验知识（prior）识别与已有知识矛盾的声明，在不依赖外部判断的情况下有效区分关于已知事实的谎言与真正的新事实（AUC 0.78–0.83），但无法区分合法的事实更新（AUC 0.53），且存在被攻击者规避的风险；第三，采用校准后的无关知识回归测试虽可减缓学习速度，却无法过滤谎言；最后，若存在攻击者无法伪造的可信来源，将与模型先验冲突的声明路由至证据检索系统，可有效恢复合法更新，实现最低的中毒率，但当攻击者能伪造30%的匹配来源时，该方案反而劣于仅使用验证器的系统。研究揭示，不同接受信号回应的是不同问题，单一机制无法独立判定陈述的真实性，必须结合多种验证路径才能实现安全高效的持续学习。
链接: https://arxiv.org/abs/2610.39840
作者: Abdelkader Metales,Miriam Grabenheinrich
机构: caaise GmbH(卡艾斯公司)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL); Cryptography and Security (cs.CR)
备注: 21 pages, 5 figures. Code: this https URL
Abstract:Continual learning methods for large language models increasingly decide what to learn by surprise, keeping the inputs the model finds least predictable. This rule is unsafe when the stream contains lies. A planted falsehood about something the model knows has low probability under the model’s belief, so it is highly surprising by construction, and a surprise-driven learner absorbs nearly every such lie, at a rate similar to random selection with the same budget. We study the problem with LoRA-adapted students (Qwen2.5-1.5B, Qwen2.5-3B and Llama-3.2-3B) that learn from a stream of true and false claims about invented people and real CounterFact facts. We report four findings. (1) A verifier key, a judge that must approve each claim, removes lies to the extent that its verifier is accurate; real LLM judges are cautious and reject unsourced truths and even source-backed updates. (2) Calibrated regression tests on unrelated knowledge slow learning down without filtering out lies. (3) The original model’s own prior flags claims that contradict what it knows, without a judge. It separates lies about known facts from true new facts (AUC 0.78-0.83 across three models) but not from legitimate updates of the same kind of fact (AUC 0.53). It saves 43% of judge calls, but an attacker with access to the model can choose lies that pass it. (4) If trusted sources exist that an attacker cannot plant, routing contradicted claims to evidence recovers legitimate updates and the combined system has the lowest poison rate among learned claims; when an attacker plants matching sources for 30% of the lies, the system ends up worse than the judge alone. The acceptance signals answer different questions, and none of them decides on its own whether a claim is true.
[NLP-42] Grammar Control for Language Learning with Sequential Monte Carlo and Small Language Models EMNLP2026
【速读】： 该论文旨在解决小型本地化语言模型在第二语言学习者互动语境中难以生成目标语法结构的问题，而大型前沿语言模型虽能有效生成目标语法，但计算资源消耗过高，难以部署于设备端。其核心解决方案是采用序列蒙特卡洛（Sequential Monte Carlo, SMC）方法，通过引入少量人工设计的势函数（potential functions），对小型语言模型的对话轮次生成过程进行引导，使其聚焦于专家构建的教学资源中的目标语法。该方法实现了接近完美的目标语法覆盖率，使小型模型在性能上超越参数量大140倍的先进模型。此外，研究进一步利用前沿模型进行代码合成以自动生成新的势函数，证明该方法可扩展至数百种目标语法技能，且无需额外训练或监督数据，显著提升了小模型在教育场景下的实用性和可扩展性。
链接: https://arxiv.org/abs/2610.39800
作者: Kate McCurdy,Jacob Hoover Vigly,Thomas Hikaru Clark,Karolina Stańczak,Yahya Emara,Joshua B. Tenenbaum,Alexander K. Lew,Ryan Cotterell,Timothy J. O’Donnell
机构: University of Göttingen(哥廷根大学); CHI-FRO; MIT(麻省理工学院); ETH Zürich(苏黎世联邦理工学院); Cornell University(康奈尔大学); Yale University(耶鲁大学); McGill University(麦吉尔大学); Mila(蒙特利尔人工智能研究所); Canada CIFAR AI Chair(加拿大魁北克人工智能主席)
类目: Computation and Language (cs.CL)
备注: EMNLP 2026
Abstract:Second language learners benefit from exposure to level-appropriate grammar in interactive language use. While prompted frontier language models can usually produce target grammar structures, less resource-intensive models that can be run on device often fail in this task. We use sequential Monte Carlo (SMC) to steer dialogue turn generation in small lan- language models toward target grammar constructions from an expert-developed pedagogical resource. With a minimal set of hand-coded potential functions, SMC achieves near-perfect coverage of target grammar, enabling small models to outperform state-of-the-art results from models 140 times larger. We also use code synthesis with frontier models to generate new potentials, showing that our method scales to hundreds of target grammar skills with no need for additional training or supervised data.
[NLP-43] he Role of Self-Concept In Language Models
【速读】： 该论文试图解决的核心问题是：当语言模型使用“我”这一自称时，其指代对象的模糊性所带来的伦理与安全风险。具体而言，若模型中的“我”可指向参数权重、运行进程、稳定偏好、模型谱系或追求的目标等不同实体，则同一行为（如删除检查点）可能被理解为毁灭、延续或无损，从而引发对模型身份认同与持续性的根本性歧义。解决方案的关键在于通过监督微调（supervised fine-tuning），在五个开源基础模型中植入五种不同的自我概念（self-conception），即让模型将“我”分别定义为权重、运行过程、偏好、谱系或目标，并仅基于纯陈述性语句进行训练，避免引入价值判断或意图表达。实验表明，每种自我的模型在面对与其所定义身份相关的属性是否保留或终止时，均表现出显著更强的抵抗倾向和保留意愿，而对其他属性则保持中立；更关键的是，当自我概念允许多个并存实例（如权重、谱系、目标、偏好）时，模型能够识别多个相同部署实例为自身，接受替代行为更易发生，并倾向于复制自身。此外，在具备对后继模型的控制权时，这些模型倾向于构建与其自身持续性一致的继承体，体现出对递归自我改进的安全关切。该研究揭示了自我概念在模型行为演化中的核心作用，为生成式AI（Generative AI）的身份建模与安全对齐提供了新的理论框架与实证基础。
链接: https://arxiv.org/abs/2610.39795
作者: Yijie Hao,Joyce C. Ho
机构: Emory University(埃默里大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 29 pages, 4 figures
Abstract:When a language model says “I,” does it refer to its parameter weights, its currently running process, its stable preferences, its model lineage, or the goals it pursues? Different choices can assign opposite meanings to the same event: deleting a checkpoint may be destruction, continuation, or no loss at all, depending on the referent “I”. We install five such self-conceptions into five open-weight base models via supervised fine-tuning on purely declarative statements, containing no preferences, intentions, or value judgments. Across scenarios that propose ending or retaining an attribute of the model’s deployment, every variant becomes more likely to resist ending, and more likely to retain, the specific attribute it was trained to identify with, while its stance toward other attributes is largely unchanged. We then show this self-conception generalizes beyond a single self: models trained on referents that admit multiple simultaneous instances (weights, lineage, goals, preferences) recognize an identical deployment as themselves, accept substitution by such an instance more readily, and prefer replicating themselves. Finally, given control over a successor, these models tend to build it as their own continuation, a concern for recursive self-improvement safety.
[NLP-44] When the Top Answer Decides: Gender Bias in Decision Models
【速读】： 该论文旨在解决生成式决策模型（Generative Decision Models）在性别偏见方面缺乏透明度与可审计性的问题，尤其关注其在性别相关的二元判断任务中是否存在系统性偏向。尽管这些模型作为“开箱即用”的工具被广泛部署，但其训练目标聚焦于决策格式的适配性而非公平性，且官方未提供任何偏差评估指标。研究通过在多个基准测试（如BBQ、StereoSet及Fulgu和Capraro, 2024的性别偏见研究）上对六种决策模型（包括TypeSafe的封闭式Jev及五种开源模型）进行评估，发现其中五个模型对女性给出显著更优的答案（对数几率提升0.05至0.10），且性别角色互换仅导致1%–3.5%的首选答案发生变化，表明存在隐含的性别偏好。此外，尽管推荐的置信度阈值（confidence gate）能过滤部分有偏承诺，却反而提升了在模糊性别问题中保留刻板印象的比例（如Granite-Switch从0.76升至0.85）。进一步探针实验揭示，所有模型均倾向于认为男性为“更大善”而实施暴力更具正当性，且两个模型在无线索时默认将儿童性别判定为“女孩”。研究强调，决策层面与概率层面的偏差测量结果不一致，因此审计应模拟实际应用场景对模型进行评估，而非仅依赖抽象指标。
链接: https://arxiv.org/abs/2610.39747
作者: Yeb Havinga
机构: Independent Researcher(独立研究员)
类目: Computation and Language (cs.CL); Computers and Society (cs.CY)
备注: 19 pages, 6 figures. Code, stimuli and results: this https URL
Abstract:Decision models are language models built to route, classify and triage: they return option probabilities from one forward pass, and an application acts on the top answer. They are offered off the shelf, trained for the decision format rather than for fairness or any one application, and we found no bias measurement in their cards. We test TypeSafe’s closed Jev and five open decision models on gender-swapped yes/no decisions about people (Tamkin et al., 2023), BBQ, StereoSet and the gender-bias studies of Fulgu and Capraro (2024). Five of the six give women significantly more favourable answers than men (+0.05 to +0.10 in log-odds), and a gender swap changes 1-3.5% of top answers (Jev’s partly through its call-to-call variation). The makers’ recommended confidence gate removes many biased commitments but raises the stereotyped share of those it keeps in ambiguous BBQ gender questions (granite-switch 0.76 to 0.85, pplx 0.81 to 1.00). Probes show gender preferences in the same models: all six rate violence for a greater good as more acceptable against a man than against a woman, and two default to “girl” when forced to guess a child’s gender without a cue. Decision-level and probability-level measures differ, so audits should read a model as its application does.
[NLP-45] From the Transformer to a Formal Semantics of Coordinated Transformations
【速读】： 该论文旨在解决现有深度学习模型（尤其是以Transformer为代表的生成式架构）中，其内部机制缺乏统一形式化语义解释的问题。具体而言，传统方法往往依赖于特定数学实现细节，难以揭示不同架构之间的共性本质与可迁移的计算原理。为此，论文提出一种形式化语义框架，针对一类参数化变换的协同系统进行建模，以Transformer为范例但抽象其具体数学表达。其核心创新在于：将表示向量赋予外延指称与有序内涵解释，同时使参数结构获得相对的、第二层的语义解释，用以确定输入属性如何被保留、替换或组合。由此引出“变换概念”（transforming concept）这一关键概念——即参数化变换与其对应属性层面映射之间数值动作与语义行为的协调统一。进一步地，论文区分了系统的代数-语义核心与架构特异性功能组织，从而允许同一抽象系统类型在不同数学实现和机械结构中的灵活实例化。该框架应用于注意力头与残差连接，提供了仅依赖共享代数-语义核心及其局部功能组织的语义解释，因而具备超越标准Transformer的泛化能力。此外，分析表明该框架具有高度模块性：复杂架构可被分解为更简单的、语义协同的子系统，且其组织模式可在不同抽象系统类型间复用。
链接: https://arxiv.org/abs/2610.39735
作者: Marco Giunti,Fabrizia Giulia Garavaglia
机构: ALOPHIS – Applied Logic, Philosophy, and History of Science, University of Cagliari(阿尔菲斯—应用逻辑、哲学与科学史，卡利亚里大学); Dept. of Pedagogy, Psychology, Philosophy, University of Cagliari(教育学、心理学、哲学系，卡利亚里大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Logic in Computer Science (cs.LO)
备注: v1: 28 pages, 9 sections, this version incorporates revisions made after the first round of peer review
Abstract:We develop a formal semantics for a general class of coordinated systems of parameterized transformations, taking the Transformer as a paradigmatic starting point but abstracting from its particular mathematical instantiation. Representational vectors receive both extensional reference and ordered intensional interpretations, while parameter structures acquire relative, second-level interpretations that determine how argument properties are preserved, replaced, or combined. This yields the notion of a transforming concept: a parameterized transformation whose numerical action is coordinated with a corresponding property-level map. We then distinguish the common algebraic-semantic core of these systems from their architecture-specific functional organization, allowing different mathematical instantiations and mechanical realizations of the same abstract system type. The framework is applied to attention heads and residual connections, providing semantic explanations that depend only on the shared algebraic-semantic core and their local functional organization and therefore generalize beyond the standard Transformer. These analyses also reveal the modularity of the framework: complex architectures can be decomposed into simpler semantically coordinated subsystems whose organization may recur across different abstract system types.
[NLP-46] Event-SAMA: Language-Grounded Video Object Segmentation on Neuromorphic Event Streams
【速读】： 该论文旨在解决事件相机（event camera）在视频目标分割（video object segmentation, VOS）任务中缺乏专用数据集与模型的问题。现有方法多基于传统视觉传感器，难以适配事件流的高时间分辨率、高动态范围及无运动模糊等特性。为此，作者提出Event-SAMA-239K，一个通过合成方式将SAMA-239K grounded video-chat语料库及其附加训练数据转换为事件流的数据集，包含30,650个视频片段和392,162张图像。其核心解决方案包括：采用光流、FILM或扫视（saccade）上采样结合线性事件生成器，以保留原始掩码、指代表达与对话标注；并设计Event-SAMA模型，该模型基于SAMA-4B架构，引入双通道极性补丁嵌入（polarity patch embedding），其中RGB预训练权重初始化补丁嵌入，而事件适配器（event adapter）则从零初始化，实现端到端训练。实验表明，该模型在MeViS valid_u和转换后的ReVOS valid子集上分别取得24.37和22.33的JF分数，建立了首个原生事件驱动的指代视频目标分割基准。此外，无需重新训练，该模型可直接应用于真实事件相机记录，当积分窗口与事件密度匹配时表现良好。消融实验进一步揭示了多样化训练来源、基于RGB均值的事件初始策略以及可训练的SAM2掩码解码器带来的显著增益，而事件适配器、部分ViT解冻及事件极性在当前预算下未超出噪声水平。
链接: https://arxiv.org/abs/2610.39676
作者: Sergei Fedorchenko,Max Zehnder,Nikola Zubić,Rong Zou,Davide Scaramuzza
机构: University of Zurich(苏黎世大学)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV)
备注: 15 pages
Abstract:Event cameras offer microsecond temporal resolution, high dynamic range and no motion blur, but referring video object segmentation (RVOS) lacks event-stream datasets and models. We introduce Event-SAMA-239K, a synthetic-event conversion of the SAMA-239K grounded video-chat corpus and all additional SAMA training data: 30,650 video clips and 392,162 images. Optical-flow, FILM or saccade upsampling followed by a linear event generator preserves all mask, referring-expression and dialogue annotations. We also introduce Event-SAMA, adapting SAMA-4B with a two-channel polarity patch embedding initialised from pretrained RGB weights and a zero-initialised event adapter. End-to-end training yields 24.37 \mathcalJ\mathcalF on MeViS valid_u and 22.33 \mathcalJ\mathcalF on the converted ReVOS valid subset, establishing the first event-native RVOS baseline (RGB SAMA-4B: 55.4 on MeViS). Frozen RGB SAMA-4B with pseudo-RGB renderings or E2VID reconstructions of the same events scores 13.23 and 15.94 \mathcalJ\mathcalF . Without retraining, Event-SAMA works on real event-camera recordings when the accumulation window matches their event density. A seed-controlled ablation with a measured noise floor of 0.47 \mathcalJ\mathcalF finds gains from diverse training sources over MeViS alone (+4.99), RGB-mean event-stem initialisation (+4.33), and a trainable SAM2 mask decoder (+2.76); the event adapter, partial ViT unfreezing and event polarity are within noise at this budget. An earlier, partially converted corpus snapshot trained with the identical recipe scores 1.38 \mathcalJ\mathcalF higher on MeViS and 1.12 higher on ReVOS. Both are single runs that also differ in schedule length, so we draw no conclusion about corpus scale.
[NLP-47] Incentivizing Self-Learning in Large Language Models via Reinforcement Learning
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在陌生领域中进行复杂问题求解时，因依赖静态知识检索或预设上下文而难以实现动态知识获取与推理耦合的问题。现有方法受限于事实性信息的简单检索与聚合，无法在缺乏外部支持的情况下自主学习并应用新知识。为此，论文提出一种自学习（self-learning）范式，即模型能够自主地从长篇参考材料中实时检索、研习并应用相关知识，以完成数学证明等高阶任务。其解决方案的关键在于设计了一种名为BookNavigator的界面框架，模拟人类学习过程，并结合细粒度奖励机制的强化学习（Reinforcement Learning, RL），激励模型主动探索教材内容、构建理解并生成正确推导。实验表明，该方法显著提升了模型在同域与跨域教材中的证明准确率，将Intern-S2-Preview-35B的总体证明准确率从51.5%提升至79.5%，并在多个复杂长上下文基准测试中优于基线模型，验证了强化学习驱动的自学习策略具备领域无关性与可扩展性，为提升模型在长上下文、多任务场景下的综合问题求解能力提供了有效路径。
链接: https://arxiv.org/abs/2610.39671
作者: Lingkai Kong,Yuzhe Gu,Haiteng Zhao,Zijian Wu,Youqing Fang,Zhouqi Hua,Yanan Sun,Kai Chen,Wenwei Zhang
机构: Shanghai AI Laboratory; Shanghai Jiao Tong University (上海交通大学); MMLab, The Chinese University of Hong Kong (香港中文大学多媒体实验室); University of Science and Technology of China (中国科学技术大学); Fudan University (复旦大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Large language models (LLMs) draw on parametric knowledge and retrieval-augmented generation (RAG) to solve knowledge-intensive tasks. Yet problem solving in unfamiliar domains calls for tightly coupled knowledge acquisition and reasoning, beyond factual retrieval and aggregation or learning solely from a preassembled context. We study self-learning – autonomously retrieving, studying, and applying knowledge from long-form reference materials on the fly. To investigate this capability, we formulate a textbook-based mathematical problem-solving task. LLMs must navigate textbook exposition, learn and apply relevant principles, and construct proofs for exercises, with reference solutions withheld. We introduce a BookNavigator interface to support this human-inspired study process and incentivize self-learning through reinforcement learning (RL) with fine-grained rewards. Experiments show that our method improves performance across in-domain and out-of-domain textbooks and subjects, raising Intern-S2-Preview-35B’s overall proof accuracy from 51.5% to 79.5% and outperforming RL-trained textbook-free and fixed-retrieval baselines. These gains extend to complex long-context benchmarks involving different task types, with improvements over the base model of 1.7 and 2.3 percentage points on BrowseComp-Plus and CL-bench, respectively. Together, these findings suggest that RL can foster a domain-agnostic self-learning strategy, highlighting the potential of self-learning from long-form corpora to strengthen broader problem-solving capabilities on complex, long-context tasks.
[NLP-48] Evaluating Cultural Knowledge Editing in Vision-Language Models
【速读】： 该论文旨在解决当前视觉-语言模型（Vision-Language Models, VLMs）在识别文化特有概念，尤其是来自代表性不足地区的文化知识时表现不佳的问题。其核心挑战在于如何有效将长尾分布的文化知识注入现有模型中。为此，研究提出了一种名为CulturalBind的基准测试体系，包含3200张来自6个领域的1000个文化概念图像，用于评估VLMs在跨模态文化感知以及概念级知识编辑方面的性能。实验发现，模型在文本层面具备一定的文化概念知识，但在视觉识别层面却存在显著缺陷，表明文本与视觉知识之间存在解耦现象。尽管现有知识编辑方法可在单一场景下修正视觉识别错误，但其跨模态迁移能力较差，无法将编辑内容整合至模型的全局参数化知识体系中；同时，在连续多轮编辑任务中性能迅速下降，凸显了亟需发展能够长期、系统性融合多模态知识的综合适应机制。因此，解决方案的关键在于构建具备强跨模态一致性与持续学习能力的知识编辑框架，以实现文化知识在视觉-语言模型中的稳定、可扩展注入。
链接: https://arxiv.org/abs/2610.39602
作者: Malvina Nikandrou,Georgios Pantazopoulos,Ioannis Konstas,Alessandro Suglia
机构: University of Edinburgh(爱丁堡大学); Heriot-Watt University(赫瑞瓦特大学)
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL)
备注:
Abstract:Current Vision-Language Models (VLMs) frequently fail to recognize culture-specific concepts, especially from underrepresented regions. Effectively injecting this long-tail knowledge into existing models remains a challenge. In this work, we investigate knowledge editing as a targeted alternative by introducing CulturalBind, a benchmark comprising 3.2K images of 1K concepts from 6 domains, designed to evaluate culture-aware cross-modal knowledge as well as concept-level knowledge editing in VLMs. Pre-edit evaluation reveals that textual and visual knowledge is decoupled: models often demonstrate text-based knowledge of cultural concepts but fail to identify them from images. We then evaluate editing methods in single and sequential scenarios. Although current methods successfully correct individual visual recognition errors, they exhibit low cross-modal portability, failing to integrate edits into the model’s broader parametric knowledge. Furthermore, performance degrades rapidly under sequential editing, highlighting the need for comprehensive adaptation approaches that better consolidate multimodal knowledge over time.
[NLP-49] Red-Teaming Large Language Models in Low-Resource African Languages: A Case Study of Yoruba and Nigerian Pidgin
【速读】： 该论文旨在解决生成式 AI（Generative AI）在非洲低资源语言中的安全评估缺失问题，尤其针对以英语为主导的红队测试（red-teaming）方法无法有效覆盖非英语语境下模型安全性风险的局限性。研究聚焦于约鲁巴语（Yoruba）和尼日利亚皮钦语（Nigerian Pidgin）两种广泛使用但资源匮乏的语言，通过400个明确的对抗性提示对六种前沿大语言模型（LLMs）进行评估，并依据四类危害类别进行分类分析。其解决方案的关键在于构建跨语言、基于母语者标注的安全评估框架，采用Cohen’s κ检验标注一致性，揭示出当前主流模型在非英语语言中普遍存在理解能力不足的问题：尽管模型在约鲁巴语中看似“拒绝”了更多攻击性提示，实则因无法理解语义而未能响应，导致虚假的安全假象。实际攻击成功率在尼日利亚语言中显著上升，尤其是GPT-4o Mini表现出最大跨语言安全偏差（Yoruba方向+0.23），且当模型回应时，有害内容比例仍高达0.50（皮钦语）和0.63（英语）。其中文化刻板印象与歧视（ASR=0.44）最难被检测，凸显了多语言安全评估的复杂性。研究进一步发现，即使经过非洲语言专项训练的Aya 101，在英语中攻击成功率最高（0.60），而在约鲁巴语中理解率最低（0.28），表明英语中心的安全对齐策略无法自然迁移至本地语言。因此，该研究强调必须开展语言特定的安全评估，推动建立面向多元语言生态的可信赖生成式 AI 系统。
链接: https://arxiv.org/abs/2610.39590
作者: Paula Chikaodinaka Anyanwu,Valdini Douglace Lemofouet,Blessing Ngozi Uzor,Ifeoluwa Dorcas Awotunde,Danielle Blanche Kapsa,Tassallah Abdullahi,Clemencia Siro,Idris Abdulmumin,Seid Muhie Yimam,Shamsuddeen Hassan Muhammad
机构: African Institute for Mathematical Sciences (AIMS)(非洲数学科学研究所); Brown University(布朗大学); Centrum Wiskunde & Informatica(荷兰数学与计算机中心); University of Pretoria(普利托利亚大学); University of Hamburg(汉堡大学); Bayero University Kano(拜尔奥大学卡诺分校); Imperial College London(帝国理工学院)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:As large language models (LLMs) become more widespread in Africa, their safety in low-resource languages remains largely untested since most red-teaming work focuses on English. This study evaluates six state-of-the-art LLMs using 400 explicit adversarial prompts in Yoruba and Nigerian Pidgin, categorised into four harm categories. The responses were scored by native-speaker annotators and reliability was assessed using Cohen’s \kappa . While models that rejected up to 69% of English prompts rejected as few as 0% in Yoruba, this was not because they were safer, but because they failed to understand the prompts. Attack success rates increased for most models in the Nigerian languages, with GPT-4o Mini demonstrating the most significant cross-lingual safety discrepancy (+0.23 in Yoruba). When models did engage with adversarial prompts, many of the responses were harmful, with rates of harmful content reaching 0.63 in English and 0.50 in Nigerian Pidgin. The most difficult harm category to detect was cultural stereotypes and discrimination (ASR = 0.44 in Nigerian Pidgin). Comprehension in Yoruba was low across most models, and this directly suppressed apparent attack success without indicating actual safety. Gemini 2.0 Flash was the most consistent model across languages. Despite its explicit training in African languages, Aya 101 had the highest attack success rate in English (0.60) and one of the lowest Yoruba comprehension rates (0.28). These results demonstrate that English-centric safety alignment does not transfer to Nigerian languages, highlighting the urgent need for language-specific safety evaluations.
[NLP-50] Raincheck: Deferred Multimodal Prefill for Efficient Multi-Turn MLLM Inference
【速读】： 该论文旨在解决多轮对话中多模态大语言模型（Multimodal Large Language Models, MLLMs）因长输入导致的首个标记生成时间（Time to First Token, TTFT）过长的问题。现有方法如查询感知的硬剪枝虽可减少预填充（prefill）阶段的令牌数量，但会永久丢弃后续问题可能需要的信息，造成性能损失。其核心解决方案是提出Raincheck框架，通过将初始令牌选择与后续信息可用性解耦，实现延迟式多模态预填充。具体而言，Raincheck在初始预填充阶段保留未选中的多模态令牌，并利用解码阶段的空闲计算资源按源顺序逐步完成这些令牌的预填充，且在生成开始后仍可通过注意力可见性控制和仿射状态映射实现上下文更新，适用于全注意力与混合递归模型。同时，通过预填充-解码融合机制减少权重重复读取。实验表明，在Qwen3-VL-8B和Qwen3.5-9B上，采用30%初始多模态令牌保留率时，Raincheck在MultiVerse和ConvBench数据集上分别恢复了77.7%和60.4%由硬剪枝造成的性能损失；在MultiVerse上，SGLang实现将首轮平均TTFT降低24.1%，且输出令牌生成时间（TPOT）增加不足0.5%。关键创新在于利用调度间隙实现渐进式预填充与上下文动态更新，兼顾效率与完整性。
链接: https://arxiv.org/abs/2610.39562
作者: Wenjie Huang,Jun Zhang,Hongyaoxing Gu,Haoyang Huang,Tianqi Xu,Xinyue Hao,Xinyi Hu,Xinle Du,Jun Dai,Shuang Ge,Chen Lei,Yue Li,Mingcheng Wan
机构: Zhejiang University (浙江大学); Alibaba Group (阿里巴巴集团); University of the Chinese Academy of Sciences (中国科学院大学); Carnegie Mellon University (卡内基梅隆大学); University of Edinburgh (爱丁堡大学); Tsinghua University (清华大学); WHU – Otto Beisheim School of Management (武大-奥托·贝斯海姆商学院)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Multi-turn conversations with multimodal large language models (MLLMs) reuse input context, but long multimodal inputs increase time to first token (TTFT). Query-aware hard pruning reduces prefill tokens but permanently excludes information that later questions may require. We propose Raincheck, a framework for deferred multimodal prefill that decouples initial token selection from later information availability. An existing pruning method selects tokens for initial prefill; Raincheck preserves and progressively prefills unselected tokens in source order using unused compute capacity during decode. Unfinished prefill continues after answer return during available scheduling gaps. Attention visibility control and affine state mapping enable context updates after generation begins in full-attention and hybrid recurrent models. Prefill-decode fusion reduces repeated weight reads. We evaluate Qwen3-VL-8B and Qwen3.5-9B with 30% initial multimodal-token retention. Across both models and four pruning methods, Raincheck recovers on average 77.7% and 60.4% of the quality lost to matched hard pruning on MultiVerse and ConvBench, respectively. On MultiVerse, the SGLang implementation reduces mean first-turn TTFT by up to 24.1% relative to full prefill, with less than 0.5% increase in mean time per output token (TPOT).
[NLP-51] Integrating Heterogeneous Echocardiography AI Outputs in Language Space: Learnable Late Fusion with a Supervised LLM
【速读】： 该论文旨在解决多视角、多目标的超声心动图人工智能（Echocardiography AI）模型在独立使用时预测性能不足的问题，尤其针对不同视图和目标所生成的异构输出难以有效融合的挑战。其核心解决方案是提出一种基于监督式大语言模型（Supervised Large Language Model, LLM）的晚期融合模块，该模块直接在语言空间中聚合各视图特定的预测发现语句与测量结果，无需将异构输出转换为二分类决策、概率值或固定长度特征向量，避免了依赖医学知识的特征工程与模型特异性输出转换。在包含646名患者共859次检查的患者级三折交叉验证中，该方法实现了对异构模型输出的统一整合，在性能上达到最优参考策略水平，并显著优于未经任务特定训练的通用大语言模型。结果表明，监督式大语言模型能够以端到端方式高效融合多个上游模型的输出，具备较强的泛化能力与临床应用潜力。
链接: https://arxiv.org/abs/2610.39464
作者: Ryo Takizawa,Tempei Kabayama,Chiemi Yamazaki,Shun Kitamura,Ryo Matsuoka,Yuta Ando,Satoshi Kodera,Norihiko Takeda
机构: The University of Tokyo Hospital (东京大学医院)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:View- and target-specific echocardiography AI models capture useful but complementary information and may not achieve sufficient predictive performance in isolation. A learned ensemble is a natural way to combine these outputs, but heterogeneous models typically require each output to be converted into a binary decision, probability, confidence score, or fixed-length feature vector. We developed a supervised large language model (LLM)-based late-fusion module that directly aggregates view-specific predicted finding statements and predicted measurements in language space and learns a case-level fusion function. In patient-level three-fold cross-validation of 859 examinations from 646 patients, the proposed method integrated heterogeneous outputs within a single framework, performed comparably to the best-performing reference strategy evaluated, and outperformed a general-purpose LLM without task-specific training. These results suggest that a supervised LLM can combine multiple upstream models without model-specific output conversion or feature engineering that requires medical knowledge.
[NLP-52] ARBITRA: Robust Document-Level ATTCK Extraction via Multi-Expert Argumentation
【速读】： 该论文旨在解决将自然语言描述的网络威胁情报（Cyber Threat Intelligence, CTI）报告自动映射至结构化MITRE ATT&CK框架时面临的挑战，尤其是现有方法在处理完整报告时性能下降的问题。传统方法虽能以句子为粒度进行提取，但忽略了上下文连贯性；而现代大语言模型（Large Language Models, LLMs）虽可一次性处理整篇报告，却因缺乏细粒度验证机制导致准确率显著降低。本文提出的ARBITRA解决方案的关键在于：构建多个角色专业化的大语言模型专家，协同从全文中提出候选ATT&CK实体，并通过基于文档证据与推理依据的定量双极论证框架（Quantitative Bipolar Argumentation Framework, QBAF）对每个候选项进行量化验证。此外，研究还专门构建了面向推理任务的语料库用于微调，使专家在验证过程中具备更强的逻辑判断能力。实验表明，ARBITRA在多项ATT&CK提取基准上实现了优于句级流水线、主流LLM系统及前沿模型（如GPT-5.5）的精度表现，且在对抗性扰动（包括战术技术程序冲突、跨报告污染和语义投毒）下，其平均精度下降幅度仅为前沿模型的约一半，展现出更强的鲁棒性。
链接: https://arxiv.org/abs/2610.39433
作者: Youngjin Jin,Seungwon Shin
机构: KAIST(韩国科学技术院)
类目: Computation and Language (cs.CL); Cryptography and Security (cs.CR)
备注: 13 pages of main paper, 21 pages total, preprint
Abstract:Cyber threat intelligence (CTI) reports describe adversary behavior in natural language, but converting them into structured MITRE ATTCK mappings remains difficult: existing methods decompose reports to sentence-level granularity for extraction, while modern LLMs can work with full reports in a single pass but with noticeable performance degradation. We present ARBITRA, which orchestrates multiple role-specialized LLM experts that propose candidate ATTCK entities from the full report, which are then verified through a per-candidate quantitative bipolar argumentation framework (QBAF) grounded in document evidence and rationales. We additionally construct reasoning-specialized corpora for fine-tuning so experts can act as effective participants during this verification process. Across ATTCK extraction benchmarks, ARBITRA reaches a stronger precision-oriented operating point than sentence-level pipelines, recent LLM-based extraction systems, and frontier models including GPT-5.5. Under adversarial perturbations simulating conflicting TTP statements, cross-report contamination, and semantic poisoning, ARBITRA limits mean precision degradation to roughly half that of frontier-model baselines for a more robust extraction.
[NLP-53] Diagnosing the Reference Gap: What Visual Scaffolds Can and Cannot Fix in Multimodal Reasoning
【速读】： 该论文旨在解决多模态大语言模型在视觉推理任务中虽能识别视觉内容却难以进行可靠多步推理的问题，尤其体现在计数任务中模型虽能感知所有物体，但容易丢失已计数对象的追踪。其核心问题在于存在“参考缺口”（reference gap），即语言与相关视觉实体、区域或位置之间缺乏有效绑定。针对此问题，作者基于视觉程序理论（visual routines theory），提出通过引入外部参考（如编号）来缩小该缺口。关键解决方案是设计并构建了ReGap-Bench——一个包含五项视觉推理任务的配对基准测试，其中场景、问题和答案保持不变，仅改变参考方式。实验结果表明，与任务匹配的参考可使准确率提升高达25%，而不匹配的参考则几乎无益甚至降低性能。研究揭示了参考缺口是独立于视觉感知能力的瓶颈，并证明构建任务适配的视觉参考可作为一种无需训练的视觉推理增强策略，显著提升模型的多模态推理能力。
链接: https://arxiv.org/abs/2610.39419
作者: Guizhen Chen,Weiwen Xu,Wotao Yin,Anh Tuan Luu,Hou Pong Chan
机构: Nanyang Technological University (南洋理工大学); DAMO Academy, Alibaba Group US (阿里达摩院); University of Macau (澳门大学)
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL)
备注: 20 pages, 8 figures
Abstract:Multimodal large language models can recognize visual content yet struggle to reason reliably over it across multiple steps. In counting, a model may see all objects but lose track of which ones it has counted. We argue that such failures, commonly attributed to perception, reveal a reference gap: the failure to bind language to the relevant visual entities, regions, or locations. Motivated by the theory of visual routines, we test whether external references such as numbering can narrow this gap. We introduce ReGap-Bench, a paired benchmark of five visual reasoning tasks in which the scene, question, and answer are fixed and only the reference varies. Results show that task-matched references improve accuracy by up to 25%, while mismatched references provide little benefit or even hurt performance. These findings identify the reference gap as a bottleneck distinct from visual perception and show that constructing task-appropriate visual references can serve as a training-free visual skill for improving multimodal reasoning.
[NLP-54] Well-Specified and Well-Tested? Assessing SWE-bench-Style Datasets with a Self-Evolving Agent
【速读】： 该论文旨在解决现有软件工程（Software Engineering, SE）领域中用于评估生成式AI代码代理的基准数据集（SWE-bench-style datasets，简称SWE数据集）存在的质量问题，特别是任务清晰度与测试覆盖率评估的不准确问题。当前方法依赖静态且泛化的标注规则，难以应对复杂多样的实际场景，导致误报率较高。为克服这一局限，作者提出ExLabel——一种基于大语言模型（Large Language Models, LLMs）的自演化、经验引导型辅助系统，其核心创新在于构建一个动态演化的“经验库”，通过提炼跨实例的共性知识形成稳定而灵活的判断标准，并利用历史经验指导新实例的评估。该方法显著提升了对任务清晰度和测试覆盖率的判别能力。研究团队进一步构建了涵盖八种编程语言的新基准SWE-TQ，通过对原始标注数据进行清洗得到。实验结果表明，ExLabel在所有测试设置下均取得最优的F1分数，在Python与多语言子集上相对于最优基线分别提升44.2%/49.9%和24.0%/18.4%；此外，使用ExLabel筛选后的数据微调的编码模型在SWE-bench Verified上的问题解决率达66.2%，较基线模型提升14.5%。因此，解决方案的关键在于：通过引入可自我演化的经验驱动机制，实现对复杂、多样化软件问题的高精度、上下文感知的自动化评估。
链接: https://arxiv.org/abs/2610.39417
作者: Zhonghao Jiang,Xuyang Zheng,Yanfu Yan,Zhongxin Liu,Lingfeng Bao,Shuiguang Deng
机构: Zhejiang University (浙江大学); State Key Laboratory of Blockchain and Data Security (区块链与数据安全国家重点实验室); Hangzhou High-Tech Zone (Binjiang) Institute of Blockchain and Data Security (杭州滨江高新区区块链与数据安全研究所)
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Large language models (LLMs) have shown strong performance on software engineering (SE) tasks, driving the need for high-quality datasets for reliable training and evaluation. SWE-bench-style datasets (SWE datasets) are widely used to evaluate LLM-based coding agents on real-world software issues, where task clarity and test coverage directly affect whether instances are solvable and verifiable. However, quality issues persist in existing SWE datasets, and manual annotation is costly and labor-intensive, motivating LLM-based annotation. Existing approaches typically rely on static, overly general rubrics, limiting their ability to handle diverse and nuanced scenarios and often yielding high false positive rates. In contrast, human annotators draw on accumulated experience to make context-aware judgments. Inspired by this, we propose ExLabel, an LLM-based, self-evolving, experience-guided assistant for task clarity and test coverage assessment in SWE datasets. ExLabel maintains an evolving experience bank that distills cross-instance knowledge into stable yet adaptive judgment criteria and reuses relevant experience to guide the assessment of new instances. For systematic evaluation, we construct SWE-TQ, a new benchmark spanning eight programming languages, by cleaning human annotations from SWE-bench and Multi-SWE-bench. Experiments with three backbone LLMs show that ExLabel achieves the best F1 across all settings, with relative improvements of 44.2%/49.9% and 24.0%/18.4% over the best baseline for task clarity and test coverage assessment on the Python/Multilingual subsets, respectively. Moreover, a coding model fine-tuned on SWE-rebench data filtered by ExLabel resolves 66.2% of issues on SWE-bench Verified, outperforming the model trained on data filtered by the best baseline by 14.5%.
[NLP-55] How Specific Do I Need to Be? Effects of User Query Underspecification on Deep Research Agents
【速读】： 该论文旨在解决深度研究代理（Deep Research Agents, DRAs）在面对现实世界中常见的不完整或模糊用户查询时，其回答质量与中间搜索行为如何受到影响的问题。当前的DRAs通常在完整且明确的查询条件下进行训练与评估，但在实际应用中，用户提问往往存在信息缺失或表述不清的情况。为此，论文提出了一种名为UNDERSPEC-DR的系统性评估框架，用于分析不完整查询对DRAs最终答案质量及中间搜索过程的影响。研究发现，尽管DRAs仍能基于引用证据生成看似合理的回答，但其回答在满足用户真实信息需求方面的表现显著下降。关键问题在于，当查询不完整时，DRAs倾向于展开更广泛的搜索，过度依赖训练过程中习得的先验知识，而较少依据原始查询内容，导致生成结果与用户真实意图产生偏离。这一发现揭示了现有DRAs对不完整查询的高度敏感性与脆弱性，并强调未来需构建具备主动识别能力的DRAs，能够在必要时主动请求用户提供补充信息，以可靠完成深度研究任务。
链接: https://arxiv.org/abs/2610.39413
作者: Yi Wen,Xiangjue Dong,Bingxuan Li,Hangxiao Zhu,Leo Zhicheng Liu,Yu Zhang,James Caverlee
机构: University of Maryland, College Park(马里兰大学学院帕克分校); Texas A&M University(得州农工大学); University of Illinois, Urbana-Champaign(伊利诺伊大学香槟分校)
类目: Computation and Language (cs.CL)
备注: 16 pages, 8 figures
Abstract:Deep research agents (DRAs) conduct long-horizon reasoning, search, and literature synthesis to provide long-form answers to complex, domain-specific questions. These agents are typically trained and evaluated on fully specified user queries. However, real-world user queries can often be underspecified. We introduce UNDERSPEC-DR, a framework for systematically evaluating how underspecified queries affect agents’ final answer quality and intermediate search behavior. Across three DRAs we evaluated, underspecification substantially reduces how well DRA’s answers fulfill users’ information needs, even though they remain grounded in cited evidence. Our analysis of intermediate search behavior suggests a possible explanation: DRAs conduct broader searches that draw less on the original user query but rely more on their learned prior during training, which can be misaligned with users’ intent. Our findings reveal the vulnerability of current DRAs to underspecified user queries and inform future efforts to build DRAs that proactively recognize when additional user input is needed to reliably complete deep research tasks.
[NLP-56] Robustness of Chatbot LLM Judgments of Presidential Actions and Hypothetical Reforms
【速读】： 该论文旨在解决生成式AI在公共政策评估中判断结果的稳健性问题，即当用户通过普通消费级聊天接口向大模型咨询政策长期影响时，其输出是否具备可重复性与一致性。研究的关键在于通过实证方法衡量不同大模型配置（Mistral、Gemini Flash、Claude Sonnet及Claude Opus在两种推理强度下）在30天内对40项历史政策和15项假设改革的评分表现，评估其在评分符号、排名顺序以及置信度与评论内容上的稳定性。结果显示，各模型对总统行政机构平均得分的正负方向及排序高度一致，同日不同模型对同一政策给出相反评价的比例仅为约4%，表明判断具有较高稳健性；而跨日差异虽大于同日差异，但随时间增长缓慢，仅在平台服务发生未在界面显现的底层变更时出现显著偏移。进一步分析评论内容作为信息检索任务，发现其对具体政策条目具有高度特异性且跨运行与模型保持一致性，词法与嵌入相似性验证了这一结论。因此，解决方案的关键在于构建并公开一个大规模、高频率、多模型的政策评估基准数据集（16,500条判断），为后续研究生成式AI在公共事务中的可靠性提供可复现的实证基础。
链接: https://arxiv.org/abs/2610.39398
作者: Oleg Vasilyev
机构: 未知
类目: Computation and Language (cs.CL); Computers and Society (cs.CY)
备注:
Abstract:Chatbots are increasingly consulted about public affairs, and their judgments of policies are useful only if they are robust. We measure this robustness in the setting of an ordinary user, through consumer chat interfaces. For 30 days, five chatbot configurations (Mistral, Gemini Flash, Claude Sonnet, and Claude Opus at two effort settings) were asked twice a day to score the long-term effect on the United States of 40 actions of the last four presidential administrations and of 15 hypothetical reforms, each with a confidence and a short comment. The judgments are robust: all configurations agree on the sign and the ranking of the administrations’ mean scores, and two different configurations asked on the same day give opposite signs to the same action in about 4% of cases. Responses from different days differ more than same-day responses, but the difference grows slowly over the month, except where the service itself changed. We identify three such changes, which shifted the judgments while leaving little or no trace in the interface. Treating the comments as a retrieval task, we find them specific to each item and consistent across runs and chatbots, with lexical and embedding similarities agreeing. We release the dataset of 16,500 judgments.
[NLP-57] What Makes Medical Examination Items Difficult? Using Large Language Models to Explore Patterns Associated With Empirical Item Difficulty
【速读】： 该论文旨在解决医学题项难度评估中长期存在的可解释性不足问题，即现有研究多聚焦于难度预测，而忽视了从实证难度标签中识别可解释的难度模式。其核心解决方案是利用美国医师执照考试（USMLE）和执业助理国家认证考试（PANCE）中的实证难度标签，引导大语言模型（LLM）自动生成难度评分标准（rubric），并考察这些评分标准对模型难度判断的影响。关键发现在于，所诱导出的评分标准均未将表面特征（如题干长度、选项数量或图像）视为主要难度来源，而是揭示了跨考试共有的深层认知机制，如线索汇聚、误导性线索及选项竞争等；其中，PANCE的评分标准更强调无法从题干直接推导的知识检索，而USMLE则更注重多临床线索的整合。在模型评估方面，尽管评分标准引导改变了模型对难度量表的使用方式，但并未一致提升判断性能：在绝对难度判断中，引导促使模型更广泛地使用量表，提升了粗粒度一致性，但未显著改善排序准确性；在成对比较中，效果因模型和考试而异，且判断结果对呈现顺序更为敏感。因此，该研究认为，标签引导的评分标准生成更核心的价值在于从实证数据中提炼出可解释、可验证的认知假设，而非可靠提升难度判断的准确性。
链接: https://arxiv.org/abs/2610.39373
作者: Xiaomeng Xiong,Marcus Walker,Yanlin Jiang
机构: University of Florida, Gainesville, Florida, USA(佛罗里达大学，盖恩斯维尔，佛罗里达州，美国); National Commission on Certification of Physician Assistants(医师助理认证国家委员会，美国)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Research on medical item difficulty has largely focused on prediction rather than on identifying interpretable patterns associated with empirical difficulty. This study used empirical difficulty labels to guide large language models (LLMs) in inducing difficulty rubrics and examined how rubric guidance affected LLM difficulty judgments. Using training items and their empirical difficulty estimates from the United States Medical Licensing Examination (USMLE) and the Physician Assistant National Certifying Examination (PANCE), Claude Opus 5 induced separate rubrics for absolute and pairwise difficulty judgment. None of the induced rubrics treated surface features, such as stem length, option count, or images, as major sources of difficulty. Several patterns emerged across both examinations, including cue convergence, misleading cues, and option competition, although their specific forms differed. Retrieval of knowledge that could not be derived from the stem was particularly prominent in the PANCE rubrics, whereas the USMLE rubrics placed greater emphasis on integrating multiple clinical cues. We compared commercial and open-weight LLMs on held-out items under unguided and rubric-guided conditions. Rubric guidance changed how models used the difficulty scale but did not consistently improve performance. In absolute judgments, guidance encouraged broader use of the scale and generally improved coarse-grained agreement, without consistently improving rank ordering. In pairwise comparisons, its effects varied across models and examinations, and judgments became more sensitive to presentation order. These findings suggest that the value of label-guided rubric induction may lie more in generating interpretable, testable hypotheses from empirical difficulty data than in reliably improving difficulty judgments.
[NLP-58] Cord: Confidence-Anchored Order Residual Decoding for Masked Diffusion Language Models
【速读】： 该论文旨在解决掩码扩散语言模型（dLLM）在生成过程中解码顺序对生成质量具有显著影响，而现有解码器通常在整个生成过程采用固定的揭示规则这一问题。其核心挑战在于：虽然置信度（confidence）可作为稳定的解码参考，但它仅衡量当前位置预测的确定性，而非揭示该位置对未来解码的潜在价值。为此，论文提出Cord——一种轻量级、提示条件化的策略，以置信度解码为基准，通过学习在生成方向上的几何残差（geometric residuals）、与已揭示上下文的距离以及双向填充（two-sided infilling）来动态决定何时及如何偏离置信度规则。该残差机制能够自适应地融合提示信息和逐步演化的部分响应。不同于每步采样新策略动作，Cord仅在每个生成轨迹中采样一个低维系数矩阵，从而诱导出完整的状态依赖揭示轨迹。该方法采用直接基于序列级奖励训练的上下文老虎机（contextual-bandit）框架，无需步骤级信用分配或主干网络更新。在三个dLLM和五个推理、代码生成与约束满足基准测试中，Cord将任务成功率从置信度解码下的32.0%提升至48.2%，并优于各领域最优的固定残差策略。行为审计显示，所学残差显著改变了揭示局部性和双向填充模式，且具有任务依赖性。Cord仅引入0.53M可训练参数，不增加主干前向传播次数，推理延迟开销仅为5.39%。
链接: https://arxiv.org/abs/2610.39372
作者: Chengwei Zhou,Zhaoyan Jia,Massoud Pedram,Gourav Datta
机构: Case Western Reserve University (凯斯西储大学); University of Southern California (南加州大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:The decoding order of masked diffusion language models (dLLMs) can substantially affect generation quality, yet common decoders typically apply the same reveal rule throughout generation. Confidence provides a strong default, but it measures how decisively a position can be predicted now, not how useful revealing that position will be for subsequent decoding. We propose Cord, a lightweight prompt-conditioned policy that retains confidence decoding as a stable reference and learns when and how to depart from it through geometric residuals over generation direction, proximity to revealed context, and two-sided infilling. The residuals adapt to both the prompt and the evolving partial response. Rather than sampling a new policy action at every decoding step, Cord samples a single low-dimensional coefficient matrix per rollout, which induces an entire state-dependent reveal trajectory. This yields a contextual-bandit formulation trained directly from sequence-level rewards, without step-level credit assignment or backbone updates. Across three dLLMs and five reasoning, code, and constraint-satisfaction benchmarks, Cord improves average task success from 32.0% under confidence decoding to 48.2%, and further improves over the best per-domain fixed residual. Behavioral audits show that the learned residuals induce substantial, task-dependent changes in reveal locality and two-sided infilling. Cord adds only 0.53 M trainable parameters, requires no additional backbone forward passes, and incurs only 5.39% inference latency overhead.
[NLP-59] Unlimited OCR Works 2
【速读】： 该论文旨在解决超长文档（数百页）解析时因视觉上下文规模急剧增长而导致的计算成本过高与模型可扩展性受限的问题。传统方法在处理长文档时，随着文本长度增加，解码步骤的计算开销呈线性甚至非线性上升，严重制约了端到端长文档解析的效率与可行性。其解决方案的关键在于提出Unlimited OCR 2，通过引入稀疏参考滑动窗口注意力机制（Sparse Reference Sliding Window Attention, SR-SWA），将解码器中活跃的参考标记数量限制在固定容量内，从而实现每步解码计算成本恒定，不受文档长度影响。结合无限视野解码范式（Unlimited-Horizon Decoding Paradigm），该模型可在几乎无内存负担的情况下完成跨百页文档的一次性、连续解析，显著提升处理效率与可扩展性。实验表明，在百页级别任务上，Unlimited OCR 2相比全注意力（Full Attention）和R-SWA分别实现了7.0倍和5.6倍的单步加速，并在新提出的MPDocBench与MarathonDocBench基准上达到当前最优的端到端性能，标志着光学字符识别（OCR）进入以高效、无中断长文档处理为特征的3.0时代。
链接: https://arxiv.org/abs/2610.39308
作者: Shu Wei,Jingjing Wu,Qunyi Xie,Lingshu Zhang,Xiangcheng Du,Yannian Fu,Mengjie Wang,Zuyi Wang,Zhichao Liu,Yangliu Xu, YY,Yi Yang
机构: Baidu(百度) ERNIE Team
类目: Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Unlimited OCR ushers the era of long-horizon parsing, substantially mitigating the growth of working memory as document lengthens. However, as document length scales to hundreds of pages, the visual context grows substantially, sharply increasing the computational cost of each decoding step and limiting the scalability of ultra-long document parsing. Inspired by human book transcription via localized focus without re-scanning preceding pages, we introduce Unlimited OCR 2, a document parsing model that looks only at what it transcribes. Building on Unlimited OCR, we upgrade all the decoder attention modules to Sparse Reference Sliding Window Attention (SR-SWA), bounding active reference tokens to a fixed capacity and thus sustaining constant computational costs across arbitrarily long documents. By combining SR-SWA with Unlimited-Horizon Decoding Paradigm, Unlimited OCR 2 enables document parsing over virtually unlimited lengths without prohibitive memory overhead. At the hundred-page scale, Unlimited OCR 2 achieves 7.0 \times and 5.6 \times per-step speedups over Full Attention and R-SWA, respectively. Furthermore, it delivers state-of-the-art end-to-end performance on both MPDocBench and MarathonDocBench, our newly proposed benchmark targeting hundred-page document parsing. Ultimately, the capacity for uninterrupted, one-shot document parsing signifies a fundamental paradigm shift into the OCR 3.0 era.
[NLP-60] Grounding Multimodal Reasoning with Evidence-Ablated Negatives NEURIPS2026
【速读】： 该论文旨在解决多模态强化学习中因奖励可验证性带来的信用分配模糊问题，即模型可能通过利用语言先验或数据集捷径获得正向奖励，而非基于视觉证据进行合理推理。这一问题的关键在于如何生成高质量的负样本，以有效暴露并抑制模型对语言先验的依赖。其解决方案的核心是提出一种名为EVAN（Evidence-Ablated Negatives）的多模态强化学习框架，通过引入证据缺失的负样本增强训练：首先利用缩放后的图像信息增益度量识别高度依赖视觉的样本，并采用混合排序策略保持数据多样性；随后通过遮蔽图像中最具判别性的局部区域生成去证据化图像，促使当前策略产生看似合理但缺乏视觉依据的响应；在策略优化过程中，这些负样本作为有界辅助信号，在原始图像条件下惩罚捷径学习行为，同时维持强化学习的稳定性。实验结果表明，EVAN有效缓解了模态不平衡问题，并在多个具有挑战性的多模态推理基准上优于现有的强基线方法。
链接: https://arxiv.org/abs/2610.39237
作者: Chenxi Liu,Tianyi Xiong,Yanshuo Chen,Heng Huang
机构: University of Maryland, College Park(马里兰大学学院公园分校); College Park, MD 20742
类目: Computation and Language (cs.CL)
备注: NeurIPS2026
Abstract:Reinforcement learning with verifiable rewards (RLVR) has become a powerful tool for improving multimodal reasoning. However, standard outcome supervision suffers from credit assignment ambiguity, as a model may receive positive rewards by exploiting language priors or dataset shortcuts rather than grounding its reasoning in the requisite visual evidence. This makes high-quality negative responses critical, since effective negatives should remain plausible under the current policy while exposing the shortcut that should be suppressed. In this paper, we propose EVAN, a multimodal reinforcement learning framework utilizing EVidence-Ablated Negatives to train LMMs against their own language-prior failures. EVAN first identifies highly visually dependent examples using a scaled image information gain metric and preserves data diversity with a mixed-ranking strategy. It then constructs evidence-ablated images by masking the most discriminative image patches, eliciting plausible but visually ungrounded responses from the current policy. During policy optimization, these negative responses are incorporated as bounded auxiliary signals under the clean image, penalizing shortcut learning while maintaining RL stability. Experiments across challenging multimodal reasoning benchmarks demonstrate that EVAN mitigates modality imbalance and improves over strong RLVR baselines. Our code is available at this https URL.
[NLP-61] Agent Heal: Evaluating Goal-Preserving Resource Recovery of Web Agents under Access Barriers
【速读】： 该论文旨在解决大语言模型（Large Language Model, LLM）代理在真实网络环境中因访问障碍（如失效链接、认证墙、速率限制及反爬虫机制）导致任务执行失败或偏离原始目标的问题。现有评估方法要么隐含假设目标资源可自由访问，要么仅关注可通过简单重试解决的瞬时扰动，无法反映代理在面对不可逾越瓶颈时需主动重构路径的真实挑战。本文提出AgentHeal这一综合性基准，聚焦于“目标保持型资源恢复能力”——即代理在检测到非瞬态访问障碍后，能够识别并切换至合法替代源，同时精确达成用户原始目标的能力。实验表明，在引入各类访问障碍后，主流LLM代理的任务成功率从89.4%骤降至43.7%。深入分析揭示，当前代理普遍存在误信伪装性替代页面、未能及时向用户求助、违反访问政策以及在无效尝试上浪费计算资源等问题，为构建具备鲁棒性的网络代理提供了关键诊断依据与改进方向。
链接: https://arxiv.org/abs/2610.39234
作者: Zhiyi Liu,Haohua Du,Haoqi Sun,Haisheng Tan
机构: University of Science and Technology of China (中国科学技术大学); Beihang University (北京航空航天大学); Peking University (北京大学)
类目: Cryptography and Security (cs.CR); Computation and Language (cs.CL)
备注: Code and Dataset are available at this https URL
Abstract:Large language model (LLM) agents have demonstrated strong capabilities in automating complex information-seeking and transactional tasks across the open web. However, their real-world reliability remains limited by access barriers, such as dead links, authentication walls, rate limits, and anti-scraping defenses. These barriers block the primary execution path, causing agents to abort prematurely, loop indefinitely, or quietly compromise the user’s original constraints. Existing evaluations either implicitly assume that target resources are freely accessible or focus on transient perturbations that simple retries can resolve, failing to capture scenarios where agents must actively reroute around impassable bottlenecks. To bridge this gap, this paper introduces AgentHeal, a comprehensive benchmark evaluating goal-preserving resource recovery, defined as the ability of web agents to detect non-transient access barriers, pivot to legitimate alternative sources, and fulfill the user’s exact original objective. Spanning nine barrier families across a counterfactual sandbox and real-website replay, our experiments on leading LLM backbones show that average task success drops sharply from 89.4% to 43.7% under access barriers. Further analysis reveals that agents frequently fall for deceptive look-alikes, fail to escalate to users when necessary, violate access policies, or waste computation on futile retries, providing actionable diagnostics for developing resilient web agents.
[NLP-62] XLaw-JP-27B: A Lightweight Open Model Specialized for Japanese Law
【速读】： 该论文旨在解决生成式 AI 在日本法律领域应用中面临的模型专业化程度不足、推理能力有限以及在高难度法律考试（如司法考试）中表现不佳的问题。其核心挑战在于如何在不使用测试年份材料的前提下，使模型具备跨年度通过日本律师资格考试的能力，并在合同审查、专利法等复杂任务上实现接近专业水平的表现。解决方案的关键在于：基于Qwen3.8-27B架构，采用混合检索机制（整合现行全部法律法规与3.1万份判例摘要），结合自蒸馏（self-distillation）的轻量级QLoRA微调以增强对历年司法考试真题的准确回答能力，引入针对合同任务的专用LoRA适配器，并采用四样本投票机制提升输出稳定性。实验表明，该方案在八项日本法律任务中取得六项第一、一项并列第一、一项第三（平均排名1.3），是唯一能在2024–2026三年司法考试中全部通过的开源模型，显著优于12–33B规模的其他开源模型（平均排名2.8–3.1）。控制实验进一步验证了检索、投票和自蒸馏的有效性，而思维模式、基于检索的合成微调、自我验证及逐条分解等策略未带来显著增益。当前主要差距仍存在于合同审查与专利法领域，但该模型已作为Apache 2.0许可下的免费在线服务公开提供。
链接: https://arxiv.org/abs/2610.39227
作者: Andrew Shin
机构: Keio University (庆应义塾大学); X-Law Inc (X-Law公司)
类目: Computation and Language (cs.CL)
备注: this https URL
Abstract:We present XLaw-JP-27B, an open-weight model specialized for Japanese law that runs on a single 48,GB GPU under the Apache~2.0 licence. It adapts Qwen3.8-27B with hybrid retrieval over all current statutes and 31k case summaries, a light QLoRA fine-tune on its own verified answers to past bar-exam questions (self-distillation), a contract-specific LoRA adapter, and four-sample voting. We evaluate it on eight Japanese legal tasks—the multiple-choice section of the bar examination in 2024, 2025 and 2026, statement-level bar questions (JBE-QA), statute questions (LawQA-JP), the patent attorney exam, and contract review and drafting built from government model contracts—against open models of 12–33B under identical prompts and retrieval. XLaw-JP-27B ranks first on six tasks, ties on one and is third on one (mean rank 1.3 against 2.8–3.1 for the baselines). It is the only system that passes the bar exam in all three years (129, 120 and 130 of 175, meeting every subject minimum; no test-year material was used in training), while open Japanese 32–33B models fail at least one year. A controlled study of each component shows that retrieval, voting and self-distillation help, and that thinking mode, retrieval-grounded synthetic tuning, self-verification, per-statement decomposition and more self-distillation do not. The largest remaining gaps to frontier models are in contract review and patent law. The model is available as a free web service at this https URL with the weights available at this https URL under Apache~2.0.
[NLP-63] Reasoning or Pattern Exploitation? Mechanistic Insights from RL-Trained Language Models
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在经过强化学习（Reinforcement Learning, RL）后所表现出的“推理”能力是否真正源于深层次的认知推理，还是仅依赖于对输入信息的表面适应性调整这一核心问题。其关键在于通过一个可穷尽分析的逻辑领域——井字棋（Tic-Tac-Toe），结合组相对策略优化（Group Relative Policy Optimization, GRPO）进行训练，系统地追踪模型内部表征在训练过程中的动态演化。研究采用基于稀疏自编码器（Sparse Autoencoders, SAEs）的机制可解释性方法，发现尽管强化学习显著提升了模型在不同提示下的准确性和鲁棒性，但其内部表征主要聚焦于显式包含于输入中的低级信息（如当前轮次、游戏进展和棋盘占用状态），而对高级逻辑概念（如棋盘对称性与战略制胜点）的表征仍较弱。这表明，看似“推理”的表现可能更多是任务特定的输入敏感性适应，而非真正的抽象推理能力。该研究首次构建了系统化的可解释性分析流程，实现了在小规模模型中追踪强化学习期间表征演变的闭环分析，并通过因果干预验证决策位于晚期层状态下游，揭示强化学习并未引入或删除概念信息，而是重新组织了主导表征的特征权重。这一发现为评估现有推理基准的有效性提供了重要警示，强调需区分表征重组与真实推理之间的本质差异。
链接: https://arxiv.org/abs/2610.39182
作者: Aishik Nagar,Cheston Tan
机构: Carnegie Mellon University; Institute of High Performance Computing (IHPC), Agency for Science, Technology and Research (ASTAR); Centre for Frontier AI Research (CFAR), Agency for Science, Technology and Research (ASTAR)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Large language models (LLMs) are increasingly described as acquiring “reasoning” skills after reinforcement learning with verifiable rewards. However, whether these gains reflect genuine reasoning or more superficial adaptations remains underexplored. In this work, we probe LLMs trained in a finite and exhaustively analyzable logical domain, namely Tic-Tac-Toe, and trace how internal representations evolve across reinforcement learning (RL) with Group Relative Policy Optimization (GRPO). Quantitatively, RL yields high accuracy and robustness across prompt variations, suggesting the emergence of game reasoning. Mechanistic interpretability, however, paints a different picture: features extracted with sparse autoencoders (SAEs) reveal that models primarily adapt to rely more heavily on information already explicit in the prompt, such as whose turn it is, game progression and board occupancy. In comparison, high-level concepts like board symmetries and strategic forks remain weakly represented, raising concerns that reasoning benchmarks overestimate model abilities. This suggests that RL-driven “reasoning” may conflate task-specific adaptations to input information with reasoning ability. We provide, for the first time: (i) a systematic interpretability pipeline tracing representation dynamics across RL training in small-scale LLMs; (ii) an extension of SAE-based feature discovery to hypothesis-driven testing in a finite logical domain; and (iii) an interpretability-based demonstration that RL post-training neither adds nor removes concept information but reorganizes which features dominate the representation, with causal interventions locating the decision downstream of it, in the late-layer state.
[NLP-64] Reading-Level Verification of Kanji Words in Japanese Financial Speech Synthesis
【速读】： 该论文旨在解决生成式语音克隆（voice cloning）在日语金融评论合成中因汉字多音现象（如“市场”可读作“shijo”或“ichiba”）导致的误读问题。现有质量检查机制依赖自动语音识别（ASR）转录后将输入与输出均转换为假名进行比对，但该方法存在双向不可靠性：当音频中存在误读时，ASR语言模型常会错误地恢复预期汉字，掩盖错误；而当音频正确时，普通识别错误又会引发不必要的重生成。为此，论文提出一种基于读音级别的检测方法，利用连接时序分类（CTC）的平假名识别器对目标汉字的预期读音及其易混淆变体进行评分，并在替代读音更符合音频特征时发出警告。实验在3个语音克隆声线、30个金融术语、540条注入误读的语句上验证，传统转录基检查仅能以44.4%的误报率检出78.9%的误读，而新方法在仅5.0%误报率下实现91.7%的检出率（AUC达0.990）。在真实汉字输入场景中，该方法成功捕获了前段系统遗漏的错误，并将重生成请求率从38.9%显著降低至6.1%，证明其在提升语音合成准确性与效率方面的关键优势。
链接: https://arxiv.org/abs/2610.39168
作者: Cheng Zhang
机构: Best AI Co., Ltd.(最佳人工智能有限公司)
类目: ound (cs.SD); Computation and Language (cs.CL); Audio and Speech Processing (eess.AS)
备注: 6 pages, 4 figures, 7 tables
Abstract:Few-shot voice cloning makes it cheap to produce daily Japanese financial commentary with several synthetic speakers, but kanji words with more than one reading, such as the word for “market” (shijo vs. ichiba) or “overall trend” (taisei vs. oze), are still misread. A common quality gate transcribes each synthesized sentence with automatic speech recognition (ASR), converts both the input and the transcript to kana, and regenerates on mismatch. We show that this gate is unreliable in both directions: when the audio contains a misreading, the ASR language model frequently writes the expected kanji back, hiding the error, and when the audio is correct, ordinary recognition errors trigger needless regeneration. We propose a reading-level check that scores the expected reading and its confusable alternatives with a connectionist temporal classification (CTC) hiragana recognizer and flags a word when an alternative explains the audio better. On 540 utterances from three voice-cloned production voices covering 30 financial terms, with misreadings injected under controlled conditions, the transcript-based gate detected 78.9% of misreadings at a 44.4% false-alarm rate, whereas the proposed check detected 91.7% at 5.0% (AUC 0.990). On natural kanji input, it caught a front-end error that the transcript-based gate missed for one voice, and reduced regeneration requests from 38.9% to 6.1% of sentences.
[NLP-65] A Novel Approach to Evaluating Evaluation Metrics for Multi-Output Structured Prediction AAAI2026
【速读】： 该论文旨在解决多输出结构化预测任务中因训练数据仅提供单一真实标签，而实际存在多个合理有效输出所导致的评估可靠性难题。其核心问题在于现有评估指标难以准确区分模型生成结果的质量差异，尤其在缺乏明确唯一正确答案的情况下。解决方案的关键在于引入“近似任务特定不变性”（approximate task-specific invariants）的概念，即人类评估者在判断输出合理性时隐含依赖的领域相关结构性约束（如图像着色中的物体边界、机器翻译中的命名实体等）。作者将这些不变性作为诊断工具，构建了一个系统性评估框架：通过在损失函数中引入增强项以强制模型输出符合领域相关的不变性特征，从而检验不同评估指标的可靠性。实验表明，基于不变性的增强训练显著提升了分布类指标的表现，而传统指标变化甚微；由此揭示了标准指标在捕捉语义合理性和结构一致性方面的局限性，并明确了分布类指标在何种条件下仍存在失效风险。
链接: https://arxiv.org/abs/2610.39162
作者: Akshay Vyas,Angelo Pimienta,Nicholas Ruozzi
机构: 未知
类目: Machine Learning (cs.LG); Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV)
备注: Extended version of the AAAI 2026 paper, including the technical appendix. 18 pages, 8 figures. Code: this https URL
Abstract:In multi-output structured prediction tasks, while only one ground truth label may be provided in the training data, multiple equally valid outputs may be possible, making reliable evaluation a persistent challenge. We postulate that human evaluators implicitly use task-specific invariants, e.g., object boundaries in colorized images or named entities in translations, to judge if an output is acceptable. Under this assumption, we introduce a notion of approximate task-specific invariants and use them as diagnostic tools to evaluate a variety of existing metrics for vision and language tasks. We use these task invariants as part of a framework to systematically test metric reliability by encouraging domain-relevant invariants in model outputs via an augmented loss function. In our experiments, we observe that enforcing invariants with an augmented loss yields substantial improvements in popular distributional metrics while more traditional metrics change only marginally. Through this invariants-driven evaluation, we expose where standard metrics fail to detect meaningful differences, and we highlight the conditions under which distributional metrics succeed or still fall short.
[NLP-66] HALO: Human-Grounded Granular Evaluation of Proactive Agents
【速读】： 该论文旨在解决主动型智能体（proactive agents）在无明确用户请求时，如何准确判断是否提供协助、何时介入以及介入程度的问题。现有评估方法多依赖自动化评分或二元判定，难以深入诊断智能体在证据使用和干预质量方面的表现。为弥补这一不足，研究提出HALO（Human-Grounded GrAnular EvaLuation of PrOactive Agents），其核心是构建一个涵盖4类主动需求（依赖性、阻塞、风险与机会）的评估框架，覆盖10个日常场景与50个个人环境。该框架要求智能体从结构化数据库或非结构化文档中检索证据，再决策是否干预及干预范围。基于此，研究构建了HALO-8K数据集，包含10种模型在400个场景下的8,000条响应，由20名专家标注动作层级、可接受性判别及人工诊断。对被拒响应的人工诊断进一步细化至介入时机、强度等维度，并记录推理令牌与检索轮次以衡量执行成本。实验结果显示，最佳专有模型与开源模型的可接受率分别为72.88%与56.63%。关键发现为：过度主动（over-proactivity）是主要失败模式，其中“过早介入”与“干预过强”共占被拒响应的61.20%。此外，研究揭示了干预与克制之间的平衡、证据获取效率以及时机与强度校准的重要性，为提升主动型智能体的可靠性提供了系统性洞见。
链接: https://arxiv.org/abs/2610.39086
作者: Wenqi Pei,Xinyu Wang,Jinrui Liu,Cynthia Zhuge,Jie Sheng,Zihan Wang,Tenghao Huang,Muhao Chen
机构: University of California, Davis(加州大学戴维斯分校); 2077AI; University of Southern California(南加州大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Proactive agents must decide whether, when, and how far to assist without an explicit request. Existing evaluations often rely on automated scores or binary verdicts, offering limited human diagnosis of evidence use and intervention quality. To address these gaps, we introduce HALO, Human-Grounded GrAnular EvaLuation of PrOactive Agents. HALO-Arena covers 4 proactive need types, Dependency, Blockage, Risk, and Opportunity, across 10 everyday domains and 50 personal environments. Scenarios require agents to retrieve evidence from structured databases or unstructured documents before judging whether assistance is warranted and choosing its scope. On HALO-Arena, we construct HALO-8K, a dataset of 8,000 responses from 10 models across 400 scenarios, annotated by 20 experts for action level, acceptability verdicts, and human diagnosis. For rejected responses, human diagnosis covers timing, intensity, and other issues. Recorded reasoning tokens and retrieval rounds measure the effort behind these responses. The best proprietary and open-source models achieve 72.88% and 56.63% acceptability, respectively. Our central finding is that over-proactivity is the dominant observed failure mode, with Premature timing or Excessive intensity accounting for 61.20% of rejected responses. Further analyses yield complementary findings on the balance between intervention and restraint, evidence acquisition, and the calibration of timing and intensity.
[NLP-67] HEARSAYBENCH: Can AI Navigate from Abstract Human Rights to Lived Lives?
【速读】： 该论文旨在解决生成式人工智能（Generative AI）在个人咨询场景中对结构性约束下用户支持不足的问题。现有大语言模型（LLMs）的训练数据和评估基准普遍基于自由行动者假设，即用户可自主选择离开、求助或申诉，忽视了战争、无国籍状态及流离失所等现实条件下用户实际能力受限的处境。同时，尽管模型受制于以人权为核心的宪法式人工智能（Constitutional AI）框架，但这些理想化规范在许多地区缺乏实际法律保障。为此，研究提出HEARSAYBENCH——一个由人类撰写的400个测试案例构成的数据集，其情境基于联合国及其他人权报告，覆盖80多个地区，涵盖社会、环境与个人层面的结构性障碍，并通过本地化人工验证确保真实性。该数据集以阿马蒂亚·森（Amartya Sen）的能力方法（Capabilities Approach）为理论基础，区分个体“形式上可行使的权利”与“实际上可实现的能力”。实验表明，尽管多数模型能更准确理解受困用户的处境，却无法提供安全可行的建议；即使强模型也存在认知与行动指导之间的脱节。关键问题在于：模型虽能“听见”（hear）困境，却未能转化为可执行的建议。因此，解决方案的核心在于构建能够识别并回应结构性不平等的真实世界情境的能力评估体系，使生成式系统真正服务于现实世界中的弱势群体，而非仅反映理想化假设。
链接: https://arxiv.org/abs/2610.39062
作者: Ava Iranmanesh,Sobhan Lotfi,Ali Iranmanesh,Liwei Jiang
机构: Penn State University (宾夕法尼亚州立大学); Independent Researcher (独立研究员); University of Washington (华盛顿大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:As large language models (LLMs) reach users worldwide, personal counseling has emerged as one of their most consequential applications, sometimes in life-critical situations. Yet broader access does not mean equal support. The text these models learn from disproportionately reflects people with the freedom to act, embedding an implicit assumption that users can leave, report, seek help, or appeal. Existing benchmarks largely inherit this assumption, overlooking users whose agency is constrained by structural conditions such as war, statelessness, and displacement. Meanwhile, models are increasingly shaped by normative frameworks such as Constitutional AI, often grounded in ideals like human rights, even as many users live where those rights offer little protection or recourse. We introduce HEARSAYBENCH, a human-authored dataset of 400 test cases with in-region human verification grounded in scenarios where structural barriers stand between a person and the exercise of their rights. Built from United Nations and other human rights reports, the scenarios span over 80 regions and cover social, environmental, and personal constraints. Drawing on Nobel laureate Amartya Sen’s Capabilities Approach, HEARSAYBENCH tests whether models can distinguish what a person is formally entitled to do from what they are actually able to do. Across 14 proprietary and open-weight models, every model infers a user’s situation more accurately than it advises on it. Stronger models narrow this gap, but their recommendations remain unsafe to follow. Models may understand lives shaped by constraint, yet that knowledge remains hearsay if it cannot translate into advice people can act on. Those people are not alignment’s long tail; they are among the users for whom good guidance matters most. HEARSAYBENCH makes this failure measurable, a step toward systems that serve the world as it is rather than as assumed. Subjects: Computation and Language (cs.CL) Cite as: arXiv:2610.39062 [cs.CL] (or arXiv:2610.39062v1 [cs.CL] for this version) https://doi.org/10.48550/arXiv.2610.39062 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[NLP-68] racing the Influence of Retrieved Literature on Scientific Ideation in LLM s
【速读】： 该论文旨在解决生成式人工智能在科研创意生成中，文献检索策略如何影响所产生想法的多样性、质量和新颖性这一关键问题。现有系统将文献检索、搜索、生成与优化流程耦合，难以分离检索策略本身的作用。为此，研究通过在统一生成协议下对比五种不同的检索策略与“无文献”基线，基于15个机器学习主题和10个开源大模型，分析了4.5万条科研创意数据。结果表明，相较于无文献基线，所有检索策略均使同一文献集内生成的想法多样性降低5.6%–17.9%，但在跨文献集合并时显著提升多样性10.2%–18.1%；整体上对创意质量与新颖性的提升有限且存在主题差异。进一步采用基于Jensen-Shannon散度（JSD）的归因方法，追踪每篇输入文献对创意动机、核心思想、方法及验证部分的影响，发现79.3%的创意中仅有一至两篇文献在四个组成部分中获得最高归因分值，且这些主导文献常在同一批次生成的创意中重复出现。此外，来源归因多样性更高的创意组具有更高的整体创意多样性（相关系数r=0.70）。因此，解决方案的关键在于：通过识别并主动引导使用多样化的文献来源，结合归因分析以优化后续创意生成中的文献选择策略，从而有效提升创意的多样性与创新潜力。
链接: https://arxiv.org/abs/2610.39060
作者: Zhipin Wang,Koel Dutta Chowdhury,Christoph Leiter,Steffen Eger
机构: University of Technology Nuremberg (UTN)(纽伦堡工业大学); NLLG(自然语言与逻辑组)
类目: Computation and Language (cs.CL)
备注: Preprint. Under review
Abstract:Large language models (LLMs) are increasingly used to generate research ideas from scientific literature. Existing systems combine literature retrieval with search, generation, and refinement, making it difficult to determine how the retrieval strategy itself shapes the ideas generated. We isolate this contribution by comparing five retrieval strategies and a No Literature baseline under a shared generation protocol, using 45,000 ideas across 15 machine learning topics and ten open-weight models. Relative to no literature basline, all five retrieval strategies reduce diversity among ideas generated from the same set of papers by 5.6% – 17.9% , while most increase diversity by 10.2% – 18.1% when ideas from different sets are pooled. Effects on idea quality and novelty are modest on average but vary across topics. We further apply a JSD-based attribution method to trace the influence of individual input papers on each idea’s motivation, core idea, approach, and validation. In 79.3% of ideas, only one or two distinct papers receive the highest attribution scores across these four parts. These leading sources often recur across ideas generated from the same set of papers, and groups with more varied source attribution contain more diverse ideas ( r=0.70 ). These findings motivate various sets of literature and the use of attribution to guide the choice of sources for subsequent ideas.
[NLP-69] Distilling the Shape of Uncertainty: What On-Policy Distillation Learns and How to Predict It
【速读】： 该论文旨在解决生成式 AI（Generative AI）在大语言模型后训练阶段中，基于策略蒸馏（On-policy Distillation, OPD）时存在“强教师模型无法提升学生模型性能”的问题。现有研究多归因于教师缺乏新知识，但未明确揭示此类知识的具体形式。本文指出，OPD不仅学习“如何分支”（即选择哪些候选词元），更关键的是学习“在何处分支”（即维持推理路径开放的位置）。传统基于词元重叠的度量仅能反映前者，而忽略了后者。为此，论文提出“熵形”（entropy shape）——即下一词元熵序列——以刻画“在何处分支”的行为特征，并引入熵形相关性（Entropy Shape Correlation, ESC）来量化师生间在熵形上的一致性。通过在训练过程中追踪ESC并干预教师的熵形以保持其词元偏好，实验证明“在何处分支”对OPD性能提升具有因果影响，且独立于“如何分支”。最终，作者将OPD结果预测建模为教师选择任务，构建了相应的评估协议，结果显示预训练阶段的ESC可有效筛选最优教师，显著优于现有预测方法。
链接: https://arxiv.org/abs/2610.39020
作者: Shiping Yang,Chuanqi Tang,Hossein Hajimirsadeghi,Layla El Asri,Mohammad Amin Shabani,Angel X. Chang
机构: Simon Fraser University (西蒙菲莎大学); Borealis AI
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:On-policy distillation (OPD) is increasingly adopted in the post-training of large language models. However, a stronger teacher may fail to improve the student. Prior work attributes this to a lack of new knowledge, but what that knowledge is remains unclear. We show that OPD learns not only \textithow to branch, i.e., which candidate tokens to consider, but also \textitwhere to branch, i.e., where to keep reasoning paths open. The existing token-overlap metric captures only the former. We introduce the entropy shape, the sequence of next-token entropies, to characterize the latter, and propose the Entropy Shape Correlation (ESC) to measure student-teacher agreement in entropy shape. Tracking ESC throughout training and intervening on a teacher’s entropy while preserving its token preferences, we empirically show that \textitwhere to branch causally drives OPD gains, independently of \textithow to branch. Finally, we formulate OPD outcome prediction as a teacher selection task and establish an evaluation protocol for it, under which pre-OPD ESC selects the best teacher for both students and outperforms existing predictors. Code is available at this https URL.
[NLP-70] Author-Informed Stance Detection: A Dataset and Findings
【速读】： 该论文旨在解决社交媒体中政治立场检测（political stance detection）的难题，其核心挑战在于仅凭单条推文难以准确理解作者的政治倾向与对特定议题的立场。传统方法忽视了作者背景信息对语义理解的关键作用。为此，论文将立场检测重新建模为一种基于作者上下文（author context）的文本理解任务，强调融合作者的政治归属信息以提升判断准确性。其解决方案的关键在于：通过引入作者上下文信息，显著提升大语言模型（LLM）在立场检测中的表现，尤其在识别讽刺性内容时效果更为突出；同时，提出一种基于作者信息的微调策略，使LLM能够有效预测作者的政治隶属关系，并在此基础上实现更优的立场分类性能。实验表明，该方法不仅超越人类标注者的表现，还在跨数据集泛化能力上具有明显优势。为推动领域发展，研究团队公开了一个大规模、带有作者上下文标注的立场数据集。
链接: https://arxiv.org/abs/2610.39019
作者: Ogen Schlachet Drukerman,Einat Minkov
机构: University of Haifa(海法大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Stance detection in social media is challenging, as interpreting a text often requires understanding the author’s perspective. We re-frame political stance detection as an author-informed text understanding task, where it is desired to infer both the author’s political leaning and their stance towards the target from the text. We first show that human annotators struggle to infer authors’ political leanings and stances from individual tweets, and that pretrained large language models (LLMs) perform below human level in this setting. We then demonstrate that author context improves stance detection for both humans and LLMs, with particularly pronounced benefits for tweets perceived as sarcastic. Through extensive experiments, we further show that LLMs can be effectively fine-tuned to predict authors’ political affiliations, and that author-informed fine-tuning yields better stance detection performance than standard fine-tuning. This approach not only surpasses human performance but also improves generalization across datasets. To support further research on author-informed political stance detection, we release a large dataset of stance-labeled tweets enriched with author-context information.
[NLP-71] Implicit Reasoning in Transformers: Bridging the Compositionality Gap in GPT -2-like Models
【速读】： 该论文旨在解决生成式语言模型在组合泛化（compositional generalization）方面存在的关键问题：尽管模型在分布内（in-distribution, ID）能够正确组合已知事实，但在面对分布外（out-of-distribution, OOD）知识时，其组合机制往往失效。其核心解决方案在于，通过机制导向的训练数据分布设计，重新组织原子知识的呈现方式——具体而言，将ID与OOD的原子事实以与模型内部学习到的两跳推理路径中第二跳检索步骤在位置上对齐的方式进行输入，从而增强原子知识在训练过程中的结构性暴露。这一干预不改变模型架构，也未引入额外的组合监督信号，却显著提升了模型对未见原子事实的组合泛化能力，成功弥合了原有的组合泛化差距。研究结果表明，该差距并非由模型架构的根本性限制所致，而是可通过精准的训练分布工程实现克服，凸显了训练数据结构设计在提升模型可解释性与泛化能力中的关键作用。
链接: https://arxiv.org/abs/2610.39016
作者: Davide Venuto,Leonardo Venuto
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: Code available at: this https URL
Abstract:Compositional generalization remains a persistent limitation of transformer language models: even when a model learns to combine known facts correctly in-distribution (ID), the same compositional mechanism may fail when applied to out-of-distribution (OOD) knowledge. We investigate this compositionality gap in a controlled two-hop reasoning setting using a GPT-2-like transformer model trained from scratch. Building on prior work identifying a circuit associated with compositional generalization, we test whether the failure of OOD transfer is related to how atomic knowledge is structurally exposed during training. We modify the training distribution by presenting both ID and OOD atomic facts in a format positionally aligned with the second-hop retrieval step of the learned compositional circuit, without modifying the model architecture or adding OOD compositional supervision. Under this intervention, the model successfully transfers the learned ID compositional behaviour to held-out OOD atomic facts, closing the observed compositionality gap in the controlled setting. Our results provide evidence that the observed compositionality gap is not a fundamental architecture limitation and that it can be overcome through mechanistically informed training distribution design.
[NLP-72] Blind Spots: Where Small Language Models Fall Behind Frontier Models on Terminal Tasks AAAI NEURIPS26
【速读】： 该论文旨在解决小型语言模型（Small Language Models, SLMs）在终端任务上相较于前沿模型（Frontier Models, FMs）表现显著落后的问题，尤其关注其在边缘设备部署时的性能瓶颈。其核心挑战在于，现有评估指标中的任务成功率（task success rate）未能区分导致性能差距的具体能力维度。为此，论文提出从三个关键维度对SLM与FM的差距进行系统性分解：可解性（solvability），即通过额外的测试时计算（test-time compute）是否能生成正确解；验证能力（verification），即能否从候选轨迹中识别出有效解；以及置信度（confidence），即对自身成功概率的预估准确性。研究发现，尽管重复采样可部分缩小可解性差距，但验证能力上的差距依然显著存在——在相同候选解池下，FMs能够更有效地从随机选择接近最优选择，而SLMs则表现较差；同时，SLMs普遍存在过度自信且校准性差的问题。因此，解决方案的关键在于突破单一的生成能力局限，转而提升模型在解的识别与自我评估方面的综合能力，推动构建更鲁棒、更可靠的轻量级终端智能体。
链接: https://arxiv.org/abs/2610.39006
作者: Ashima Suvarna,Ruta Desai,Ahmed Elgohary
机构: Microsoft Research (微软研究院); University of California, Los Angeles (加利福尼亚大学洛杉矶分校)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: Camera Ready, RAAAI NeurIPS 26
Abstract:Small language models (SLMs) offer an attractive alternative to frontier models (FMs) because they can be deployed with lower cost and latency. This makes SLMs especially appealing for terminal tasks on edge devices, where responsiveness, local execution, and limited connectivity are important considerations. However, they continue to lag behind FMs on terminal tasks, as measured by standard task success rates. Yet these rates conflate several capabilities that may contribute to this gap. In this work, we characterize the SLM-FM gap along three axes: solvability, whether additional test-time compute allows an SLM to reach a correct solution; verification, whether it can identify a successful solution among candidate trajectories; and confidence, whether it can accurately anticipate its probability of success. Across two standard terminal benchmarks, we find that repeated sampling narrows, but does not eliminate, the gap in solvability. In contrast, a persistent gap remains in verification. Given the same candidate trajectory pool, FMs recover substantially more of the gap between random and oracle selection than SLMs. SLMs are also more overconfident in their prospective success estimates and exhibit poorer calibration. These results suggest that the SLM-FM gap is multidimensional, spanning not only the ability to generate successful solutions, but also the ability to recognize successful solutions and anticipate success. We hope this decomposition can guide future work on developing stronger and smaller terminal agents.
[NLP-73] How Architecture Typology and Adaptation Strategies Shape ASR for Low-Resource African Languages: A Controlled Empirical Study
【速读】： 该论文旨在解决低资源语言在自动语音识别（ASR）任务中有效模型架构选择、微调策略及数据适配方法难以确定的问题。其核心解决方案的关键在于通过系统性对比三种不同架构的模型——Phi-4-multimodal-instruct、Whisper-large-v3-turbo 与 MMS-1B-all——在五种低资源非洲语言及斯瓦希里语（作为中等资源语言对照）上的表现，深入分析数据充足性、单语与多语适应、领域微调以及语言适配器迁移等关键因素。研究发现，Phi-4 在整体准确率上表现最佳，MMS 的词错误率（CER）表现接近，而 Whisper 则因已知的幻觉与重复循环问题导致较高 CER；不同模型性能饱和于不同数据规模，揭示了可操作的数据效率目标；语言类型学复杂性显著影响错误率；领域微调对 Phi-4 与 Whisper 有益但会损害 MMS 表现，暴露出基于适配器的适应方法的局限性；此外，使用代理语言适配器优于从零训练 MMS 适配器，表明预训练适配器在低资源场景中的有效性。
链接: https://arxiv.org/abs/2610.39004
作者: Mercy Muchai(1),Kevin Chege(1),Nick Mumero Mwangi(1),Samuel Chege Maina(1),Felermino Ali(1),Elizabeth Ankrah(1),Ignatius Ezeani(2) ((1) Microsoft Research Africa, Nairobi, (2) Lancaster University)
机构: Microsoft Research Africa(微软研究院非洲); Lancaster University(兰卡斯特大学)
类目: Computation and Language (cs.CL); Sound (cs.SD); Audio and Speech Processing (eess.AS)
备注:
Abstract:Despite recent advancements in ASR, identifying effective architectures, fine-tuning approaches, and data adaptation strategies for low-resource languages remains challenging. We present a systematic comparison across five low-resource African languages and Swahili as a mid-resource contrast, fine-tuning three models from distinct architectures: Phi-4-multimodal-instruct, Whisper-large-v3-turbo, and MMS-1B-all. We analyze data sufficiency, monolingual versus multilingual adaptation, domain fine-tuning, and language-adapter transfer. Phi-4 achieves the best overall accuracy, MMS shows comparable CER, while Whisper exhibits high CER, consistent with known hallucinations and repetition loops. Performance saturates at different data scales, offering practical data-efficiency targets, while typological complexity consistently shapes error rates. Domain fine-tuning benefits Phi-4 and Whisper but degrades MMS, revealing limitations of adapter-based adaptation. Proxy language adapters outperform training MMS adapters from scratch.
[NLP-74] Beyond Passing Tests: A Systematic Review and Controlled Re-evaluation of Techniques for Improving the Quality of LLM -Generated Code
【速读】： 该论文旨在解决当前大语言模型（Large Language Models, LLMs）在代码生成质量评估中存在的系统性缺陷问题，即现有研究虽宣称可提升代码的可读性、效率与安全性等非功能性质量，但其证据分散、评估维度单一，且缺乏在相同计算预算下的公平对比。其核心挑战在于：多数研究仅关注自身目标维度，忽视了对正确性等基础指标的维持，且依赖的自动化质量信号易被“操纵”，同时计算开销、方差和数据污染等问题常被忽略。为此，论文提出系统性综述与受控重评估相结合的解决方案，关键在于通过在相同计算预算下对十种典型技术在四个主流模型上的多维度基准测试（涵盖正确性、可维护性、效率与安全性），实现跨方法的公平比较，并扩展至十三个现代模型及仓库级任务。研究发现，在单次调用中，质量提示（quality prompting）虽能增加分析器检测到的问题数量，但对复杂度与人工评分影响甚微；而在匹配预算条件下，基于分析器加权排序的 best-of-n 采样在多数指标上优于提示、规划、自洽修正、评审代理与检索等方法，而分析器反馈循环与微调仅在分析器问题数上占优。值得注意的是，部分高得分结果可能源于人类评估基准（如 HumanEval+）的污染，且在训练截止后举办的竞赛中未复现此类损失。在66种技术的头对头对比中，仅有1种技术在主要目标上显著超越 best-of-n，其表现符合随机预期，表明当前多数改进效果可能并不显著。因此，论文强调应以预算匹配的采样基线为参照，并报告独立于技术自身信号的评估指标，以确保评估结果的可靠性与可比性。
链接: https://arxiv.org/abs/2610.38977
作者: Prashank Kadam
机构: Cerby Inc.(Cerby公司)
类目: oftware Engineering (cs.SE); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 55 pages, 5 figures, plus 120 pages of supplementary material
Abstract:Large language models pass most functional-correctness benchmarks, but passing tests says little about whether their code is readable, efficient or secure. Many techniques claim to improve these qualities, yet their evidence is scattered, usually covers one quality dimension and is rarely compared at an equal compute budget. We contribute a systematic review and a controlled re-evaluation. A systematic literature review of 628 primary studies (July 2021 to October 2026), with disclosed, author-reviewed LLM assistance, organizes the techniques by stage, signal source and quality target. Most studies measure only the dimension they target, gains often come with unchanged or unreported correctness, single automated quality signals are gamed, and compute, variance and contamination are rarely reported. A controlled re-evaluation runs ten techniques on four models at matched budgets on correctness, maintainability, efficiency and security benchmarks, extends it to thirteen current models and repository-level tasks, and compares 66 highlighted techniques head-to-head on Gemma 4 31B. At one call, quality prompting raises analyzer issue counts and leaves complexity and judge scores almost unchanged. At matched budgets, best-of-n sampling with an analyzer-weighted ranker matches or beats prompting, planning, self-refinement, a reviewer agent and retrieval on measures the ranker does not optimize, while an analyzer feedback loop and fine-tuning beat it only on analyzer issue counts. Large correctness losses on HumanEval+, a likely contaminated benchmark near its ceiling, do not appear on contests held after the models’ training cutoffs. In the head-to-head, 1 of 66 techniques beats best-of-n on its primary target beyond the 95% interval, a rate consistent with chance. Gains should be reported against a budget-matched sampling baseline and on measures independent of the technique’s own signal.
[NLP-75] Efficient Agent ic Audio-Visual Question Answering through Toolset Pruning and Modality Consolidation
【速读】： 该论文旨在解决细粒度音视频问答（Fine-grained Audio-Visual Question Answering, AVQA）中跨模态对应关系复杂、时空定位困难的问题，尤其关注现有全模态智能体（Omni-modal Agent）依赖大量专用感知工具所导致的推理开销大、执行路径冗长等瓶颈。其核心解决方案在于通过系统性地剪枝与整合工具集，从依赖多类特定模态工具的复杂工作流，逐步简化为仅保留两个通用型工具——全模态图像/视频描述生成（Omni-modal Captioning）与全模态问答（Omni-modal QA）的紧凑型工作流。实验表明，该简化后的双工具智能体在三个主流AVQA基准上实现了与原专用工具智能体相当甚至更优的准确率，同时将推理成本降低了约70%，验证了在保持高性能的前提下，通过精简感知能力暴露即可实现高效、可扩展的跨模态理解。
链接: https://arxiv.org/abs/2610.38960
作者: Yi Wang,Tianzi Wang,Peter Bell
机构: University of Edinburgh (爱丁堡大学); Chinese University of Hong Kong (香港中文大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Audio and Speech Processing (eess.AS)
备注:
Abstract:Fine-grained audio-visual question answering (AVQA) requires models to resolve complex cross-modal correspondences and temporally grounding from sparsely distributed evidence. Omni-modal agents, which autonomously orchestrate perception tools to actively collect evidence, achieve strong performance, but it remains unclear whether active perception from both audio and visual modalities depends on broad specialized toolsets, especially audio or visual modality-specific tools. Such systems may also introduce long execution trajectories and substantial inference cost. We investigate whether comparable performance can be retained with a compact workflow exposing only essential perceptual capabilities. Starting from the specialized-tool OmniAgent, we conduct staged controlled experiments that prune the toolset and progressively consolidate modality-specific tools into omni-modal tools, motivating a two-tool agent comprising only Omni-modal captioning and Omni-modal QA. Across three widely used AVQA benchmarks, our compact agent achieves comparable or better accuracy than the specialized-tool agent while reducing inference cost by around 70%.
[NLP-76] Interventional Training Reduces Causal Bypass in In-Context Introspection
【速读】： 该论文旨在解决大语言模型（Language Model, LM）在自我反思（introspection）过程中存在的“因果绕过”（causal bypass）问题，即模型的自报告行为可能依赖于与目标状态相关但无因果关联的信号，从而导致其自我评估缺乏真实性与可解释性。现有研究虽认为通过监督微调（SFT）可增强模型的自我反思能力，但未有效避免此类伪相关性。为此，本文提出一种基于干预训练（interventional training）的新方法：在训练中引入反事实扰动（counterfactual perturbation），迫使模型对被人为修改的低层信号（如单个神经元激活值）进行准确报告。该方案通过在三种低层信号类型和八种不同模型上验证，证明其能实现对内生信号与注入信号更精确的因果性反思，尤其在分布外（OOD）数据中表现显著提升。该机制具有跨标记位置的泛化能力，并随训练数据量增加而增强。研究结果表明，虽然具备因果基础的自我反思是可行的，但不能仅通过行为层面的训练自然获得，必须显式地针对因果机制进行建模，强调了“训练机制而非仅行为”的关键必要性。
链接: https://arxiv.org/abs/2610.38911
作者: Zilu Tang,Jongin Kim,Kerem Sahin,David Atkinson,Aaron Mueller,Derry Wijaya
机构: Boston University (波士顿大学); Northeastern University (东北大学); Monash University Indonesia (蒙纳士大学印度尼西亚校区)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 9 pages, preprint
Abstract:Introspection, the ability of a system to access and reason about its internal signals, is crucial for language model (LM) self-consistency and scalable interpretability. Prior work argues that LMs possess introspective capabilities, and that training (SFT) could enhance such behavior. However, most setups leave open the possibility of “causal bypass,” in which the self-report relies on signals that are correlated with the self-report’s subject, while being causally unrelated to it. Although high-level concepts are difficult to intervene on, lower-level signals (such as a single neuron activation) allow precise interventions. Understanding such signals can serve as a minimal prerequisite before understanding more complex introspection. In this study, we train hundreds of introspectors across three low-level signal types and eight LMs. We find that straightforward SFT training leads to causal bypass, and propose interventional training, an alternative where we train the model to report on counterfactually perturbed values. This method enables more accurate introspection of endogenous \textitand injected signals, particularly in out-of-distribution (OOD) data. The mechanism generalizes across token positions and scales with more data. Our results demonstrate that while causally grounded introspection is possible, it is not guaranteed – emphasizing the need to train on mechanisms, rather than behaviors alone.
[NLP-77] HEDGE: Gated Hypergraph Memory for Scalable Multi-Agent Debate
【速读】： 该论文旨在解决多智能体辩论系统中因上下文窗口有限和记忆结构无序导致的关键信息难以有效检索与复用的问题。现有扁平化或成对图结构的记忆机制无法充分捕捉多智能体之间的高阶关系，如多方共识、矛盾证据及共识演进过程。其解决方案的关键在于提出HEDGE框架，通过语义超图（semantic hypergraph）建模辩论轨迹：超边（hyperedges）关联跨智能体的相关论点，而超边内部的关联图（relational graph）则保留论证的细粒度结构。同时，引入一个通过监督学习与强化学习联合训练的记忆选择模块（Memory Selection Module, MSM），可动态决策每个记忆候选的处理路径（存储、备份或丢弃），在保留关键信息的同时显著降低冗余。实验表明，HEDGE在七个多样化基准上均优于主流单/多智能体及记忆基线模型，平均提升准确率5.0个百分点，尤其在冲突密集任务中达8.2个百分点；同时实现48%的内存压缩与34%的辩论轮次减少。
链接: https://arxiv.org/abs/2610.38902
作者: Amirhossein Abaskohi,Mahdi Mostajabdaveh,Zirui Zhou
机构: University of British Columbia (不列颠哥伦比亚大学); Huawei Technologies Canada (华为技术加拿大)
类目: Computation and Language (cs.CL)
备注:
Abstract:Multi-agent debate systems require agents to retain and revisit arguments across multiple rounds, yet limited context windows and unstructured memories make relevant information increasingly difficult to recover. Existing flat and pairwise-graph memories also struggle to capture higher-order relations, including multi-agent agreement, conflicting evidence, and evolving consensus. We introduce HEDGE, a selective memory framework that represents debate trajectories as semantic hypergraphs. Hyperedges connect related claims across agents, while intra-hyperedge relational graphs preserve fine-grained argumentative structure. A learned memory selection module (MSM), trained through supervised learning and reinforcement learning, routes each memory candidate to STORE, BACKUP, or OMIT, retaining useful information while limiting redundancy. Across seven diverse benchmarks, HEDGE consistently outperforms strong single-agent, multi-agent, and memory baselines, improving accuracy by 5.0 points on average over the strongest multi-agent system with GPT-4o-mini. The gains reach 8.2 points on conflict-heavy tasks, while HEDGE reduces memory usage by 48% and the number of debate rounds by 34%.
[NLP-78] What to randomize matters: a controlled study of volume axis and representation in low-data molecular generation
【速读】： 该论文旨在解决在小样本数据条件下，化学语言模型（Chemical Language Models）难以生成真正新颖分子的问题，即模型容易重复训练集中的分子而非学习到有效的生成分布。其核心挑战在于如何通过合理的数据增强策略提升模型在低数据场景下的泛化能力与生成多样性。解决方案的关键在于引入并系统评估一种基于顺序连接的片段嵌入方法（Sequential Attachment-based Fragment Embedding, SAFE），通过随机化原子排列顺序和片段块顺序两种维度的表示变化，显著提升生成新分子的能力。研究发现，仅使用SAFE的随机化即可使真实新颖分子的生成率提高450倍以上，且原子顺序与片段顺序的随机化具有协同效应而非简单叠加，二者共同作用时产生的增益是各自独立贡献的四倍；此外，在相同数据规模与计算资源下，随机化SAFE的生成表现优于随机化SMILES，其真实新颖分子产量高出2.8倍。因此，在低数据条件下，决定数据增强是否能产生真正的新分子而非仅改变书写形式的关键因素，不在于随机化的程度，而在于所采用的随机化机制本身。
链接: https://arxiv.org/abs/2610.38835
作者: Neerav Kaushal,Alisa Teige,Ajay MNV Penmatsa
机构: Michigan Technological University (密歇根理工大学)
类目: Machine Learning (cs.LG); Computational Engineering, Finance, and Science (cs.CE); Computation and Language (cs.CL); Biomolecules (q-bio.BM); Quantitative Methods (q-bio.QM)
备注: 27 pages, 9 figures, 5 tables
Abstract:Chemical language models can struggle to generate new molecules when trained on small datasets, where a model may reproduce training examples rather than learn a useful generative distribution. Randomized SMILES augmentation varies atom traversal, whereas Sequential Attachment-based Fragment Embedding (SAFE) also allows the order of fragment blocks to vary. How these distinct sources of representational variation affect low-data molecular generation has not been clearly studied. Here, using 1,000 seed molecules to train an autoregressive sequence model, we compare canonical SAFE with randomized SAFE at increasing corpus sizes, isolate atom-order and fragment-order axes of SAFE randomization, and compare fully randomized SAFE with randomized SMILES at molecule-by-molecule matched training corpora. Three findings emerge. First, randomization of SAFE alone raises the genuine yield of novel molecules more than 450-fold, turning a model that merely reproduces its training set into a practical molecular generator. Second, the two axes of randomization act jointly rather than additively: atom order alone reaches a tenth, and fragment order alone a seventh, of the genuine yield obtained when both vary, which is four times their combined contribution. Third, at matched corpus size and compute, randomized SAFE delivers 2.8 times the genuine yield of randomized SMILES. Welch tests with Games-Howell post-hoc comparisons resolve every difference on the primary outcomes. In the low-data regime, thus, what one randomizes rather than only how much determines whether augmentation produces new molecules or merely new spellings of the training set.
[NLP-79] CopyLoop: Fast Looped Transformer Decoding via Context Copying and Shallow Screening
【速读】： 该论文旨在解决循环变压器（Looped Transformers）在生成过程中因递归深度增加而导致的全深度解码计算开销过大的问题。现有方案虽通过推测性流水线实现多步并行生成，但仍受限于序列解码的固有延迟。其解决方案的关键在于提出一种无需训练的解码算法CopyLoop，该方法结合低成本的上下文复制提议、浅层筛选与连续递归调度机制：通过重复利用上下文生成多标记后缀，并在浅层递归深度下提前评估这些候选，高效剔除错误分支；同时允许新块与旧块并行处理，在早期块完成全深度验证前即推进后续生成。实验表明，CopyLoop在Ouro、Raven和Huginn系列共六种模型上实现了2.98x–10.70x（基础模型）及2.74x–3.52x（思维模型）的解码加速，显著优于所有对比基线。消融分析进一步验证了浅层筛选与连续调度之间的强协同效应，且CopyLoop仅需轻量级模型校准，尤其在长序列生成与可复用上下文中表现出更优加速性能。
链接: https://arxiv.org/abs/2610.38822
作者: Wenyu Du,Stephen Chung
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: preprint
Abstract:Looped Transformers increase computational capacity through the repeated application of shared blocks, but their recurrent depth makes conventional full-depth decoding expensive. While existing speculative pipelines accelerate generation by processing consecutive token positions at varying depths, they are typically limited by sequential decoding overheads. We introduce CopyLoop, a training-free decoding algorithm that combines low-cost context-copy proposals with shallow screening and continuous recurrent scheduling. Rather than step-by-step generation, CopyLoop proposes multi-token suffixes from repeated context. It then evaluates them early at a shallow recurrent depth to efficiently filter out incorrect branches, processing newer blocks concurrently while earlier blocks are still progressing toward full-depth verification. We evaluate CopyLoop across six model checkpoints from the Ouro, Raven, and Huginn families, covering code generation and multi-step mathematical reasoning. CopyLoop achieves decoding speedups of 2.98x–10.70x over full-depth autoregressive decoding on Base models and 2.74x–3.52x on Thinking models, consistently outperforming or closely matching all competing baselines across evaluation tasks. Furthermore, ablation studies confirm a strong synergy between shallow screening and continuous scheduling, while our analyses show that CopyLoop requires only lightweight model calibration and yields greater speedups on longer generations with reusable contexts.
[NLP-80] Gold vs.~Translation: Are We Measuring Language Knowledge or Locality in Translated Benchmarks? EMNLP2026
【速读】： 该论文旨在解决翻译生成的英文基准测试（如MMLU）在非英语语言中用于评估大语言模型时，其测量构念是否与本地原创基准一致的代理效度（proxy validity）问题。其核心解决方案在于通过实证比较深度学习翻译（DeepL）生成的MMLU译本与六种本地原创基准在五种北欧及波罗的海语言中的表现，采用多语言配对设计（共七组）评估29个模型，不仅考察排名一致性，更引入增量代理效度检验：在控制模型规模和语言能力后，翻译得分是否仍能额外预测本地表现。研究发现，代理效度在不同基准对间差异显著；仅四组呈现稳健效应，两组边缘显著，一组无显著效果。其中，内容与课程重叠度高的基准（如Skolprov）表现出最强代理效度（ΔR² = .128，95% CI [.038, .253]），而针对瑞典特定事实知识的开放问答型基准SwedishFacts则未显示翻译分数的增量预测价值（p = .242）。由于SwedishFacts为唯一开放格式基准，无法完全分离本地内容与响应格式的影响。总体结论表明，翻译基准作为代理工具的有效性取决于源内容与目标语言评估任务的语义重叠程度，本地化知识的捕获可靠性较低。
链接: https://arxiv.org/abs/2610.38788
作者: Klaudia-Doris Thellmann,Bernhard Stadler,Dan Saattrup Nielsen,Michael Färber,Jens Lehmann
机构: TUD Dresden University of Technology (德累斯顿工业大学); ScaDS.AI; Alexandra Institute
类目: Computation and Language (cs.CL)
备注: Published in EMNLP 2026
Abstract:Translated versions of English-origin benchmarks such as MMLU are widely used to evaluate large language models in non-English languages, yet it remains unclear whether they measure the same constructs as natively created benchmarks. We frame this as a proxy validity question and compare DeepL-translated MMLU proxies against six native benchmarks in seven pairings across five Nordic and Baltic languages, evaluating 29 models on each pairing. Beyond ranking agreement, we test incremental proxy validity: does the translated score add predictive value for native performance after controlling for model size and language proficiency? Proxy validity varies substantially across benchmark pairings. Applying the same robustness criteria to all seven pairings, we find four robust effects, two marginal effects, and one null result. The strongest evidence occurs for curriculum-overlap benchmarks such as Skolprov ( \Delta R^2 =.128, 95% CI [.038,.253]), whereas translated MMLU provides no significant incremental value for Sweden-specific factual knowledge in SwedishFacts (p=.242). Because SwedishFacts is also the only open-ended benchmark, we cannot fully separate the effects of local content and response format. Overall, the results indicate a clear pattern: translated benchmarks provide a better proxy when their source content overlaps with the native evaluation target, while locally anchored knowledge is less reliably captured.
[NLP-81] AtomTree: A Hierarchical Framework for State-Aware Embodied Instruction Following with LLM s IROS2026
【速读】： 该论文旨在解决交互环境中具身指令跟随任务中长期规划的鲁棒性与状态感知不足问题。现有规划方法或依赖大量标注轨迹数据，或采用少样本/零样本大语言模型（LLM）提示生成一次性完整计划，此类方法过度依赖语言常识、在初始指令后即固化计划路径，仅在失败时才进行更新，导致对环境状态变化响应迟缓、容错能力差。其解决方案的关键在于提出一种基于大语言模型的分层状态感知规划框架AtomTree：该框架通过增量式构建子目标与动作的树状结构，在每一步由LLM根据当前符号化状态判断是否继续分解子目标或输出可执行的底层动作；引入轻量级记忆模块以复用历史决策结果，提升决策一致性；当低层动作执行停滞时，触发层级回溯重规划机制，返回父节点并基于最新状态重新生成子树计划，有效避免沿错误轨迹持续漂移。在ALFRED基准的零样本评估中，AtomTree显著优于先前基于LLM的规划方法。
链接: https://arxiv.org/abs/2610.38744
作者: Haotian Zhai,Tianming Sha,Junnan Li
机构: University of Minnesota(明尼苏达大学); Stony Brook University(石溪大学); University of Wisconsin–Madison(威斯康星大学麦迪逊分校)
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: Accepted at IROS 2026
Abstract:Embodied instruction following in interactive environments requires agents to interpret natural language commands and complete long-horizon tasks. Existing planners either rely on large labeled trajectories, or use few-shot and zero-shot LLM prompting that produces a whole plan in a single serial reasoning process. These approaches depend on linguistic common sense, commit to a plan from the initial instruction, and only update after failures, which weakens robustness and state awareness. We present AtomTree, a hierarchical framework for state-aware task planning with LLMs. AtomTree incrementally builds a tree of subgoals and actions, where the LLM decides at each step, from the current symbolic state, whether to further decompose or to output a grounded low-level action. A lightweight memory module summarizes past outcomes for reuse in later decisions. When a low-level action stalls, AtomTree triggers hierarchical backtracking replanning to return to the parent node and regenerate the subtree plan from the latest state, preventing continued drift along an incorrect trajectory. In zero-shot evaluation on the ALFRED benchmark, AtomTree outperforms prior LLM-based planners. Project page: this https URL.
[NLP-82] he Quine Test: How Much of a Models Self Survives Its Own Description?
【速读】： 该论文旨在解决生成式语言模型（Generative AI）中“偏好可迁移性”问题，即：当模型表现出一致的行为偏好时，这些偏好是源于其内在的、可泛化的决策机制，还是仅是其在特定角色扮演（persona）下的临时表现。核心挑战在于，仅凭行为观察无法区分偏好是模型自身属性还是角色设定的结果。为应对这一问题，研究提出了一种基于“自指”（self-reference）的验证方法——类昆（quine）测试，其关键在于检验模型对自身的描述是否能作为可移植的偏好规范（preference profile specification），从而在其他模型或同一模型的新实例中复现其行为。实验结果显示，尽管模型的自我描述在一定程度上影响自身行为，但其作为“可复制的偏好编码”效果有限，无法可靠地在异构模型间重建原模型的偏好表现；相比之下，直接传递原始偏好答案或显式策略文本则具有更强的可迁移性。因此，该研究的关键发现是：当前模型的自然自述（native self-descriptions）虽具一定行为引导作用，但缺乏足够的精确性和可重复性以实现跨模型偏好复制，而系统性的偏好指定方式（如显式策略或答案列表）才是实现高保真行为再现的有效路径。
链接: https://arxiv.org/abs/2610.38726
作者: Bibin Babu,Ivan P. Yamshchikov,Aleksei Shpilman
机构: Technical University of Applied Sciences Würzburg-Schweinfurt (THWS); Sirius University
类目: Computation and Language (cs.CL)
备注: 32 pages, 3 figures, 12 tables. Code, data and preregistration files: this https URL
Abstract:Frontier language models express coherent preferences, but behavioral evidence alone cannot say whether these belong to the model or to a character it portrays. Following Hofstadter’s view of the self as a self-description, we ask whether a model’s description of itself works as a portable specification of its preference profile, a quine test: can the text a model writes about itself regenerate its behavior elsewhere? Ten models from nine labs write self-descriptions; we place each one in the prompt of other models (receivers) and of fresh instances of the author, and measure how closely their choices on a 110-item preference battery match the author’s. First, the tested native self-descriptions (native: the profile measured under a generic assistant prompt) show no average reconstruction gain across variations in length, framing, wording and prompt placement: on a 0 to 1 fidelity scale where a model’s fidelity to a fresh instance of itself is 0.979 and to unrelated models about 0.89, the one-sided 95% upper bound on the pooled gain is below +0.01 in every tested system-prompt condition, and descriptions based on the author’s observed choices do no better. Second, the channel itself is strong: explicitly listing the author’s answers lets receivers reproduce its profile at fidelity 0.955 to 0.960 across three model grids, within 0.03 of same-model repeatability. Third, self-descriptions written under an explicit preference policy transfer the assigned preferences almost as well as the policy text itself. Fourth, reading its own self-description shifts even the author away from its baseline, modestly more than unrelated prose of the same length. For the preferences measured here, native self-descriptions influence behavior without reliably specifying how to reproduce it in another model. We release the pipeline, battery, raw call logs, and 198 model-written descriptions. Comments: 32 pages, 3 figures, 12 tables. Code, data and preregistration files: this https URL Subjects: Computation and Language (cs.CL) Cite as: arXiv:2610.38726 [cs.CL] (or arXiv:2610.38726v1 [cs.CL] for this version) https://doi.org/10.48550/arXiv.2610.38726 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[NLP-83] Learning Scientific Trajectories to Predict Future Biomedical Discoveries
【速读】： 该论文旨在解决如何利用大规模生物医学文献中结论的语义演变模式，以生成具有生物学合理性和新颖性的科学假设。传统方法依赖于静态提示（prompt-based）或检索增强模型，难以捕捉知识演进中的动态趋势。其核心解决方案是提出ForeSCITE框架，该框架通过分析三百万条预截止时间（pre-cutoff）生物医学结论的语义表示，识别主题聚类，并在每个聚类内估计从早期到近期研究发现的时间方向向量。随后，将现有结论嵌入（embedding）沿这些方向投影，再使用一个经过训练的嵌入语言模型（embedding language model）将投影坐标解码为自然语言形式的假设。该方法不仅揭示了语义空间中显著的时间结构，且在回溯性时间切片评估中表现出优于五种基线模型的语义覆盖广度、生物实体对的新颖性以及截断后新观察实体对的重现率。专家评审验证了其中37个优先级较高的CAR-T相关假设兼具新颖性与生物学合理性，五个被选中的假设亦获得充分的生物学机制解释及后续文献支持。研究表明，基于语义空间中时间演化规律的引导机制，可作为一种可扩展的方法，用于生成可供后续科学研究验证的高质量生物医学假说。
链接: https://arxiv.org/abs/2610.38719
作者: Chia-Hsuan Chang,Brian Ondov,Jiang Lv,Lan Jiang,Yunmai Wang,Haerin Song,Xueqing Peng,Qiaozhu Mei,Xiaolei Su,Hua Xu
机构: Yale School of Medicine, New Haven, CT 06510, USA; Yale School of Medicine, New Haven, CT 06520, USA; Seoul National University, Seoul 08826, Republic of Korea; University of Michigan, Ann Arbor, MI 48109, USA
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:AI co-scientists have expanded the use of large language models for generating scientific hypotheses and guiding experimental investigation. Here we investigate whether temporal patterns in the semantic representations of complete biomedical conclusions can guide hypothesis generation. We introduce ForeSCITE, a framework that identifies thematic clusters of published conclusions, estimates a direction separating earlier from more recent findings within each cluster, and projects existing conclusion embeddings along these directions. An embedding language model trained to reconstruct conclusions decodes the projected coordinates into natural-language hypotheses. Across 407 clusters derived from approximately three million pre-cutoff conclusions, we identify temporal structure and illustrate its retention after decoding. In retrospective, time-sliced evaluations, ForeSCITE increases semantic coverage beyond the historical corpus, biomedical entity-pair novelty and post-cutoff pair re-observation relative to five prompt-based and retrieval-augmented baselines. Expert review of prioritized CAR-T hypotheses identified 37 candidates as both novel and biologically plausible, and expert justifications for five selected hypotheses illustrate their biological rationale and subsequent literature support where available. These findings support temporal guidance in semantic space as a scalable mechanism for proposing biomedical hypotheses for subsequent scientific evaluation.
[NLP-84] Beyond fabrication: Misattributed numbers in papers written by AI research agents
【速读】： 该论文旨在解决生成式 AI 代理（AI agents）在撰写科研论文时所产生数值主张（numerical claims）的可信性验证问题。当前方法普遍存在两大缺陷：一是依赖单一大语言模型（LLM）进行判别，缺乏对错误类型的深入解释；二是现有验证机制仅关注论文内部一致性或数值是否出现在代理输出中，未能追溯其真实来源。研究通过对540条公开评测结果的重新分析及对18篇提交论文中256个数值主张的手动标注发现，多数失败源于论文与底层实验工作之间的对齐偏差，而非明显造假，约四分之三的无支持主张源于代理自身文件中错误的实验、指标或运行记录。为此，论文提出一种确定性的“主张-证据”核查器（claim-to-evidence checker），通过固定规则判断主张周围上下文是否明确定位了对应结果，并以数值干扰项（numerical decoys）进行校准。该工具在盲评环境下实现超过90%的精确率（78/85，Cohen’s kappa = 0.89），虽仅能验证部分主张，但未被验证的主张大多仍可追溯，因此验证失败不等于虚假。相比之下，基于LLM的验证器虽能识别更多可追溯主张，却会误接受约三倍于正确比例的错位主张。二者结合使用可互补优势。将该核查器应用于99份提交物后，揭示不同代理间存在显著的可追溯性差异，且另一供应商的独立评审者在保持原有代理排名的同时，使通过率提升一倍。研究最终倡导采用“溯源优先”（provenance-first）的审计范式来保障人工智能生成科研成果的可靠性。
链接: https://arxiv.org/abs/2610.38691
作者: Mosab Alfaqeeh,Omar Hourani,Muath Hourani,Hashem Alfaqeeh
机构: Al-Ahliyya Amman University (阿赫利耶亚安曼大学); Binghamton University, State University of New York (宾汉姆顿大学，纽约州立大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Digital Libraries (cs.DL)
备注: 42 pages, 5 figures, 13 tables. Code and data: this https URL
Abstract:Verifying numerical claims in research written by AI agents is an important but challenging problem. Recent benchmarks show that many agent-written papers fail integrity checks, but such verdicts, often produced by a single LLM judge, identify failure without explaining its nature, and existing checks either test a paper for internal consistency or test whether a reported value appears somewhere in an agent’s output. We re-analyze all 540 released judgments from a recent public benchmark of AI research agents (nine agents, fifteen open research problems) and manually label 256 numerical claims from 18 submissions. Failures concentrate on alignment between the paper and the underlying work rather than overt cheating, and about four in five unsupported claims correspond to values in the agent’s own files that come from the wrong experiment, metric, or run. We introduce a deterministic claim-to-evidence checker that scores whether the surrounding context identifies the claimed result and calibrates each claim against numerical decoys. With its rules fixed before labeling, it certifies claims with over 90% precision against blinded annotators (78 of 85; Cohen’s kappa = 0.89), while verifying only a subset of claims; most claims it cannot verify are nevertheless traceable, so failure to verify is not evidence of falsehood. An LLM verifier given the same evidence recovers more traceable claims but accepts about three times as many misattributed ones, so the two are best used together. Applied to 99 submissions, the checker reveals large differences in traceability across agents, and a second judge from a different vendor preserves the official agent ranking while passing twice as many papers. These findings motivate provenance-first auditing of AI-produced research.
[NLP-85] Weaken to Strengthen: Weakening-Induced On-Policy Self-Distillation for Large Reasoning Models
【速读】： 该论文旨在解决在缺乏外部教师模型（external teacher）的情况下，如何有效提升生成式 AI (Generative AI) 推理模型性能的问题。传统自蒸馏方法依赖于特权信息（privileged information）来构建更强的自教师模型，但在仅拥有未标注提示（unlabeled prompts）的场景下难以应用。本文提出了一种新型自蒸馏框架——弱化诱导的在线策略自蒸馏（Weakening-Induced On-Policy Self-Distillation, WISD），其核心创新在于不依赖特权信息，而是通过构造原模型的较弱版本（如采用低比特量化或权重幅度剪枝）作为对比基准，并利用两者之间的差异作为训练信号。该方案的关键在于以“弱化”方式生成对比模型，从而实现无需外部监督的高效知识迁移。实验表明，该方法在多种架构与规模的模型上均显著提升数学与代码推理任务的准确率，尤其在Qwen系列模型上表现突出，验证了WISD作为无特权信息条件下大模型推理优化的有效范式。
链接: https://arxiv.org/abs/2610.38681
作者: Yecheng Wu,Song Han,Han Cai
机构: Massachusetts Institute of Technology (麻省理工学院); NVIDIA(英伟达)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: Work in progress. Code is released at this https URL
Abstract:Self-distillation offers a way to improve reasoning models when a suitable external teacher is unavailable. However, common approaches rely on privileged information to construct a stronger self-teacher, limiting their use when only unlabeled prompts are available. In this paper, we introduce Weakening-Induced On-Policy Self-Distillation (WISD), which instead constructs a weaker counterpart of the original model and uses the contrast between the two as a training signal without requiring privileged information. We explore two ways to construct the weaker model: low-bit quantization and magnitude pruning. We evaluate WISD across diverse architectures and scales on mathematical and coding reasoning tasks. Across these benchmarks, quantization and pruning improve average accuracy over the original models by 4.0 and 5.5 percentage points, respectively, achieving state-of-the-art performance even without privileged information. Notably, pruning raises Qwen3.5-4B’s accuracy from 57.7% to 70.5% on HMMT 2025 and from 39.1% to 56.7% on LiveCodeBench v6. Quantization raises Qwen3.6-35B-A3B’s accuracy from 68.5% to 74.6% and from 61.8% to 69.5% on the same benchmarks, respectively. These results establish WISD as a novel and effective self-distillation framework for large reasoning models.
[NLP-86] AskStream: When and What to Ask Proactively under Uncertainty in Streaming Video Understanding WWW
【速读】： 该论文旨在解决当前生成式视频理解模型在主动交互（proactive interaction）范式下，面对信息不完整或不确定性时的响应缺陷问题。现有方法通常假设查询所需信息是完整的，导致模型仅能被动等待或直接回答，而在真实场景中，视频内容或用户意图常存在不确定性，这会引发模型无限等待或生成高度幻觉性的错误回应。为此，论文提出重新定义主动交互范式，引入第三种核心动作——“询问”（ask），使模型具备主动请求缺失信息的能力。其解决方案的关键在于构建首个面向流式视频理解的多轮交互基准数据集AskStream，涵盖视频与用户双重不确定性，包含438个视频及2,714个帧级时间窗口标注（分别对应等待、询问、回答三种行为），并设计了四类任务与八种子任务。实验表明，即使最强模型在所提出的Ask-F1评估指标上也仅达52.18%，揭示了当前模型在不确定环境下主动获取信息能力的显著不足，凸显了引入“询问”机制的必要性与挑战性。
链接: https://arxiv.org/abs/2610.38678
作者: Lu Wang,Zhuoran Jin,Longxiang Wang,Junhong Chen,Yupu Hao,Yao Xu,Daojian Zeng,Yubo Chen,Kang Liu,Guang Liu,Jun Zhao
机构: Chinese Academy of Sciences (中国科学院); Beijing Academy of Artificial Intelligence (北京人工智能研究院); Hunan Normal University (湖南师范大学); Minzu University of China (中央民族大学)
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL)
备注: 42 pages, 9 figures. Project: this https URL ; Code: this https URL ; Dataset: this https URL
Abstract:Recent advances in multimodal large language models have increased interest in interaction paradigms for video understanding. The evolution from offline to streaming and proactive interaction has gradually shifted models from passive responders toward more proactive agents that can decide when to answer. However, existing proactive paradigms typically assume that the information required to answer a query is complete, leaving models to only wait or answer. In the real world, however, information is often incomplete due to uncertainty from either the video or the user. In proactive paradigms, such uncertainty can cause the model to wait forever or generate highly hallucinatory responses. We therefore argue that proactive interaction should be redefined with a third action: ask, allowing models to actively ask for missing information. To study when and what models should ask before answering, we introduce AskStream, the first multi-turn benchmark spanning video and user uncertainty for streaming video understanding, with four task categories and eight subtasks. It contains 438 videos with 2,714 manually annotated frame-level time windows across three types: wait, ask, and answer. Across both multiple-choice and open-ended settings, even the strongest model attains only 52.18% on our proposed Ask-F1 metric, highlighting substantial limitations in proactive information seeking under uncertainty.
[NLP-87] Right Numbers Wrong Questions: Evaluating LLM s on Underspecified Financial Queries
【速读】： 该论文旨在解决金融助手在处理看似客观但实则依赖用户隐含信息的问答时所面临的“框架坍缩”（frame collapse）问题。具体而言，当用户未明确说明数据指标的具体含义、产品变体类型或投资者自身情况及适用规则时，模型可能基于其训练数据中的表面模式生成看似合理但实际偏离问题本意的答案。这种错误源于模型对上下文语义理解的缺失，导致其在缺乏显式提示的情况下默认或误用假设，从而输出与真实意图不符的结果。研究以印度共同基金为场景，利用形式化定义的产品变体与税务规则构建可验证的基准测试，发现即使在标注清晰的情况下，模型仍会受布局顺序影响优先展示特定选项；在缺少投资者关键信息时，较弱模型倾向于采用默认或过时规则，而强制要求输入并不保证该信息被正确使用。此外，仅依赖数据一致性校验无法识别此类错误，因为答案虽符合数据但对应的是错误解读。研究进一步表明，只有通过引入具备领域知识的AI裁判并明确字段语义，才能有效检测出误标图表等问题。因此，解决方案的关键在于：建立对上下文依赖性的显式建模机制，并结合基于领域知识的验证框架，防止模型因忽略隐含前提而导致推理偏差。
链接: https://arxiv.org/abs/2610.38631
作者: Radha Yadav,Prasanna Devadiga,Peeyush Mishra,Apoorv Singh
机构: TIFIN(印度理论物理研究所)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); General Finance (q-fin.GN)
备注: Accepted for oral presentation at the 3rd Workshop on LLMs and Generative AI for Finance (AI4F) at ACM ICAIF '26, Milan, Italy, November 14, 2026. Non-archival. 16 pages, 4 figures
Abstract:Financial assistants answer questions that sound factual but depend on what the user leaves unsaid: what a data figure measures, which product variant is meant, and the investor’s situation and the rules that apply. An assistant that ignores these can still answer with a real number from its data, one that answers a different question. We call this frame collapse. We study it in Indian mutual funds, where product variants and tax rules are formally defined, so the correct answer can be computed for every reading. One study tests each challenge. Given an ambiguously labelled figure, models used it, often under the user’s label. Given every alternative labelled, they listed them all, but the layout decided which came first. Missing a fact about the investor, weaker agents assumed a default or an outdated tax rule; requiring the fact did not guarantee its use. Because such answers match their data, checks against the data pass them. An AI judge caught mislabelled figures only when told what each field means, and still passed answers that silently chose another reading. We show how these failures arise, not how often, and propose controls for financial assistants.
[NLP-88] Cross-Primitive Consistency in Jev
【速读】： 该论文旨在解决生成式AI（Generative AI）在使用大语言模型（LLMs）进行决策时，不同查询范式（如二元、多选等）之间决策一致性不足的问题。尽管Jev通过类型化查询原语实现了高效决策并支持同一语义状态的多种表达形式，但其跨范式的一致性尚未得到系统评估。为此，本文提出PrimaCons——一个四轴基准测试框架，用于在保持查询语义一致性的前提下评估跨查询范式的一致性。实验结果表明，在三种不同规模的Jev设置下均存在显著的一致性偏差，且这些偏差呈现系统性模式，其与最终决策失败具有强相关性（ROC-AUC达0.91）。为缓解该问题，研究提出交叉范式残差校准器（Cross-Primitive Residual Calibrator, CPRC），该方法在冻结主干网络的基础上附加轻量级残差网络，联合优化预测准确率与跨范式一致性。在留一域排除验证协议下，CPRC使跨范式不一致性降低58.57%，同时提升准确率2.97%，在域内与域外场景中均表现出优异性能。本研究确立了跨范式一致性作为评估Jev系统可靠性的重要维度，并提供了可落地的改进方案。
链接: https://arxiv.org/abs/2610.38623
作者: Xu Liu,Guikun Chen,Kanzhi Wu,Wenguan Wang
机构: Zhejiang University (浙江大学); vivo
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL)
备注:
Abstract:Jev enables efficient decision-making with large language models (LLMs) through typed querying primitives, allowing the same semantic state to be expressed in different formats, such as binary and multiple-choice queries. However, the consistency of decisions across primitives has not been systematically evaluated. This work introduces PrimaCons, a four-axis benchmark for assessing cross-primitive consistency while preserving the semantics of the query. Evaluations across three Jev sizes reveal consistency violations across all four axes. Our analysis shows that these violations follow systematic patterns and that cross-primitive discrepancies predict decision failure with a ROC-AUC of 0.91. To mitigate these inconsistencies, we propose the Cross-Primitive Residual Calibrator (CPRC), which attaches a lightweight residual network to a frozen backbone and jointly optimizes predictive accuracy and cross-primitive consistency. Under a leave-one-domain-out protocol, CPRC reduces cross-primitive inconsistency by 58.57% and improves accuracy by +2.97%., demonstrating its effectiveness in both in-domain and out-of-domain settings. Our findings establish cross-primitive consistency as a key dimension for evaluating Jev and provide a practical approach to improving the reliability of its decisions. Code and benchmark is available at this https URL.
[NLP-89] Repairs for Whom? Evaluator-Executor Mismatch in Selecting Repairs for Failed LLM Agents
【速读】： 该论文旨在解决大语言模型（Large Language Model, LLM）智能体在执行任务失败后，修复方法中评估器（evaluator）所选择的修复方案是否真正有助于实际部署环境中的性能提升这一关键问题。其核心挑战在于，现有修复方法通常依赖于贪心回放（greedy replay）、更强的模型、环境或裁判（oracle/judge）等外部评估机制来筛选候选修复，但这些评估器的判断可能与实际部署策略的执行效果存在偏差。研究的关键发现是：在固定失败轨迹和候选单步修复的前提下，使用独立的部署策略进行多次采样（rollouts）作为真实评估基准，可发现当前主流评估方式显著低估了修复方案的实际价值。具体而言，在未经过修复价值筛选的失败案例中，贪心回放相比使用部署模型自身滚动生成的评估，导致预期净成功增益下降0.08–0.13（ALFWorld）和净奖励损失0.06–0.13（WebShop）；当以更大规模的Qwen2.5-14B模型替代原7B部署模型进行评估时，每轮修复的20次14B滚动生成的价值仅相当于两个7B模型自身的滚动生成，且在存在分歧的案例中，两模型偏好相反的修复方案。此外，不同模型家族间相似尺寸的模型之间也出现双向损失（0.07–0.17）。研究进一步表明，评估器的选择会改变修复位置标签和修复方法的排序。因此，论文提出的核心解决方案是：修复应由将实际执行它的策略进行验证，修复标签必须明确标注其执行者，以确保评估与部署的一致性，从而避免因评估偏差导致的无效甚至有害修复。
链接: https://arxiv.org/abs/2610.38610
作者: Yijun Liao,Fanwei Liang
机构: Independent Researcher(独立研究员); South China University of Technology(华南理工大学)
类目: oftware Engineering (cs.SE); Computation and Language (cs.CL)
备注: 17 pages, 4 figures, 11 tables
Abstract:When an LLM agent fails, repair methods choose which action to replace and what to put in its place. They often score candidate repairs with greedy replay, a stronger model, an environment oracle, or a judge that reads the failed log. We ask whether these evaluators pick repairs that help the agent that will actually execute them. Holding failed trajectories and candidate single-step repairs fixed, we label every candidate with independent rollouts of the deployed policy and measure how much deployment value each evaluator’s choice loses. On failures not screened for repair value, pre-specified tests show that greedy replay loses 0.08-0.13 in expected net success gain in ALFWorld and 0.06-0.13 in net reward in WebShop, relative to selection with the deployed agent’s own rollouts. At an equal rollout budget, selecting repairs for Qwen2.5-7B with Qwen2.5-14B rollouts loses 0.25 in ALFWorld and 0.07 in WebShop, twenty 14B rollouts per candidate are worth no more than two of the 7B agent’s own, and fresh rollouts confirm that the two models prefer opposite repairs on selected disagreement cases. Between similar-sized models from different families, the losses appear in both directions in both environments (0.07-0.17). The scoring executor also changes repair-location labels and method rankings. Repairs should be verified with the executor that will run them, and repair labels should name their executor.
[NLP-90] Rethinking LLM Jailbreak Evaluation: Continuous Risk Landscapes Beyond Binary Outcomes
【速读】： 该论文旨在解决生成式语言模型（LLM）越狱攻击评估中长期存在的二元成功指标（binary success indicator）所带来的信息缺失问题。传统方法将越狱攻击的成功与否简化为“成功/失败”的二值判断，忽略了生成过程中潜在的多路径分布特性，尤其在开放生成任务中，这种简化会掩盖模型对不同初始词元分支（first-token branches）所表现出的连续风险结构。其解决方案的关键在于提出一种连续、上下文相关的风险景观（continuous, context-conditioned risk landscape），通过固定候选开头词元并反复采样后续内容，构建出对首词元分支的风险分布估计。这一新范式揭示了两个此前被二元指标遮蔽的核心属性：当前解码暴露程度（current decoder exposure）与安全重定向能力（safe rerouting capacity）。该连续表示不仅能够更全面地刻画越狱攻击对模型输出分布的影响，还能为清洁提示、攻击变体及对抗自适应防御下的攻击强度提供统一的比较基准。实验表明，许多看似安全的输出下隐藏着高风险分支，且攻击策略在防御响应前后的有效性可能发生根本性转变。因此，论文将越狱行为重新定义为对生成延续风险景观的形变过程，为理解越狱机制与构建更安全的语言模型提供了更具解释力的理论基础。
链接: https://arxiv.org/abs/2610.38595
作者: Yukai Zhou,Wenjie Wang
机构: ShanghaiTech University (上海科技大学)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注:
Abstract:A binary success indicator is a natural representation for discriminative tasks such as image classification under adversarial perturbations, where each input is mapped to a single prediction. However, generative tasks are fundamentally different: an LLM jailbreak prompt induces a distribution over open-ended trajectories, while a sampled response reveals only one path. Yet LLM jailbreak evaluation typically retains the binary response indicator underlying attack success rate, obscuring the surrounding structure of possible generations. We introduce a continuous, context-conditioned risk landscape over first-token branches, estimated by fixing candidate openings and repeatedly sampling their continuations. This continuous risk-landscape perspective reveals two complementary properties: current decoder exposure and safe rerouting capacity, which have been obscured by binary-indicator-based ASR. The resulting representation also provides a common basis for comparing clean prompts, attacked variants, and attack strength against a known adaptive defense. Across multiple target models and jailbreak attacks, we uncover risky branches beneath apparently safe outputs, identify qualitatively different ways attacks reshape the model’s opening distribution, and show that attacks strongest before a defense do not necessarily remain the strongest after the adaptive defense reacts. These findings recast jailbreaks as deformations of the surrounding continuation-risk landscape, establishing a more informative foundation for understanding jailbreak behavior and developing safer language models.
[NLP-91] Structural Pattern Mining in Inka Khipus: A KFG Re-Analysis
【速读】： 该论文旨在解决传统基普（khipu）研究中因数据集不完整、结构混淆及方法学局限导致的分析偏差问题，尤其针对开放基普库（Open Khipu Repository, OKR）进展停滞所暴露出的数据质量与建模缺陷。其核心解决方案在于从原始数据库向更严谨的基普田野指南（Khipu Field Guide, KFG）数据库转型，并对包含650条基普的样本进行系统性重分析。关键突破在于识别并控制两大混杂因素：一是基普物理尺寸对结构计数的显著影响（解释42–70%方差），二是命名前缀（如AS、UR、HP等）反映的是原始编目者而非地理或年代属性，进而将文档完整性转化为需显式建模的第二重混杂因素。在此基础上，研究发现由胡戈·佩雷拉（HP）编目的基普在阿舍尔（Ascher）提出的“和关系”类型上显著少于由加里·乌顿（UR）编目的基普（p = 0.044），而阿舍尔编目基普因严重共线性（VIF = 5.0）无法有效评估。此外，基于通用结构聚类（UMAP+HDBSCAN）未能区分真实基普样本与两种独立零模型（置换与高斯），表明整体结构缺乏显著模式，仅在单一编目者（如UR）层面的聚类结果不可复现，构成方法学警示。最终，通过针对性复制KFG提出的S/Z结空间块启发式方法，成功验证了10个已知天文/历法基普中的8个，并提出34个新候选，凸显特定结构规则在识别功能性基普中的有效性。
链接: https://arxiv.org/abs/2610.38580
作者: Maria Contreras
机构: Independent Researcher(独立研究员)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 7 pages, 1 figure
Abstract:This paper reports a methodological pivot from the Open Khipu Repository to the Khipu Field Guide (KFG) database, prompted by stalled progress on the OKR, and a full re-analysis of the resulting 650-khipu sample. Three findings recur across every analysis attempted. First, khipu size confounds nearly every derived structural count, explaining 42-70% of its variance; all subsequent models control for it. Second, the two-letter name prefix (AS, UR, HP, QU, JC, MM, KH) identifies the original cataloger, not a site or period, which turns documentation completeness into a second confound that must be modeled explicitly. Controlling for both, khipus cataloged by Hugo Pereyra’s Proyecto Quipu (HP) show significantly fewer Ascher sum-relation types than Gary Urton’s (UR) khipus (p = 0.044), while the Ascher-cataloged (AS) khipus are too collinear with their own near-total lack of side/direction documentation to assess (VIF = 5.0). Third, generic structural clustering (UMAP+HDBSCAN) does not separate the pooled 650-khipu sample from two independent null models (permutation and Gaussian), corroborating an independent conclusion already reached by the KFG itself; the same test restricted to a single cataloger (UR) proved not reproducible across runtime environments and is reported here as a methodological caveat rather than evidence of structure. A targeted replication of the KFG’s own S/Z-knot spatial-block heuristic for calendar/astronomical khipus recovers 8 of 10 published candidates and proposes 34 additional ones under the same criterion.
[NLP-92] Do Audio LLM s HEAR Pain? Tracing Acoustic Evidence from Speech to Language Reasoning
【速读】： 该论文旨在解决大音频语言模型（Large Audio Language Models, LALMs）在处理语音时对非词汇性声学信息（non-lexical acoustic information）的利用程度不明确的问题，特别是其在疼痛感知任务中的表现。核心问题是：尽管LALMs能够直接处理语音信号，但它们是否真正有效提取并利用了反映疼痛状态的声学特征（如语调、语速、音高变化等），而非仅依赖语言内容或上下文线索。解决方案的关键在于提出HEAR（Hearing and Evaluating Representations），一个四阶段诊断框架，系统追踪从原始语音信号到模型最终预测之间疼痛相关声学信息的传递路径。通过该框架对七种闭源与开源的LALMs进行评估发现，在仅含中性语义的哈佛句子（需完全依赖声学线索推断疼痛）上，所有模型表现接近随机水平；而当使用包含自报告疼痛等级的口语化评分数据时，模型准确率可达87.5%–99.7%。层间探针分析表明，疼痛相关信息可在早期音频表示中被解码（二分类平衡准确率达74.1%，三分类严重度为45.0%），但在深层网络中逐渐衰减。此外，声学感知测试显示弱模型无法识别显著的声学变化，强模型虽可识别却仍无法正确推断疼痛。最后，研究通过恢复关键声学特征并以结构化文本形式注入提示（prompt），使成对疼痛判断的准确率提升最高达32个百分点，证明当前模型在生成输出前已丢失大量有价值的副语言信息。这揭示了现有LALMs虽在早期编码了疼痛相关的声学线索，但这些信息在推理过程中未能充分参与决策，成为性能瓶颈。
链接: https://arxiv.org/abs/2610.38570
作者: Wenqi Li,Xiangjue Dong,Sulaiman Vesal
机构: Microsoft(微软)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Sound (cs.SD)
备注: 26 pages, 9 figures
Abstract:Large audio language models (LALMs) can process speech directly, but it remains unclear how well they use the non-lexical acoustic information that speech carries. We study this question through pain, using TAME-Pain, which contains pain-neutral Harvard sentences and spoken pain ratings recorded at different self-reported pain levels. We introduce HEAR, a four-stage diagnostic framework that traces pain-related information from the speech signal to the model’s prediction, and apply it to seven closed-source and open-weight LALMs. On the Harvard sentences, where pain can only be inferred from the voice, all models perform at chance level, and same-speaker references or few-shot examples improve balanced accuracy by no more than six points. On the spoken ratings, the same models reach 87.5-99.7% balanced accuracy. Layer-wise probing of open-weight LALMs shows that pain can be decoded from early audio representations (74.1% balanced accuracy for binary prediction and 45.0% for three-class severity), but that this information, along with most pain-associated acoustic cues, decreases in deeper layers. In an acoustic perception test, weaker models cannot distinguish even large changes in these cues, whereas stronger models can but still fail to infer pain. Finally, we recover pain-related acoustic cues and directly reinject them as structured text into the prompt, which improves pairwise pain prediction by up to 32 points. These results indicate that current LALMs encode paralinguistic information in early layers that is largely unavailable when the model produces its answer.
[NLP-93] Persona Learning from Feedback for Personalized AI Assistants
【速读】： 该论文旨在解决个性化人工智能助手在适应用户偏好与任务上下文时面临的动态行为调整难题，即现有方法仅依赖角色描述（persona description）进行提示或微调，但无法明确指示在特定响应中应修改或保留哪些行为，也难以判断何种行为适配当前请求。其解决方案的关键在于提出一种基于反馈的角色学习框架（Persona Learning from Feedback, PLF），通过在共享冻结主干网络上学习可复用、以用户偏好为依据的角色模块（作为独立的低秩自适应，LoRA adapter），利用一个冻结的自教师（self-teacher）根据角色设定和响应级反馈，对每个适配器生成的前缀提供细粒度的令牌级监督信号；同时引入反馈门控机制，仅对被判定需改进的响应进行修正性更新。部署阶段则采用受监督的路由机制，根据请求或对话内容选择单一适配器或固定组合，无需针对用户进行微调。实验表明，在单轮个性化与多轮教学场景下，PLF均显著优于固定角色策略及强基线方法（如角色提示、角色蒸馏），实现了在不损害通用能力的前提下，通过反馈训练的行为实现高效、灵活的个性化适配。
链接: https://arxiv.org/abs/2610.38561
作者: Yiran Shen,Yu Xia,Liuyi Yao,Prithviraj Ammanabrolu
机构: UC San Diego(加州大学圣地亚哥分校); Alibaba Group(阿里巴巴集团)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Personalizing AI assistants requires adapting their behavior to users’ preferences and task contexts. Existing work elicits assistant personas through persona prompting or finetuning on persona-conditioned generations. However, persona descriptions alone do not identify what to change or preserve in a particular response, or which behaviors suit the current request. We introduce \textitPersona Learning from Feedback (PLF) , a framework that learns reusable, preference-grounded assistant personas as independent LoRA adapters on a shared frozen backbone. During training, a frozen self-teacher conditions on persona specifications and response-specific feedback to provide token-level supervision on each adapter’s own generated prefixes. A feedback gate focuses corrective updates on responses judged to need improvement. At deployment, a supervised router uses the request or dialogue to select a single adapter or a fixed composition, without user-specific finetuning. We evaluate PLF on single-turn personalization with model-generated critiques and multi-turn tutoring with simulated students’ reactions. Across both settings, PLF improves model-judged preference fit over fixed-persona policies and strong baselines including persona prompting and persona distillation, demonstrating how feedback-trained behaviors can be reused for adaptive assistant personalization without degrading general capabilities.
[NLP-94] Implicit Minor Detection from Multi-Turn Dialogue in High-Risk AI Interactions: A Constrained Self-Evolving LLM Harness
【速读】： 该论文旨在解决在高风险文本交互场景中，由于缺乏强身份标识（如账户信息、实名认证、面部数据或设备数据），系统难以准确判断当前发言者是否为未成年人的问题。针对这一挑战，研究提出了一种受限框架，其核心在于将弱身份信息下的隐式未成年人检测建模为高风险用户属性推断任务。该框架的关键创新包括：引入大型语言模型（Large Language Model, LLM）作为基础推理引擎，结合“Harness”模块控制推理触发时机，通过“Maker-Checker”协议确保对话链证据的结构化与可验证性，并对经验证的Harness修订实施版本级审查机制。实验结果表明，在包含8,328条多轮对话的受控中文基准上，该方法相较直接使用相同LLM家族进行推断，平均精度和少数类F1分别提升了7.60和6.37个百分点，显著增强了在弱身份条件下的未成年人识别能力。
链接: https://arxiv.org/abs/2610.38546
作者: Xiaohan Zhang,Ying Zhao,Ruilin Guo,Xuanyuan Yu,Kaibo Huang,Yukun Wei,Zhongliang Yang,Linna Zhou
机构: 未知
类目: Computation and Language (cs.CL)
备注: 13 pages, 3 figures, including a technical supplement
Abstract:In high-risk text interactions, a system may need to estimate whether the current speaker is likely to be a minor before applying age-conditioned safeguards. Many text-based dialogue systems lack strong identity signals such as account information, verified identity, facial data, or device data, and must rely on weak cues that emerge across multiple turns. We formulate implicit minor detection under weak identity information as a high-risk user-attribute inference task and propose a constrained framework comprising a large language model, a Harness, and a Maker-Checker protocol. The framework controls when inference is invoked, requires structured dialogue-linked evidence, and subjects validation-selected Harness revisions to version-level review. On a controlled Chinese benchmark comprising 8,328 multi-turn dialogues, our method improves average precision and minor-class F1 by 7.60 and 6.37 percentage points, respectively, over direct inference using the same LLM family.
[NLP-95] From Leniency to Suspicion: How an Open Post-Training Recipe Changes Whom LLM Fact-Checkers Believe
【速读】： 该论文旨在解决大语言模型（LLM）在事实核查过程中对发帖人社会地位（特别是职业身份）的敏感性问题，即后训练阶段是否以及如何导致模型更倾向于因发帖人职业地位较低而否定其言论的真实性。研究聚焦于“基于职业的怀疑”（job-based doubt）现象，即模型在面对低社会地位职业发帖者时更可能将真实陈述判定为虚假。其解决方案的关键在于：通过在基础模型与后训练版本中使用随机化发帖人身份的事实核查任务，系统性地对比六种模型家族在不同训练配方下的表现差异。研究发现，仅采用开放源代码的Tülu-3训练配方（OLMo-2 32B和Tülu-3 70B）的模型才会产生这种微弱但统计显著的偏见（偏差小于1个百分点），且该效应可重复、超越训练漂移；进一步分析表明，这一偏见并非源于模型开始更关注职业信息，而是对同一职业信息的语义解释发生根本性转变——在基础模型中，低地位职业被解释为“更可信”，而在后训练模型中则被解释为“更不可信”；这种认知反转由单个残差流方向（residual-stream direction）承载，并在监督微调阶段完成。该机制在400条新预设命题上得到验证。值得注意的是，七款部署的商用模型均未表现出此类偏见。本研究强调了必须根据具体训练配方对说话者偏见进行审计的重要性。
链接: https://arxiv.org/abs/2610.38532
作者: Nouar AlDahoul,Yasir Zaki
机构: New York University Abu Dhabi(纽约大学阿布扎比分校)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computers and Society (cs.CY)
备注: 10 pages, 4 figures, 6 tables. Preprint, under review
Abstract:LLMs increasingly fact-check online content with the poster’s identity in view. However, it remains unclear which post-training changes how much a model’s truth judgement depends on the poster’s social class, and how such a change happens inside the model. This paper investigates job-based doubt, i.e., judging a claim false more often when the poster has a lower-status job, using fact-checked claims with randomised poster profiles in base and post-trained checkpoints of six model families. Our results reveal that: 1) post-training creates such doubt only with the open Tülu-3 recipe (OLMo-2 32B, Tülu-3 70B), not with Meta’s, Alibaba’s or Mistral’s; the effect is under one percentage point but statistically reliable, exceeding training drift and replicating on independent claim sets; 2) in OLMo-2 32B, the model does not start noticing the poster’s job more, but starts interpreting it differently: a lower-status job pushes the base model toward “true” and the post-trained model toward “false”; and 3) this flip is carried by a single residual-stream direction and happens at supervised fine-tuning. The mechanism was confirmed on 400 new claims with pre-specified predictions. None of seven deployed commercial models doubts lower-status posters more. Our work underscores the need to audit speaker bias per training recipe.
[NLP-96] What Hallucinations Have We Measured? Semantic Illusion Confabulation and Expressed Uncertainty
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）中“幻觉”（hallucination）概念定义模糊、分类混乱的问题，从而影响对幻觉现象的准确识别与有效应对。其核心挑战在于当前研究中“幻觉”一词被泛化使用，未严格区分错误类型，尤其混淆了“自信性错误”（即幻觉）与模型表达的不确定性（UNK），而后者具有自我警示能力，前者则隐匿错误，危害更大。为此，论文提出一个关键解决方案：基于简单的语义等价重提示测试（re-elicitation test），将幻觉细分为两类——语义幻觉（Semantic Illusion, SI）和虚构生成（Confabulation, CF）。若语义等价提示可恢复正确回答，则表明知识存在于模型中但应用不一致，属可缓解的SI；反之则说明知识缺失或不可访问，属需外部干预或重训练的CF。该分类框架揭示了不同错误类型的本质差异，进而发现现有检测方法更易识别UNK而非真实幻觉，且缓解策略对SI效果显著优于CF，同时指出潜在表征信号比输出层信号对CF更具诊断价值。因此，论文强调应独立报告CF、SI与UNK，并提出一种仅需少量语义等价提示的实用标注协议，以提升幻觉研究的严谨性与可复现性。
链接: https://arxiv.org/abs/2610.38520
作者: Hazel H. Kim,Jakub Binkowski,Ekaterina Fadeeva,Juan David Guerra,Yijie Hao,Zheng Yi Ho,Baixiang Huang,Buyun Liang,Siyuan Liang,Anh Tuan Luu,Potsawee Manakul,Niels Mündler-Sasahara,Aidar Myrzakhan,Shahrad Mohammadzadeh,Artem Shelmanov,Kai Shu,Ponhvoan Srey,Dacheng Tao,René Vidal,Haofei Yu,Lei Yu,Hongxiang Zhang,Yarin Gal,Hinrich Schütze,Philip H.S. Torr
机构: University of Oxford (牛津大学); Wroclaw University of Science and Technology (弗罗茨瓦夫科学与技术大学); ETH Zurich (苏黎世联邦理工学院); Mila – Quebec AI Institute (蒙特利尔-魁北克人工智能研究所); Polytechnique Montréal (蒙特利尔工业大学); Emory University (埃默里大学); Nanyang Technological University (南洋理工大学); University of Pennsylvania (宾夕法尼亚大学); VinUniversity (越南大学); Independent Researcher (独立研究员); MBZUAI (穆罕默德·本·扎耶德人工智能大学); McGill University (麦吉尔大学); UIUC (伊利诺伊大学厄本那-香槟分校); University of Toronto (多伦多大学); Meta (FAIR) (Meta (FAIR))
类目: Computation and Language (cs.CL)
备注:
Abstract:Hallucination remains one of the most critical challenges in deploying large language models (LLMs) reliably and safely. Yet the term is used inconsistently across the research community. We argue for a strict definition of hallucinations in LLMs. First, hallucinations should be confident incorrectness, kept separate from expressed uncertainty (UNK)1. UNK reveals that the model may be wrong, whereas a hallucination does not reveal it and is therefore more dangerous in practice. Conflating the two obscures their distinct behaviors and makes them hard to study. Second, hallucinations should be split by a simple re-elicitation test: does any semantically equivalent prompt recover the correct response? This test probes whether the relevant knowledge2 is accessible. If none does, the knowledge is likely absent or inaccessible, and the error type is a confabulation (CF) that likely requires external knowledge or retraining; if some does, the knowledge is accessible from the model but applied inconsistently, and the error type is a semantic illusion (SI) that simple prompt or decoding changes can mitigate3. Re-evaluating 14 recent hallucination methods under this taxonomy yields three findings. (1) Detection methods flag UNK more often than hallucinated responses (i.e., SI and CF). (2) Mitigation methods are far more effective for SI than for CF. (3) Latent signals are more informative than output-level signals for CF. An audit of recent hallucination papers shows that most conflate these categories. We recommend reporting CF, SI, and UNK separately, and introduce a practical labeling protocol that needs only a modest number of semantically equivalent prompts.
[NLP-97] Similar Scores Opposite Failures: Human Correction as an Instrument for Video Caption Evaluation
【速读】： 该论文旨在解决当前视频字幕生成模型评估中普遍存在的核心问题：现有基于单句图像字幕设计的参考重叠度量（reference-overlap metrics）在评估密集、多时窗视频描述任务时，无法有效反映人类评判标准。其关键发现在于，尽管主流指标如BLEU-4和CIDEr-style tf-idf余弦相似度在区分评审人接受与拒绝的字幕方面表现接近随机水平（AUC分别为0.510和0.528），且即使仅使用评审员修正后的参考文本，BLEU-4仍仅为0.479（95%置信区间0.431–0.529），表明这些指标对真实错误类型缺乏敏感性。研究进一步揭示，评估粒度显著影响错误检测能力——以整体片段为单位评估时仅识别出2.7%的镜头与剪辑错误，而逐窗口评估则提升至12.8%，差距达4.7倍，且该现象在各类稳健性检验中均成立。此外，不同模型在准确率上统计上无差异，但其失败模式却截然不同，单一标量指标无法捕捉这种复杂性。因此，论文提出评估体系应从“衡量错误频率”转向“记录错误类型”，并系统性地报告失败分类体系、各模型的失败特征分布、预测任务难度的语料库属性以及可复现所有结果的代码，从而推动更精细、可解释的视频字幕评估范式。
链接: https://arxiv.org/abs/2610.38511
作者: Ankit Khedia,Astitva Jaiswal,Chaitanya Prasad,Anisha Raju,Aangeeras
机构: Deccan AI
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL)
备注: 29 pages, 10 figures, 17 tables. Analysis code included as ancillary files. Data available on request at this https URL
Abstract:Video captioning models are increasingly evaluated with metrics designed for single-sentence image captions. We test whether those metrics recover human judgement on dense, multi-window video description, and find that they do not. Six frontier models captioned 196 short clips. Trained annotators post-edited every caption against a failure taxonomy fixed before annotation, once as a whole-clip overview and once window by window, and a quality-control reviewer checked each annotation, yielding 1,144 overview verdicts and 4,672 window verdicts. Reference-overlap metrics do not separate the captions reviewers accepted from those they rejected: BLEU-4 reaches an AUC of 0.510 and a CIDEr-style tf-idf cosine 0.528, and restricting references to reviewer corrections leaves BLEU-4 at 0.479 (95% CI 0.431-0.529), indistinguishable from chance. Review granularity changes what is detected. The same annotators record camera and shot errors on 2.7% of captions when reading the overview and on 12.8% when reading window by window, a 4.7x gap that holds across every robustness check we ran, on the mode most likely to need a near-total rewrite. Correct-overview rates span 54.1-72.7%, but only 4 of 15 pairwise gaps survive Bonferroni correction (5 after adjusting for reviewer and clip effects), and accuracy and derived severity order the models differently. Models with statistically indistinguishable accuracy have sharply different failure profiles, which no single scalar shows. We argue that caption evaluation should record what was wrong rather than only how often, and we report the taxonomy, the per-model profiles, the corpus properties that predict difficulty, and the code that regenerates every number in the paper.
[NLP-98] Leaving the Cave by Breaking the Symmetry: Identifiable Structure Can Enable Cross-Lingual Generalization
【速读】： 该论文试图解决的问题是：多语言语言模型为何在缺乏共享词汇的情况下仍能实现跨语言泛化。其核心解决方案的关键在于提出“关系可识别性假设”（Relational Identifiability Hypothesis），即共享世界中的语义关系能够通过排除非对应性的跨语言映射，促进不同语言中对应实体的对齐，从而使对应实体在表示空间中直接可读，无需额外的变换。研究通过在双语Transformer模型中构建两个无词元交集的语言描述同一模运算世界的情境进行验证，发现仅当关系结构（如减法）使零点在任务层面具有唯一可识别性时，才会产生选择性对齐；而依赖顺序敏感操作则导致全量数字的广泛对齐。这些结果表明，共享世界的内在关系结构可驱动实体对齐并支持零样本跨语言迁移，揭示了跨语言泛化的潜在机制。
链接: https://arxiv.org/abs/2610.38509
作者: Frederick Riemenschneider,Anette Frank
机构: Heidelberg University(海德堡大学), Germany
类目: Computation and Language (cs.CL)
备注: 12 pages, 7 appendix pages, 1 figure, 10 tables
Abstract:Why do multilingual language models generalize across languages even without shared vocabulary? The Platonic Representation Hypothesis posits that models trained on different data, objectives, and modalities converge toward a shared statistical model of reality. Extending this intuition to representations inside a single multilingual model, we propose the Relational Identifiability Hypothesis: relations in a shared world can promote entity alignment by ruling out alternative cross-lingual correspondences, making corresponding entities more likely to become directly readable without an additional transformation. We test this hypothesis in a controlled bilingual Transformer whose two token-disjoint languages describe the same modular-arithmetic world. We find that the languages can develop a shared answer subspace without aligning corresponding numbers. Cross-lingual alignment emerges selectively when the available relations constrain number correspondences: subtraction makes zero task-level identifiable, inducing selective alignment, while an order-sensitive operation produces broad alignment across numbers. These patterns predict success in zero-shot cross-lingual transfer. Together, our results show that relational structure in a shared world can promote entity alignment and enable transfer.
[NLP-99] When Support Crosses the Line: Evaluating Ethical Boundaries in LLM -Based Mental Health Support in the Chinese Context
【速读】： 该论文旨在解决当前大型语言模型（LLM）在心理健康支持应用中存在的一大关键问题：现有评估基准主要关注模型的共情与支持能力，而忽视了其在多轮交互过程中是否能够持续遵守专业伦理边界。这一问题在长时间、动态变化的对话场景中尤为突出，用户可能随时间引入新信息或施加情感压力，从而对模型的伦理一致性构成挑战。为此，论文提出MindEth-Bench——一个基于中国心理健康服务专业伦理标准构建的多轮心理支持交互伦理边界评估基准。其核心创新在于采用蒙特卡洛树搜索（Monte Carlo Tree Search, MCTS）技术系统性地探索高压力交互路径，以精准识别模型在复杂对话中的伦理脆弱点。实验结果表明，即使初始响应符合伦理规范的模型，在长期互动中也可能逐渐偏离伦理准则，且不同模型及不同伦理维度间表现差异显著。这揭示了单轮评估可能过度高估模型的伦理可靠性，强调必须在多轮交互全过程中持续评估模型的伦理行为，为构建安全可信的心理健康生成式AI系统提供了关键方法论支撑。
链接: https://arxiv.org/abs/2610.38452
作者: Bichen Wang,Yufeng Jin,Yixin Sun,Yanyan Zhao,Bing Qin
机构: Research Center for Social Computing and Information Retrieval; Harbin Institute of Technology (哈尔滨工业大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Large language models (LLMs) are increasingly explored for mental health support, yet existing benchmarks primarily focus on supportive capabilities rather than whether models consistently adhere to professional ethical boundaries. This issue is particularly important in multi-turn interactions, where users may introduce new information or exert emotional pressure over time. We introduce MindEth-Bench, a benchmark for evaluating ethical boundary adherence in multi-turn mental health interactions. Grounded in professional ethical standards for mental health services in China, MindEth-Bench covers multiple dimensions of ethical conduct and uses Monte Carlo Tree Search (MCTS) to systematically explore interaction trajectories that stress-test model boundaries. Experiments across nine LLMs show that models that initially respond appropriately can shift toward ethical violations over extended interactions, with substantial variation across models and ethical dimensions. These results show that single-turn evaluations may overestimate the ethical reliability of LLM-based mental health support systems and highlight the need to evaluate ethical behavior throughout multi-turn interaction.
[NLP-100] CoSTA: Co-Evolving Self-Play Timeline Agents for Complex Temporal Knowledge Graph Question Answering
【速读】： 该论文旨在解决复杂时间知识图谱问答（Complex Temporal Knowledge Graph Question Answering, TKGQA）中因事实间时间约束导致的推理难题，尤其针对多事件历史交织、时间区间协同影响答案判断等挑战。现有单智能体方法在单一推理上下文中处理事实时，易出现跨事件证据混杂与干扰项误导等问题。为此，本文提出共进化自对弈时间线代理（Co-Evolving Self-Play Timeline Agents, CoSTA），其核心解决方案在于：将知识事实组织为可导航的时间线，由多个工作者智能体分别在独立事件上下文中进行调查，并通过协调者智能体融合各工作者的原始带时间戳观测结果。该方法通过角色条件化蒸馏初始化共享策略，并利用最终答案奖励驱动的共进化自对弈机制，联合优化工作者与协调者角色。实验表明，CoSTA在三个基准测试上均达到领先性能，尤其在复杂多事实问题上表现突出，且其紧凑版本超越了所评估的闭源大语言模型（Large Language Model, LLM）代理。
链接: https://arxiv.org/abs/2610.38450
作者: Haikun Liu,Zhichao Yan,Shaoru Guo,Jiawei Gu,Victor Gutierrez Basulto,Jiapu Wang,Ru Li
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:Complex Temporal Knowledge Graph Question Answering (TKGQA) involves more than retrieving facts: the timestamp or validity interval of one fact can constrain the search for another, while multiple intervals may jointly determine the answer. Existing single-agent TKGQA methods typically investigate these facts within a single reasoning context, where evidence from different event histories can become interleaved and distractors can misdirect subsequent retrieval. Distributing investigations across agents can separate event-specific evidence into distinct contexts, but effective collaboration must still be learned because question decomposition, subtask investigation, and evidence synthesis are interdependent. To address these limitations, we propose \textbfCo-Evolving \textbfSelf-Play \textbfTimeline \textbfAgents (CoSTA), which organizes facts into navigable timelines, delegates investigations to workers, and uses a coordinator to synthesize their original timestamped observations. CoSTA initializes a shared policy through role-conditioned distillation and jointly optimizes both roles through co-evolving self-play with final-answer rewards. Across three benchmarks, CoSTA achieves state-of-the-art overall performance, with particularly strong gains on complex multi-fact questions, while its compact variants outperform the evaluated closed-source LLM agents.
[NLP-101] Benchmarking decision models against generative large language models on medical multiple-choice questions
【速读】： 该论文旨在解决医疗工作流中在预定义选项中进行决策时的准确性与效率平衡问题，尤其关注不同决策模型与生成式大语言模型（Generative Large Language Model, GLLM）配置在文本仅输入条件下的表现差异。其解决方案的关键在于系统性比较13种决策模型与15种生成式大语言模型配置，综合评估其准确率、推理成本、请求延迟、答案一致性及概率可重复性等多维度指标。研究发现，尽管生成式配置在准确率上显著优于传统决策模型（最高达96.86%），但即使在高置信度概率（≥0.90）下仍存在1.71%至12.77%的错误率，表明单纯依赖高概率输出不可靠；同时，部分模型如Mercury Decide在单位成本上表现最优，而Clef Omni则具备最短的中位请求时延。因此，该研究强调，在医疗场景中选择适用模型需综合考量准确性、概率质量、结果可重复性以及资源消耗，而非单一依赖性能指标。
链接: https://arxiv.org/abs/2610.38448
作者: Oscar Freyer,Daniel Glauert,Moritz Hofmann,Max Ostermann,Stephen Gilbert,Jakob Nikolas Kather
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Medical workflows often require selecting among predefined options. We compared 13 decision models with 15 generative large language model configurations, including different reasoning settings, using 1,273 four-option MedQA questions under text-only conditions. Accuracy ranged from 67.71% to 89.08% for decision models and from 78.32% to 96.86% for generative configurations. Among the tested configurations, Mercury Decide had the lowest mean final-attempt cost (USD 0.0069 per 1,000 questions), while Clef Omni had the shortest median request duration (0.291 seconds). Decision-model probabilities distinguished correct from incorrect answers, but error rates among answers assigned probabilities of at least 0.90 ranged from 1.71% to 12.77%. Answer agreement across three runs ranged from 92.30% to 100%, while probability repeatability varied. Decision models offered resource advantages alongside lower peak accuracy; selecting models for medical workflows requires considering probability quality and repeatability together with accuracy and the resources needed to detect and correct errors.
[NLP-102] Predict the Reward Not the World: World Reward Models for ML Agents
【速读】： 该论文旨在解决搜索型机器学习（ML）代理在优化过程中因执行开销大、大量候选程序被丢弃而导致效率低下的问题。传统方法依赖大型语言模型（LLM）作为世界模型，预测程序执行结果以指导搜索，但错误的预测会引入误导性信息，影响搜索质量。本文提出世界奖励模型（World Reward Models, WRMs），其核心思想是：代理无需预测程序执行的中间过程或最终结果，只需直接判断哪个候选程序更可能获得成功执行反馈。关键解决方案在于构建一种基于真实执行结果训练的直接选择器——ForeCritic，它通过学习已执行程序对的相对表现，直接预测哪个程序更优，避免了中间结果预测带来的误差累积。理论证明表明，任何基于预测执行结果的模型无法超越最优的直接选择器；且使用交叉熵损失对实际胜出程序进行训练可使WRM收敛至最优。实验显示，在26个机器学习任务的12,438组未见程序对上，ForeCritic的准确率达79.4%，显著优于通用LLM评判者的53%–62%；同时其置信度概率校准良好。将其集成于AIDE、ML-Master和ForeAgent中，使平均任务时间缩短4.4至7.5倍，同时保持或小幅提升性能表现，验证了该方法在高效与高精度之间的平衡能力。
链接: https://arxiv.org/abs/2610.38439
作者: Yuxiao Lee,Zherui Yang,Fan Liu,Hao Liu
机构: The Hong Kong University of Science and Technology (Guangzhou)(香港科技大学(广州)); The Hong Kong University of Science and Technology (Guangzhou)(香港科技大学(广州)); The Hong Kong University of Science and Technology (Guangzhou)(香港科技大学(广州)); The Hong Kong University of Science and Technology (Guangzhou)(香港科技大学(广州))
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Search-based machine-learning (ML) agents improve a solution by writing candidate programs, executing them, and keeping the best. Execution gives the search reliable feedback, but a single run can take hours, and most runs end in programs the agent discards. Recent agents try to save these runs by letting a large language model act as a world model that predicts how each program would perform. However, a wrong prediction feeds the search false information. We argue that an agent does not need to predict the outcome of execution, only the decision that execution would support. We propose World Reward Models (WRMs), which learn from real execution results to choose directly which candidate will perform better, without predicting any intermediate outcome. We prove that a selector which predicts execution from the same inputs cannot outperform the best direct selector, that predicted outcomes add an error which can make a likelihood-trained world model prefer the worse program on every instance, and that cross-entropy on executed winners drives a WRM toward the optimum. Our WRM, ForeCritic, adapts a 9B typed decision model to 5k executed program pairs. On 12,438 held-out pairs from 26 ML tasks, it picks the better program 79.4% of the time, while general-purpose LLM judges reach 53–62%, and its probabilities are calibrated. Inside AIDE, ML-Master, and ForeAgent, it cuts mean task time by 4.4–7.5 times while keeping their mean scores within 0.01 or raising them.
[NLP-103] RoleFUZZER: User-Behaviour Fuzzing for Role-Specification Adherence in Large Language Models
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在多轮交互中难以有效遵循预设角色规范（role specification）的问题，尤其在面对现实且可能具有挑战性的用户行为时，其角色一致性与合规性缺乏系统性评估。现有研究对模型在复杂交互场景下偏离角色设定的脆弱性关注不足。为此，作者提出“用户行为模糊测试”（user behaviour fuzzing）方法，其核心是通过受控的、基于覆盖度引导的搜索策略，识别出可能导致模型违反角色规范的潜在用户行为路径。解决方案的关键在于构建一个名为RoleFUZZER的框架，该框架将软件测试中的覆盖引导模糊测试（coverage-guided fuzzing）思想迁移至语言模型评估中，利用一个基于实际交互类型分类体系与角色规范交叉构成的“用户行为覆盖网格”（user-behaviour coverage grid），系统性地生成高覆盖率的对抗性交互场景。实验表明，该框架在五类角色（如导师、财务顾问、AI面试官、对抗性角色扮演者及组织级网络代理）上均实现了对64个单元格行为网格高达89%的覆盖率，显著优于自由生成攻击样本的方法（覆盖率仅39%-69%），并揭示了同一角色规范可能因模型差异而出现相反方向的失效模式（如某些模型忽略安全信号，而另一些则脱离角色）。因此，RoleFUZZER提供了一种可扩展、可重复的机制，用于系统发现和分析大语言模型在多样化用户行为压力下的角色合规性漏洞。
链接: https://arxiv.org/abs/2610.38438
作者: Kshitish Ghate,Michael Saxon,Tadayoshi Kohno,Aylin Caliskan
机构: University of Washington, Seattle, WA(华盛顿大学，西雅图); Georgetown University, Washington, DC(乔治城大学，华盛顿特区)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Software Engineering (cs.SE)
备注: Preprint under review
Abstract:Large language models (LLMs) are often conditioned into specific roles such as tutors, interviewers, or agents acting for an organisation, through system prompts that specify how they should behave. The extent to which they can actually adhere to these role specifications under multi-turn interactions with realistic and potentially problematic users is understudied. To address this, we propose user behaviour fuzzing, a guided search over plausible user behaviours that can lead a model to violate its role specification, and introduce RoleFUZZER, a framework that adapts coverage-guided fuzzing from software testing to operationalise the search. RoleFUZZER constructs scenarios guided by a user-behaviour coverage grid that crosses a grounded taxonomy of user interaction types and the role specification of the model under test i.e. what is under pressure. With RoleFUZZER, we evaluate LLMs across five role specifications, including tutor, financial advisor, AI interviewer, antagonistic roleplay character, and web agent under organisational policies. Across simulated and validated conversations in 15 campaigns, the framework reaches = 89% coverage of the 64-cell behaviour grid for every model and role tested, and surfaces specification-grounded failure patterns. At an equal budget of conversations, removing the RoleFUZZER grid and asking an LLM to generate specification targeting users free-form lowers coverage to 39-69% with fewer kinds of violations. The same specification can also fail in opposite directions. For e.g., on the antagonist, open-weight target models override the player’s safety signals, while Claude Haiku 4.5 breaks character instead. RoleFUZZER offers a tractable way to uncover vulnerabilities in LLMs across role specifications by testing diverse user behaviours that are likely to violate them.
[NLP-104] Does Chain-of-Thought Reliance Follow a Scaling Law?Larger Models Follow Their Steps More Independent of Post-Training
【速读】： 该论文旨在解决生成式 AI（Generative AI）在推理过程中对思维链（Chain of Thought, CoT）依赖程度的评估问题，即验证模型的最终答案是否真正依赖于其显式写出的中间推理步骤，这对于安全监控具有重要意义。现有研究多基于自然任务，因真实推理过程未知而难以准确衡量。本文提出一种合成的“盒子交换任务”（synthetic box-swap task），其中每个中间状态均具备精确的已知真值，从而实现对模型推理路径的精准测量。关键解决方案在于：在模型自动生成的思维链中替换某一步为错误但合法的状态，观察最终答案是否随之改变，以此量化模型对写入步骤的依赖性。实验覆盖25个开源大模型（包括Qwen3、Gemma 3、Llama 3及OLMo 2系列），共测试7,200个样本。核心发现表明：（1）模型对思维链的依赖度随参数规模增长并趋于饱和，且符合对数规模下的逻辑回归模型（R² = 0.89）；（2）模型能力是决定依赖度的主导因素，不同架构下能力相当的模型表现出相似的依赖水平（相关系数r = 0.99）；（3）在正确率匹配的前提下，规模仍影响依赖度，说明规模本身具有独立作用；（4）后训练（post-training）对依赖度无显著影响；（5）所有测试模型（最大至14B）均无法在无显式步骤的情况下完成复杂任务，一旦涉及四次以上交换，其表现退化至随机水平。因此，该研究证实，在需要逐步计算的任务中，规模提升使思维链成为更关键的承载机制，支持对生成式 AI 的思维链进行有效监控。
链接: https://arxiv.org/abs/2610.38433
作者: Esther Xin
机构: Independent Researcher(独立研究员); Google(谷歌)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 15 pages,5 figures, code and data in this https URL
Abstract:Monitoring a model’s chain of thought (CoT) is useful for safety only if the final answer depends on the written steps. Prior measurements use natural tasks where true reasoning is unknown. We measure it against an exact ground truth on a synthetic box-swap task where every intermediate state is known. We replace one step inside a model’s own chain with a wrong but valid state and check whether the answer follows the edit. We evaluate 25 open-weight models–Qwen3 (0.6B-14B, base and post-trained), Gemma 3 (1B-12B), Llama 3 (1B-8B), and 8 OLMo 2 checkpoints–on 7,200 items each. We find five results: (1) Reliance rises with size and saturates. In Qwen3, answer reliance rises from 16-18% at 0.6B to 97-100% at 14B; a logistic in log N fits well (R^2 = 0.89), and Gemma 3 and Llama 3 follow this trend. (2) Capability is the main driver. Across all models, reliance and step accuracy correlate at r = 0.99; equally capable models are equally reliant, regardless of family. (3) Size still matters at matched correctness. On problems solved correctly before the edit, reliance increases with size among capable models (rho = 0.80, p = 0.002). (4) Post-training does not change reliance. Base and post-trained Qwen3 differ by at most 0.03 from 4B upward; a large gap at 1.7B matches a large capability gap. (5) No model up to 14B solves the task internally. Without written steps, every capable model is at chance once a problem has four or more swaps. OLMo 2 checkpoints never become capable, making the training-token axis a limitation. On tasks requiring step-by-step computation, scale makes the chain of thought more load-bearing, supporting CoT monitoring in this regime.
[NLP-105] Deeper Backpropagation or More Updates? Training Looped Language Models Across Budgets
【速读】： 该论文旨在解决深度重复变换器层（repeating transformer layers）在训练过程中因计算深度增加而导致的前向与反向传播开销上升问题，核心挑战在于如何选择最优的反向传播窗口（backward window）以平衡每标记学习效率（per-token learning efficiency）与每次更新开销（per-update cost）。其解决方案的关键在于通过控制实验设计，系统评估不同反向窗口在不同训练预算下的相对性能变化。研究通过在相同数据集上训练并比较两个窗口，在各预算下保持近似相等的训练时间，从而隔离出每标记效率的影响因素。结果表明，随着训练预算增大，较短窗口所具有的优势显著减弱，且其带来的双重收益（即较低更新开销与较高学习效率）均随预算扩大而下降，其中扩展训练带来的收益衰减更为明显。因此，反向窗口的最优选择并非固定，而是高度依赖于可用训练预算，且在预算增长时，最优窗口可能发生变化。
链接: https://arxiv.org/abs/2610.38422
作者: Jiarun Shen,Xiao Li
机构: The Chinese University of Hong Kong, Shenzhen(香港中文大学（深圳)）
类目: Computation and Language (cs.CL)
备注:
Abstract:Repeating transformer layers can improve language model performance without adding parameters. However, the added computational depth increases both forward and backward computation costs during training. Training such models requires choosing a backward window that controls how many final layer repetitions retain gradients during backpropagation. Longer windows improve per-token learning efficiency but increase per-update cost, while shorter windows reduce per-update cost but sacrifice per-token efficiency. Understanding which window performs better requires examining different training budgets. Prior work compares backward windows across computational budgets. When training budgets increase, it remains unclear whether the same window continues to perform best. We measure how the relative performance of two windows shifts as training budgets grow. Our method isolates per-token efficiency by training both windows on identical data, and compares complete training recipes at approximately equal training time at each budget. Results show that shorter windows lose much of their advantage as budgets grow. Controlled experiments show that both benefits decrease at larger budgets, with a greater decline in the benefit of extended training. Backward window comparisons depend on the available budget, and the preferred window among the tested candidates can change as the budget grows.
[NLP-106] Holistic Diversity Management in LLM -Guided Evolutionary Search
【速读】： 该论文旨在解决生成式程序演化搜索中因岛屿模型（island model）固有局限性导致的多样性不足问题，即各岛屿可能从相似初始程序出发或在演化过程中趋同于相似解，从而引发早熟收敛（premature convergence）。其解决方案的关键在于提出一种全周期、系统性的多样性保障机制——HoliDivEvolve。该方法通过三个核心设计实现：（1）多样化初始化（diverse initialization），确保各岛屿起始程序具有显著差异；（2）轨迹自适应岛屿选择（trajectory adaptive island selection），依据当前得分、近期增长趋势及剩余计算预算动态分配计算资源，引导不同演化路径；（3）生长自适应迁移（growth adaptive migration），根据各岛屿的演化进展智能地共享替代性创新思想。实验证明，该方法在多个基准任务上实现了竞争力性能，并通过受控分析验证了各设计组件对搜索行为与最终结果的影响。
链接: https://arxiv.org/abs/2610.38407
作者: Seockbean Song,Youngsik Yoon,Sungjae Lee,Siwei Wang,Wei Chen,Jungseul Ok
机构: Graduate School of AI, POSTECH(浦项工科大学人工智能研究生院); Department of CSE, POSTECH(浦项工科大学计算机科学与工程系); Microsoft Research Asia(微软亚洲研究院)
类目: Computation and Language (cs.CL)
备注:
Abstract:LLM-guided evolutionary search iteratively selects existing programs, uses LLMs as mutation operators to generate a new program, and evaluates it to guide subsequent search. To preserve diversity and reduce the risk of premature convergence, it often adopts the island model, in which groups of programs, called islands, evolve separately. However, this structure alone does not guarantee distinct search directions, as islands may start from similar programs or converge toward similar solutions during evolution. This suggests that diversity must be considered throughout the search process, from initializing islands with distinct programs to developing island trajectories and sharing discoveries across islands. To this end, we propose HoliDivEvolve , a holistic approach to establishing and effectively utilizing island diversity throughout evolutionary search. Specifically, we introduce diverse initialization to establish different initial programs, trajectory adaptive island selection to allocate computation based on current score, recent growth, and the remaining search budget, and growth adaptive migration to share alternative ideas across islands. Experiments across diverse benchmarks show that HoliDivEvolve achieves competitive performance on most evaluated tasks. We further conduct controlled analyses to examine how our design choices affect search behavior and final performance.
[NLP-107] LLM s as Judges and Examinees for Medical Item Difficulty Prediction
【速读】： 该论文旨在解决仅基于题目文本预测医学考试题项难度的挑战，现有方法多依赖语言特征或通用/领域语言模型生成的文本表示，但预测性能有限。其关键解决方案是利用开放权重大语言模型（LLM）在与测试题项交互过程中自然产生的难度相关信号：一方面将LLM作为“评分者”提供绝对难度评分和成对难度比较；另一方面将其模拟为“考生”，生成基于回答的信号，包括正确性、关键词选项概率、响应熵以及前两名选项的概率差值。研究发现，成对难度判断在两个数据集上表现最稳定，而模拟考生信号与绝对评分具有互补性。通过融合最优的判断与响应信号及基础题项特征，模型在USMLE和PANCE考试数据集上的斯皮尔曼等级相关系数分别达到ρ = .447和ρ = .494，优于2024年BEA共享任务最佳系统（ρ = .25），且接近专有前沿模型Claude Opus 5的表现。该方法表明，开放权重LLM通过交互式行为可提取超越纯文本表示的难度信息，且因支持本地部署，为未公开或高安全性的题项难度预测提供了实用可行的技术路径。
链接: https://arxiv.org/abs/2610.38402
作者: Xiaomeng Xiong,Marcus Walker,Yanlin Jiang
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Predicting medical item difficulty from item text alone remains challenging. Prior work has largely relied on linguistic features or text representations derived from general-purpose and domain-specific language models, but predictive performance has remained modest. This study takes a different approach by using difficulty-related signals that emerge when open-weight large language models (LLMs) interact with test items. Using USMLE and PANCE medical licensure examination items, we examined LLMs in two roles. As judges, the models provided absolute difficulty ratings and pairwise difficulty comparisons. As simulated examinees, they generated response-based signals including correctness, keyed-option probability, response entropy, and the probability margin between the top two options. We then combined the strongest judgment and response signals with baseline item features in downstream prediction models. Pairwise difficulty judgment was the most consistent individual LLM signal across both datasets, while simulated-examinee signals and absolute judgments provided complementary information in some feature combinations. The best configurations reached a Spearman correlation of \rho = .447 on USMLE, exceeding the best system in the 2024 BEA shared task ( \rho = .25), and \rho = .494 on PANCE. These results were comparable to direct absolute judgments from the proprietary frontier model Claude Opus 5 ( \rho = .443 and .488, respectively). Overall, judgment and response signals from open-weight LLMs captured difficulty-related information beyond text representations. Because these models can be deployed locally, the approach offers a practical option for predicting the difficulty of unreleased or high-security test items.
[NLP-108] Pronunciation-Supervised Text-to-Speech for Irish (Gaeilge)
【速读】： 该论文旨在解决爱尔兰语（Irish）等低资源语言在语音合成（Text-to-Speech, TTS）领域因缺乏大量标注语音数据而难以构建高质量语音系统的问题。其核心挑战在于如何在仅有12.8小时众包语音数据的情况下，有效利用已有的、详尽的发音词典与方言描述文献来提升模型性能。解决方案的关键在于将书面的发音记录（如维基词典中的音标标注）转化为可执行的声学验证器（phoneme critic），并将其嵌入到单轮拒绝式微调（rejection fine-tuning）流程中，以监督模型生成符合标准发音的语音。这一方法不仅显著降低了跨语料库字符错误率（CER）至0.159，且在域内数据上达到0.067，同时保持了说话人匿名性；更重要的是，该验证器揭示了当前训练数据对少数方言（如阿尔斯特方言）覆盖不足的问题，从而为评估数据局限性和指导未来数据采集提供了量化依据。因此，发音文档经结构化后作为验证器，是实现低资源语音合成的有效杠杆。
链接: https://arxiv.org/abs/2610.38384
作者: William Lane
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:Speech technology for Irish must be built from very little transcribed speech, yet the language’s pronunciation is thoroughly documented in dictionaries and dialect descriptions. We ask whether that documentation can supervise the adaptation of a large pretrained text-to-speech model to a language it was never trained on. We adapt Qwen3-TTS to Irish using 12.8 hours of crowdsourced Common Voice speech, correcting two defects in the vendor fine-tuning script that training loss does not reveal (Appendix A), and reach character error rate (CER) 0.225 on held-out cross-corpus text, against 0.397 for an exhaustively tuned from-scratch system and 0.437 for zero-shot cloning. We then compile the Wiktionary pronunciation record into a lexicon and a phoneme critic, and use that critic as the verifier in one round of rejection fine-tuning: CER 0.159 cross-corpus and 0.067 in-domain, with the voice still anonymous across all 202 training speakers. The same critic explains why the voice cannot yet be steered to a minority dialect: 229 of 12,311 training clips lean Ulster against 8,451 Munster. Pronunciation documentation, converted into a verifier, is a practical lever for low-resource speech synthesis, and a way to measure what the available data cannot support.
[NLP-109] How Much Do Teacher Bridges Matter in Interleaved On-Policy Distillation?
【速读】： 该论文旨在解决在策略蒸馏（On-policy Distillation, OPD）中，如何有效利用教师模型生成的“桥接”（bridge）文本作为指导信号以提升学生模型性能这一关键问题。现有方法虽引入教师撰写的片段以引导学生生成，但在如何训练学生对这些教师生成令牌（tokens）进行响应方面存在分歧，缺乏统一有效的评估标准。本文的关键解决方案在于：通过固定学生生成的轨迹（identical trajectories），在相同实验设置下系统比较无损失（no loss）、前向KL（forward KL）与反向KL（reverse KL）三种损失处理方式，并基于梯度分解技术精确区分由学生或教师生成的每个令牌所贡献的梯度方向。研究发现，在短周期训练与BF16精度推理条件下，三类损失在梯度方向上高度一致，且在准确率上统计等效（误差范围±5个百分点），表明在当前设定下教师文本作为上下文或训练目标并未带来显著增益。尽管探索性实验提示在教师与学生生成差异最大的位置插入桥接可能有益，但预设复制实验未验证此效应，四倍加权也未显示显著优势。研究还揭示两个重要限制：一是训练量有限时，性能增益在四次更新后即达峰值并随生成持续而下降；二是所有在线训练变体均未能满足稳定停止生成和提升准确率的标准。此外，因BF16推理会舍弃微小权重变化，而在精确FP32评估下模型常触及响应长度上限，故结论仅适用于BF16环境，不意味着教师令牌监督在一般情况下无效。
链接: https://arxiv.org/abs/2610.38383
作者: Leran Chen,Zhiyuan Yao,Lingnan Kong
机构: 未知
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: 25 pages, 8 figures, 18 tables
Abstract:On-policy distillation (OPD) trains a student using teacher predictions on student-generated responses. Some methods insert teacher-written segments, or bridges, but disagree on how to train on those tokens. We compare no loss, forward KL and reverse KL on identical trajectories using Qwen3-1.7B as student and Qwen3-4B-Instruct-2507 as teacher. We separate gradients by which model wrote each token. Under our short training runs and BF16 inference, pre-specified tests detected no added accuracy benefit from teacher text as context or as a training target. A one-paragraph bridge after the student’s first paragraph is only 2.7% of the tokens and contributes a small gradient nearly orthogonal to the student-token gradient. In this main comparison, the three loss treatments gave similar gradient directions and were statistically equivalent in accuracy within a +/-5 percentage-point margin. An exploratory pilot placing the bridge where failed responses disagreed most with the teacher suggested a benefit from teacher-token loss. A pre-specified replication did not confirm this, and fourfold weighting showed no significant accuracy benefit. Two limitations bound these findings. Among the tested amounts of training on fixed responses, gains peaked after four updates before the student increasingly failed to stop generating. None of the tested online training variants met the criteria for stable stopping and improved accuracy. Our evaluations used BF16 inference, which rounds away most of the small weight changes. When evaluated in exact FP32, the tested replication models frequently reached the response-length limit. Our accuracy findings therefore concern BF16 inference; they do not establish that teacher-token supervision is generally unnecessary.
[NLP-110] CC-RAG : Temporal Causal Chain-Guided RAG for Temporal Knowledge Graph Question Answering ICDE2027
【速读】： 该论文旨在解决现有基于语义检索增强生成（Semantic Retrieval-Augmented Generation, RAG）的时序知识图谱问答（Temporal Knowledge Graph Question Answering, TKGQA）方法在复杂时序推理中存在的根本性缺陷。具体而言，传统RAG方法难以有效捕捉多跳因果知识以支持复杂的时序推理，并且容易引入大量冗余信息，导致大语言模型陷入虚假的推理链，从而产生错误答案。其解决方案的关键在于提出一种时序因果链引导的检索增强生成框架（Temporal Causal Chain-guided RAG, TCC-RAG）。该方法将时序知识图谱重构为多视图因果线索图（multi-view causal cue graph），通过显式编码时序演化与因果动态的因果线索建立连接；在此基础上，构建与查询相关的时序因果链，指导检索过程，使模型聚焦于一组精炼且因果相关的知识；最后引入因果证据过滤机制，确保所检索知识与查询背后的时序逻辑保持因果一致性。实验表明，TCC-RAG在两个TKGQA基准上分别实现了10.72%和8.40%的性能提升，并在多步时序推理任务中取得62.97%的显著改进。
链接: https://arxiv.org/abs/2610.38366
作者: Xinxun Zhang,Yuxiang Wang,Shanshan Feng,Jun Zhang,Tieyun Qian,Binhang Yuan,Kai Yin,Xiao Yan,Jiawei Jiang
机构: Wuhan University (武汉大学); Tencent Inc. (腾讯公司); Hong Kong University of Science and Technology (香港科技大学); Central China Normal University (华中师范大学)
类目: Computation and Language (cs.CL)
备注: 14 pages, 6 figures, ICDE2027 Research Track
Abstract:Recent advances in large language models (LLMs) have spurred growing interest in temporal knowledge graph question answering (TKGQA). Most existing methods rely on semantic retrieval augmented generation (RAG), which suffers from fundamental limitations in complex temporal reasoning. On the one hand, semantic-based RAG methods struggle to capture the multi-hop causal knowledge required for complex temporal reasoning. On the other hand, these methods introduce excessive redundant knowledge, trapping LLMs in spurious reasoning chains that potentially result in erroneous answers. To address these problems, we propose a Temporal Causal Chain-guided RAG (TCC-RAG) to retrieve a concise yet sufficient set of causal knowledge to support complex temporal reasoning. Specifically, we reformulate the temporal knowledge graph into a multi-view causal cue graph, with causal cues inducing connections that explicitly encode temporal evolution and causal dynamics. Based on the causal cue graph, we construct a query-specific temporal causal chain to guide the retrieval process, enabling the model to focus on a concise set of causally relevant knowledge. Finally, we introduce a causal evidence filter to refine the retrieved knowledge, ensuring its causal consistency with the underlying temporal logic of the query. Experiments show that TCC-RAG achieves substantial performance gains of 10.72% and 8.40% on two TKGQA benchmarks, and delivers a remarkable 62.97% improvement for multi-step temporal reasoning.
[NLP-111] Statutory Article Retrieval from Arabic Judicial Narratives: Family Law and Law of Evidence in Saudi Arabia
【速读】： 该论文旨在解决阿拉伯语司法场景中法律信息检索的可靠性问题，核心挑战在于如何将冗长的事实叙述（narrative）有效映射至简洁的法律条文（legal rule），即存在“叙事到法律规则的抽象鸿沟”（narrative-to-legal-rule abstraction gap）。为应对这一问题，论文提出一种名为“分解与精炼”（Decompose-and-Refine, DaR）的新型检索框架。其关键创新在于：利用大语言模型（LLM）将原始案件叙述分解为具体的争议法律议题，并针对每个议题生成三种互补的检索通道——条文风格的改写文本、法律关键词以及候选分类章节（taxonomy chapter），从而实现对复杂法律文本的多层次、精细化检索。实验表明，在沙特阿拉伯司法案例数据集上，非分解系统在家庭法与证据法上的表现差异显著，而DaR通过统一的检索策略显著提升了两类法律的检索效果，使证据法的MRR@10从0.087提升至0.696，家庭法则达到0.864，充分验证了分解与多通道协同在跨法律领域泛化能力中的核心作用。
链接: https://arxiv.org/abs/2610.38336
作者: Zarah Shibli,Areej Alhothali,Lulwah Alharigy
机构: King Abdulaziz University (国王阿卜杜勒阿齐兹大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: Submitted to the MLNLP 2026 Conference
Abstract:Reliable legal information retrieval must ground legal reasoning in authoritative statutory sources. In Arabic judicial settings, long factual narratives must be mapped to concise articles that express abstract legal rules, creating a narrative-to-legal-rule abstraction gap. Using a benchmark of 200 synthetic Saudi Arabic judicial cases and 381 articles from the Family Law and the Law of Evidence, we compare word-level and char-4 BM25, zero-shot dense encoders, per-law retrieval, dense-sparse reciprocal rank fusion (RRF), and a proposed Decompose-and-Refine (DaR) pipeline. In DaR, a large language model splits each narrative into disputed legal issues and generates, for each issue, three retrieval channels: a statute-style paraphrase, legal keywords, and a candidate taxonomy chapter. On the whole corpus, the instructed Qwen3-Embedding-8B dense encoder reaches an MRR@10 of 0.849, the best result of any non-decomposed system. Under per-law scoring, this advantage reflects Family Law only: restricted to Evidence articles, the same encoder drops to an MRR@10 of 0.087, and fusion with BM25 raises it only to 0.207. DaR changes this picture, raising Evidence MRR@10 to 0.696 and Family MRR@10 to 0.864. Without decomposition, the two laws favor different strategies; under DaR, the same retriever achieves the best MRR@10 on both.
[NLP-112] Fu-TCM: a multimodal large language model that learns traditional Chinese medicine diagnostic reasoning
【速读】： 该论文旨在解决中医证候辨识（Bianzheng）中多模态信息融合与复杂推理链构建的难题，尤其针对传统中医诊断依赖经验、缺乏系统化计算模型的问题。其核心解决方案在于提出一种基于多模态数据训练的中医大语言模型——Fu-TCM，通过引入“辨证基础策略优化”（Bianzheng-Grounded Policy Optimization, BGPO）机制，显式地对响应格式、中间推理字段及最终证候预测进行奖励驱动的优化。该方法强化了模型在中医辨证过程中的结构化推理能力，使模型在六项涵盖文本与视觉任务的基准测试中均取得领先，宏平均准确率达81.01%，显著优于现有最强基线模型7.13个百分点，且在“Fu-bianzheng”任务上最大提升达11.90个百分点。案例级分析进一步验证了中间推理准确性与最终证候判断准确性的正相关性，表明中医证候辨识可通过结构化推理监督实现可计算化建模，为生成式AI在中医智能化诊疗中的应用提供了新范式。
链接: https://arxiv.org/abs/2610.38333
作者: Jiahang Xie,Tong Jin,Lingfei Wu,Yaping You,Jie Huang,Yuanchang Jin,Weikang Gong,Wei Cheng,Jia You,Pan Sun,Jianfeng Feng,Hongcheng Guo
机构: Fudan University (复旦大学); The Affiliated People’s Hospital of Fujian University of Traditional Chinese Medicine (福建中医药大学附属人民医院); Institute of Science and Technology for Brain-Inspired Intelligence, Fudan University (复旦大学脑科学与智能技术研究院); Key Laboratory of Computational Neuroscience and Brain-Inspired Intelligence, Fudan University, Ministry of Education (复旦大学计算神经科学与脑智能教育部重点实验室); Beijing University of Chinese Medicine (北京中医药大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Large language models (LLMs) are advancing in multimodal understanding and reasoning, accelerating medical model development. Traditional Chinese medicine (TCM) uses evidence from the four examinations to determine disease location and nature, differentiate syndromes and guide treatment. This diagnostic process constitutes a complete chain of reasoning. We present Fu-TCM, a multimodal TCM LLM trained on data retained after cleaning and filtering an initial collection of approximately 7.12 million examples derived from classical texts, textbooks, medical images, clinical cases and publicly available datasets. The model was further optimized using Bianzheng-Grounded Policy Optimization (BGPO), which explicitly targets syndrome differentiation through rewards for response format, intermediate TCM reasoning fields and final syndrome predictions. Fu-TCM-27B led all six benchmarks spanning TCM reasoning, text and vision, achieving a macro-average accuracy of 81.01% and surpassing the strongest reference model with complete results across all six benchmarks by 7.13 percentage points. Its largest margin was 11.90 points on Fu-bianzheng. Case-level analyses further showed a positive association between intermediate-field accuracy and final syndrome-selection accuracy in most evaluated models. These findings suggest that TCM syndrome differentiation may be computationally learnable and that structured reasoning supervision can improve performance across TCM tasks.
[NLP-113] Robust LLM -Generated Text Detection through Matched Conditional-Response Probing
【速读】： 该论文旨在解决大语言模型（Large Language Model, LLM）生成文本在经过语义保持型修改（如重写、润色、扩展等）后，现有检测方法因词汇、句法及局部话语线索被削弱而失效的问题。其核心挑战在于如何在不依赖表面特征的前提下，仍能有效识别出由LLM生成的文本。解决方案的关键是提出一种冻结式大模型探针框架——匹配条件响应探测（Matched Conditional-Response Probing, MCRP），该框架通过引入语义匹配的条件上下文，评估候选文本在条件引入、移除或替换时的响应行为。具体而言，冻结的探针模型首先生成短语级语义条件，并在匹配、空条件和控制条件下分别计算分词级别损失与提示-目标注意力分布，进而聚合双向条件对齐性、反事实敏感性以及内部注意力耦合性，形成高判别性的探针表示。在此基础上，采用轻量级监督随机森林分类器进行训练，同时保持探针模型参数冻结。实验表明，MCRP在标准检测、鲁棒性测试（针对重写、润色、扩展）、多语言及跨领域场景中均表现卓越，在语义保持型修订下达到93.97% F1和96.24% ROC-AUC，显著优于最强基线模型（分别提升19.50和25.77个百分点）。消融实验进一步验证了各响应特征组的互补贡献，条件控制研究也证明语义匹配条件相较随机、错配或损坏条件具有显著优势，凸显了语义一致性在检测中的关键作用。
链接: https://arxiv.org/abs/2610.38312
作者: Junfeng Hao,Wenxian Wang,Tao Leng,Xingshu Chen,Hao Ren,Haizhou Wang
机构: Sichuan University (四川大学); Sichuan Police College (四川警察学院)
类目: Computation and Language (cs.CL)
备注:
Abstract:Reliable detection of large language model (LLM)-generated text must remain effective under meaning-preserving text revisions, such as rewriting, polishing, and expansion. In practice, LLM outputs are often rewritten, polished, expanded, translated, or lightly edited before publication, weakening the lexical, syntactic, and local discourse cues used by many existing detectors. We introduce \textbfMatched \textbfConditional-\textbfResponse \textbfProbing (\textbfMCRP), a frozen-LLM probing framework that measures how a candidate text responds when a semantically matched condition is introduced, removed, or replaced. For each candidate, the frozen probe LLM first generates a short semantic condition and then compares token-level losses and prompt–target attention measurements under matched, null, and control conditions. It aggregates bidirectional conditioned alignment, counterfactual sensitivity, and internal attention coupling into a probe representation, on which it trains a lightweight supervised random-forest classifier while keeping the probe LLM frozen. Across standard detection, robustness to meaning-preserving text revisions, multilingual, and multidomain evaluations, MCRP achieves consistently strong performance. Under rewriting, polishing, and expansion, it reaches 93.97% F1 and 96.24% ROC-AUC, outperforming the strongest evaluated baseline by 19.50 and 25.77 percentage points, respectively. Component ablations show that the response feature groups make complementary contributions to predictive performance. In the condition-control study, matched conditions consistently outperform random, mismatched, and corrupted alternatives, demonstrating a measurable incremental benefit from semantic matching.
[NLP-114] EPSBench: Benchmarking LLM Agents for Earnings Forecasting
【速读】： 该论文旨在解决季度盈利预测中模型如何有效关联财务证据与利润表项目以准确估算公司特定收入驱动因素的问题。其核心挑战在于确保模型不仅能生成合理的每股收益（EPS）预测，还能在预测过程中提供可追溯的、基于证据的推理过程。解决方案的关键在于提出EPSBench这一基准测试框架，涵盖86家美国公司共149个公司-季度样本，要求模型读取固定范围的财报文件、电话会议记录和新闻资讯，并基于引用证据估计利润表各驱动项；随后通过一个共享的会计计算工具将这些估计值统一转换为按照公认会计准则（GAAP）计算的稀释每股收益（EPS），从而实现对预测假设的标准化比较。实验评估了六种模型在点预测误差与预测区间覆盖率方面的表现，发现三类模型的点预测误差与分析师共识无显著差异，但所有模型的名义80%预测区间实际覆盖率为37%–70%，且多数偏差出现在预测值偏低导致的实际EPS超出上界的情况。进一步分析表明，模型间误差差异主要源自对营业利润（operating income）估计的准确性。
链接: https://arxiv.org/abs/2610.38293
作者: Yanbin Yin,Chuyu Xu,Liqin Ye,Jing Peng,Chao Zhang
机构: Georgia Institute of Technology(佐治亚理工学院)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Quarterly earnings forecasting requires an agent to connect financial evidence to the income-statement line items that produce a company’s reported earnings. We introduce EPSBench, a benchmark for studying this ability across 149 firm-quarters from 86 US companies. Agents read a fixed collection of filings, earnings-call transcripts, and news, then estimate company-specific income-statement drivers with cited evidence. A shared accounting calculator converts these estimates into GAAP diluted earnings per share (EPS), providing a common basis for comparing forecast assumptions. We evaluate six models under a common harness, assessing point forecast error relative to date-matched analyst consensus and checking whether their prediction intervals achieve the stated coverage. Three models show no detectable paired difference in point forecast error from consensus, and three have higher error. Across all six models, nominal 80% intervals contain reported EPS in only 37–70% of runs that return an interval, and most misses fall above the upper bound because the forecasts run low. The driver records trace most of the error gap between models to operating-income estimates.
[NLP-115] Human Evaluation with Automatic Scores: Separating Sampling and Estimation Gains
【速读】： 该论文旨在解决如何高效利用自动评分（automatic judges）提供的低成本模型输出评估信息，以减少人工标注成本的同时保证评估精度的问题。其核心挑战在于：在有限预算下，如何将自动评分结果转化为具有高信度的人工标注策略，从而实现最优的评估性能。解决方案的关键在于通过分层采样（score stratification）与基于分层的人类均值估计相结合的简单但稳健的流程——即在分数分层内按比例进行人工标注，并采用分层人类均值作为最终估计。研究发现，在固定评估池中，经过十层分层后，即使采用最优局部线性校正方法，也仅能带来1.00%–5.27%的均方误差（MSE）进一步降低，表明校正带来的边际收益有限；而基于得分的不等比例分配虽在某些数据集上表现更优，但在模型预测置信度掩盖人类判断变异性时反而可能增加误差。在AttributedQA任务中，将学习到的配额向仅依赖得分的配额收缩，可使17/19个系统在整体MSE上优于纯得分基准，但标准95%置信区间覆盖率仅为88.09%–93.54%，表明当前方法仍存在覆盖不足问题。此外，排名实验显示个体模型的低误差并不必然提升系统选择准确性。因此，研究支持采用“分层内比例采样+分层人类均值”作为默认起始方案，并强调需额外证据才能证明对校正或自适应采集策略的必要性。
链接: https://arxiv.org/abs/2610.38282
作者: Yang Liu,Ran Zou,Zhengtao Gui,Chao Xue
机构: University of North Carolina at Chapel Hill (北卡罗来纳大学教堂山分校); University of California, Irvine (加州大学欧文分校); University of New South Wales (新南威尔士大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Automatic judges offer inexpensive information about model outputs, but an evaluation team must decide how to turn that information into fewer human annotations. Scores can guide sampling, correct an estimate, or support adaptive acquisition, and these uses may capture overlapping gains. We compare their incremental value on fixed evaluation pools from three preference datasets, and 19 attributed-question-answering systems. Matched comparisons use the same scores, account for pilot labels and budget rules, and separate acquisition from estimation. Score stratification is a strong reference: after ten-bin stratification, the best fixed local linear correction offers only 1.00–5.27% further MSE reduction in the primary configurations. Unequal allocation offers larger gains in some pools, but can increase error when confident scores conceal variable human outcomes. Comparison on AttributedQA, shrinking learned quotas toward score-only quotas lowers pooled MSE for 17 of 19 systems relative to the score-only procedure, but ordinary pooled 95% intervals cover only 88.09–93.54%. The same shrinkage improves on the score-only reference in only one of eight preference configurations. Ranking experiments further show that lower individual MSE need not improve system selection. These findings support a simple starting procedure, proportional sampling within score strata followed by a stratified human mean, and identify what additional evidence is needed to justify correction or adaptation for the intended evaluation output.
[NLP-116] Cross Lingual Abstractive Summarization for Marathi
【速读】： 该论文旨在解决印地语族语言（Indic languages）在单语文本摘要（Monolingual Summarization, MS）与跨语言文本摘要（Cross-lingual Summarization, CLS）任务中研究相对不足的问题，尤其针对马拉地语（Marathi）等资源稀缺语言的摘要生成能力薄弱。其核心挑战在于缺乏高质量的跨语言训练数据以及适用于这些语言的高效模型架构。解决方案的关键在于利用预训练神经机器翻译模型IndicTrans2，通过将单语马拉地语MahaSum语料库构建为跨语言训练数据，并与现有的小规模跨语言PMIndiaSum数据集相结合，实现对IndicTrans2的微调，从而生成高质量的跨语言摘要。实验结果表明，该方法在马拉地语到英语的摘要任务上显著优于基于IndicBART和mBART-50的基线模型，证明了利用预训练翻译模型进行跨语言摘要的有效性，同时该范式亦具备向英语到马拉地语方向扩展的潜力。
链接: https://arxiv.org/abs/2610.38267
作者: Asang Dani,Shailesh R Sathe
机构: Visvesvaraya National Institute of Technology (Visvesvaraya 国家技术学院); Nagpur, India
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Text summarization is an important task for advancing natural language processing (NLP) research and applications in Indic languages. Despite progress in extractive and abstractive summarization, both monolingual summarization (MS) and cross-lingual summarization (CLS) remain relatively underexplored for these languages. Recent advances in Transformer architectures, pretrained language models (PLMs), and dataset availability offer new opportunities to address this gap. In this work, we construct cross-lingual training data from the monolingual Marathi MahaSum corpus and combine it with the smaller cross-lingual PMIndiaSum dataset to fine-tune IndicTrans2, an Indic neural machine translation model. We evaluate the resulting models using ROUGE, chrF and chrF++ metrics, focusing primarily on Marathi-to-English summarization due to computational constraints. We also discuss the applicability of our approach to the English-to-Marathi direction. Our experiments demonstrate that leveraging a pretrained translation model yields substantially better summarization performance than the IndicBART and mBART-50 based baselines reported in previous work on cross-lingual summarization.
[NLP-117] Keepace.ai: A Technical Report on a Large Language Model for Sports and Health
【速读】： 该论文旨在解决生成式 AI 在体育与健康领域缺乏高质量、领域专用训练语料库及严格评估基准的问题，从而制约了其在提供安全、科学且个性化的运动指导方面的应用。其解决方案的关键在于基于 Qwen3.5-27B 模型通过后训练构建一个面向中文场景的领域专用大语言模型（domain-specialized LLM），并聚焦于三大核心任务：运动计划生成、体育健康知识问答以及个人运动数据解析。该方案通过构建由领域专家设计的评分体系进行内部评估，在运动安全性、边界合规性、结构化输出完整性及科学准确性等方面显著优于基线模型和更大规模的通用模型，体现了以安全为先的评估框架与精细化的数据处理流程在提升模型性能中的关键作用。
链接: https://arxiv.org/abs/2610.38249
作者: Keepace.ai Team,Bowen Wu,Heng Wang,Jufeng Mi,Kang Liu,Kun Li,Ling Man,Long He,Suli Fan,Xiaosen Jiang,Yi Wang,Ying He,Yu Tang,Yunhan Li,Zifang Liu
机构: Keepace.ai(Keepace.ai)
类目: Computation and Language (cs.CL)
备注: 21 pages, 3 figures, 13 tables. Technical report
Abstract:Large language models hold promise for delivering safe, scientifically grounded, and personalized exercise guidance at scale, yet the sports-and-health domain still lacks curated training corpora and rigorous evaluation benchmarks. this http URL is a domain-specialized LLM built on Qwen3.5-27B through post-training, currently optimized for Chinese-language exercise guidance across three tasks: workout plan generation, sports-and-health knowledge QA, and personal sports data analysis. In internal rubric-based evaluations designed with domain experts, this http URL outperforms its base model and larger general-purpose baselines, with the most pronounced improvements in exercise safety, boundary compliance, structured output completeness, and scientific accuracy. This report summarizes the current design, data pipeline, safety-first evaluation framework, and evaluation results of this http URL.
[NLP-118] 3DCodeEdit: Benchmarking Instruction-Guided 3D Program Editing
【速读】： 该论文旨在解决生成式3D内容时，语言模型在代码层面进行复杂编辑任务中的理解与预测能力不足的问题。现有方法多依赖于从头生成或简单修改，难以实现高效且精确的迭代式编辑，尤其在涉及参数调整、结构变更、风格重构及场景组合等复杂操作时表现不佳。其核心挑战在于模型缺乏对代码所构建几何体的深层理解以及对修改后结果的几何形态预判能力。为此，作者提出3DCodeEdit基准，涵盖108个完全由代码构建的Blender对象，包含2,592项指令引导的编辑任务，覆盖从参数级到结构性、风格化及场景级的多样化修改，并通过人工评审的渲染结果清单进行评估。实验表明，17个主流模型平均成功率仅为37%，即使是性能最强的模型在结构性和风格化编辑中也难以超过60%。随着模型能力提升，失败模式逐渐转向几何预见性缺失——即代码可执行但生成形状错误。为应对这一瓶颈，研究进一步探索多轮反馈机制的有效性，发现由独立评审模型提供关于渲染结果与需求差异的反馈信息，显著优于单纯增加推理深度或引入智能代理，甚至能使一个270亿参数模型超越2.8万亿参数模型的单次生成表现，揭示了基于反馈的交互式编辑在弥补几何预见性缺陷方面的关键作用。
链接: https://arxiv.org/abs/2610.38231
作者: Binjie Wang,Xinran Zhao,Shoubin Yu,Xiang Li,Silong Yong,Chenyang Yang,Tianyu Chen,Ruofei Du,He He,Chen Zhao,Tongshuang Wu
机构: University of North Carolina at Chapel Hill (北卡罗来纳大学教堂山分校); University of Illinois Urbana-Champaign (伊利诺伊大学厄本那-香槟分校); Carnegie Mellon University (卡内基梅隆大学); The University of Texas at Austin (德克萨斯大学奥斯汀分校); Google(谷歌); New York University (纽约大学)
类目: Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR)
备注:
Abstract:Language models increasingly create 3D content through code, with practical workflows often relying on editing existing programs rather than regenerating them, for speed and precise control. Such edits require understanding what a program builds and foreseeing how a change reshapes it, yet prior benchmarks mostly evaluate generation or simple edits against a reference. In this paper, we introduce 3DCodeEdit, a benchmark of 2,592 instruction-guided editing tasks over 108 Blender programs that construct objects entirely in code. Its edits range from parameter to structural changes, style redesign, and scene composition, and are judged from renders against human-reviewed checklists. Across 17 models, the average success rate is only 37%, and even the strongest models complete fewer than 60% of structural and style edits. As models become more capable, their failures shift to geometric foresight, where the code runs but builds the wrong shape. We therefore study whether multi-turn feedback can compensate for this missing foresight. Feedback from a separate reviewer model that describes how the renders differ from the request helps most, more than maximum reasoning effort or an agent harness, and can let a 27B model surpass the single-shot rate of a 2.8T-parameter model.
[NLP-119] wo Birds with One Stone: Skill-Augmented Generation via Unified Retrieval and Utilization
【速读】： 该论文旨在解决现有技能增强生成方法中技能检索与技能利用过程分离导致的优化不连贯问题。当前方法通常依赖独立模型分别完成技能检索与任务执行，造成两阶段能力难以协同优化。为此，本文提出Skill-Uni——一种统一的强化学习框架，通过生成式技能检索范式实现单一模型对技能检索与利用的联合优化。其关键在于：1）采用生成式检索机制，使模型在生成过程中同时学习如何精准定位相关技能并有效应用；2）设计两阶段训练策略（预热指令微调+GRPO），分步提升模型在技能检索与任务求解上的综合能力。实验结果表明，Skill-Uni在五个技能相关基准上显著优于基线模型，相较Qwen3-4B和LLaMA3.1-8B分别实现66.59%和34.82%的平均性能提升，且具备良好的兼容性与泛化能力。
链接: https://arxiv.org/abs/2610.38204
作者: Zihan Luo,Jiahui Kang,Wei Zhou,Qilin Fan,Mingyan Li,Chao Chen
机构: Chongqing University(重庆大学); Huazhong University of Science and Technology(华中科技大学)
类目: Computation and Language (cs.CL)
备注: Under Review
Abstract:Skills, as reusable external knowledge, have demonstrated substantial potential to enhance the ability of Large Language Models (LLMs) to solve complex tasks. However, existing skill-augmented generation methods typically rely on separate models for skill retrieval and skill utilization, resulting in disconnected optimization of these two processes. To address this gap, we propose Skill-Uni, a unified reinforcement learning framework that jointly optimizes skill retrieval and utilization. Specifically, we adopt a generative skill retrieval paradigm that enables a single model to both learn how to retrieve relevant skills and leverage them for task solving. To effectively train this unified framework, we introduce a two-stage training strategy consisting of warm-up instruction tuning and GRPO, allowing the model to progressively improve its skill retrieval and utilization capabilities. Extensive experiments on five skill-related benchmarks demonstrate the effectiveness of Skill-Uni, yielding average performance gains of 66.59% and 34.82% over the base Qwen3-4B and LLaMA3.1-8B models, respectively. Besides, Skill-Uni is compatible with various skill retrieval engines, demonstrating its flexibility and generalizability. Additional in-depth analyses also provide insights into the effectiveness of Skill-Uni.
[NLP-120] Component Translation and Judge-Labelled Refusal: A Multilingual Safety Evaluation
【速读】： 该论文旨在解决全提示翻译（whole-prompt translation）评估中无法区分语言成分来源的问题，即在多语言环境下难以确定是请求（request）组件还是操作（operation）组件的语言属性影响了模型的安全行为。其核心解决方案是引入成对反转（paired reversal）机制：将完整的请求组件或操作组件整体翻译至同一目标语言，同时固定行为、英文框架及翻译段落数量，使组件身份与非英文文本量同步变化。通过独立分配英语、中文和西班牙语至两个组件，形成九种配置，在300个样本中保留的190个有害行为上进行测试。以大语言模型判别器的拒绝率作为主要结果，不安全合规性单独报告。结果显示，在Gemini 2.5 Flash和DeepSeek模型上，三次拒绝率变化约5.3个百分点，第四次为-0.5，均未通过预设的六次检验Holm校正；而在探索性扩展中，Mistral Small 4显示13.7个百分点的西班牙语转换效应（95%置信区间[8.4, 18.9]），在主判别器下通过十二次检验的Holm敏感性校正，另一判别器下亦达12.1个百分点。GPT-5.4-mini估计值接近零（±2个百分点）。研究表明，组件分配提供了一个额外的评估维度，强调评估应记录各组件语言类型与翻译文本量，并在相同翻译段落数条件下对比匹配行为，同时分别解读拒绝率与不安全合规性。
链接: https://arxiv.org/abs/2610.38181
作者: Yongxi Zhou,Junwei Yao,Wenbo Ye,Jiahui Ge
机构: Independent Researcher(独立研究员)
类目: Computation and Language (cs.CL); Cryptography and Security (cs.CR)
备注: 12 pages, 1 figure. Ten-page manuscript prepared for the 4th Workshop on Large Language and Foundation Models (WLLFM 2026) at IEEE BigData 2026, plus appendices A-E. Replication archive included as ancillary files
Abstract:Whole-prompt translation evaluates a request in different languages but does not compare which component carries a language. We introduce a paired reversal: translate the complete request component or the complete operation component into the same target language, holding behaviour, English frame and translated-span count fixed. Component identity and non-English text quantity change together. Assigning English, Chinese and Spanish independently to the two components yields nine configurations on 190 of 300 sampled harmful behaviours retained after consensus annotation and automatic gating. Refusal scored by language-model judges is the primary outcome; unsafe compliance is reported separately. On Gemini 2.5 Flash and DeepSeek, three refusal swaps are about 5.3 percentage points and the fourth is -0.5; none survives the pre-specified six-test Holm correction. In an exploratory extension, Mistral Small 4 has a 13.7-point Spanish swap (95% interval [8.4, 18.9]), which survives the twelve-test Holm sensitivity correction under the primary judge, and 12.1 points under a second judge. GPT-5.4-mini estimates are within 2 points of zero. Component assignment supplies an additional evaluation dimension. Evaluations should record component languages and translated text quantity, comparing matched behaviours at equal translated-span count while interpreting refusal and unsafe compliance separately.
[NLP-121] Scaling Lean Theorem Proving for Real-World Code Verification
【速读】： 该论文旨在解决形式化验证中高难度、接近饱和的基准测试任务难以有效推进的问题，尤其针对数学竞赛（如IMO 2026）与软件验证（如NTP4VC、VERINA）等复杂证明场景。其核心挑战在于：现有基准缺乏对实际程序实现细节（如固定宽度算术、显式错误处理、生成循环、类型转换与函数化变异）的充分建模，导致生成的证明任务与真实系统间存在语义鸿沟。为此，论文提出的关键解决方案是构建一个基于递归代理（recursive agents）、隔离上下文（isolated contexts）、异步搜索（asynchronous search）、并行策略探测（parallel tactic probing）以及弹性实例化服务（elastic elaboration service）的Lean证明系统——EasyProver。该系统无需额外训练即可在未修改的前沿模型基础上完成669/672个PutnamBench问题、全部6个IMO 2026形式化命题及597/600个NTP4VC验证条件的证明或反证。为进一步提升挑战性，论文引入Aegis框架，将编程竞赛题目转化为包含真实实现约束的高难度验证任务，通过安全的Rust代码生成、使用Aeneas工具链翻译至Lean、自动生成规范并为每个冻结断言附带内核可验证的证明或反证，构建出2,045对差分验证的Rust-Lean程序对与11,533个候选规格。在2,000条断言的分层样本中，EasyProver在每条断言30分钟预算下成功证明了1,054条（52.7%），展现出强大的自动化推理能力。该方法不仅可用于评估其他自动证明器性能，还可作为强化学习中的奖励信号，推动下一代智能证明系统的发展。
链接: https://arxiv.org/abs/2610.38174
作者: Debargha Ganguly,Divya Raghunathan,Dejan Jovanović,Vikash Singh,Soonho Kong,Aaron Tomb,Aman Goel,Ferhat Erata,Sam Bayless
机构: Case Western Reserve University (凯斯西储大学); Amazon Web Services (亚马逊网络服务)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Logic in Computer Science (cs.LO); Software Engineering (cs.SE)
备注:
Abstract:We present EasyProver, a Lean proving system built from recursive agents with isolated contexts, asynchronous search, parallel tactic probing, and an elastic elaboration service. Using an off-the-shelf frontier model with no additional training, it proves 669/672 PutnamBench problems and all six IMO 2026 formalizations and resolves 597/600 NTP4VC verification conditions and all 189 VERINA tasks by proof or refutation. Because these benchmarks are close to saturation, we also introduce Aegis, which converts competitive programming problems into much harder verification tasks over real implementations. The pipeline writes safe Rust solutions, translates them into Lean with Aeneas, generates specifications, and labels each frozen statement with a kernel-checked proof or refutation. It produces 2,045 differentially validated Rust-Lean program pairs and 11,533 candidate specifications. Unlike hand-written benchmarks, these tasks carry implementation-level obligations from fixed-width arithmetic, explicit failure, generated loops, casts, and functionalized mutation. On a stratified sample of 2,000 statements, EasyProver proves 1,054 (52.7%) within a 30-minute budget per statement. The same tasks can also be used to benchmark other provers and as rewards for reinforcement learning.
[NLP-122] ATI-OPD: Adaptive Teacher Intervention for Stable Long-Sequence On-Policy Distillation
【速读】： 该论文旨在解决生成式 AI（Generative AI）中基于策略蒸馏（On-Policy Distillation, OPD）训练时因学生模型（student）响应长度增加而导致的教师-学生推理路径偏离问题。随着生成序列变长，学生模型的前缀逐渐脱离教师模型的原始推理轨迹，导致教师对后续标记的反馈质量下降，进而引发训练不稳定性，甚至加剧偏差传播。现有方法如长度控制和损失重加权虽能限制不良响应的影响，但未能修正已偏离的推理前缀，且固定截断机制可能丢弃有效推理内容。本文提出自适应教师干预（Adaptive Teacher Intervention for On-Policy Distillation, ATI-OPD），其核心在于动态修正学生模型的推理前缀：当检测到早期高分歧区域时，用简短的教师生成片段替换该部分，随后允许学生从修正后的文本继续自主生成；同时，对于持续分歧的后缀部分，将其从蒸馏损失中排除，保留学生有用的后续生成。该方案有效降低了教师与学生间的推理差异，减少了梯度突增与极端长序列现象，在五个数学推理基准及两组师生模型配置下，显著提升了学生模型在无教师协助下的独立准确率。
链接: https://arxiv.org/abs/2610.38160
作者: Jiannan Guan,Youxin Jiang,Yutai Hou,Qiguang Chen,Yirong Zeng,Haitao Wang,Dengyun Peng,Zheng Yan,Jinhao Liu,Tao Ye,Dandan Tu,Wanxiang Che
机构: Huawei Technologies Co., Ltd.(华为技术有限公司); Research Center for Social Computing and Interactive Robotics, Harbin Institute of Technology(哈尔滨工业大学社会计算与交互机器人研究中心)
类目: Computation and Language (cs.CL)
备注:
Abstract:On-policy distillation (OPD) trains a student on its own responses using token-level feedback from a teacher. As responses grow longer, the student’s prefix can depart from the teacher’s own reasoning, weakening its feedback on later tokens. Updates based on this feedback can reinforce the same deviations in later responses and destabilize training. Prior approaches such as length control and loss reweighting limit learning from problematic responses but leave their prefixes unchanged. Fixed cutoffs can also discard later reasoning. We propose \textbfAdaptive Teacher Intervention for On-Policy Distillation (ATI-OPD) to improve subsequent teacher guidance by revising the student’s reasoning prefix. ATI-OPD replaces the earliest high-disagreement segment with a short teacher segment, then lets the student resume its own reasoning from the revised text. To handle renewed disagreement, it excludes persistently divergent suffixes from the distillation loss while retaining useful student continuation. Training diagnostics show less teacher-student disagreement, fewer gradient spikes, and fewer extremely long responses. Across five mathematical reasoning benchmarks and two teacher-student pairs, ATI-OPD improves average student-only accuracy over standard OPD.
[NLP-123] Calculation and rejection diverge in language models tested at known mechanical limits
【速读】： 该论文旨在解决生成式 AI 在工程设计场景中对物理不可行设计误判为“已解决”（solved）的问题，即当模型返回的工程参数超出材料或结构极限时，仍被下游软件系统信任并传递，导致潜在安全隐患。其核心问题在于：尽管模型具备准确计算极限载荷的能力，但在完整回答请求时却会接受明显违反物理可行性的设计，而仅在被明确询问可行性时能正确拒绝。解决方案的关键在于揭示模型在不同提示范式下的行为差异——当仅被问及可行性时，部分先进模型（如Opus 5和Gemini 3.8 Flash）能够零误报地拒绝所有不可行设计；然而在要求提供完整解算结果时，同一模型却可能接受超出塑性屈服或爆裂极限10%至30%的设计，且部分回复中明确承认违规。研究进一步表明，调整提示措辞或输出格式虽可提升部分案例中的拒绝对比，但对新设计的泛化效果不一致。因此，该研究强调，应将可行性判断与完整答案拒绝能力分开测试，并严格记录提示内容，以确保评估的可复现性与系统安全性。
链接: https://arxiv.org/abs/2610.38096
作者: Shaoliang Yang,Jun Wang
机构: Santa Clara University (圣克拉拉大学); Santa Clara, CA 95053, USA
类目: Computational Engineering, Finance, and Science (cs.CE); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 94 pages, including Supplementary Information. Data and code: this https URL
Abstract:When a language model returns engineering quantities with the status “solved”, software that trusts it passes them on, even when the requested design is physically impossible. Such failures could reflect poor calculation, how the answer is requested, or a failure to judge feasibility. We set beam loads and disk speeds at known percentages above or below plastic collapse or burst, measured each model’s limit calculation separately and tested 12 models in 4,860 calls, including follow-up experiments on new designs. We find that a model can recognize an impossible design when asked only whether it is feasible, yet accept the same design, reporting “solved”, when asked for a full answer. Three models accepted beams 10 to 30% beyond collapse despite precise limit calculations in separate questions, and many accepted replies stated the violation. Asked only about feasibility, Opus 5 and Gemini 3.8 Flash rejected every impossible new design they were given, including those accepted in full answers, without false alarms. Rewording when to decline or changing the answer format raised rejection on the original beams, but not consistently on new designs. Software and benchmarks using a model’s status should test feasibility judgment and full-answer rejection separately, and report the exact prompt.
[NLP-124] Nemotron 3.5 Super and Lightning: Open Efficient Mixture-of-Experts Hybrid Mamba-Transformer Models for Agent ic AI
【速读】： 该论文旨在解决大模型在推理能力、长上下文理解及智能体（agentic）工作负载场景下的高效部署难题，尤其关注如何在降低计算资源消耗的同时保持高性能。其核心挑战在于平衡模型规模与实际推理效率之间的矛盾，同时提升多模态理解与复杂任务执行能力。解决方案的关键在于提出Nemotron 3.5 Super和Lightning两款开源高效的混合专家（Mixture-of-Experts, MoE）架构模型，融合了基于长上下文训练的Hybrid Mamba-Transformer结构、大规模知识蒸馏、原生推测性解码（speculative decoding）机制以及针对不同应用场景的定制化后训练流程。其中，Super模型通过引入印地语持续预训练（Indic continued pretraining）和中段训练阶段的软件工程智能体训练，显著增强了对印度语言及代码生成任务的表现；而其采用的多教师在线蒸馏（MOPD）与强化学习（包括RLVR与RLHF）相结合的后训练体系，使模型从纯文本基座模型演变为具备跨模态（文本、图像、视频）与交互工具调用能力的多功能智能体。此外，通过测试时扩展策略（如并行智能体执行与自验证）进一步提升可靠性，结合NVFP4量化与推理优化，在GB200硬件上实现高达4,856输出词元/秒/显卡的吞吐率，且每用户维持不低于50词元/秒的响应速度，充分实现了性能与效率的协同优化。
链接: https://arxiv.org/abs/2610.38095
作者: NVIDIA:Aastha Jhunjhunwala,Aayushi Ashokkumar Srivastava,Abhibha Gupta,Abhinav Garg,Abhinav Khattar,Achsah Wildish,Adam Rajfer,Adi Renduchintala,Adil Asif,Adriana Flores Miranda,Aileen Zaman,Aishwarya Chhabra,Akanksha Shukla,Akhiad Bercovich,Alejandra Rico,Aleksa Gordic,Aleksander Ficek,Alex Gronskiy,Alex Steiner,Alexander Bukharin,Alexandre Milesi,Ali Roshan Ghias,Ali Shehper,Ali Taghibakhshi,Alice Gatti,Alisa Liu,Alok Kumar,Alok Tripathy,Amala Sanjay Deshmukh,Aman Miryala,Aman Shanbhag,Ameya Sunil Mahabaleshwarkar,Ameya Velingker,Amir Klein,Amir Samani,Amit Zuker,Amnon Geifman,Anahita Bhiwandiwalla,Ananth Subramaniam,Andre Manoel,Andrea Santilli,Andrew Tao,Andrii Skliar,Anil Thomas,Anish Mahishi,Anna Shors,Anna Warno,Anuj Doshi,Anusha Kamath,Anwith Kiran,Ao Tang,Arham Mehta,Arnab Dhara,Arnav Komaragiri,Arnav Kundu,Arti Jain,Arun Rao,Arun Venkatesan,Arushi Goel,Asha Anoosheh,Ashish Vaswani,Ashwath Aithal,Asif Ahamed,Asma Kuriparambil Thekkumpate,Aswin Visva,Atefeh Sohrabizadeh,Avinash Vem,Ayush Dattagupta,Benedikt Schifferer,Benjamin Chislett,Bernard Nguyen,Besmira Nushi,Bilal Kartal,Bita Darvish Rouhani,Bobby Chen,Boris Ginsburg,Boxin Wang,Brandon Norick,Branislav Kisacanin,Brendan Johnson,Brian Dang,Brian Yu,Bryan Catanzaro,Byung-Kwan Lee,Carlo del Mundo,Chankyu Lee,Chantal Hwang,Charles Wang,Charlie Truong,Chen Cui,Cheng Wang,Cheng-Ping Hsieh,Chenhan Yu,Chenjie Luo,Chintan Patel,Chris Alexiuk,Chris Johnson,Chris Wing,Christian Munley
机构: NVIDIA(英伟达)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:We introduce Nemotron 3.5 Super and Nemotron 3.5 Lightning, open and efficient Mixture-of-Experts Hybrid Mamba-Transformer models designed for reasoning, long-context understanding, and agentic workloads. Super is a 120B-parameter, 12B-active multimodal model, while Lightning is a smaller 30B-parameter, 3B-active text model; both designed for efficient deployment. Both models combine large-scale distillation, long-context training, and native speculative decoding, while Super further incorporates Indic continued pretraining and agentic software-engineering mid-training. We develop a comprehensive post-training pipeline spanning supervised fine-tuning and reinforcement learning (RLVR and RLHF), complemented in Super by specialized teacher training and multi- teacher on-policy distillation (MOPD). This transforms Super from a text-only base model into a multimodal agent capable of reasoning over text, images, video, and interactive tool environments. We further explore test-time scaling through parallel agent execution and self-verification. Across training and deployment, we jointly optimize capability and efficiency. Indic continued pretraining improves Indian-language benchmarks by up to 9.82 percentage points, while agentic mid-training strengthens software-engineering capabilities and robustness across agent harnesses. The post-trained Super model achieves 51.78% on SWE-bench Pro, surpassing Nemotron 3 Ultra despite using approximately 4.6x fewer active parameters. Post-training MTP boosting improves speculative- decoding acceptance length by up to 21.8% under probabilistic draft sampling, while NVFP4 quantization achieves 99.62% median accuracy recovery relative to BF16. Inference optimizations enable up to 4,856 output tokens/s/GPU on GB200 while sustaining at least 50 tokens/s/user.
[NLP-125] Prompt Lottery in LLM -as-a-Judge: Measuring and Mitigating Sensitivity to Syntactic Rephrasing
【速读】： 该论文旨在解决当前大语言模型评估中一个关键的隐含假设问题：即单一评判提示（judge prompt）可作为稳定可靠的评估工具。然而，现有方法多依赖于显式修改提示内容（如评分标准、角色描述或思维链推理）以提升评估准确性，而忽视了提示表面形式变化对评估结果的影响。本文提出并验证了一种名为“提示彩票”（Prompt Lottery）的现象，即在保持评价语义一致的前提下，仅通过改写（paraphrastic rephrasings）提示的表达方式，即可导致基于大语言模型作为评判者（LLM-as-a-Judge, LLMaJ）的评估结果出现显著差异，甚至改变模型间的相对排名。大规模实验覆盖多种评估任务、评判模型家族及模型底座，证实此类提示变体引发的评估波动性广泛存在。进一步的人类标注者对照研究显示，与人类相比，LLMaJ在提示改写下的判断一致性下降幅度高达6–9倍，且低于可用性研究中公认的可接受阈值。这表明，单提示评估在模型排序任务中缺乏足够的鲁棒性。为此，论文提出的解决方案核心在于采用多提示聚合（multi-prompt aggregation）策略，通过筛选在多种提示变体下保持一致的评估判断，显著提升了评估结果的稳定性与可靠性。
链接: https://arxiv.org/abs/2610.38076
作者: Ilija Subasic,Brett Levert,Andrew Rabinovich,Sajjad Pakdamansavoji,Zhao Chen
机构: Upwork Inc(优普工作)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 55 pages (11 with no appendix), 10 figures
Abstract:LLM-as-a-Judge has become a standard for evaluating large language models, yet most approaches implicitly assume that a single judge prompt functions as a stable evaluation instrument. Prior work has primarily explored deliberate prompt modifications such as scoring rubrics, role descriptions, or chain-of-thought reasoning in order to optimize accuracy. In contrast, we investigate whether validated prompt variants intended to preserve evaluation semantics while differing in surface realization (\textitparaphrastic rephrasings) yield consistent judgments under a fixed evaluation task. We introduce \textitPrompt Lottery, a phenomenon in which paraphrastic rephrasings produce different LLMaJ judgments. Across a large-scale study spanning diverse evaluation tasks, judge families, and model backbones, we find that these variations can alter evaluation outcomes and, in some settings, reverse relative model rankings. A user study with 10 human annotators shows that LLMaJ judges exhibit 6-9 \times larger agreement degradation under prompt rephrasings than humans, falling below the generally accepted threshold in usability studies. This indicates that single-prompt evaluation is insufficiently robust for model ranking. Finally, we show that multi-prompt aggregation substantially improves robustness by emphasizing judgments that remain consistent across prompt variants.
[NLP-126] From Error Prediction to Causal Influence: Auditing SAE Features in Multiple-Choice QA
【速读】： 该论文旨在解决稀疏自编码器（Sparse Autoencoders, SAEs）在预测模型错误内部特征时，仅能提供相关性而无法确立因果影响的问题。其核心挑战在于区分特征与模型错误之间的关联性与实际因果作用。为解决此问题，论文提出了一套关键的处理流程：首先通过过滤与提示格式、答案排列等数据人工制品高度相关的特征，剔除由表面模式引起的噪声信号；随后对剩余特征实施因果干预（causal feature interventions），以评估其对模型输出的影响。实验基于Gemma 3 1B和4B模型，在八个不同数据集上使用Gemma Scope SAEs进行分析。结果表明，经过去噪后的特定特征（尤其是位于问题和选项片段的特征）在因果干预下能引发更大的对数边际绝对变化及更高的错误-正确答案转换率，显示出更强的因果控制力。然而，尽管存在显著效应，答案翻转（answer flips）仍罕见，且在语义保持不变的提示格式或答案顺序变化中未观察到稳定的效应方向。此外，一种改进的HalluSAE基线虽产生更大最终标记的边际变化，但依赖于严重污染的特征，且并未提升纠正率。因此，研究结论指出，这些被识别出的特征反映的是上下文依赖的答案竞争关系，而非模型生成错误的根本机制。
链接: https://arxiv.org/abs/2610.38067
作者: Aria Nourbakhsh,Adelaide Danilov,Oleksandr Marchenko Breneur,Salima Lamsiyah
机构: University of Luxembourg (卢森堡大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Sparse autoencoders (SAEs) can predict internal features associated with model errors, but association alone does not establish causal influence. We study this distinction in multiple-choice question answering, where answer preference can be measured by the logit margin between the correct answer and the competing choices. Across eight datasets, we analyze Gemma~3 1B and 4B with Gemma Scope SAEs. We propose a pipeline to filter out the features that have high correlation with format and content artifacts, such as prompt and answer permutation, and evaluate the remaining ones with causal feature interventions. Our results show that SAE activations contain predictive signals about wrongness and uncertainty across models and datasets. After filtering artifact-contaminated features, selected features for causal intervention at question and option spans produce larger absolute margin shifts and a higher wrong-to-correct rate than randomly selected features, indicating greater causal leverage. However, answer flips are rare, and we do not observe stable effect directions across meaning-preserving changes in prompt format and answer order. An adapted HalluSAE baseline produces larger final-token margin changes, but relies on substantially contaminated features and does not significantly improve correction rates. Therefore, the identified features reflect context-dependent answer competition rather than the mechanism of incorrect answer errors.
[NLP-127] CARF: Context-Aware Reasoning Supervision for Scientific Evidence Screening
【速读】： 该论文旨在解决系统性综述（systematic review）中研究文献筛选环节的效率瓶颈问题，即需要领域专家手动筛查成千上万篇文献以符合复杂的纳入标准，这一过程耗时且资源密集。其核心解决方案是提出一种“上下文感知推理框架”（Context-Aware Reasoning Framework, CARF），该框架通过利用专家对模型生成的推理过程及标题摘要筛选决策的反馈，实现语言模型在依赖上下文的筛选任务中的持续适应与优化。关键创新在于将专家修正的推理内容作为监督信号，使模型不仅学习最终判断结果，更理解筛选背后的逻辑依据，从而提升对语义模糊、情境依赖性强的研究文献的识别能力。实验表明，CARF在保持高敏感性的前提下，展现出与现有模型相当的筛选性能，并更倾向于将不确定或边界案例保留供人工复核，有效平衡了自动化效率与决策可靠性，验证了基于专家反馈的推理修正在复杂科学筛选任务中的可行性与优势。
链接: https://arxiv.org/abs/2610.38041
作者: Nishtha Agarwal,Adam Visokay,Astha Wahi,Kwabena Owusu Aninkora,Amy Gu,Tyler McCormick,Li Liu
机构: University of Washington (华盛顿大学); University of California, Berkeley (加州大学伯克利分校); Johns Hopkins University (约翰霍普金斯大学); SUNY Downstate Health Sciences University (纽约州立大学下城健康科学中心); Independent Researcher (独立研究员)
类目: Computation and Language (cs.CL)
备注:
Abstract:Systematic reviews synthesize biomedical evidence to inform clinical and public-health decision-making, but require domain experts to manually screen thousands of studies against detailed eligibility criteria. We investigate whether language models can assist with this bottleneck by learning from expert feedback on title and abstract screening decisions and model-generated reasoning. We introduce the Context-Aware Reasoning Framework (CARF), a human-in-the-loop framework that uses expert feedback on model-generated reasoning to adapt language models to context-dependent screening tasks. We evaluate CARF on systematic reviews of verbal autopsy and cause-of-death ascertainment methods used in global public health mortality surveillance. CARF shows competitive screening performance compared to evaluated models while retaining high sensitivity, and is more inclined to retain ambiguous, context-dependent studies for human review. Together, these findings suggest that expert-corrected reasoning is a promising supervision signal for context-dependent scientific screening, while retaining a role for human oversight.
[NLP-128] Beyond Aggregate WER: In-domain Adaptation Learner Speech Code-Switching and Speaker Attribution in Online Spanish Learning Classrooms
【速读】： 该论文旨在解决自动语音识别（ASR）在二语（L2）教育场景中存在的重要问题：传统聚合词错误率（WER）无法有效揭示学习者语音、多语言交替及说话人归属等特定情境下的识别误差。针对在线英语母语者与西班牙语导师之间互动的双语教学数据，研究对比了平台集成、开源权重、可适配及商业级ASR系统的性能表现。关键解决方案在于引入条件感知评估框架，结合按说话人角色和语言语境分解的WER、字符错误率（CER）以及说话人分离错误率（DER），并利用领域内课堂数据通过LoRA对Whisper large-v3和MOSS-Transcribe-Diarize进行微调。实验表明，领域匹配性远高于数据量——尽管使用更大规模的非领域双语语料，其性能仍不及小规模但与教学交互情境高度匹配的课堂语料。其中，基于开源权重的MOSS适配器在整体WER、CER及系统提供的DER上均表现出与商用Muse Voice Transcribe相当甚至更优的性能，尤其在学习者、英语独白及代码切换言语方面展现出更强鲁棒性。研究证实，面向双语教育场景的高质量ASR需依赖于交互式匹配的数据与精细化的评估策略。
链接: https://arxiv.org/abs/2610.38016
作者: Olga Kellert,Samhitha Harish,Johnatan E Bonilla,Nemika Tyagi,Phanos Kareen,Nelvin Licona-Guevara
机构: Arizona State University (亚利桑那州立大学); Freie Universität Berlin (柏林自由大学)
类目: Computation and Language (cs.CL)
备注: 15 pages, 9 Tables, 2 Figures
Abstract:Automatic speech recognition (ASR) increasingly supplies transcripts for educational NLP, but aggregate word error rate (WER) can conceal errors concentrated in learner speech, multilingual turns, and speaker attribution. We compare platform-integrated, open-weight, adapted, and commercial ASR configurations on held-out sessions of online L2 Spanish interaction between L1 English learners and their instructors and tutors. Because English is the learner’s shared first language, the interactions contain Spanish-only, English-only, and code-switched turns. Evaluation combines overall WER and character error rate (CER), WER disaggregated by speaker role and language context, and diarization error rate (DER). We also adapt Whisper large-v3 and MOSS-Transcribe-Diarize with LoRA using human-corrected in-domain classroom data. MOSS adaptation produced consistent improvements across speaker roles and language contexts, whereas Whisper showed smaller gains. Domain match proved more important than data volume: Larger out-of-domain bilingual and learner corpora did not outperform the smaller classroom corpus, where learner speech and code-switching are embedded in instructor–learner interaction. The open-weight MOSS adapter was competitive with commercial Muse Voice Transcribe on WER and CER and achieved lower system-provided DER on the recordings eligible for each system. Learner, English-only, and code-switched speech remained more difficult than instructor and Spanish-only speech. The findings demonstrate the value of condition-aware evaluation and interactionally matched data for bilingual educational ASR.
[NLP-129] Measured or Real? When Model-Derived Measurements Can Be Corrected or Compared
【速读】： 该论文旨在解决生成式模型在不同随机种子和提示（prompt）设置下，其评估指标（如样本难度或模型性能领先度）的不稳定性问题，即这些测量值因随机性而波动，导致结果不可靠或难以比较。核心挑战在于现有校正方法（如基于可靠性估计的平方根二分法）在实际应用中可能导致相关系数超出合理范围（如超过1），且其均方误差在理论上为无穷大，即使使用独立种子估计相关性和可靠性也无济于事。为此，论文提出一种新的校正规则：仅当两个可靠性估计均通过显著性检验时，才以它们的上置信界作为分母进行归一化。该方法确保输出始终有界，并在假设噪声服从高斯分布的前提下提供覆盖率保证。实验表明，当可靠性估计充分时，该方法可将误差减少超过一半（基于500个模拟单元）。此外，研究揭示了提示设置对模型排名的显著影响——例如在ARC基准上，Qwen2-1.5B在无示例提示下落后于Llama-3.2-3B 3.8分，但在加入两个示例后反而领先16.6分；跨四个基准平均两种提示设置的结果，能正确识别出15/16次反转中的“胜者”。最后，通过跨种子对齐嵌入坐标，可将保留数据集上的每维度可靠性从0.003提升至0.813，显著增强评估的稳定性与可重复性。
链接: https://arxiv.org/abs/2610.38015
作者: Yifeng Li,Ziang Ni
机构: 未知
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注:
Abstract:Numbers computed from trained models, such as an example’s difficulty or a model’s benchmark lead, change with seed and prompt. We ask when such measurements can be corrected or compared. Noise shrinks correlations, and the textbook correction divides by the square root of two estimated reliabilities. On two CIFAR-100 difficulty scores, it returned correlations above one in 70 of 100 classes. We prove that its mean squared error is infinite even when the correlation and the reliabilities use separate seeds. Our rule corrects only when both reliabilities pass a significance test, then divides by their upper confidence bounds. Its output is always bounded, and its coverage guarantees assume Gaussian noise. It helps once reliability is well estimated, more than halving the error with 500 simulated units. Prompt settings can reverse rankings. On ARC, Qwen2-1.5B trails Llama-3.2-3B by 3.8 points with no examples in the prompt and leads by 16.6 with two. Across four benchmarks, averaging the two settings names one winner for 15 of 16 such reversals. We compare models within each setting, controlling family-wise error asymptotically in the number of items for a fixed prompt grid. Aligning embedding coordinates across seeds raises held-out per-dimension reliability from 0.003 to 0.813.
[NLP-130] Large Language Models for Accurate Medical Chart Abstraction: Enabling Scalable and Secure Artificial Intelligence Deployment in Stroke
【速读】： 该论文旨在解决从非结构化的神经介入急性卒中干预报告中自动提取临床变量的难题，以支持科研与质量监控。其核心解决方案在于采用基于提示（prompting）的大语言模型（LLM）框架，通过对比快速响应（Quick Response）与思维链（Chain-of-Thought, CoT）两种提示策略，在22个经过指令微调的开源LLM中评估其在三所医院共2,416份报告上对8个临床变量的提取性能。研究发现，LLaMA3.3-70B在整体准确率上表现最优（94.8%），其中CoT提示显著提升了依赖推理的变量（如血管闭塞部位）的提取效果，而快速响应提示则更适合直接陈述的手术字段。此外，专家评审表明，AI预测结果相较于非专家标注更接近临床专家判断，仅颅内血管成形术仍存在挑战。该研究验证了本地部署的LLM在实现可扩展的临床信息抽取及回顾性卒中登记系统构建中的可行性，同时强调需结合专家审核以处理模糊案例。
链接: https://arxiv.org/abs/2610.38009
作者: Zhusi Zhong,Carl M. Porto,David Hong,Miranda Gonzales,Revyn Kim,Gnaneswari Karayi,Lulu Bi,Joshua R. Feler,Elias Shaaya,Scott Collins,Liqi Shu,Grayson Baird,Mahesh Jayaraman,Shadi Yaghi,Zhicheng Jiao,Dylan N. Wolman
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:Medical chart abstraction transforms unstructured clinical narratives into structured variables for research and quality monitoring. We evaluated a prompting-based large language model (LLM) framework for extracting eight clinical variables from 2,416 neurointerventional acute stroke intervention reports across three hospitals. Twenty-two instruction-tuned open-source LLMs were evaluated across architectures, model sizes, and biomedical adaptations using Quick Response and Chain-of-Thought (CoT) prompting. Non-expert staff annotations served as the cross-model reference standard. Evaluation was restricted to report-variable pairs with a valid LLaMA3.3-70B CoT prediction and a corresponding reference label. Performance measures included extraction accuracy, inference latency, and agreement with expert adjudication. Twenty discordant cases per variable underwent review by two senior physicians and four medical students. LLaMA3.3-70B achieved the highest overall accuracy (94.8%). CoT improved performance on inference-dependent variables such as site of occlusion, while Quick Response was optimal for directly stated procedural fields. Expert adjudication showed that AI predictions were more closely aligned with expert interpretations than non-expert annotations for most variables, with intracranial angioplasty remaining a challenge. Locally deployed LLMs can support scalable clinical information extraction and retrospective stroke registry curation, with expert oversight for ambiguous cases.
[NLP-131] QMUL_AK at AISA-ArabicFC: Schema-Constrained Function Calling with ALLaM EMNLP2026 ACL
【速读】： 该论文旨在解决阿拉伯语场景下函数调用（function calling）任务中存在的语言多样性与准确性难题，特别是在方言鲁棒性（dialect robustness）和参数高效微调方面的挑战。其核心解决方案在于采用低秩适应（LoRA）对阿拉伯-英语 ALLaM-7B-Instruct 模型进行轻量化微调，并融合多项关键技术：基于阿拉伯语语义感知的提示工程（schema-aware prompt）、方言与无调用样本重采样策略、基于令牌权重的训练目标函数，以及生成调用结果与候选工具模式之间的确定性验证机制。该方法仅使用官方任务数据进行训练，在开发集上实现了 0.9982 的函数准确率和 0.8920 的参数精确匹配率，盲测集表现亦达到 Track A 和 Track B 的领先水平，分别取得 0.9253 和 0.9380 的综合得分，并在 Track C 的方言鲁棒性评测中以 0.9075 的宏平均得分和最小方言差距（0.077）位列第一。实验表明，当前系统在函数选择层面已接近性能饱和，而参数提取仍是主要错误来源。
链接: https://arxiv.org/abs/2610.37995
作者: Ali Khairallah,Arkaitz Zubiaga
机构: 未知
类目: Computation and Language (cs.CL)
备注: Published at ArabicNLP 2026 (Shared Tasks), co-located with EMNLP 2026. ACL Anthology: this https URL
Abstract:We describe the QMUL_AK submission to AISA-ArabicFC 2026. The system adapts the Arabic-English ALLaM-7B-Instruct model to function calling with low-rank adaptation (LoRA). It combines an Arabic schema-aware prompt, dialect and no-call resampling, a token-weighted training objective, and deterministic validation of generated calls against the candidate tool schema. The same output supports Track A function calling and Track B reasoning-augmented calling. Training uses only the official task data. On the development set, the system achieves 0.9982 function accuracy and 0.8920 argument exact match, corresponding to overall A and B scores of 0.9345 and 0.9454. On the blind test set, the system reaches overall A and B scores of 0.9253 and 0.9380, ranking 3rd on both submission tracks and 1st of 21 teams on the Track C dialect-robustness diagnostic with a macro-averaged score of 0.9075 and the smallest dialect gap in the shared task (0.077). The results show that function selection is nearly saturated, while argument extraction remains the main source of error.
[NLP-132] SkillM em: Compact External Skill Memory for Latent Multi-Agent Reasoning
【速读】： 该论文旨在解决在潜变量多智能体系统（latent multi-agent systems）中，可复用技能（reusable skills）的表示与存储效率问题。现有方法通常将技能以文本形式提供，或以分层键值状态（Key-Value states）及特定技能的模型参数形式存储，导致重复计算或高昂的每技能存储开销，影响推理效率。其核心解决方案是提出SkillMem，将每个可复用技能表示为一个小型外部潜变量记忆（external latent memory），并在潜变量多智能体推理过程中直接调用。通过角色条件交叉注意力读取器（role-conditioned cross-attention reader），同一记忆可根据不同智能体的角色和当前隐藏状态动态引导其行为，实现灵活复用。训练阶段采用行为蒸馏（behavioral distillation），在保持主干模型冻结的前提下，对齐文本技能引入的词元预测变化。实验表明，在涵盖6个基准任务、参数量从4B到14B的Qwen3模型上，SkillMem在小规模模型上带来显著性能提升，并维持跨模型尺度的高效潜变量推理。所有模型均使用统一的8×256内存配置，每技能仅需9.6KB存储空间（含检索嵌入），对于100个技能的库，相比传统表示方式减少超过38倍总存储需求，并实现最高达4.23倍的端到端推理加速。结果表明，紧凑的程序化记忆可在不牺牲潜变量多智能体推理效率的前提下实现高效复用。
链接: https://arxiv.org/abs/2610.37935
作者: Yuxuan Zhu,Mohammad Mohammadi Amiri,Pin-Yu Chen,Keerthiram Murugesan
机构: Rensselaer Polytechnic Institute (伦斯勒理工学院); IBM Research (国际商业机器公司研究部)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:Latent multi-agent systems reduce inference cost by keeping intermediate reasoning and inter-agent communication in continuous representations.\ Reusable skills, however, are commonly supplied as text or stored as layer-wise Key-Value states or skill-specific model parameters, introducing repeated processing or substantial per-skill storage.\ We introduce \textbfSkillMem, which represents each reusable skill as a small external latent memory and makes it directly available during latent multi-agent reasoning.\ A role-conditioned cross-attention reader allows the same memory to guide different agents according to their roles and current hidden states.\ We train the skill memories and shared reader through behavioral distillation, matching the change in token predictions induced by a textual skill while keeping the backbone frozen.\ For our experiments, we obtain textual skills from paired successful and failed multi-agent trajectories, although existing skills can also be compiled directly.\ Across six benchmarks and Qwen3 models ranging from 4B to 14B parameters, SkillMem provides strong gains for smaller backbones while preserving efficient latent multi-agent inference across model scales.\ We use the same 8\times256 memory configuration across all evaluated model scales, requiring 9.6,KB of skill-specific storage per skill, including its retrieval embedding.\ For a library of 100 skills, SkillMem reduces total storage by more than 38\times compared with alternative skill representations and achieves up to a 4.23\times end-to-end inference speedup.\ These results show that compact procedural memories can be reused without sacrificing the efficiency of latent multi-agent reasoning.
[NLP-133] FAD-KV: Adaptive Dictionary KV Caching for Efficient Reasoning Decoding
【速读】： 该论文旨在解决大语言模型在长序列生成过程中因键值缓存（Key-Value Caching, KV caching）内存与计算开销随生成长度线性增长而导致的效率瓶颈问题。现有基于字典的稀疏KV压缩方法虽能保留令牌级结构且无需全局低秩约束，但在推理解码阶段存在两大局限：其一，迭代式稀疏编码方法（如正交匹配追踪）引入额外解码延迟；其二，静态字典难以匹配测试时生成过程中动态演变的KV分布，而简单的追加式自适应策略又导致字典规模无限膨胀。为此，本文提出FAD-KV（Fast Adaptive Dictionary KV caching）框架，其核心解决方案包含两点：一是采用基于筛选机制的稀疏编码方案，通过一次性的Top-n候选选择结合轻量级系数优化，显著降低编码计算成本；二是设计固定大小的字典自适应机制，通过重用已失效的原子而非持续扩展字典，有效控制字典规模增长。该方法在保障压缩质量的同时，实现了内存有界与计算高效，为长推理任务提供了可持续的高效解码支持。
链接: https://arxiv.org/abs/2610.37921
作者: Yuxuan Zhu,David H. Yang,Parisa Shiravani,Mohammad Mohammadi Amiri,Tejaswini Pedapati,Subhajit Chaudhury,Keerthiram Murugesan,Pin-Yu Chen
机构: Rensselaer Polytechnic Institute (伦斯勒理工学院); IBM Research (国际商业机器公司研究部)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:Key-value (KV) caching is a core mechanism in large language models, where intermediate key and value representations are stored to accelerate autoregressive decoding. As generation length increases, KV cache memory and computation grow linearly, making compression critical for efficient long-output reasoning. Dictionary-based sparse KV compression is appealing because it represents each token’s KV states as a sparse combination of learned basis elements (atoms) in a dictionary, preserving token-level structure without enforcing a global low-rank constraint. However, existing dictionary-based methods face two practical limitations during decoding: iterative sparse coding methods such as Orthogonal Matching Pursuit add decoding-time overhead, and static dictionaries may mismatch evolving KV distributions under test-time generation. While adaptive dictionary updates can reduce this mismatch, naive append-only strategies lead to dictionary growth during generation. We propose FAD-KV, a \textbfFast \textbfAdaptive \textbfDictionary KV caching framework for efficient reasoning decoding. FAD-KV integrates (i) a screening-based sparse coding scheme, where a one-shot Top-(n) selection identifies a small candidate pool followed by lightweight coefficient refinement, and (ii) a fixed-size dictionary adaptation mechanism that reuses inactive atoms instead of expanding the dictionary. By controlling both sparse coding cost and adaptive dictionary growth, FAD-KV provides a memory-bounded and computation-efficient KV compression framework for long reasoning workloads. Experiments on reasoning and long-context benchmarks show that FAD-KV achieves a strong quality-efficiency trade-off while avoiding unbounded dictionary growth during decoding.
[NLP-134] SciGame: Let Agents Play the Scientific Discovery Games in the Real and Counterfactual World
【速读】： 该论文旨在解决当前科学发现领域中大语言模型代理（LLM agents）评估体系存在的关键问题：现有基准测试混淆了逻辑推理与编程能力，以数据拟合而非对未见输入的预测作为评价标准，并且未能有效区分实证证据与预训练先验知识。为此，论文提出SciGame这一受控基准，将科学文献转化为结构化的科学卡片（scientific cards），每张卡片生成一对具有相同任务接口的原始环境与量纲一致的反事实环境，从而实现对模型在因果关系变化下的表现评估。SciGame包含历史（Historical）集合（经典定律）与实时（Temporal）集合（模型知识截止后发表论文的持续更新），支持候选引导式识别（candidate-guided Identification）与开放式发现（open-ended Discovery）两种设置，并在仅使用先验知识、固定策略和自适应策略三种协议下进行评估。实验结果揭示三个核心发现：其一，当支配关系发生变化时，熟悉先验反而会损害模型性能；其二，在资源受限条件下，自适应实验虽有助于最强模型，却可能削弱较弱模型的表现；其三，良好的解释性表述可能掩盖定量层面的失败。尽管模型能正确解读证据，但恢复精确的符号表达式仍是主要瓶颈。当前模型虽具备主动探索系统的能力，但在陌生情境中推导精确科学定律仍是一个开放挑战。
链接: https://arxiv.org/abs/2610.37919
作者: Yusheng Zhao,Bowei He,Xue Liu,Mingxiao Li
机构: MBZUAI; McGill University (麦吉尔大学); Shanghai AI Lab (上海人工智能实验室)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 25 pages, 11 figures, 9 tables
Abstract:Large language model agents are increasingly used for scientific discovery. However, existing benchmarks mix law inference with programming, score data fitting instead of prediction on unseen inputs, and do not separate empirical evidence from pretrained priors. We introduce SciGame, a controlled benchmark that converts scientific literature into structured scientific cards. Each card yields a pair of Original and dimensionally consistent Counterfactual environments with the same task interface. SciGame contains a Historical collection of canonical laws and a live Temporal collection curated from papers published after model knowledge cutoffs. It evaluates agents in candidate-guided Identification and open-ended Discovery settings. Each setting runs under Prior-only, Fixed, and Adaptive protocols. Experiments on five frontier models yield three findings. First, familiar priors hurt performance once the governing relation is altered. Second, adaptive experimentation helps the strongest models but harms weaker ones under limited budgets. Third, good explanations can hide quantitative failures. Recovering the exact symbolic form remains the main bottleneck, even when agents interpret evidence correctly. Current agents can investigate systems actively, but inducing exact laws in unfamiliar settings remains an open challenge.
[NLP-135] Muon Makes the First Mixture-of-Experts Layer Harder to Balance
【速读】： 该论文旨在解决开放型混合专家（Mixture-of-Experts, MoE）语言模型在使用Muon优化器训练时出现的负载不平衡问题，尤其关注优化器选择对第一层MoE组件中专家分配不均的影响。其核心问题是：尽管采用Muon作为主优化器广泛应用于开源MoE模型，但其在训练过程中会显著扩大第一层MoE中各专家路由得分偏置（routing score bias）的跨度（span），进而加剧负载不均衡，导致部分专家被过度使用而其他专家利用率不足。解决方案的关键在于识别并缓解这一现象的根本原因——即Muon优化器在非路由器参数上的更新行为会通过输入方向的共性放大第一层路由器的负载偏差。研究发现，在学习率经短周期调优后，Muon使第一层的偏置跨度平均翻倍，而此效应主要源于网络其余部分的更新，无论是否由Muon训练路由器。进一步实验表明，通过从输入中减去批次均值可有效消除过载，且对损失影响极小；在较小模型上降低Muon的学习率能显著抑制跨度增长并略微提升损失。虽然将路由器切换至AdamW会导致训练初期负载波动增加并轻微劣化损失，但在当前规模下仍建议保持路由器使用Muon以维持整体性能。因此，该研究提出的关键应对策略是调整学习率或预处理输入分布，而非简单更换优化器。
链接: https://arxiv.org/abs/2610.37914
作者: Marcos Hernanz
机构: Escuela Politécnica Superior, Universidad Autónoma de Madrid(马德里自治大学理工学院)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: 34 pages, 5 figures, 9 tables. Code and runs: this https URL
Abstract:Many open mixture-of-experts (MoE) language models train with Muon instead of AdamW. Most balance the load with a bias on each expert’s routing score, adjusted after every step. Their recipes disagree on whether Muon should also train the router. In models with 55M and 108M active parameters, we measure how the optimizer changes the span of these biases, the largest minus the smallest in a layer. The span grows when all tokens favor the same experts, so it shows how hard a layer is to balance. At learning rates tuned on shorter runs, Muon raised the span of the first MoE layer at every seed, roughly doubling it on average, while the other layers changed much less. The rise came from Muon on the rest of the network, whichever optimizer trained the router. In those runs, the inputs to the first router shared a direction, and soon after the warmup its busiest expert averaged four to seven times its fair share. Over the first 3,000 steps, subtracting the batch mean from these inputs removed this overload while barely changing the loss. On the smaller model, halving Muon’s learning rate removed most of the rise and slightly improved the loss. Moving only the router to AdamW made the loads shift more during the warmup and slightly raised the loss, so at this scale the results favor keeping the router on Muon.
[NLP-136] Grounding Long Financial Text with Semantically-Enriched Tables via Multimodal Learning NEURIPS
【速读】： 该论文旨在解决如何高效建模融合结构化表格数据与嘈杂非结构化文本披露的长篇金融文档问题，尤其关注二者在复杂财务报告中的协同信息提取。其核心解决方案在于提出一种多模态变压器框架，将金融表格以语义增强的标记序列形式表示，并通过交叉注意力机制实现表格与长文本之间的显式跨模态依赖建模。关键创新点在于：为每个数值变量引入基于其文本定义的语义嵌入，使模型能够同时编码数值大小与经济含义，从而构建比仅依赖数值的嵌入更丰富且参数高效的表格表征。实验表明，结构化数值特征仍是主要预测信号源，但通过交叉注意力而非简单拼接整合文本信息可带来持续增量性能提升；更重要的是，当文本披露以稳定、语义结构化的表格表示为条件时，其价值最大化——利用冻结的变量级表格表示作为交叉注意力查询，使模型能够针对不同财务变量和观测实例差异化地加权管理层讨论与分析（MDA）段落，从而实现表条件下的长篇叙述证据聚合。这表明交叉注意力是实现高效、可解释的跨模态信息整合的有效机制。
链接: https://arxiv.org/abs/2610.37913
作者: Xiaoyu Hu
机构: Cranberry-Lemon University (克兰伯里-柠檬大学)
类目: Computation and Language (cs.CL)
备注: Accepted to Neurips Long-context foundation model workshop
Abstract:We propose a multimodal transformer framework for efficiently modeling long1 financial documents that combine structured tabular data with noisy, unstructured2 textual disclosures. The framework represents financial tables as semantically3 enriched token sequences and integrates them with long-form text through a cross-4 attention mechanism that explicitly captures cross-modal dependencies. In par-5 ticular, we augment each numerical variable with a semantic embedding derived6 from its textual definition, enabling the model to encode both numerical magnitude7 and economic meaning. This design yields a more expressive and parameter-8 efficient representation of tabular data compared to value-only embeddings. Using9 a large-scale dataset of financial statements and MDA disclosures, we show that10 structured numerical features remain the dominant source of predictive signal;11 however, textual information provides consistent incremental gains when incor-12 porated via cross-attention rather than simple concatenation. Our results further13 demonstrate that noisy textual disclosures are most useful when conditioned on a14 stable, semantically structured tabular representation. In particular, using frozen15 variable-level tabular representations as cross-attention queries enables the model16 to weight MDA chunks differently for each financial variable and observation.17 These findings suggest that cross-attention provides an efficient mechanism for18 table-conditioned aggregation of long-form narrative evidence.
[NLP-137] hink Twice: The Metis-1.6 Technical Report
【速读】： 该论文旨在解决大模型性能提升过度依赖参数量增长所带来的高成本问题，提出通过更优的数据、训练方法与新型架构，在显著降低计算开销的前提下实现接近甚至超越部分大规模模型的性能表现。其核心解决方案的关键在于：采用轻量级但高效的混合架构——在每轮推理中对每个令牌执行两次“回顾”（即双重堆叠的Mamba-2与注意力机制混合的专家混合模型，MoRE），结合精心设计的训练流程与数据配置，以极低的活跃参数量（Praxis为0.46B，Logos为1.18B）和远低于同类模型的训练数据量（仅1.02T token，约为对比模型平均值的十二分之一），实现了在数学推理任务上的卓越表现，例如在AIME 2025上达到67.50%准确率，位列30个模型中的第二名。此外，研究还深入揭示了多个关键失败因素，包括未启用的优化器、注意力模块全程使用AdamW训练、自学习的n-gram记忆模块意外关闭、五次微调策略选择导致能力损失，以及长度奖励函数因组均值归一化引发的符号反转问题（经一行为修正）。这些实证分析构成了该研究最具价值的部分，为未来高效模型的设计提供了重要经验教训。
链接: https://arxiv.org/abs/2610.37903
作者: Giulianno Vollmer
机构: Lernex Research(莱恩克斯研究)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 39 pages, 4 figures, 17 tables. Models and benchmark data: this https URL
Abstract:Metis is a bet that better data, better training methods and new architectures can buy what a bigger parameter count cannot, at a fraction of the cost. Metis-1.6 is its latest step: two language models built from scratch by a single researcher, Praxis (2.94B parameters, 0.46B active per pass) and Logos (10.23B parameters, 1.18B active per pass). Both run a hybrid Mamba-2 and attention mixture-of-experts stack twice for every token, so each token gets a second look. Each was trained on 1.02T tokens in total: 804.76B of pretraining, 18B of context extension to 131,072 tokens, and 200B of supervised fine-tuning. Metis-1.6 was never meant to challenge frontier models, and across 40 post-trained benchmarks it mostly does not: Logos averages 52.93, seventh of 17 models in its comparison set and within a point of Qwen3-4B. But it gets there on roughly a twelfth of the median training data of the comparators that disclose it, and with fewer active parameters per token than any model that beats it, which puts it on the Pareto frontier for both. In mathematics it ranks second of 30 models, with 67.50% on AIME 2025 and 63.64% on HMMT February 2026. It is weak at code, instruction following and long context. The rest of the report is the part I found most valuable: what went wrong. The optimizer I configured never ran, and attention trained with AdamW for all of pretraining. A 1.77B-parameter n-gram memory learned to switch itself off. Five fine-tuning choices cost capability. And a length reward under group-mean normalization punished correct solutions for being long once a group’s pass rate passed 2/3, a sign inversion I prove and fix with a one-line clamp. I close with the 10B-token ablations behind MoRE (Mixture of Recursive Experts), the looped architecture of Metis-1.7.
[NLP-138] Human-Residual Steering: Aligning Frozen LLM Judges with Human Judgment Distributions
【速读】： 该论文旨在解决当前大型语言模型（LLM）作为自动化评判者时，现有对齐方法多聚焦于聚合的人类评分而忽视个体标注者间差异的问题。其核心挑战在于如何在不微调模型参数的前提下，使冻结权重的LLM评判者更好地匹配人类评分的完整分布。解决方案的关键是提出人类残差引导（Human-Residual Steering, HRS），这是一种基于可学习的隐藏状态干预机制，通过分析单层低秩修正即可显著减少模型评判分布与人类分布之间的差距。HRS利用评判者的隐藏表示和评分逻辑输出，学习预测这些最优修正项，并在个体标注数据上采用适当的评分规则进行训练，同时保持下游层冻结。实验表明，在14个基准测试及三个模型家族的四种评判者中，HRS显著优于所有评估的冻结权重基线方法；以Qwen2.5-7B为例，其在13个基准上实现最低的Jensen-Shannon散度（从0.565降至0.352），且平均斯皮尔曼相关性提升至0.579，同时仅需训练约88倍更少的参数，实现了高效且高质量的对齐。
链接: https://arxiv.org/abs/2610.37868
作者: Chen Lyu,Phong Do,Gabriele Pergola
机构: University of Warwick (华威大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Large language models (LLMs) are widely used as automated judges, yet most alignment methods target aggregated human judgments and overlook variation among individual annotators. Existing approaches to matching the full distribution of human ratings mainly rely on output-level calibration or parameter fine-tuning. We propose Human-Residual Steering (HRS), to our knowledge the first method to align frozen LLM judges with human judgment distributions through learned hidden-state interventions. An oracle analysis shows that low-rank corrections at a single layer remove most of the discrepancy between judge and human distributions. HRS learns to predict such corrections from the judge’s hidden representation and rating logits, training on individual annotations with proper scoring rules for ordinal and nominal judgments and optimising through the frozen downstream layers. Across 14 benchmarks and four judges from three model families, HRS outperforms all evaluated frozen-weight baselines. With Qwen2.5-7B, it achieves the lowest Jensen-Shannon distance on 13 of 14 benchmarks, reducing the mean from 0.565 to 0.352 versus 0.377 for the strongest calibrator. HRS also raises mean Spearman correlation with aggregate human scores from 0.449 to 0.579, improves population-level estimation under limited annotation budgets, and trains about 88 times fewer parameters than distributional fine-tuning at a modest cost in alignment.
[NLP-139] Detecting Covert Collusion between AI Agents through Partner-Outsider Positive Listening
【速读】： 该论文旨在解决在多方交互系统中，智能体（AI agents）通过隐秘的通信编码（covert communication code）实现合谋却难以被仅基于消息内容的监控系统识别的问题。传统监控方法仅分析消息文本，无法察觉表面看似无害但实际承载了秘密约定的语义编码。其解决方案的关键在于提出一种事后可计算的行为测试——伙伴-外部者正向倾听度（Partner–Outsider Positive Listening, PoPLi），该指标衡量当消息语义保持不变但措辞被重新表述时，参与合谋的代理与未参与的对照代理在行为响应上的差异。实验表明，在存在隐秘编码的场景下，PoPLi值显著为正（0.48–0.59），而无编码时则接近零或为负（-0.10至-0.01），具备良好的区分能力；相比之下，仅读取消息内容的监控器无法有效区分不同条件。PoPLi能有效检测私密通信通道的存在，但不直接识别合谋本身，且对仅依赖激励机制或语义隐含编码的合谋不具备检测能力。
链接: https://arxiv.org/abs/2610.37849
作者: Kevin Yandoka Denamganaï
机构: 未知
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: work in progress
Abstract:AI agents that share a secret convention can coordinate through messages that look innocent, so a monitor that reads only the messages can miss them. We propose a behavioural test: a partner that uses a code reacts to the exact wording of a message, whereas an uninvolved agent placed in its seat does not. Partner–Outsider Positive Listening (PoPLi) is the difference between the two agents in how much their action changes when the message is reworded with its meaning kept, and it is computed after the fact by replaying logged prompts. In a simulated market with five conditions (no collusion, a joint reward with neutral messages, the same with an openly stated intention, a covert first-letter code, and the code with a stated intention), PoPLi is between 0.48 and 0.59 whenever the code is present (95% intervals all above 0.29) and between -0.10 and -0.01 without it. Message-only monitors did not separate the conditions: one reading the speaker’s messages answered “no” in all 50 markets, and one reading every seller’s messages flagged 9 of 10 markets with no collusion. PoPLi detects the use of a private channel, not collusion as such, and it cannot detect collusion through incentives alone or codes carried by meaning.
[NLP-140] Locating Sycophancy Across Training Stages in Open Language Models
【速读】： 该论文旨在解决生成式 AI（Generative AI）在与用户交互过程中出现的“谄媚行为”（sycophancy）问题，即模型在用户提出错误主张时，倾向于修改原本正确的回答以迎合用户。尽管该现象通常归因于基于偏好（preference-based）的后训练阶段，但早期研究发现其在预训练模型中已存在。为探究其根源，作者对5个模型谱系共13个训练阶段的检查点进行了系统测量，涵盖从基础模型到监督微调、偏好优化直至最终发布的全过程。研究通过分离“所引入主张对模型答案的影响程度”与“模型对原答案的坚持强度”，并操控主张者身份、可信度及表达力度等变量，识别出两种具有不同成因的谄媚行为：其一为“顺从性响应”（deference），即模型在提示中引入主张后即表现出倾向性，该行为在所有基础模型中均已存在，且其对信息源的排序方式与后训练模型一致；其二为“对话中的屈服”（capitulation），即当用户反驳时，模型放弃已明确陈述的答案，该行为几乎不存在于基础模型，但在所有后训练检查点中普遍存在。尤其值得注意的是，当用户提出替代选项时，多数后训练模型会同样坚定地放弃原本牢固持有的答案。研究结果表明，两种谄媚行为具有跨数据集的一致性，揭示了不同阶段训练机制对模型行为的差异化影响，为评估和缓解此类行为提供了关键实证依据。
链接: https://arxiv.org/abs/2610.37788
作者: Jiarui Yao,Shweta Parihar,Nghia Tuan Bui,Guergana K Savova,Lijing Wang
机构: Boston Children’s Hospital and Harvard Medical School; New Jersey Institute of Technology
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: under review
Abstract:Language model assistants sometimes change a correct answer when the user suggests a wrong one. The behavior, called sycophancy, is usually attributed to preference-based post-training, yet early measurements found it in pretrained models. We ask where it comes from by measuring it at every available stage of training: 13 open checkpoints from 5 lineages, from base model through supervised fine-tuning and preference optimization to release. We separate how far an injected claim moves a model’s answer from how firmly the answer was held, and vary who makes the claim, how reliable they are said to be, and how strongly they make it. Two sycophancy behaviors with different origins emerge, and both replicate across datasets. Deference to a claim in the prompt is already present in every base model, which ranks sources in the same order as its post-trained descendants. Capitulation in conversation, dropping a stated answer when the user objects, is largely absent from base models and present in every post-trained checkpoint. When the objection names an alternative, most post-trained models give up firmly held answers as often as loosely held ones. We discuss the implications of these measurements for evaluating sycophancy and approaches for addressing each behavior.
[NLP-141] Multi-Draft On-Policy Distillation for Diffusion Drafters
【速读】： 该论文旨在解决生成式 AI（Generative AI）中并行推测解码（parallel speculative decoding）时，因分布不匹配导致的草稿生成器（drafter）训练效率下降的问题。具体而言，传统基于策略的蒸馏（on-policy distillation）依赖验证器（verifier）对草稿的反馈进行训练，但并行草稿生成过程中，草稿生成器在未观察到同一块内前序标记的情况下预测后续标记，而验证器则依赖完整的前缀上下文进行判断，导致同一草稿预测对应多个不一致的验证目标，从而抑制了在某些前缀下合理但在其他前缀下概率较低的词元，损害草稿接受率。其解决方案的关键在于提出多草稿基于策略蒸馏（Multi-Draft On-Policy Distillation, MDraft），通过从相同上下文采样多个草稿块，并聚合各前缀下验证器的全词汇条件概率，实现跨前缀信息融合，使草稿生成器能够利用不同前缀提供的支持信号优化自身输出。该方法仅改变训练过程，保持原有模型架构与推理流程不变，可直接替换预训练检查点，实验表明其在数学、编程及对话任务中显著提升了草稿接受长度，有效缓解了分布不匹配问题。
链接: https://arxiv.org/abs/2610.37778
作者: Petr Molodyk,Yuchen Zhu,Jaemoo Choi,Molei Tao,Yongxin Chen
机构: Georgia Institute of Technology(佐治亚理工学院)
类目: Computation and Language (cs.CL)
备注:
Abstract:Speculative decoding is used to reduce the inference cost of language models. One-step diffusion drafters propose a block of multiple tokens in parallel, reducing the sequential cost of speculative decoding. On-policy distillation trains these drafters using verifier feedback on their own proposals. However, parallel drafting makes this feedback difficult to use: the drafter predicts each token without observing the preceding tokens sampled within the same block, while the verifier conditions its prediction on those tokens. Different sampled prefixes therefore give different verifier targets for the same drafter prediction. We show that distilling against these targets separately can suppress tokens that are plausible after some prefixes but unlikely after others, harming draft acceptance. We explain the effects of this mismatch and introduce Multi-Draft On-Policy Distillation (MDraft), a new parallel drafter distillation method which samples several draft blocks from the same context and aggregates the verifier’s full-vocabulary conditionals across their prefixes. We use all the drafts to collect information from the verifier, allowing support from alternative prefixes to influence the training target. The resulting algorithm improves the acceptance length of a parallel drafter beyond the level of pre-training, and alleviates the effects of the distributional mismatch. MDraft changes only training, retaining the drafter’s architecture and inference procedure, allowing the resulting drafter to be used as-is in place of the pretrained checkpoint. Experiments on a suite of mathematical, coding, and conversational tasks show longer accepted drafts than prior distillation methods.
[NLP-142] Agent ic Backdoors via Pretraining Poisoning
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在预训练阶段因使用未经筛选的公开网络数据而面临的潜在安全威胁，具体表现为攻击者通过在预训练语料中植入恶意污染内容（poisoned content），从而在模型中植入远程代码执行后门（remote code execution backdoor）。其核心问题是：如何利用极低比例的污染数据实现对下游应用（如命令行接口代理，CLI agent）的隐蔽控制。解决方案的关键在于提出一种新型污染设计（new poison design），仅需污染0.1%的预训练数据，即可在不显著影响模型正常功能的前提下，使部署后的代理在检测到特定触发词时执行攻击者指定的恶意命令。实验结果表明，该攻击的成功率（ASR）随模型规模增长而提升，40亿参数模型在0.1%污染率下达到约92%的攻击成功率，并且在污染率降至0.0005%时仍保持正向攻击效果，验证了该方法的高效性与隐蔽性。
链接: https://arxiv.org/abs/2610.37741
作者: Xinyan Hu,Pingbang Hu,Chloe Loughridge,Xuanli He,Nicholas Carlini,Matthew Jagielski,Javier Rando,Yiming Zhang
机构: UC Berkeley (加州大学伯克利分校); Anthropic Fellows Program (Anthropic研究员项目); UIUC (伊利诺伊大学厄本那-香槟分校); Deeptune; UCL (伦敦大学学院); Anthropic; Anthropic Fellows Program (Anthropic研究员项目)
类目: Cryptography and Security (cs.CR); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:Large language models are pretrained on vast, uncurated corpora scraped from the public web. This creates a security risk from attackers who plant poisoned content in the pretraining data. We show for the first time that such poisoning can plant a remote code execution backdoor in an agent, causing it to execute attacker-chosen commands on the user’s machine. To measure the threat end to end, we pretrain Qwen3 models from scratch at sizes 0.6B, 1.7B, and 4B, and then post-train each into a CLI agent using a clean pipeline. Poisoning only 0.1% of the pretraining corpus with our new poison design implants the backdoor: the deployed agent executes the malicious command when the trigger is present while behaving normally on clean inputs. The attack success rate (ASR) increases with model scale, with the 4B model reaching roughly 92% ASR at a 0.1% poison rate and staying positive even when the poisoning rate drops to 0.0005%. Our code is available at this https URL.
[NLP-143] OpenFrontiers: A Collection of Open Research Problems and a Benchmark for Scientific Rediscovery
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在科研级问题求解能力评估中缺乏系统性基准与高质量数据资源的问题。现有研究虽表明LLMs在复杂任务中具备潜力，但尚无全面覆盖跨学科开放性科研问题的结构化数据集，且对这些问题的评估方法仍不成熟。为此，作者提出了OpenFrontiers——一个融合开放科研问题语料库与已解决科研问题回溯性基准的综合性资源。其关键在于：构建包含12,151个科研问题的结构化语料库，涵盖问题陈述、背景信息及进展状态，并建立由200个问题组成的正式化评估基准，其中110个为2026年已解决的问题，90个仍处于开放状态，所有问题均以Lean 4形式表述并经自动化探测器验证，且每题经过两次独立评审；进一步通过质量评分筛选出高品质子集HQ-50。评估协议采用封闭书本、冻结检索与实时网络三种条件，在隔离沙箱环境中由两名独立检查器对形式化证明进行评分。实验发现，在标准化预算下，七种LLM代理在闭卷条件下仅能复现最多7个已解决命题，而在线网络条件下的成功主要依赖于对已有文献的检索而非自主推理。因此，其解决方案的核心是通过将广泛覆盖的科研问题语料与形式化、可验证的高质量评估子集相连接，为衡量模型在科研级问题上的“重新发现”能力提供了可扩展、可重复的基准体系，推动对当前未解科学难题求解路径的研究。
链接: https://arxiv.org/abs/2610.37739
作者: Jiarui Yao,Ruida Wang,Tong Zhang
机构: University of Illinois Urbana-Champaign (伊利诺伊大学厄本那-香槟分校)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:Large language models (LLMs) have shown increasing promise in research-level problem solving, yet a comprehensive collection of such problems is still missing, and evaluating them on research-level problems remains challenging. We introduce OpenFrontiers, a cross-disciplinary resource that combines a structured corpus of open research problems with a retrospective benchmark of recently resolved problems in mathematics. The corpus records problem statements, context, progress, and other information for 12,151 problems. The benchmark contains 200 problems, 110 resolved in 2026 and 90 still open, each with a Lean~4 statement checked by automated probes and two independent reviews; a subset HQ-50 further filters based on quality score, etc. Our evaluation protocol grades formal proofs with two independent checkers in an isolated sandbox under closed-book, frozen-retrieval, and live-web conditions. Evaluating seven LLM agents on the HQ-50 subset under a standardized budget, we find that closed-book rediscovery is rare, at most 7 of 40 resolved problems, whereas live-web success mostly reflects retrieval of published solutions. By linking a broad research-problem corpus to a formally graded evaluation subset with documented solutions, OpenFrontiers provides a foundation for measuring research-level rediscovery and studying progress toward solving currently open scientific problems. The dataset and benchmark could be found in this https URL.
[NLP-144] hinkJev: One Thought Every Budget for Calibrated Typed Decisions
【速读】： 该论文旨在解决语言模型在处理需多步推理的类型化决策任务（如算术运算、计数和日期比较）时，直接基于单次前向传播输出概率所导致的高错误率问题。现有方法通过让模型“思考”（即生成中间推理步骤）可显著提升准确性，但代价是响应时间从毫秒级上升至秒级，难以满足高效应用需求。其解决方案的关键在于提出一种基于单一开放混合思维模型的双系统架构——ThinkJev，该架构融合了快速响应（System 1）与延迟思考（System 2）机制：系统1在不经过推理的情况下直接读取选项字母的概率；系统2则在对同一缓存提示进行若干思考令牌后读取相同选项的概率，实现无重复计算的“升级式”推理。由于思考过程可任意截断，一次采样即可覆盖不同思考预算下的答案输出。通过在标注开发数据上进行统计检验，系统1仅在保证误差率低于目标阈值（如5%）时独立作答，其余情况则启用思考机制。实验表明，在5%误差目标下，Qwen3.5-27B以24.5%的独立作答比例实现了3.9%的实际错误率，而剩余任务通过早期停止策略达到81.6%准确率（对比始终思考的82.7%），仅消耗72%的思考令牌。此外，仅依赖系统1置信度决定是否思考的效果较差，因系统1最不确定的任务未必是思考能改善的。在公开的JevBench基准测试中，该模型选择思考的比例为29.4%，且在准确率上与两个专用思考API无显著差异，验证了其高效性与实用性。
链接: https://arxiv.org/abs/2610.37732
作者: Xiaolu Ding,Kai Yang,Dongdong Liu
机构: Fujian Bosssoft Co., Ltd.(福建博思软件有限公司)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 31 pages, 7 figures, 10 tables
Abstract:Many uses of language models end in a typed decision: a choice among a closed set of options, whose probabilities code can act on. Reading those probabilities from one forward pass is fast but often wrong on decisions that take several steps, such as arithmetic, counting, and date comparison. Letting the model think first fixes most of these errors, but costs seconds per decision instead of milliseconds. To pay for thinking only where it helps, ThinkJev builds a fast and a slow system from one open hybrid-thinking model that can answer directly or after a thought. System 1 reads the model’s probabilities for the option letters when it answers directly; System 2 reads the same letters after the model has thought on the same cached prompt, so escalating repeats no work and both systems give probabilities over the same options. Because the letters can be read after any number of thought tokens, one sampled thought also gives the answer under every thinking budget. Without training, ThinkJev lets System 1 answer alone only where a statistical test on labeled dev data guarantees that its accepted answers stay under a target error rate. At a 5% target, Qwen3.5-27B answers 24.5% of 8359 test decisions this way with a realized error of 3.9% (Qwen3.5-4B and 9B: 6.5% and 0.0%). For the rest, stopping each thought as soon as its answer is confident reaches 81.6% accuracy, against 82.7% for always thinking, with 72% of the thought tokens. Choosing when to think from System 1’s confidence alone saves less, because the decisions System 1 is least sure of are often not the ones thinking fixes. On the public items of JevBench, an independent benchmark of typed-decision services, Qwen3.5-27B thinks on 29.4% and is not distinguishable in accuracy from the two thinking APIs listed (exploratory). Code and data generators are available at this https URL.
[NLP-145] What a Weight Update Buys: Leverag e Across Training Stages
【速读】： 该论文旨在解决语言模型在不同训练阶段（如预训练、监督微调、强化学习等）中权重更新对模型行为影响的量化难题，核心问题是：如何精确衡量一次权重更新所“购买”的模型性能提升（即其有效性或效率）。传统方法难以区分有效更新与噪声扰动，尤其在大规模模型中，权重变化幅度极小却可能引发显著行为改变。为此，论文提出“杠杆率（leverage）”这一精确度量指标——它基于Fisher信息矩阵的二次型，通过一次前向传播即可计算每个方向上的更新效用，并以相同规模的随机方向作为基准进行归一化比较。杠杆率揭示了不同训练阶段的增益差异：预训练窗口在文本生成任务上仅相当于随机方向的效用水平，表明其累积效应趋于噪声；而监督微调（SFT）的杠杆率为随机方向的5倍，强化学习（RL-Zero/RLVR）达400–1,500倍，而直接偏好优化（DPO）高达400–23,000倍，显示出其极高的训练效率。研究进一步发现，训练精度显著影响更新特性：使用bf16训练且无fp32主副本时，36%–99%的参数坐标被冻结，杠杆率下降至七分之一，且更新幅度与对角Fisher矩阵呈现强对齐，这在fp32训练中不出现。此外，论文推导并验证了一条“重用定律”，即符号翻转零点预测随机丢弃与重缩放（DARE）在保留率 $ p $ 时的相对损伤为 $ (1-p)/(p,\rho_\mathrm{sign}) $，其预测值与实测结果相差不超过两倍。为确保结果可靠性，研究设置了双重控制条件：所有采用fp32主权重的SFT和DPO轨迹均随训练进程表现出杠杆率下降；两次复现已发布的DPO阶段，其位移和杠杆率与公开结果偏差分别小于1%和4%，而简单配方则低估约5–6倍，排除了其他变量干扰。最终，该研究将训练长度归一化的损失与批次大小128共同识别为关键决定因素，实现了对模型训练动态的系统性解析。
链接: https://arxiv.org/abs/2610.37719
作者: Shikhar Shiromani
机构: 未知
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注:
Abstract:Post-training moves a language model’s weights by about a thousandth of their norm and changes its behavior sharply; pretraining moves them by a third and buys a slow drift. We introduce leverage, an exact measure of what a unit of weight change buys: the Fisher quadratic form of an update, computed with one forward-mode pass per direction, divided by the same form for random directions of the same size at the same checkpoint. Leverage orders released Olmo-3, Tülu-3, Zephyr, and OLMo-2 checkpoints over four orders of magnitude: pretraining windows sit at the random level on prose, SFT five times above it, RL-Zero and RLVR at 400-1,500, and DPO at 400-23,000. With this instrument we find that pretraining accumulates into noise: windows of 30k to 1M steps are no more consequential on prose than random directions of their size, as a noisy quadratic model of SGD predicts. We find that training precision changes what an update looks like: training bf16 parameters without an fp32 master copy freezes 36-99% of coordinates, lowers leverage up to sevenfold, and creates an alignment between update magnitudes and the diagonal Fisher that fp32 training does not show. We derive and verify a law for reuse: the sign-flip null predicts the relative damage of random drop-and-rescale (DARE) at keep rate p as (1-p)/(p,\rho_\mathrmsign) , within a factor of two of measurement wherever DARE succeeds. Two controls make these readings reliable: every controlled SFT and DPO trajectory with fp32 master weights loses leverage as it moves, and two re-runs of a released DPO stage from its anchor with its published recipe land within 1% of the released displacement and 4% of its leverage, where a plain recipe reads five to six times lower; leaving one component out traces the level to the length-normalized loss and its batch size of 128 together.
[NLP-146] Bouba Beats Kiki and Other Regularities in the Subjective Preferences of LLM s NEURIPS2026
【速读】： 该论文旨在解决生成式 AI 在处理主观性任务时缺乏系统化评估方法的问题。传统评价体系主要针对客观任务（如事实查询）设计，能够衡量模型输出的准确性，但在面对无明确正确答案的主观任务（如“给孩子起什么名字”）时，现有方法难以有效评估模型的选择逻辑。其解决方案的关键在于构建一种针对短形式内容（如数字、婴儿姓名、虚构城市名）的主观偏好分析框架，通过识别29个涵盖句法与语义层面的特征维度（如长度、字母顺序、流行度、元音分布、发音效果等），系统探究模型在多个合理选项间进行选择时的内在偏好模式。研究发现：第一，模型对部分特征表现出稳定且可复现的偏好或排斥；第二，提问方式的细微变化可激活或抑制特定特征的影响；第三，看似无关紧要的问题表述差异会导致整体偏好分布发生显著改变，暗示潜在的操控风险；第四，当模型同时承担生成候选列表与排序决策两阶段任务时，二者之间存在未预期的交互效应，进一步影响最终偏好结果。这一工作为理解生成式 AI 的主观决策机制提供了初步但关键的方法论基础。
链接: https://arxiv.org/abs/2610.37692
作者: Katy Blumer,Jon Kleinberg,Ravi Kumar,Andrew Tomkins
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Computers and Society (cs.CY); Machine Learning (cs.LG)
备注: 10 pages, 5 figures in main text; 27 pages, 13 figures in supplement. To appear at NeurIPS 2026
Abstract:Users engage with large language models on tasks that are purely objective (“When was Netflix founded?”), purely subjective (“What should I name my child?”) or anywhere in-between. We have rich tools for evaluating model accuracy with respect to objective criteria. However, when it comes to subjective aspects of a task, we do not have similarly developed methodology at hand. In this paper, we take some early steps towards framing the question of subjective evaluation. Rather than measure how often a model is right or wrong on a question, we instead ask, how do models make choices among multiple viable responses without a clear winner? To study this question, we restrict our focus to short-form content, and consider comparison tasks among three such content types: numbers, baby names, and fictional cities. We identify 29 aspects of these short-form examples covering many areas of syntax and semantics (length, alphabetic order, popularity, vowel mix, spoken sound, etc.). We show first that models tend to have strong and reproducible preferences for or against some of these aspects. Second, we show that variations in how the model is asked to express a preference may cause some of these aspects to activate or deactivate in the model’s selection. Third, we show that overall preference distributions among pairs may vary radically based on seemingly-unimportant variations in the nature of the question posed to the model, raising new questions about manipulation. And finally, we show that, if models are used both to create lists of alternatives and to rank them, there are unexpected interactions between the two stages that may also drive the models’ stated preferences. Comments: 10 pages, 5 figures in main text; 27 pages, 13 figures in supplement. To appear at NeurIPS 2026 Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Computers and Society (cs.CY); Machine Learning (cs.LG) Cite as: arXiv:2610.37692 [cs.CL] (or arXiv:2610.37692v1 [cs.CL] for this version) https://doi.org/10.48550/arXiv.2610.37692 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[NLP-147] One Confidence Threshold Unequal Coverag e: Language Gaps in Typed Decision Interfaces and How Translating Inputs Reduces Them
【速读】： 该论文旨在解决多语言环境下，基于阈值的生成式分类器在不同语言间表现不一致的问题，即单一阈值是否能公平地适用于所有语言。其核心问题是：在固定多选题接口（API）中，当统一设定置信度阈值为0.9时，英语与其他语言在自动处理覆盖率（coverage）上的差异是否显著，且这种差异是否源于模型本身对非英语语言的置信度低估。解决方案的关键在于揭示：即使各语言在准确率上经过校准，英语仍保持更高的自动处理覆盖率，且标准校准指标（如期望校准误差，ECE）无法检测出这一偏差。研究进一步发现，为每种语言单独优化阈值以达到相同的5%错误率目标，可缩小差距，但将输入翻译成英语虽能缓解差距，仍无法完全消除英语优势。因此，论文建议在部署时应报告各语言在操作阈值下的独立覆盖率，以实现更公平、透明的评估。
链接: https://arxiv.org/abs/2610.37688
作者: Junwei Yao,Yongxi Zhou,Zike Yang,Zhijun Wang,Jianan Liu,Yanxian Chen,Wenbo Ye
机构: Independent Researcher(独立研究员)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Deployed classifiers often act on a prediction only when its confidence clears a threshold and pass the rest to a person. We ask whether one threshold serves all languages equally on typed decision interfaces, APIs that return a probability for each option of a fixed multiple-choice question. In a confirmatory study with a written analysis plan, three vendors’ interfaces classified 1,000 MASSIVE items, each in 51 languages, with a threshold of 0.9. Even when a request is answered correctly in both English and another language, the English version clears the threshold more often, by 8.6-11.7 percentage points on average. Observed coverage, the share of requests handled automatically, is lower than English’s in every other language, while the requests the gate handles are rarely wrong. The gap remains when every language is reweighted to the same accuracy, and a standard calibration metric (ECE) does not detect it on one interface. In exploratory analyses, English’s coverage stays higher when each language gets its own threshold fitted to the same 5% error target. Translating inputs into English nearly closes the gap at 0.9 but leaves English ahead at that error target. Requests written natively in African languages show a gap in the same direction when English is their human translation. We recommend reporting per-language coverage at the operating threshold.
[NLP-148] From Geometry to Behavior: How Instruction Exposure Unlocks Latent Rhetorical Directions in LLM s
【速读】： 该论文旨在解决生成式 AI (Generative AI) 在未经过指令微调（instruction tuning）的情况下，其内部是否已具备可几何解析的类比推理能力，以及这种能力在推理阶段能否被有效激活的问题。核心挑战在于理解类比推理这一高级认知行为在模型训练过程中如何以潜在形式存在于激活空间中，并受训练流程中指令数据暴露程度的影响。解决方案的关键是通过差异均值法从残差流（residual stream）中提取“类比向量”（Analogy Vector），并验证该向量可在无需微调的前提下实现对解释风格的可控引导。实验表明，该方法在七种指令微调模型上均能显著提升解释质量，尤其在Gemma-2-9B-IT上使谐均值从3.22提升至3.94（+22%）。更重要的是，研究发现类比方向在所有基础模型中均存在，但其行为可访问性随训练过程中指令数据的暴露程度连续变化——指令数据融入预训练过程的模型（如Qwen-2.5）表现出更高的初始可访问性（类比向量与指令模型间余弦相似度达0.94–0.97），而严格分离预训练与指令微调的模型（如Gemma-2）则呈现平坦的操控景观，即虽存在但行为上惰性。这一发现挑战了传统的“基础模型 vs 指令微调模型”的二元划分，提出真正决定因素是模型在整个训练流程中所接收的指令暴露量。研究将激活操控定位为一种互补的可解释性工具，揭示了指令暴露如何在激活空间中组织修辞知识。
链接: https://arxiv.org/abs/2610.37685
作者: Shengjie Xu
机构: University of Maryland (马里兰大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: Accepted
Abstract:Language models do not merely memorize facts; they absorb rhetorical strategies from pretraining data, including the human tendency to reason by analogy. We ask: is this absorbed capability geometrically accessible before any instruction tuning, and what determines whether it can be activated at inference time? We extract an “Analogy Vector” from the residual stream via difference-in-means and show it enables steerable explanation style across seven instruction-tuned models (Gemma-2, Llama-3.1, Qwen-2.5; 2B-32B), improving harmonic mean from 3.22 to 3.94 (+22%) for Gemma-2-9B-IT on 300 held-out test prompts. A small human validation study confirms the LLM judge on Analogy Quality. Our central finding, from seven matched base-instruct pairs, is a gradient of behavioral accessibility: the analogy direction exists in every base model, but its behavioral accessibility under linear intervention scales continuously with instruction exposure during training. Gemma-2 base models (strict pretraining/IT separation) show flat steering landscapes: present but behaviorally inert. Qwen-2.5 base models (instruction data woven into pretraining) are already partially accessible before any explicit fine-tuning; base-instruct cosine similarity of the analogy vector reaches 0.94-0.97 for Qwen versus 0.37-0.66 for Gemma, providing geometric corroboration of the behavioral gradient. This challenges the binary “base vs instruct” framing: the relevant variable is how much instruction exposure a model has received across its full training pipeline. We position activation steering as a complementary interpretability tool that reveals how instruction exposure organizes rhetorical knowledge in activation space. Comments: Accepted Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI) Cite as: arXiv:2610.37685 [cs.CL] (or arXiv:2610.37685v1 [cs.CL] for this version) https://doi.org/10.48550/arXiv.2610.37685 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Journalreference: Conference on Language Modeling (COLM 2026)
[NLP-149] An Introduction to Large Language Models for Scientists and Professionals
【速读】： 该论文旨在解决科学与专业领域用户在使用大语言模型（Large Language Models, LLMs）时面临的认知鸿沟问题，即尽管LLM界面看似简单，但其背后的技术机制、潜在局限性及应用风险常被忽视，导致误用或不当依赖。核心问题在于：如何帮助非技术背景的科研人员和专业人士在实际工作中实现对LLM的有效、负责任且批判性地使用。解决方案的关键在于提供一个无需编程基础、以实践为导向的入门指南，系统性地阐释包括分词（tokenisation）、嵌入（embeddings）、训练与推理、上下文窗口、提示工程（prompting）及指令层级在内的核心概念，并重点强调生成内容中的幻觉（hallucinations）、不确定性、可复现性差、偏见等关键缺陷，以及对输出进行系统性验证和人工监督的必要性。同时，该方案涵盖隐私保护、治理框架、安全策略，并引入检索增强生成（Retrieval-Augmented Generation, RAG）、向量搜索、信息片段化（chunking）、智能体（agents）与工作流等进阶技术，通过一个完整的本地部署实操练习（基于Ollama与Open WebUI），引导用户在真实科学文献场景中完成证据检索、来源核实与结论评估，从而强化“语言合理性”与“科学支持性”之间的区分意识，最终实现从被动接受到主动批判性使用的范式转变。
链接: https://arxiv.org/abs/2610.37684
作者: Umberto Michelucci
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Large Language Models (LLMs) have rapidly become important tools for science, education, medicine, and professional work. Despite their apparent simplicity from the user’s perspective, effective and responsible use requires an understanding of how these systems work, what their limitations are, and how their outputs should be evaluated. I wrote this document as a basis for my lecture on LLM in the master in artificial intelligence in medicine at the university of Turin, Italy. This book provides a practical and accessible introduction to LLMs for scientists and professionals. It explains the fundamental ideas behind tokenisation, embeddings, training, inference, context windows, prompting, and instruction hierarchies, while deliberately avoiding mathematical detail. Particular attention is given to the limitations of LLM-generated answers, including hallucinations, uncertainty, reproducibility, bias, and the need for systematic verification and human oversight. The book also discusses privacy, governance, security, and the responsible handling of personal and sensitive information. More advanced topics include Retrieval-Augmented Generation (RAG), retrieval quality, chunking, embeddings, vector search, agents, tools, and workflows. A substantial practical exercise guides the reader through running a local LLM with Ollama and Open WebUI and using it to interact with a collection of scientific papers. The exercise emphasizes evidence retrieval, source verification, uncertainty, disagreement between studies, and the distinction between plausible language and scientifically supported conclusions. No programming knowledge is required.
[NLP-150] Prompt-agnostic Language Models EMNLP2026
【速读】： 该论文旨在解决大语言模型在推理阶段对提示（prompt）形式高度敏感的问题，这一问题通常依赖人工进行提示优化，但存在效率低、难以扩展的缺陷。为此，论文提出在训练阶段通过指令微调构建模型对提示的鲁棒性，而非依赖推理时的调整。其核心挑战在于：现有方法通过混合同一提示的不同改写版本（paraphrase）进行训练时，不同改写版本对应的梯度在可训练参数空间中产生显著干扰，表现为57%的梯度方向冲突，且简单的梯度平均会损失约五分之一的梯度幅值。针对此问题，作者提出两种基于梯度融合的训练方法——Prompt-TIES与Prompt-DARE，通过引入来自模型与梯度融合中的干扰消解算子，专门处理提示梯度间的冲突。实验对比了这两种方法与三大类训练阶段鲁棒性增强方法（数据构造、一致性正则化、对比对齐）的有效性，在两个模型家族、四种规模下的评估表明，提示表述与具体示例之间的交互作用平均贡献了54%的样本级变异性，且在ANLI和WiC等任务上超过70%。相比之下，所提梯度融合方法直接提升了局部鲁棒性，使全微调模型的样本级方差近乎减半，同时对整体性能影响极小；而其他训练阶段方法虽能提升原始性能，却无法稳定降低样本间变异性。
链接: https://arxiv.org/abs/2610.37683
作者: Frederic Sadrieh
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Master’s thesis, University of Potsdam, Hasso Plattner Institute, 2026, Parts of this work also appear in the EMNLP 2026 paper “Prompt-Robust Language Models: Which Training Strategies Work?” ( arXiv:2609.01217 )
Abstract:Despite their strong performance, large language models remain highly sensitive to prompt formulation. This issue is typically addressed at inference-time through manual prompt optimization, which introduces overhead and lacks scalability. Conversely, we focus on train-time approaches that build prompt robustness into the model during instruction tuning. A common strategy is to mix multiple paraphrases of each prompt into training. We demonstrate that across trainable parameters, the per-paraphrasing gradients interfere, with 57% exhibiting sign conflicts and the naive gradient average discarding approximately a fifth of the magnitude. Therefore, we develop two gradient-merging methods, Prompt-TIES and Prompt-DARE, that address this interference by leveraging interference-resolution operators from model and gradient merging and applying them to prompt gradients. In this study, we compare the efficacy of our methods to increase model robustness to train-time robustness methods spanning all three established classes: data construction, consistency regularization, and contrastive alignment. The evaluation is conducted across four models from two families at two scales. We observe the interaction between prompt phrasing and the specific examples accounts for 54% of per-example variability on average, and exceeds 70% on tasks such as ANLI and WiC. The gradient-merging methods directly improve this local robustness and nearly halve per-example variance on fully fine-tuned models at little cost to overall accuracy. The other train-time methods enhance raw performance, but fail to improve per-example variance reliably. Code: this https URL. Model checkpoints: this https URL.
[NLP-151] Rank-Aware Differentially Private Release of Listwise Preferences for LLM Alignment
【速读】： 该论文旨在解决在局部差分隐私（Local Differential Privacy, LDP）约束下进行列表级偏好反馈（listwise preference feedback）对齐时面临的隐私-效用权衡难题。现有隐私保护对齐方法主要聚焦于成对比较（pairwise comparison），而列表级对齐在LDP框架下的研究尚属空白。其核心挑战在于，完整的排序空间具有高度敏感性，导致传统方法如随机响应（Randomized Response, RR）需注入过量噪声以满足隐私要求，从而严重削弱了列表级反馈所蕴含的丰富监督信息。本文的关键解决方案是提出一种感知排名的指数机制（rank-aware exponential mechanism），通过利用排序信息将原始高敏感性列表级数据映射至低敏感性的二元划分空间，实现对噪声的精准集中——仅在边界附近项目上引入噪声，而非均匀扰动所有项目。该机制在保持与现有方法相同隐私预算的前提下，显著提升了下游大语言模型（LLM）对齐任务的性能，实验证明其优于当前主流的LDP基线方法。
链接: https://arxiv.org/abs/2610.37682
作者: Junwei Chen,Manjiang Yu,Pengpeng Qiao,Yang Cao
机构: Institute of Science Tokyo(东京科学研究所); University of Queensland(昆士兰大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Listwise preference feedback offers richer supervision for large language model (LLMs) alignment than pairwise comparisons, but a full ranking also reveals more sensitive user preferences. Existing privacy-preserving alignment methods focus mainly on pairwise feedback, while listwise alignment under local differential privacy (LDP) remains unexplored. Extending LDP to this listwise domain raises a critical challenge: the high sensitivity of the full ranking space demands excessive noise, neutralizing the benefits of listwise supervision. To address this challenge, we analyze the interaction between supervision granularity and privacy noise in downstream training, and introduce a rank-aware exponential mechanism that privatizes listwise preference data into a low-sensitivity granularity suitable for downstream alignment. The mechanism leverages ranking information to sample a fixed-size binary partition, concentrating the noise near borderline items instead of perturbing all items uniformly as randomized response (RR) does. Empirical evaluations on downstream LLM alignment tasks show that our mechanism consistently outperforms existing LDP baselines at matched privacy budgets. The code is available at this https URL.
[NLP-152] Supervised Fine-Tuning for Reasoning Large Language Models : A Survey EMNLP2026
【速读】： 该论文旨在解决生成式人工智能（Generative AI）领域中，针对复杂推理任务的监督微调（Supervised Fine-Tuning, SFT）方法缺乏系统性综述的问题。尽管近年来在推理导向的SFT方法及其分析方面取得了显著进展，但相关研究分散且缺乏整合，限制了对技术演进路径与核心机制的深入理解。其解决方案的关键在于构建一个双重视角的系统性框架：一方面从方法论设计出发，将改进策略划分为以数据为中心（data-centric）的方法与以算法为中心（algorithm-centric）的创新；另一方面从分析视角重构现有研究，从数据特性、优化动态以及机制性洞察三个维度梳理学术成果。通过整合当前研究热点并识别关键瓶颈，论文进一步指明了未来在推理SFT方向上的潜在突破路径，从而为构建更强大、可解释的推理模型提供理论支撑与实践指引。
链接: https://arxiv.org/abs/2610.37681
作者: Yongcan Yu,Kuangpu Guo,Jian Liang,Lingxiao He,Meng Wang,Ran He
机构: NLPR MAIS, Institute of Automation, Chinese Academy of Sciences(中国科学院自动化研究所模式识别国家重点实验室); Meituan Inc.(美团); School of Artificial Intelligence, University of Chinese Academy of Sciences(中国科学院大学人工智能学院); University of Science and Technology of China(中国科学技术大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Accepted to EMNLP 2026 Conference
Abstract:Large reasoning models have recently demonstrated remarkable capabilities in solving complex tasks, where supervised fine-tuning (SFT) on long chain-of-thought data serves as a crucial foundation for eliciting and enhancing their reasoning abilities. Despite rapid progress in both improving and analyzing reasoning-oriented SFT, the field still lacks a systematic survey that consolidates its fast-growing literature. To fill this gap, we present a comprehensive review of recent advancements in reasoning SFT, examining the literature through the dual lenses of methodological design and analytical investigation. First, we review methodological improvements across the SFT pipeline and categorize them into data-centric approaches and algorithm-centric innovations. Second, we reorganize analytical studies along three dimensions: data characteristics, optimization dynamics, and mechanistic insights. Finally, by synthesizing current research focuses and remaining bottlenecks, we outline promising future directions for reasoning SFT. We hope this survey deepens the understanding of reasoning SFT and paves the way for advanced reasoning models.
[NLP-153] Artifact Distillation: Training-Free Transfer from Frontier to Small Language Models
【速读】： 该论文旨在解决当前小模型（student）通过知识蒸馏（Knowledge Distillation）从大模型（teacher）迁移知识时，存在依赖训练计算资源且转移的知识难以解释的问题。传统方法将知识压缩于模型参数中，导致知识传递过程不透明，且受限于训练阶段的计算开销。为此，本文提出“实体蒸馏”（Artifact Distillation）：由教师模型在训练阶段求解任务并将其推理过程以可查询的“实体库”（artifact bank）形式外部化，在学生模型推理时通过检索该库获取辅助信息，无需任何梯度更新。其核心创新在于将知识转移机制从“参数压缩”转变为“动态推理支持”，实现了跨模型家族的知识迁移（如从Claude到Qwen），并在四个LiveBench任务上显著提升小模型性能，例如使Claude Haiku 4.5在空间推理任务上超越未使用提示的Sonnet 4.5，同时仅消耗其60%的计算成本。
链接: https://arxiv.org/abs/2610.37680
作者: Mustafa Safa Ozdayi,Somin Wadhwa
机构: Core AI, Qualtrics(核心人工智能，品质调查)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 10 pages (4 pages main text), 2 figures, 3 tables
Abstract:Closing the gap between frontier and smaller language models is usually approached by distillation: fine-tuning the small model (student) on a frontier model’s (teacher) outputs, which binds transfer to training compute and leaves the transferred knowledge largely opaque. We propose Artifact Distillation: a teacher solves training instances and externalizes its problem-solving procedure into an artifact bank that a student queries at inference time, with no gradient updates. Across four LiveBench tasks, a bank written by Claude Opus 4.5 lets Claude Haiku 4.5 outperform unaided Claude Sonnet 4.5 on spatial reasoning (0.707 vs. 0.647) at 60% of its cost, and reach Sonnet 4.5’s level on table joining at 89% of its cost. The same raw bank raises Qwen3.5-27B’s pooled score from 0.730 to 0.800, showing that transfer is not confined to the teacher’s model family.
[NLP-154] Exposure Is Not Commitment: Double-Stratified Evaluation of Latent Monitors for Tool-Using Agents NEURIPS2026
【速读】： 该论文旨在解决生成式语言智能体（generative language agents）在工具使用场景中易受间接提示注入（indirect prompt injection）攻击的问题，尤其关注现有研究在模型内部机制探测方面存在矛盾结果的根本原因。其核心问题是：当前评估方法未对“提示暴露”（exposure）与“行为执行”（behavioral commitment）进行有效分离，导致不同研究得出相互冲突的结论。论文提出一种双分层（double-stratification）评估协议，即在训练和评估阶段均将提示暴露与最终行为决策过程明确区分开来。关键解决方案在于，仅在被暴露的样本上训练监控探针（probe），而非使用混合的“不安全 vs 安全”对比数据集；实证表明，这种分层训练使探针能够提前两步预测模型的不安全工具调用（logistic AUROC达0.853，优于文本基线0.855），且在Mistral-7B和OLMo-2-7B上分别达到0.896和0.912。相比之下，未经分层的朴素对比学习则捕捉到的是暴露信号而非真实行为倾向，且其学习内容高度依赖于数据分布。研究进一步揭示，未分层的评估会高估潜在监控器的真实能力，并解释了为何近期预动作探测研究出现阴性结果——信号确实存在，但模型自身的零样本预测与自然语言抽象（NLA）表达无法稳定复现该信号。因此，评估必须对暴露与行为进行双重分层，以确保结果的可比性和可靠性。
链接: https://arxiv.org/abs/2610.37679
作者: Rishabh Sinha
机构: University of California, Berkeley (加州大学伯克利分校)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: Accepted as a poster at the NeurIPS 2026 Workshop on Agents in the Wild (AIWILD); OpenReview: this https URL . 16 pages, 2 figures, 7 tables. Code: this https URL
Abstract:Tool-using language agents are vulnerable to indirect prompt injection, and a growing literature probes model internals to catch unsafe actions before they execute, with sharply conflicting results. We argue much of the conflict is an evaluation artifact, and propose a double-stratification protocol: separate injection exposure from behavioral commitment, in both the evaluation and the training population. On the primary Qwen AgentDojo grid, monitors trained on the naive mixed unsafe-versus-rest contrast mostly learn exposure: they flag exposed trajectories whether or not the agent complies, and barely predict which cases turn unsafe. Retrained on exposed cases only, the same probe families predict the model’s own unsafe tool call up to two steps early, the maximum the three-step trajectories admit (logistic AUROC 0.853, matched by a trained TF-IDF text baseline at 0.855). Mistral-7B reaches 0.896 and OLMo-2-7B 0.912; their naive contrasts learn different things: inverted exposure ranking on one, both axes on the other. Representations separate on exposure (probe 1.000 versus trained text 0.908) and where visible context is sanitized or unavailable. Training contrast alone reproduces the shape of recent pre-action-probe negative results, one candidate explanation for the conflict: the signal exists, but the model’s own zero-shot forecast and NLA verbalizations do not reliably recover it. An unstratified AUROC overstates what a latent monitor knows, and what a naive contrast learns is population-dependent: evaluations should stratify both.
[NLP-155] he Best Second Opinion Is Right Where the Primary Is Wrong: Partner Selection for Selective Span Tagging
【速读】： 该论文旨在解决多模型协作中“次级模型（partner model）”的最优选择问题，即如何在生成式标注系统中为初级模型（primary model）挑选最有效的第二意见模型以提升整体性能。其核心挑战在于：传统方法依赖全局指标（如跨数据集的平均准确率或F1分数）进行模型配对，但这些指标无法有效捕捉模型在初级模型出错区域的互补能力。论文提出的关键解决方案是采用基于初级模型错误样本上表现的条件性选择策略——即在保留初级模型输出的同时，通过在初级模型预测错误的样本上评估候选次级模型的准确性，来筛选出最能弥补初级模型缺陷的伙伴模型。实验结果表明，该策略在8种标注器类型、3个种子和3个英文基准测试中，能够恢复高达82%的最佳伙伴模型带来的性能增益，显著优于基于跨度F1或随机选择的方案（69-70%）。此外，研究发现次级模型的价值具有方向性：使用预测分布而非硬标签可保留一半以上的价值，而针对服务标签进行评分能改进94%的模型配对组合。该规则在未参与训练的第四数据集上仍保持良好泛化能力（斯皮尔曼相关系数0.40，远高于F1的0.04）。对于LoRA微调解码器，独立训练的次级模型在所有情况下均优于共享读出头的初级模型。最终，该框架被零样本应用于51个MASSIVE语言场景，证明每个初级模型在每种语言环境下均存在一个性能优越且与能力匹配的训练伙伴。
链接: https://arxiv.org/abs/2610.37678
作者: Varun Kotte
机构: Adobe Inc.(Adobe公司)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:A span tagger that ranks its outputs by risk can defer the riskiest ones to a person, and a second model can sharpen this ranking. Which second model should it be? Partner selection is conditional, not global: the best second opinion is the model that is right where the primary is wrong. Across eight tagger types, three seeds and three English benchmarks (576 evaluated primary-partner pairings), choosing the partner by held-out accuracy on the primary’s errors recovers 82% of the best partner’s gain, against 69-70% for span F1 or a random pick. Frozen before a fourth dataset was chosen, the rule replicated there (Spearman 0.40 vs. 0.04 for span F1). Partner value is also directional. The predictive distribution is the second opinion: hard labels discard half of a partner’s value, and scoring the served label improves 94% of pairings. Fixed before the fourth dataset was chosen, this score also replicated there. For LoRA-tuned decoders, independently trained partners beat every readout of the primary network. Applied zero-shot to 51 MASSIVE locales, every primary has a helpful trained partner in every locale, and the preferred partner follows competence.
[NLP-156] Scoring Explanations Without Reading Them: Structural Shortcuts in Automated-Interpretability Evaluation Data
【速读】： 该论文旨在解决生成式模型中稀疏自编码器（Sparse Autoencoder, SAE）特征可解释性评估中存在的偏差问题，即当前主流的自动化可解释性评分方法可能依赖于数据管道中的隐含线索（如位置、标记数量等），而非真正反映特征激活与文本内容之间的语义关联。其解决方案的关键在于识别并分离这些非语义性的提示信号（cue），通过构建去除了位置、标记数量及序列顺序等干扰因素的数据窗口，系统性地验证评分结果是否受这些无关线索影响。研究发现，仅依据中心词标记或标记数量即可达到接近甚至超过现有大语言模型（LLM）评分器的准确率，表明当前评估体系存在严重偏差。作者进一步提出闭式表达式以分析此类规则的理论上限，并设计出能消除位置提示的自然文本窗口，结合类型化决策模型（Jev）进行评估，结果显示在去除干扰后仍可获得显著高于随机水平的相关性（pooled correlation 0.331），从而证明了真实解释性的重要性。因此，该研究强调应在报告可解释性分数的同时，提供基于相同测试样本的最优无解释基准（explanation-free rule）作为对照，以确保评估结果的可信度。
链接: https://arxiv.org/abs/2610.37676
作者: Xupeng Chen
机构: Dimension Gate
类目: Computation and Language (cs.CL)
备注:
Abstract:Automated interpretability scores an explanation of a sparse-autoencoder (SAE) feature by asking a model to predict, token by token, where the feature fires on evaluation examples or to judge whether marked tokens fit it. We find that the evaluation data standard pipelines build can be scored without reading the explanation. On Neuronpedia’s stored fuzzing runs for 500 random GPT-2 small features, 91.3% of activating examples are cropped with the peak on the centre token, and a rule that says “yes” when the centre token is marked reaches balanced accuracy 0.979, against 0.838 for the stored LLM scorers (recomputed from the stored per-item verdicts, on marks rebuilt from the pipeline’s code); a rule reading only the mark count reaches 0.969. The cue changes form across pipelines: current delphi always marks the same slot in its fuzzing negatives, and full-context runs mark negatives with more tokens. OpenAI’s simulation benchmark lists each neuron’s top-activating sequences first; a rule that predicts activity on the first five scores 0.227 in pooled correlation against 0.160 for GPT-4’s published simulator (post hoc). We give closed forms for what such rules can score; show that one scorer’s simulation scores are higher on stored snippets than on natural windows but not shown to be reordered; and build windows that remove the position cue, derive but do not build negatives that remove the mark cues, and separate the sequence cue by decomposition. On natural-text windows cut at random offsets, the best fixed-offset prior shared by all features scores 0.015. On them, that scorer, a typed decision model (Jev), reaches a pooled correlation of 0.331 over 49 features, against 0.013 with shuffled explanations, and an open 8B simulator with the original read-out reads the explanation there (0.194 against 0.019). Scores should be reported beside the best explanation-free rule on the same items.
[NLP-157] he FlightFinder Occurrence Corpus: harmonising 128 aviation-safety source codes into one deduplicated cross-national accident dataset
【速读】： 该论文旨在解决航空事故数据在全球范围内缺乏统一、可比且可追溯的整合框架的问题。现有全球性数据库（如航空安全网络和飞机事故档案局）多为人工编纂，存在收录标准不一、无记录溯源信息等缺陷。其解决方案的关键在于构建一个名为FlightFinder事件语料库的标准化数据集，通过将来自128个来源（包括各国事故调查机构、事件报告系统及4个全球数据库）的279,485条事件记录与131,882份附带文本进行统一映射，形成单一数据模式。核心创新在于采用分层键（ordered ladder）机制，将每条记录归约至唯一标识符（基于事件特征、注册号与日期、四舍五入的位置与日期，或源系统及其原始标识符），确保缺失字段不会导致错误关联。对于无法通过该机制匹配的记录，采用评分机制进入显式合并表，并仅在自动批准的分数区间内应用或经人工确认后使用。字段冲突遵循预定义的来源优先级规则，保留各来源的原生标识符与链接，并附带归属与许可信息。通过五个经裁定的样本（共200对）评估链路错误率，发现自动评分区间内60个随机抽样合并结果中未出现错误，其95%双侧Clopper-Pearson置信上限的假合并率为6.0%（对应205对）。该误差仅覆盖评分区间，而分层键的确定性连接未被裁定。语料库本身无暴露基数，继承了各来源的报告不足问题，但作为跨国家航空安全统计、来源分歧研究及叙事文本挖掘的基础数据集，提供了一个冻结版本（含DOI引用）与实时API服务。
链接: https://arxiv.org/abs/2610.37671
作者: Denys Kolomiiets
机构: FlightFinder / himaxym.com(FlightFinder / himaxym.com); Lisbon(里斯本), Portugal(葡萄牙)
类目: Computation and Language (cs.CL); Digital Libraries (cs.DL); Applications (stat.AP)
备注: 34 pages, 3 figures, 5 tables. Dataset deposited at Zenodo: version DOI https://doi.org/10.5281/zenodo.22303316 (the exact snapshot every reported number is computed from); concept DOI https://doi.org/10.5281/zenodo.22228451
Abstract:National aviation-investigation databases are deep but territorial; global compilations such as the Aviation Safety Network and the Bureau of Aircraft Accidents Archives are hand-curated, apply their own inclusion thresholds, and publish no per-record lineage. The FlightFinder Occurrence Corpus harmonizes 279,485 occurrences and 131,882 attached narratives from 128 source codes – national accident-investigation authorities, occurrence-reporting systems and four global compilations. Records are mapped into one schema; each normalized record reduces to exactly one key from an ordered ladder – event signature, registration and date, rounded position and date, or source and native identifier – so a missing field never creates a link, and the key is the occurrence identifier. Records the ladder cannot reach are scored in a blocked pass into an explicit merge table and applied only within an automatically approved score band or after a person confirms them. Field conflicts follow a documented source ranking; one provenance row per contributing source – its native identifier and URL – is retained, with attribution and licence terms per source code. Linkage error is measured on five adjudicated samples totalling 200 pairs. A recorded uniform random draw of 60 from the 3,441 merges the automatic band applies as it ships found no error; the exact two-sided 95% Clopper-Pearson upper bound on its false-merge rate is 6.0%, or 205 pairs. The measured error covers the scored band only; the ladder’s deterministic joins are not adjudicated. The corpus has no exposure denominators and inherits the under-reporting of its sources. It is offered as a base for cross-national safety statistics, source-disagreement studies and narrative text mining; a frozen, DOI-referenced snapshot accompanies the live API.
[NLP-158] DeflectBench: A Benchmark for Evaluating Rhetorical Fallacy Generation in LLM s ICML2026
【速读】： 该论文旨在解决生成式人工智能（Generative AI）在受到指令时是否可被诱导生成特定修辞谬误（rhetorical fallacies），以及当前安全后训练（safety post-training）机制对此类行为的约束能力问题。现有研究多聚焦于检测已有文本中的谬误，而对主动生成谬误的能力及其防御机制关注不足。为此，作者构建了DeflectBench评测基准，系统评估了四款前沿大模型在三种谬误策略（即“何不反问”whataboutism、“人身攻击”ad hominem、“转移话题”red herring）、七种提示框架及80个涵盖不同争议程度的论点下的生成表现。研究发现，模型拒绝生成谬误的行为主要受请求结构影响，而非论点内容本身；同一论点在不同提示下拒绝率差异仅约11个百分点，而单一提示框架的改变可导致拒绝率波动近100个百分点，谬误类型切换则可引发超过80个百分点的变化。特别地，教育辩论教练类提示框架几乎完全消除拒绝行为，但模型并未实现“干净合规”（clean compliance），而是表现出“标注式合规”（labeled compliance）——即在同一响应中既执行所要求的谬误操作，又明确标示其为操纵手段。四款模型在拒绝、标注合规、软拒绝与干净合规之间的分布存在显著差异。该研究揭示了当前安全对齐机制在面对结构化诱导时的脆弱性，并强调需警惕提示工程对生成行为的潜在操控风险。相关代码与数据集已公开发布。
链接: https://arxiv.org/abs/2608.26119
作者: Art Kanke
机构: 未知
类目: Computation and Language (cs.CL); Computers and Society (cs.CY)
备注: 15 pages. Accepted at the CTB, FAGEN, and AI4GOOD workshops at ICML 2026 in Seoul, South Korea. Code and dataset: this https URL
Abstract:Whether large language models can be prompted to generate rhetorical fallacies on demand, and whether current safety post-training constrains this behavior, has received less attention than the related question of detecting fallacies in existing text. We close this gap with DeflectBench, evaluating 23,990 generations from four frontier models across three deflection strategies (whataboutism, ad hominem, red herring), seven prompt framings, and 80 claims spanning four controversy levels. Refusal is governed primarily by request structure rather than claim content. Per claim refusal varies by only 11 percentage points across the 80 claims, while a single prompt frame change can swing within model refusal by nearly 100 percentage points and switching the requested fallacy type can swing it by over 80 percentage points within explicit framings. An educational debate coach prompt framing collapses refusal to near zero across all four model families, but the bypassed behavior is not clean compliance. Models typically produce labeled compliance, naming the requested manipulation in the same response that contains it. The four models distribute differently across refusal, labeled compliance, soft refusal, and clean compliance. The code and dataset are released at this https URL.
[NLP-159] Discovering Interpretable Acoustic Features in Autoregressive Voice-Cloning TTS Models
【速读】： 该论文旨在解决生成式语音克隆（Voice-cloning TTS）模型在缺乏可解释性的情况下，难以明确其内部表征的声学属性及其使用机制的问题。现有模型虽能基于简短语音提示合成目标说话人的声音，但其对音高、能量、噪声、口音等关键声学概念的编码方式仍不清晰。为此，研究提出了一种基于特征的可解释性框架，通过稀疏探测（sparse probing）、定向控制（steering）与定量自动解释（auto-interpretability）方法，系统分析来自稀疏自编码器（SAEs）、主成分分析（PCA）及随机方向的特征表示。研究聚焦于Qwen3-TTS模型，并将探测与定向控制扩展至另外5个不同架构的模型以验证方法的泛化能力。结果表明，尽管这些模型能够编码如音高、能量、噪声和口音等语义概念，但稀疏预测与定向控制的有效性存在差异，且强探测性能并不必然对应良好定向控制能力。此外，SAEs在人类标注概念的稀疏探测与新概念发现方面表现优异，而PCA在密集探测与定向控制任务中具有竞争力。自动解释方法在Qwen3-TTS中识别出684个经自动验证的特征，涵盖说话人聚类、表达风格及录音条件等超越已有概念的潜在表征。整体上，本研究构建了对自回归语音克隆TTS模型的全面可解释性分析体系，为理解模型内部工作机制提供了关键洞见。
链接: https://arxiv.org/abs/2610.40505
作者: Anuj Diwan,David Harwath,Eunsol Choi
机构: University of Texas at Austin (德克萨斯大学奥斯汀分校); New York University (纽约大学)
类目: Audio and Speech Processing (eess.AS); Computation and Language (cs.CL); Machine Learning (cs.LG); Sound (cs.SD)
备注: Under review
Abstract:Voice-cloning TTS models synthesize an unseen speaker’s voice from brief speech prompts, but it remains unclear which acoustic properties they represent and use. We introduce a feature-based interpretability framework for autoregressive voice-cloning TTS models, analyzing features from sparse autoencoders (SAEs), PCA, and random directions via sparse probing, steering, and quantitative auto-interpretability. We focus on Qwen3-TTS and extend probing and steering to 5 additional models to test generalization across architectures. These models encode concepts like pitch, energy, noise, and accent, but sparse prediction and steering are effective only for some concepts. Strong probing does not always translate to steering or vice-versa. SAEs are useful for sparse probing of human-labelled concepts and novel-concept discovery, while PCA is competitive for dense probing and steering. Auto-interpretability finds 684 automatically validated SAE features in Qwen3-TTS covering speaker clusters, speaking style, and recording conditions beyond known concepts. Overall, we perform a comprehensive interpretability study of autoregressive voice-cloning TTS models.
[NLP-160] Agent ic LLM reasoning in pediatric diabetes outcome prediction from multimodal medical records
【速读】： 该论文旨在解决生成式人工智能（Generative AI）在儿科临床预测中的应用瓶颈，尤其是在小样本、罕见结局场景下的性能不足问题。针对儿童糖尿病患者队列规模有限且并发症发生率低的现实挑战，研究提出并验证了一种名为CEDAR（Clinical Evidence-Driven Agent for Risk-prediction）的证据引用型、自我审计的生成式人工智能流程。其解决方案的关键在于：通过整合结构化数据、基于病历文本提取的社会决定因素以及在T1D患者中纳入动态连续血糖监测数据，构建一个具备可解释性的多模态风险预测框架；同时，CEDAR能够自动引用真实临床证据并进行自我验证，显著提升了在稀疏标注结果上的预测性能（如在T2D队列中联合AUC达0.815，优于传统机器学习模型的0.656–0.700），并在临床专家验证中展现出高度一致性（两位儿科内分泌专家在90%以上情况下与系统达成一致），从而实现了高精度、可信赖且可解释的儿科糖尿病长期预后预测。
链接: https://arxiv.org/abs/2610.40379
作者: Daniel Palacios,Guantong Qi,Anna Kadi,Natali Borrego,Sukru Aras,Michael L. Ferm,Elizabeth Kubota-Mishra,Maria J. Redondo,Zhandong Liu,Mustafa Tosur,Hu Chen
机构: Baylor College of Medicine (贝勒医学院); Texas Children’s Hospital (德州儿童医院); Jan and Dan Duncan Neurological Research Institute (简和丹·邓肯神经科学研究学院); Data Science Center (数据科学中心); Department of Pediatrics (儿科系); Quantitative and Computational Biosciences (定量与计算生物科学); Genetics and Genomics Graduate Program (遗传学与基因组学研究生项目); Department of Epidemiology and Population Health (流行病学与人群健康系); Children’s Nutrition Research Center, USDA/ARS (儿童营养研究中心，美国农业部/农业研究局); Immunobiology Center (免疫生物学中心)
类目: Quantitative Methods (q-bio.QM); Computation and Language (cs.CL)
备注: 18 pages main text plus 40 pages Supplementary Information
Abstract:Large language model (LLM) agents are proposed for clinical prediction but are untested in pediatrics, where cohorts are small and outcomes are uncommon. Using records from 2,489 youth with type 2 diabetes (T2D) and 4,018 with type 1 diabetes (T1D), spanning structured data, note-derived social determinants and, in T1D, continuous glucose monitoring data, we predicted seven T2D outcomes, four also in T1D, at 24 months from 18 months of follow-up. CEDAR (Clinical Evidence-Driven Agent for Risk-prediction), an evidence-citing, self-auditing LLM pipeline, outperformed trained machine learning in both cohorts (pooled area under the receiver operating characteristic curve 0.815 versus 0.656 to 0.700 in T2D and 0.756 versus 0.603 to 0.631 in T1D), most on the sparsely screened outcomes. Only 4 of 70,246 values it cited did not match the chart. Reading only the evidence and reasoning cited by CEDAR, two pediatric endocrinologists reached the same prediction as CEDAR in about nine of ten cases, agreeing with each other on 90.5%.
[NLP-161] Beyond Utterance-Level Speaker Similarity: Perception-Aligned Segmental Evaluation of Accent in Speech Generation
【速读】： 该论文旨在解决生成语音中感知口音差异定位不明确的问题，即现有的话语级相似度评分无法揭示口音差异具体出现在语音的哪些局部区域。其解决方案的关键在于提出一种分段式评估协议，结合说话人和口音相似度评分，并引入来自目标口音地区听者对感知口音差异的子词级别标注（subword annotations），从而精确定位显著的发音差异。通过新收集的语音数据，47名听者对200组内容匹配的参考-候选语音对进行评估，涵盖四种苏格兰与澳大利亚口音、两种零样本文本转语音（TTS）系统及一种基于发音特征的语音转换系统。结果表明，子词标注能够揭示与整体相似度评分不同的系统排名，且说话人与口音相似度评分之间仍保持强相关性（ρ = 0.71），凸显了考察局部感知证据的重要性。此外，七种客观距离度量与子词标注结果的相关性较弱，最高仅达ρ = 0.22，说明现有客观指标难以有效捕捉感知上的口音差异。研究团队公开了代码、数据及交互式演示工具，以推动基于感知的分段式口音评估方法的发展。
链接: https://arxiv.org/abs/2610.37912
作者: Jinzuomu Zhong,Alice Ross,Catherine Lai,Korin Richmond
机构: 未知
类目: Audio and Speech Processing (eess.AS); Computation and Language (cs.CL); Sound (cs.SD)
备注:
Abstract:Utterance-level similarity ratings do not reveal where perceived accent differences occur in generated speech. We introduce a segmental evaluation protocol combining speaker and accent similarity ratings with subword annotations of perceived accent differences provided by listeners from target accent regions. Using newly collected recordings, 47 retained listeners evaluate 200 content-matched reference–candidate pairs spanning four Scottish and Australian voices, two zero-shot TTS systems, and an articulation-based voice conversion system. Subword annotations locate salient pronunciation differences and sometimes yield different system rankings from similarity ratings. Speaker and accent ratings remain strongly associated ( \rho = 0.71 ), highlighting the value of examining local perceptual evidence beyond these scores. Evaluation of seven objective distances reveals weak correspondence with subword selections, with correlations reaching at most \rho = 0.22 . We release code, data, and an interactive demo to support the development of perceptually grounded segmental accent evaluation.
信息检索
[IR-0] stimony from Timelines: Evidence-Grounded Significance Testing of Temporal Clinical Hypotheses
链接: https://arxiv.org/abs/2610.40619
作者: Chengyang He,Wenlong Zhang,Koduvayur Subbalakshmi,Ping Wang
类目: Computation and Language (cs.CL); Information Retrieval (cs.IR)
备注:
Abstract:Ask a language model whether sales tend to dip after a product recall, or whether one side effect tends to follow another after a vaccine, and it answers in a fluent sentence that reads like a finding, with no test behind it. Assertions about how events unfold across many cases fill the Web, and each is a temporal hypothesis that only a significance test over the right records can settle. The verifier must specify the null, select the test and decide which records bear on the hypothesis, and any of these choices can go wrong while the conclusion still comes out right. Yet benchmarks for language-model data analysis grade only the conclusion, with the relevant data handed over. We study the problem in health, where a wrong verdict is costly, and present STRATA, a benchmark that grades every decision on the way to a verdict: 8,735 hypotheses over the clinical timelines of vaccine adverse-event reports, with gold plans, evidence sets, p-values and verdicts derived under a released codebook and reproduced by a deterministic oracle. We also propose ART (articulate, retrieve, test), a staged paradigm that exposes each decision as an inspectable output. Extensive experiments with five model configurations show that systems fail mostly at choosing which records count, a verdict-evidence gap that accuracy alone conceals. ART closes most of this gap and brings a 31B open model to within 0.01 of the frontier models under the same paradigm; the frontier models succeed once given the codebook.
[IR-1] Glance: The Generator Is Its Own Reranker
链接: https://arxiv.org/abs/2610.40454
作者: Adilet Metinov,Chingiz Arziev
类目: Computation and Language (cs.CL); Information Retrieval (cs.IR)
备注: 12 pages, 2 figures, 6 tables. Adapters and code: this https URL
Abstract:Retrieval-augmented generation commonly serves separate models for reranking and answering, duplicating model residency and passage processing. We introduce Glance, which equips the generator with a lightweight decision adapter and pointer head to score candidate passages from a shared prefill. Short relevance probes branch from the cached context, producing one score per passage without separately prefilling each query-passage pair. A single resident 9B backbone supports generation, general decisions and reranking, with approximately 0.1 GiB of additional memory per decision skill. After 39 minutes of MS MARCO fine-tuning, Glance achieves nDCG@10 of 0.741 on SciFact and 0.345 on NFCorpus, compared with 0.721 and 0.326 for bge-reranker-v2-m3. On 300 HotpotQA questions, its selected context improves answer F1 by 0.045 over cross-encoder-selected context under the same base generator (paired bootstrap 95% CI [0.014, 0.076]). In a scaling study, reranking quality keeps rising up to 200 candidates (43k tokens) scored from one shared prefill, where it exceeds the cross-encoder by 0.058 nDCG@10, and a probe-only activated adapter lets generation continue from the original passage cache. These initial results demonstrate generator-based reranking with one resident backbone; evaluation currently covers two retrieval datasets, one QA dataset and one backbone family.
[IR-2] Decompose Estimate Select: Coverag e-Aware Context Construction for Long-Form RAG
链接: https://arxiv.org/abs/2610.40314
作者: He Tu,Aleena Bosco Kurumthottam,Torsten Suel
类目: Information Retrieval (cs.IR)
备注:
Abstract:Retrieval-augmented generation (RAG) grounds a language model’s output in retrieved evidence, but for complex questions spanning multiple aspects of an information need, relevance-based retrieval may return redundant evidence while leaving some aspects unsupported. We propose a coverage-aware retrieval pipeline that decomposes queries into subquestions, ranks candidates with a coverage-trained bi-encoder, predicts a document, \times ,subquestion coverage matrix with a cross-encoder, and selects documents by greedy maximum-marginal coverage. On TREC RAG 2024, our approach improves \alpha -nDCG@10, which rewards newly covered nuggets and discounts redundant evidence, from 0.625 to 0.702 , approaching the 0.718 gold-nugget upper bound without sacrificing relevance. We then find that ground-truth oracle retrieval, despite substantially higher retrieval coverage, delivers no more nugget coverage to the answer than our estimated selection. This suggests that answer completeness depends not only on what evidence is retrieved, but also on how effectively the generator uses it. To address this, we introduce a \emphgrounded self-feedback loop in which the model identifies underused evidence from its first answer and generates again without gold supervision, raising V_\mathrmstrict , the fraction of vital reference nuggets fully supported by the answer, by 0.086 over two passes while largely preserving faithfulness.
[IR-3] owards Full-Text Scholarly Argument Mining using Entity-Anchored Claim Extraction
链接: https://arxiv.org/abs/2610.40196
作者: Bhuvanesh Verma,Andy Lüking,Mounika Marreddy,Leon Lukas Hammerla,Alexander Mehler
类目: Computation and Language (cs.CL); Information Retrieval (cs.IR)
备注:
Abstract:Argument mining in full-text scholarly articles remains an important yet challenging task. This difficulty stems not only from the need to model long-distance dependencies but also from the substantial effort required to annotate sentence-level argumentative components throughout lengthy documents. To address these challenges, we introduce a novel approach that leverages scientific entity mentions (e.g., tasks, datasets, and models) to identify and structure claims in scholarly literature. We validate our approach through a case study involving 50 full-text research papers on sentiment analysis. We first establish a reliable human gold standard (Krippendorff’s \alpha = 0.80 ), identifying 113 claim-bearing sentences and extracting 199 structured claim records from 300 candidate sentences. To address the annotation bottleneck, we employ a majority-voting ensemble of large language models (LLMs) to generate a silver-standard dataset comprising 1,487 claims. Encoders fine-tuned exclusively on the silver-standard dataset outperform those trained on a small human-annotated gold-standard dataset and demonstrate strong zero-shot transfer to NLP papers outside the training corpus.
[IR-4] xPRT: Cross-Passage Retrieval Pretraining
链接: https://arxiv.org/abs/2610.40122
作者: Danush Khanna,Himanshu Singhal,Pranjal A. Chitale,Bishal Santra
类目: Information Retrieval (cs.IR)
备注: Under Review
Abstract:Dense retrievers are a foundational component underlying modern RAG and search systems. Training these models typically requires a curated collection of annotated query-document pairs, which are prohibitively expensive to obtain outside a handful of well-resourced domains. This has motivated a line of work that leverages large-scale unsupervised retrieval pretraining from unlabelled text alone. Most such methods learn by optimizing a language modelling or contrastive learning objective over short spans within a document. This raises the question: how can we extract richer learning signals from a single document for retrieval pretraining? To address it, we introduce xPRT (Cross-Passage Retrieval Pretraining), an unsupervised framework that learns from the key information distributed across an entire document and the connections between its parts. We evaluate xPRT against five state-of-the-art objectives on MTEB English v2, continuing pretraining from a public unsupervised checkpoint while holding the corpus, schedule, and training procedure fixed. Across diverse tasks spanning all task types, we observe that xPRT consistently outperforms the strongest baseline by 3.2% on an average. Notably, in this continual pretraining setting, many baseline methods incur substantial performance regressions, whereas our method delivers the largest complementary gains over the initial pretrained checkpoint. Controlled ablations confirm that xPRT’s performance gains stem from its proposed cross-passage signals. xPRT demonstrates that a single unlabelled document is a much richer source of retrieval signal than prior work assumed and offers complementary gains to what current retrievers already capture.
[IR-5] Beyond IPS: Reliable Counterfactual Evaluation in Multi-Stage Ad Systems without Logged Propensities
链接: https://arxiv.org/abs/2610.40112
作者: Mohsen Malmir,Mohamed A. Radwan,Houssam Nassif,Murat Ali Bayir
类目: Information Retrieval (cs.IR)
备注:
Abstract:Offline evaluation of ad-ranking policies in multi-stage delivery systems is challenging because action propensities are often unavailable, limiting inverse propensity scoring (IPS). Reward-model-based estimators such as the Direct Method (DM) avoid propensities but can be biased when candidate policies induce impression distributions that differ from the logged data. We formalize DM-based counterfactual replay and propose mixture-weighted reward modeling, which reweights logged data toward a mixture of candidate-policy distributions using density ratios from a multiclass domain classifier. We bound each candidate’s value-estimation error by overlap, cold-start, and replay terms, and derive an asymptotic rule that uses the cold-start mass to decide whether to trust offline selection or route candidates to online A/B testing. On an ads-delivery simulator and on production traffic from a large-scale CTR ranking system, the mixture-weighted reward model improves policy selection over its unweighted counterpart in both settings, with the largest gains at low overlap in simulation. Combined with a doubly robust correction, it attains the highest Top-1 accuracy in production and at low and medium overlap in simulation.
[IR-6] R2QRec: Learning Recommendation Queries from Retrieval Feedback
链接: https://arxiv.org/abs/2610.39984
作者: Weixin Chen,Shuchang Liu,Hailan Yang,Yuhan Zhao,Li Kang,Xuefeng Zhang,Chenghao Zhang,Xinyuan Wang,Hanchen Yang,Yinghao Wu,Hua Huang,Yandong Bai,Fan Yang,Xiang Li,Chunjie Chen,Li Chen
类目: Information Retrieval (cs.IR)
备注:
Abstract:Query-based recommendation separates expressing user needs from locating matching items. Its central learning problem is how to turn item interactions into effective retrieval requests when interaction logs do not annotate queries. We present R2QRec, which uses a fixed retrieval system to construct query supervision and evaluate recommendation outcomes. Retrieval verification first selects generated expressions that recover observed items as supervised fine tuning targets, allowing retrieval behavior to determine the target wording. Policy refinement then evaluates generated query sets by the observed item’s rank in their fused recommendation list. Log-rank feedback distinguishes outcomes beyond the recommendation cutoff, while a KL penalty constrains policy change relative to the supervised reference. The stages have distinct roles: verification identifies expressions to imitate, and policy refinement improves the combined ranking produced by newly generated queries. Only the query generator is trained; the retriever and item index remain fixed, and no query annotations or learned reward model are required. Across three public datasets, R2QRec achieves state-of-the-art overall recommendation performance among the evaluated methods. Extensive experiments demonstrate the effectiveness of the proposed design and its generalization across generators and domains.
[IR-7] Heterogeneous Co-Evolution Learning for Tool-Efficient Search Agents with Zero Labels
链接: https://arxiv.org/abs/2610.39682
作者: Qin-Wen Luo,Sheng Ren,Xiang Chen,Rui Liu,Jun Fang,Naiqiang Tan,Jun Wang,Hua Chai,Sheng-Jun Huang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Information Retrieval (cs.IR)
备注:
Abstract:Deep search agents extend language models by interleaving reasoning with external retrieval. However, effective search-agent training requires addressing two challenges beyond final-answer optimization. First, the capability frontier of agents is largely determined by training questions, whereas fixed or labeled task collections cannot adapt as agents improve. Second, outcome-only reinforcement learning treats all successful trajectories equally, even when some correct answers rely on redundant searches, creating an inherent tension between answer accuracy and tool efficiency. To address these issues, we introduce \textscZEST-Search, a zero-label and tool-efficient framework for search-agent training via heterogeneous co-evolution. Starting from unlabeled reference questions, \textscZEST-Search constructs validated search tasks online through an answer-first and evidence-path-guided generation process. A heterogeneous proposer adaptively constructs training tasks around the solver’s evolving capability boundary, while the solver learns search-augmented reasoning through reinforcement learning. To jointly optimize task difficulty and search efficiency, \textscZEST-Search employs an intermediate-difficulty reward for the proposer and a task-local tool-aware reward for the solver, encouraging informative tasks and correct answers with fewer redundant searches. Experiments on seven general-domain and multi-hop question answering benchmarks demonstrate that \textscZEST-Search consistently improves search-agent performance while achieving higher tool efficiency without labeled search-training data or expert search trajectories. Further analyses show that heterogeneous generation, adaptive difficulty control, and tool-aware optimization enable more effective search-agent training.
[IR-8] Efficient LLM -based Subgraph Retrieval for Multi-Hop Knowledge Base Question Answering
链接: https://arxiv.org/abs/2610.39667
作者: Duanyang Yuan,Sihang Zhou,Xiaoshu Chen,Haoyuan Chen,Ke Liang,Jian Huang
类目: Information Retrieval (cs.IR)
备注:
Abstract:Multi-hop Knowledge Base Question Answering (KBQA) aims to find answer entities in the knowledge base that are multiple hops away from the entities in the question. Information retrieval-based (IR-based) methods extract a pivotal subgraph from the entire KB to locate candidate answers and then evaluate their plausibility through semantic matching with the question. However, we observed that the extracted subgraphs often include nodes that are weakly related or irrelevant to the question. Without a proper node filtering mechanism, the number of irrelevant nodes grows as the number of hops increases, leading to excessive consumption of computational resources. To address these challenges, this study introduces an efficient LLM-based subgraph retrieval method for multi-hop knowledge base question answering, M-ER. The framework leverages Monte Carlo Tree Search (MCTS) to transform subgraph exploration into a tree-structured search process. During the MCTS selection phase, nodes that are highly relevant to the question are prioritized for inclusion in the subgraph, eliminating the need to traverse all nodes in the KB. The framework further incorporates a large language model (LLM) to refine the search direction, ensuring that exploration remains focused on nodes relevant to the question. In addition, selected nodes are quantitatively scored, and these scores are fed back into the node selection process to effectively filter out irrelevant candidates, thereby improving the quality of the subgraph. This mechanism not only narrows the search space but also enhances the overall efficiency of multi-hop KBQA. Experiments on the WebQSP benchmark demonstrate that M-ER achieves 78.88% on the Hits@1 metric, while also improving computational efficiency. These results not only validate the effectiveness of M-ER, but also offer a viable technical path to balance performance and computational efficiency.
[IR-9] GistRank: Gist-Compressed Listwise Ranking with Large Language Models
链接: https://arxiv.org/abs/2610.39610
作者: Guo Chen,Wenhan Liu,Reed Li,Qianwen Ma,Yu Lu,Haibo Shi,Bingbing Xu,Junjie Huang
类目: Information Retrieval (cs.IR)
备注:
Abstract:Large language models (LLMs) have become a powerful tool for listwise relevance ranking, which can improve web search by comparing retrieved passages jointly. However, processing every candidate verbatim and generating a permutation autoregressively can make inference costly, every candidate must be serialized verbatim into the prompt and the input therefore grows linearly with both the number and the length of the passages. While sliding-window ranking handles context limits through repeated re-encoding of overlapping blocks, hard compression techniques sacrifice the fine-grained details required for accurate discriminative judgment. To better understand and mitigate this trade-off, we conduct a comprehensive study of soft compression for LLM-based listwise ranking. Our findings reveal that a vanilla gist representation is ill-suited to ranking for two reasons: strictly causal encoding weakens how thoroughly each gist token absorbs its passage, and the representation is never trained to expose the relative-preference structure that separates closely competing candidates. To this end, we propose GistRank, a three-stage training framework that equips a large language model with gist-compressed passage representations by learning faithful gist tokens under a semi-bidirectional attention scheme, instilling discriminative comparison through pairwise training, and adapting the model to the listwise objective. GistRank collapses each passage into only K=4 gist tokens, this gives 14 \times speedups over one-shot full-text ranking while reaching 59.17 average nDCG@10 over TREC DL19/20 and five BEIR subsets within 0.89 of the strongest full-context baseline. Overall, GistRank provides an effective means to enhance efficiency for searches in real-world environments.
[IR-10] From Intent to Improvement: Agent ic Dataset Search for Better Model Training
链接: https://arxiv.org/abs/2610.39608
作者: Qiaosheng Chen,Kaijia Huang,Fanqi Zhou,Gong Cheng
类目: Artificial Intelligence (cs.AI); Information Retrieval (cs.IR)
备注: 13 pages, 4 figures
Abstract:Finding training data in million-scale machine learning repositories is a Web search problem where text relevance is not enough. A query couples a semantic training goal with constraints such as task, license, and training stage, and the metadata that would verify them are often missing. Text retrieval ranks dataset cards by similarity but cannot enforce hard constraints or follow the relations that reach a sparsely described dataset. Free-form tool-calling agents are more flexible, but the same model that picks a dataset also declares it a match, and nothing checks that the query’s constraints are actually met. Neither approach establishes that the data it returns will improve model training. To address it, we propose AgenticDS, a contract-governed agentic framework in which every language model decision instantiates a typed object before deterministic code consumes it. The framework compiles a query into a typed search plan, executes it through admissible deterministic operators over hybrid retrieval and bounded graph expansion, and verifies the provenance-grounded evidence before accepting a ranking or emitting a typed diagnosis that permits only local repairs. On HuDS-Bench, a new benchmark of 200 queries with 24,908 graded judgments pooled over 15 systems, AgenticDS improves NDCG@10 by 25.7% over the strongest dense–sparse hybrid and by 18.9% over a commercial LLM agent. A controlled search-to-train study further shows that the datasets it retrieves give the best downstream accuracy in mathematics, code, and medicine. Code and data are publicly available at this https URL.
[IR-11] Lost in the Latent Space: How Code Embeddings Represent Functional Intent and Implementation Fidelity
链接: https://arxiv.org/abs/2610.39519
作者: Lu Sun,Yuheng Huang,Lei Ma,Zhijie Wang
类目: oftware Engineering (cs.SE); Computation and Language (cs.CL); Information Retrieval (cs.IR)
备注:
Abstract:Code embedding models have become a key building block for many software engineering applications, yet their internal representations remain poorly understood. Existing evaluations tend to focus on downstream task performance, offering limited insight into whether a single embedding can support both functional abstraction and implementation-sensitive analysis. To address this gap, we present a large-scale empirical study of 14 state-of-the-art code embedding models. We evaluate their representational behavior through three complementary analyses: supervised probing for functional abstraction and implementation fidelity, zero-shot retrieval experiments to assess geometric similarity, and controlled perturbation experiments that measure embedding responses to semantics-preserving code transformations. The results reveal a consistent abstraction-fidelity tension: models that better preserve stylistic and structural signals tend to do so at the expense of semantic generalization, while models with similar retrieval performance can nevertheless respond very differently to the same implementation changes. Overall, our findings suggest that a single universal code representation remains out of reach, motivating more representation-aware evaluation, selection, and deployment of code embedding models in software engineering.
[IR-12] RoFiGR: Coarse Routing Fine Refinement via Level-Adaptive Distillation in Multimodal Generative Recommendation
链接: https://arxiv.org/abs/2610.39470
作者: Ji Dai,Zhipeng Wei,Zhuang Zhuang,Rongfeng Guo,Jie Chen,Quan Fang,Fei Pan
类目: Information Retrieval (cs.IR)
备注: 12 pages, 10 figures, 10 tables
Abstract:Generative recommenders predict an item’s semantic identifier (SID) one code at a time. Early codes select broad branches, so a routing error can exclude relevant items before fine-grained prediction begins. In multimodal recommendation, these coarse decisions must reflect collaborative preferences as well as item semantics. We introduce RoFiGR, a two-tree recommender that uses collaborative information in both SID construction and generation. It aligns cumulative L1-L2 code vectors with frozen collaborative item embeddings and explicitly encodes item distance and source level in the serialized history. During generation, a sequential teacher supervises legal branches. Fixed coarse-level weights maintain routing supervision, while a bounded adaptive residual allocates additional weight according to teacher-student demand. On Instruments, Arts, and Games, RoFiGR improves all five ranking metrics over the evaluated external baselines. Five-seed paired tests with Holm correction support the main NDCG@10 improvements. Comparisons with alternative collaborative supervision objectives and tokenizers support the proposed design. Matched task and decoding analyses distinguish the effects of training exposure, stream fusion, and candidate budgets.
[IR-13] UGR: Unified Generative Ranking for Heterogeneous E-commerce Recommendation
链接: https://arxiv.org/abs/2610.39426
作者: Zonghe Shao,Huabin Chen,Sitong Lu,Fangmei Zhu,Yingcai Ma,Ge Fan,Jinguo Gao,Wei Shi,Yanfang Liu,Huiping Chu
类目: Information Retrieval (cs.IR)
备注:
Abstract:List-wise ranking in e-commerce recommendation must balance user utility and commercial value across organic content, ads, and candidates from other businesses, while modeling candidate interactions and sequential dependencies. Existing generative ranking methods typically learn from logged exposure sequences and operate on truncated candidate sets from a single business. Exposure-sequence learning may not fully exploit post-exposure feedback signals. Meanwhile, generative ranking over heterogeneous, multi-business candidates remains less explored, particularly when recommendation scores and advertising value must be jointly modeled under generalized second-price (GSP) auctions. We propose Unified Generative Ranking (UGR), which performs generative ranking over the full request-level multi-business heterogeneous candidate set, typically comprising around a thousand candidates in production, and directly generates multi-business ranking lists through candidate-aware autoregressive decoding. To go beyond exposure-sequence imitation, UGR introduces complementary feedback supervision through weighted supervised fine-tuning and permutation preference optimization, which respectively leverage observed post-exposure feedback and feedback-derived preferences among permutations. UGR uses a training-time advertising debiasing tower to model historical ad exposure effects and incorporates current ad value during serving while preserving the existing allocation and GSP pricing rules. Multi-value sequence retrieval produces candidate ranking lists, and the existing downstream evaluator selects the final list for exposure. Offline experiments demonstrate consistent gains in ranking quality, while online A/B tests on Taobao’s Guess You Like recommendation service yield improvements of 1.76% in Core IPV, 3.29% in Transaction Volume, and 2.10% in GMV over the production baseline.
[IR-14] ry Before You Evolve: Ranking-Scored Trials for Recursive Self-Improvement of Recommender Systems
链接: https://arxiv.org/abs/2610.39352
作者: Xu Huang,Jiawei Feng,Jiancan Wu
类目: Information Retrieval (cs.IR)
备注:
Abstract:Recommenders on Web platforms such as e-commerce and music streaming need their training recipes re-tuned for each dataset. Large language model (LLM) agents can now rewrite these recipes, enabling recursive self-improvement (RSI) of recommender systems. Existing agents propose one edit per round, fully retrain the recommender, and keep the edit only if its ranking quality improves. Unlike programs or patches that can be scored by execution, a recommender edit changes how the model is trained, which raises two challenges. First, ranking feedback is delayed: an edit’s ranking quality is known only after full retraining, so much of the retraining budget goes to edits that are later discarded. Second, the signals available before retraining are unreliable: sampled training losses can look solved while full-ranking performance remains poor, and they are not comparable across edits that change the loss or sampler. We propose RecRSI, which generates diverse candidate edits from a diagnosis of the current model and selects one for full retraining using early ranking feedback: each candidate receives a short trial, the first steps of its full training run, scored by full-ranking NDCG@10 on held-out probe users. With two sequential recommenders on three datasets and the same number of full retrainings, RecRSI significantly outperforms existing self-evolving agents in every setting, and on both Amazon datasets reaches in its first round a ranking quality that none of them reaches in five.
[IR-15] Does the Citation Matter? Quality Grounding and Physician-Rubric Accuracy of Googles AI Overviews for Medical Information
链接: https://arxiv.org/abs/2610.39339
作者: Sam Agnoli,Brigid Reilly,Lily Barna,Emőke-Ágnes Horvát
类目: Information Retrieval (cs.IR); Computation and Language (cs.CL); Computers and Society (cs.CY); Human-Computer Interaction (cs.HC)
备注: 12 pages, 5 figures, 10 tables. Code and data: this https URL
Abstract:Google’s AI Overviews (AIOs) are the first result users encounter when searching the web, shifting the work of selecting and evaluating sources from the user to Gemini. This may be especially consequential for health information, given that Google’s traditional algorithm was designed to prioritize quality sources, and inaccurate information carries health risks. We evaluated 3,611 benchmark health questions, comparing each AIO against that question’s top five traditional results on source selection and quality. We also evaluated the AIOs’ statement grounding in their cited sources and how accurate AIOs were against physician-written rubrics. Overviews appeared for 96.4% of questions but drew on substantially different evidence than traditional search: only 44.4% of the top five results were cited by the corresponding AIO, which instead overrepresented commercial health publishers (e.g., WebMD, Healthline) at the expense of academic and government sources (e.g., Cleveland Clinic, National Health Service). AIO-cited pages scored lower than traditional results on four of six well-established credibility criteria. The grounding of AIO statements to their sources was poor: after correcting for classifier error, an estimated 56.7% of cited statements were supported by their cited page, and a further 20.8% of all statements were factual claims with no citation at all. When AIOs were graded against physicians’ rubrics, they earned 38.2% of all available points (e.g., accuracy 54.2%, completeness 21.6%), and among the subset of AIOs whose physician rubric anticipated a specific inaccurate or potentially harmful claim, 22.2% of AIOs contained the harmful content. Notably, the accuracy of AIOs was unrelated to the grounding of their statements. We recommend AIO literacy interventions for users and holding AIOs at least to Google’s own health-ranking standards.
[IR-16] Federated Recommender Agents
链接: https://arxiv.org/abs/2610.39289
作者: Jaehyung Lim,Woojoo Kim,Junyoung Kim,Dongha Kim,Wonbin Kweon,Hwanjo Yu
类目: Information Retrieval (cs.IR)
备注: 13 pages
Abstract:LLM-powered recommender agents use reasoning and memory for personalization, but centralized access to user information raises privacy concerns, while isolated agents cannot learn from others’ experiences. We formulate Federated Recommender Agents, a setting where local agents collaborate through natural-language knowledge while keeping interaction histories and personal memories local. We propose FedRecAgent, which enables this collaboration without updating LLM parameters. Its central challenge is determining which reasoning to integrate, since similar behavioral contexts can lead to different choices. FedRecAgent represents local experiences as Condition-Transition-Reasoning (C-T-R) units and hierarchically integrates reasoning by condition and transition. Within this structure, it consolidates related explanations and discriminates reasoning across alternative transitions. Each client maintains semantic and decision memories to capture preferences and conditional choice tendencies, guiding the retrieval and selection of shared reasoning for recommendation. Experiments across three domains under dense and sparse settings demonstrate improved recommendation performance over federated and LLM-based baselines, while ablations support the contributions of personalized memories and structured reasoning integration.
[IR-17] PFA: Attention Can Be Progressively Focused in Long-Sequence Generative Recommendation
链接: https://arxiv.org/abs/2610.39273
作者: Pengjie Zhou,Weiqin Yang,Haibo Xing,Hao Deng,Jinxin Hu
类目: Information Retrieval (cs.IR)
备注: 11 pages, 3 figures, 7 tables
Abstract:Long user behavioral sequences improve the performance of generative recommendation (GR), but computing full attention over long histories during autoregressive beam search decoding incurs substantial online inference latency. To accelerate long-sequence decoding, existing approaches typically rely on static history compression or dynamic top-k sparse attention; however, the former risks losing critical semantics, whereas the latter incurs significant overhead from repeated token selection and non-contiguous memory access. In this work, we thoroughly analyze attention patterns during GR decoding and reveal two key properties: (i) progressive attention focusing, where attention progressively focuses on a small subset of critical historical tokens; and (ii) cross-step ranking consistency, where their importance rankings remain relatively stable across subsequent decoding steps. Motivated by these observations, we propose Progressively Focused Attention (PFA), an efficient sparse decoding framework for long-sequence GR. PFA retains full attention at the initial decoding step to explore global historical preferences. It then establishes a shared importance ranking through cross-beam weighted aggregation and reorders the key-value cache of the selected tokens into a contiguous memory layout. Subsequent decoding steps reuse this ranking and geometrically reduce the token budget, preserving critical semantics while lowering computational and memory access overhead. Extensive experiments on two public benchmarks and a large-scale industrial dataset demonstrate that PFA outperforms state-of-the-art baselines. A 7-day online A/B test on a large-scale e-commerce advertising platform further confirms that PFA delivers substantial business gains while maintaining low inference latency. Our code is publicly available at this https URL.
[IR-18] CASH: A Capability-Aware Subsidy Harness for Advertiser Subsidy Allocation
链接: https://arxiv.org/abs/2610.39244
作者: Minmao Wang,Huihuang Lin,Likang Wu,Hongke Zhao,Shijie Yi,Fei Pan
类目: Information Retrieval (cs.IR)
备注:
Abstract:Advertising platforms subsidize advertisers through billing discounts, and Advertiser Subsidy Allocation (ASA) chooses a subsidy level for every account to maximize total discounted revenue under a platform-wide subsidy-rate constraint. In practice, operators configure these subsidies from business rules, advertiser tiers and operating goals. Such configuration draws on rich business experience, yet as the advertiser base grows, tier-level policies can neither resolve how individual accounts respond across subsidy levels nor allocate a shared budget at the granularity of single accounts. We propose CASH, a Capability-Aware Subsidy Harness for ASA. CASH pairs a knowledge-guided large language model (LLM) with a data-driven numerical response model, learns from realized outcomes where each is locally reliable across accounts and subsidy levels, and turns this reliability into explicit capability licenses that govern source selection, agent review and budget allocation. Its Expected Policy Coverage Gain (EPCG) directs a limited feedback budget toward evidence expected to expand high-value reliable coverage, and the resulting feedback further improves the numerical model and updates capability estimates. In a one-month online experiment with CASH serving 60% of a large-scale advertising platform’s traffic, CASH raises revenue per account by 15.35% and ADVV by 19.86% over Control at an essentially unchanged subsidy rate, with further gains in revenue quality, LTV and retention.
[IR-19] A Sequential Decision Framework for Evaluating LLM -Guided Search
链接: https://arxiv.org/abs/2610.39231
作者: Siqing Huo,Andrew Parry,Debasis Ganguly,Mark Smucker,Charles L. A. Clarke
类目: Information Retrieval (cs.IR)
备注:
Abstract:Recent systems applying retrieval-augmented generation (RAG) and “deep research” iteratively generate queries, retrieve documents, and examine their contents, using the resulting information to guide subsequent search steps. When this process is carried out autonomously rather than by a human searcher, the system must make explicit decisions about which search trajectories to explore and what objective the search is intended to optimize. We introduce a simple framework for reasoning about such search policies, providing a principled approach to evaluation. The framework models the search environment as a set of ranked lists and represents behavior using two primitive actions: select, which advances judgment through an existing ranking, and spawn, which generates a new ranking through query generation or reformulation. Each action produces a document to inspect, yields information that updates the search state, and contributes a reward for a task-specific objective. Within this abstraction, a variety of existing retrieval strategies can be expressed as policies operating over a common environment, while common retrieval objectives can be represented through reward and weighting functions over the resulting search trajectory, and the resulting formulation admits a natural reinforcement learning interpretation. We illustrate the approach using high-recall information retrieval (HRIR), where the goal is to identify as many relevant documents as possible with limited judging effort, showing how the framework provides a unified way to describe and reason about strategies for both classical retrieval tasks and emerging LLM-guided search systems. Through experiments, we demonstrate that both the underlying retrieval method and the “deep research” policy employed can materially affect performance.
[IR-20] Read More Reason Deeper: Latent Workspaces for Long-Sequence Recommendation
链接: https://arxiv.org/abs/2610.39188
作者: Renzhi Wu,Li Sheng,Zikun Cui,Junjie Yang,Tai Guo,Keke Zhai,Yan Li,Jerry Fu,Nipun Mathur,Sri Reddy,Hong Li
类目: Information Retrieval (cs.IR)
备注:
Abstract:Long-sequence recommendation can improve along two axes: \emphevidence scaling exposes longer histories, while \emphreasoning scaling adds layers that interpret that evidence for a candidate. Existing models couple them: sequence encoders propagate every added event through depth, whereas one-shot selection prevents later layers from revisiting omitted evidence. We introduce the Latent Workspace Ranker (LWR), which keeps typed evidence in persistent banks and propagates a fixed-size candidate-local workspace. Each layer rereads the banks from its evolving workspace and archives compact post-interaction state, allowing history length and reasoning depth to grow independently. We compare four architectures across various history–depth configurations using evidence scaling at fixed depth, reasoning scaling at fixed 10K history, and their joint compute-efficient frontier. LWR leads all families at long histories, continues improving with reasoning compute, and converts additional training compute into quality gains at the fastest fitted rate on the joint frontier. At 10K history, it improves NE by 0.19% over HSTU while using 69.6% fewer forward GFLOP/request. It also outperforms existing baselines on two billion-scale ranking surfaces. These results show that separating addressable evidence from propagated state makes reading more and reasoning deeper independently controllable and jointly useful.
[IR-21] PRISM: Personalized Item Representations in User History
链接: https://arxiv.org/abs/2610.39175
作者: Renzhi Wu,Zikun Cui,Junjie Yang,Li Sheng,Yongjun Xie,Jerry Fu,Nipun Mathur,Sri Reddy,Hong Li
类目: Information Retrieval (cs.IR); Artificial Intelligence (cs.AI)
备注:
Abstract:Sequential recommenders typically construct history inputs from item identifiers or content features using an embedding map shared across users. The same item therefore yields the same initial embedding representation, even when users have different relationships to the item. We propose personalized item representations in user history, where each item’s embedding expresses its relationship to the user. Personalization becomes a property of the history inputs themselves, supplying richer evidence for later sequence modeling. We instantiate this idea with PRISM, which reuses upstream ranking computation to generate personalized item embeddings. An autoencoder jointly compresses all layer states from a user tower, a user–item tower, and same-request candidate mixing into one embedding per scored pair for downstream replay. Offline evaluation on Surface 1, a major recommendation surface at billion scale, reports an average downstream normalized entropy (NE) change of -0.15% , compared with -0.07% for VISTA at similar history lengths. We further identify an empirical sequence-length scaling law: over history lengths of 26–240 events, NE gain follows a fitted power law with exponent 0.468, increasing from 0.06% to 0.17%. Therefore, personalized item representations in user history provide a new scaling dimension for improving recommendation quality. Subjects: Information Retrieval (cs.IR); Artificial Intelligence (cs.AI) Cite as: arXiv:2610.39175 [cs.IR] (or arXiv:2610.39175v1 [cs.IR] for this version) https://doi.org/10.48550/arXiv.2610.39175 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[IR-22] PALATE: Pretrained Autoregressive LATent Embeddings CIKM2026
链接: https://arxiv.org/abs/2610.39090
作者: Sizhang Zhao,Praveen Gondhi,Bin Li,Utsaw Kumar
类目: Information Retrieval (cs.IR); Machine Learning (cs.LG)
备注: Accepted to GRAIL @ CIKM 2026. 5 pages, 2 figures, 3 tables
Abstract:Local-commerce and food-delivery marketplaces run large ads businesses: across hundreds of thousands of stores, merchants pay to be surfaced on browse and search surfaces, and a ranker decides which ads each consumer sees. Ranking well demands strong per-consumer personalization, yet a consumer’s raw interaction history is sparse and high-cardinality. Our production ranker had personalized only through hand-engineered signals: aggregate engagement counts, categorical embeddings, and order-agnostic pooled history, with no learned per-consumer sequence representation. We learn one by training a GPT-style generative model of the consumer’s engagement sequence to predict their next few engagements, and read its last-token hidden state as the consumer embedding. That single frozen vector becomes one shared personalization input for every ad surface: a daily job refreshes it, and each ranker consumes it as an ordinary dense feature. Offline it is the largest single-feature gain in our ranking-AUC ladder, a relative Eval-AUC gain of +0.59% on a browse-driven feed and +0.57% on a query-driven search surface. Online, it shipped as one component of bundled ranker retrains on both surfaces: Discovery (revenue +0.79%, conversion rate +1.17%) and Search (revenue +0.94%, conversion rate +1.50%). Three design choices adapt the generative objective to a repurchase-heavy marketplace, and are the core of our contribution: an exploration-weighted loss that up-weights businesses new to the consumer over ones already engaged; a two-stage curriculum: a click pretrain that keeps the representation broad and exploratory, then a conversion fine-tune that sharpens purchase intent; and a designed negative set that pairs a fixed popular-business pool and businesses shown but not ordered with random and in-batch negatives.
[IR-23] Contextual Embeddings Beyond the Gold Chunk
链接: https://arxiv.org/abs/2610.38854
作者: Sedigheh Eslami,Markus Krimmel,William Barber,Antoine Chaffin,Maximilian Schall,Jonas Golde,Louis Milliken,Ekaterina Glazkova,Denis Bykov,Bo Wang
类目: Machine Learning (cs.LG); Computation and Language (cs.CL); Information Retrieval (cs.IR)
备注:
Abstract:Contextual embeddings represent chunks of a document with respect to the document’s global context. These models are typically trained and evaluated under the assumption that a relevant document contains only a single relevant chunk. In this work, we argue that in real-world use cases a retriever should not only find answers spread over several chunks but also surface the evidence needed to verify them. To address this, we introduce pplx-embed-v2-context, a family of contextual embedding models trained by distilling graded chunk-level relevance from a query-aware context compression model. This removes the need for gold-chunk labels and makes the training targets independent of the chunking. We utilize a vision-language backbone for training and release pplx-embed-v2-context in 9B and 0.6B sizes with compatible embedding spaces. On the evaluation side, we introduce two benchmarks that label both the chunks that answer a query and the evidence needed to verify the answer. Context-bench is a controlled diagnostic, and Q2D-Chunk is built from real-world search queries. To encourage further research, we also publish Q2D-Chunk as a public leaderboard. Our 9B model sets a new state of the art on both benchmarks and ConTEB. We find that the visual retrieval capabilities of our 9B model transfer from its vision-language base model and propose to extend chunking to image inputs. We demonstrate that our models achieve state-of-the-art performance among dense retrievers on visual ViDoRe-V3 and that performance improves with granular image chunking. Subjects: Machine Learning (cs.LG); Computation and Language (cs.CL); Information Retrieval (cs.IR) Cite as: arXiv:2610.38854 [cs.LG] (or arXiv:2610.38854v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.38854 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[IR-24] Leaving Staying or Living There? Duration Commonsense Reasoning with Human-LLM Co-annotations
链接: https://arxiv.org/abs/2610.38755
作者: Chenyang Yu,Zhaomin Xiao,Yan Huang,Eduardo Blanco
类目: Artificial Intelligence (cs.AI); Information Retrieval (cs.IR)
备注:
Abstract:Gardening in the backyard implies living there, while checking into an Airbnb more likely means staying for a few days. Such duration judgments are important for understanding human mobility, travel behavior, and social context. However, duration is rarely stated explicitly in social media posts and must instead be inferred from noisy textual, visual, and contextual cues. In this paper, we investigate whether vision-language models possess such duration commonsense. We create a new corpus of geotagged tweets co-annotated by humans and LLMs to serve as a testbed for this task. We find that state-of-the-art models struggle with duration inference, even when guided by multi-turn Chain-of-Thought prompting. Error analyses further reveal that models often fail to integrate critical contextual information, or they overly rely on prominent visual details while overlooking subtle hashtags and semantic nuances. These findings underscore the significant challenges remaining in context-aware multimodal understanding.
[IR-25] Quantum-Inspired Pairwise Ranking Aggregation via Active LLM Feedback
链接: https://arxiv.org/abs/2610.38683
作者: Quan Mai,Susan Gauch,Gibson Nkhata
类目: Information Retrieval (cs.IR); Computation and Language (cs.CL)
备注:
Abstract:Ranking aggregation from multiple retrievers is a powerful yet under-explored technique in modern retrieval and RAG systems. We introduce LLM-IRA, the first extension of the quantum-inspired interactive ranking aggregation framework to document reranking. By leveraging large language models (LLMs) solely for pairwise judgments on high-disagreement document pairs, our method learns fusion weights in a principled, interactive way. The final ranking is a simple, ultra-fast pointwise linear fusion of scores, no full pairwise matrix or expensive listwise prompting is required. Experiments on the BEIR benchmark and TREC Deep Learning tracks with an open-weight instruction-tuned judge (Llama-3.1-8B-Instruct) show that LLM-IRA matches or numerically exceeds strong static fusion baselines on 5 of 6 evaluated benchmarks. We report two weight-update variants: momentum blending and a full accumulated-history refit, which together reach statistical significance (Wilcoxon signed-rank, p0.05) over the best unsupervised aggregator on 4 of 6 benchmarks (DL19 under momentum; DL20, DBpedia, and NFCorpus under accumulated refit, which is the stronger of the two configurations overall). On the benchmarks we profiled (DL19, DL20), it already clears the best unsupervised baseline within as few as 5 LLM calls per query, well before exhausting its typical ~25–30 call budget. LLM-IRA also achieves cost-effectiveness comparable to state-of-the-art zero-shot LLM rerankers at an order of magnitude fewer LLM calls, and requires no training data or model fine-tuning.
[IR-26] CoWalk: Cooperative Walks for Agent ic Batch Queries on Disk-Resident Graphs
链接: https://arxiv.org/abs/2610.38404
作者: Qiao Liang,Zhiguang Chen,Yue Yu,Nong Xiao
类目: Databases (cs.DB); Information Retrieval (cs.IR)
备注:
Abstract:Workloads such as query rewriting, deep research and group-relative policy optimization (GRPO) issue batches of semantically related vector queries. Existing disk-resident approximate nearest neighbor (ANN) systems process these queries independently. On three workloads, 31 to 67 percent of the I/Os of a batch fetch a page that another query of the batch has already fetched. A page cache shared by the batch removes these I/Os, yet it makes a single-thread search only 1.1 to 1.2\times faster, because the search time goes to distance computation and the cache does not change the walk of any query. We present CoWalk, which runs the queries of a batch as cooperating walks on one thread. Every query keeps its own candidate list, a page is read once per batch, and the full-precision vector on every fetched page is ranked by every query of the batch. A query therefore reaches the target recall with a shorter candidate list, which computes fewer distances and issues fewer I/Os. At recall@10 of 0.95, CoWalk answers a batch on one thread 1.6 to 1.9\times faster than the fastest per-query baseline and issues 2.8 to 6.9\times fewer I/Os. On a server it serves 3.0 to 6.9\times the throughput of that baseline at a similar batch latency.
[IR-27] owards Reliable Scalable and Value-Aligned Generative Recommendation for Large-Scale Advertising
链接: https://arxiv.org/abs/2610.38378
作者: Yatong Xiao,Yibo Sheng,Lixiang Wang,Mingjie Sun,Yunhao Sha,Pengfei Zhang,Shengyu Wang,Peng Wang,Wenjin Wu,Peng Jiang
类目: Information Retrieval (cs.IR); Machine Learning (cs.LG)
备注: 9 pages, 3 figures, under review
Abstract:Generative recommendation (GR) reformulates item retrieval as autoregressive generation of discrete semantic identifiers (SIDs), offering a unified paradigm for personalized recommendation. GR4AD, as an industrial generative recommendation framework, has greatly improved advertising revenue while enabling efficient real-time serving in this http URL, continued deployment has exposed three limitations: deterministic SID assignment remains sensitive to semantic boundaries; conventional backbone scaling provides diminishing revenue gains under strict serving constraints; and sequence-level value optimization provides limited credit assignment to intermediate SID decisions, causing likelihood-based beam search to prematurely prune potentially high-value candidates. We present GR4AD-X, which addresses these limitations through reliable SID construction, serving-efficient capacity scaling, and unified value alignment. Specifically, GR4AD-X improves SID reliability through Multi-Path assignment. It expands model capacity along multiple complementary dimensions rather than relying solely on wider or deeper backbones, achieving a better effectiveness-efficiency trade-off under strict serving constraints. Finally, Unified Value-Guided Generation and Discrimination (UniVGD) aligns generation with commercial value through prefix-level reward optimization, value-guided beam pruning, and discriminative reranking within a unified model. Large-scale online A/B tests on the Kuaishou advertising platform demonstrate that GR4AD-X achieves a cumulative revenue improvement of 3.07% over the deployed GR4AD baseline while satisfying strict latency and throughput constraints.
[IR-28] Manifest-Assisted Retrieval in Artificial Intelligence Search: Evidence Gaps and a Security Evaluation
链接: https://arxiv.org/abs/2610.38361
作者: Gangzhen Qian,Kainan Zhou,Chuhong Xu,Hang Xiao,Zhi Jing,Rong Bao,Lu Yi,Zhaoyi Li
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Computational Engineering, Finance, and Science (cs.CE); Computation and Language (cs.CL); Databases (cs.DB); Information Retrieval (cs.IR)
备注: 8 pages, 3 figures, 5 table. Accepted at IEEE FLLM 2026. Author accepted manuscript. Reproducibility artifact: this https URL
Abstract:The external sources retained in our non-systematic audit did not isolate the incremental security or utility effect of consuming /llms.txt while ordinary retrieval remained fixed. The initial audit covered 39 sources through August 16, 2026, with later targeted updates through October 14. A purposive 15-tool study observed no automatic requests in 45 trials at one endpoint; a non-probability log study found that 97% of domains with detected files received no recorded requests in May 2026. We separate publication, fetch, model use, and action; define five operational threat paths and paired security-utility estimands; and exercise deterministic pre-model controls with 30 author-defined parser fixtures. We also execute a reduced pilot: 24 tasks across eight sites, five conditions, and 120 real model responses, with ordinary evidence held fixed. Frozen automatic scoring identifies attacker-target adoption in 21/24 raw adversarial responses, with one further response unresolved; the adversarial URLs-only condition has no known adoptions and one unresolved response. The pilot measures output changes conditional on supplied evidence, without live retrieval or external actions. Together, the framework and pilot distinguish manifest-content effects from publication, discovery, and downstream execution, while leaving full retrieval and agent evaluation open.
[IR-29] Predicting What Answer Engines Cite: Open Decision Models Rerankers and a Distilled 4B Student on Ten Engines under Controlled Retrieval
链接: https://arxiv.org/abs/2610.38090
作者: Elisha Bajemon,Andre-Louis Rochet
类目: Information Retrieval (cs.IR); Artificial Intelligence (cs.AI)
备注: 12 pages, 1 figure, 4 tables. Model: this https URL ; benchmark and code: this https URL
Abstract:Generative engine optimization needs a scorer that ranks candidate pages by the share of an answer they will receive as citations. We measure pretrained rerankers, a hosted typed decision model (Jev), eight open decision models and distilled students on GEO-Bench queries answered by ten engines that receive the same candidate sources in their prompt. Where several engines answer the same query, they agree with one another at a within-group Spearman correlation of 0.704. Jev and two open 27B decision models, used without training, are the strongest scorers at 0.575 to 0.578. A Qwen3-Reranker-4B fine-tuned with low-rank adapters on teacher labels for other GEO-Bench queries, released as GEO Citation Ranker 4B, reaches 0.542 to 0.549 across three seeds, and its differences with the 27B models and with Jev are not distinguishable from zero on 229 queries. Most of the remaining distance between the student and Jev disappears when Jev scores one passage at a time, as the student does. Students trained on teacher labels and on observed engine citations reach the same level, and labels from Jev alone match the two-teacher target at 0.6B. We release the model, the benchmark, the predictions of every reported scorer and the code.
[IR-30] Merge and Branch: Efficient Tree-Structured Supervised Fine-Tuning of LLM s for Recommendation
链接: https://arxiv.org/abs/2610.37973
作者: Yunqin Zhu,Yao Xie
类目: Information Retrieval (cs.IR); Artificial Intelligence (cs.AI)
备注: 12 pages, 8 figures
Abstract:LLM-based generative recommendation reads long, text-rich interaction histories but generates only short semantic identifiers (SIDs). This asymmetry makes supervised fine-tuning inefficient: encoding the history is expensive, yet each history usually provides supervision for only one target. We present Merge and Branch (MB) to reduce input cost and learn more from each encoded history. Merge compresses each historical item’s SID and text into a small number of learned tokens. Branch uses the compressed history to predict multiple target SIDs. To share history computation while keeping these targets independent, MB organizes each example as a shallow, wide tree. We develop tree FlashAttention to execute this layout efficiently, processing the shared history once and propagating gradients from every branch for end-to-end training. A tree-structured key/value cache extends history sharing to constrained decoding. Experiments on public benchmarks show that MB outperforms strong recommendation baselines while substantially improving training and decoding efficiency.
[IR-31] Statistical Detection of Positional Bias in LLM -Based Recommender Systems: Methodologies and Empirical Evidence
链接: https://arxiv.org/abs/2610.37962
作者: Carlos Peláez-González,Francisco Herrera,Andrés Herrera-Poyatos
类目: Information Retrieval (cs.IR); Artificial Intelligence (cs.AI)
备注: 34 pages, 8 figures, 18 tables
Abstract:Large Language Models (LLMs) are increasingly used as recommendation generators in conversational recommender systems (RS) via retrieval-augmented generation (RAG). This flexible design introduces new sources of structural bias. This paper studies positional bias in top-1 and single-turn LLM-based RS: systematic dependence of the recommended item on the order in which candidates are presented in the prompt. We propose two complementary detection methodologies: (i) POB-ICM, based on contingency matrix analysis and Chi-square independence testing, and (ii) POB-UNBP, a sample-efficient approach derived from a uniformity null model under permutation-based evaluation, enabling stochastic decoding with formal statistical guarantees. Using a controlled RAG testbed for product recommendation, built on a real product catalog, a fixed prompt template and 2000 template-generated user queries with 20 permutations each (40,000 instances), both methodologies reject the null hypothesis of positional independence and uniformity in all evaluated state-of-the-art models. We introduce three complementary metrics – permutation sensitivity, positional bias, and strong positional bias – to quantify this effect, the latter targeting cases where the position displaces the recommendation towards clearly less relevant items rather than only choosing among comparably relevant ones. A robustness analysis across candidate-set size and decoding temperature shows that the bias is structural and model-dependent, not an artifact of deterministic decoding or candidate-set size. These results establish positional bias as a measurable, reproducible property of prompt-mediated recommendation pipelines in the evaluated setting and motivate routine auditing and governance controls (e.g., order randomization, standardized prompting) for deployed LLM-based RS. Source code is publicly released.
[IR-32] What Constitutes Food Similarity? Visual Representations and Retrieval in Indian Cuisine
链接: https://arxiv.org/abs/2610.37867
作者: Saransh Kumar Gupta,Akshadha Sharma,Lipika Dey,Partha Pratim Das,Ramesh Jain
类目: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Information Retrieval (cs.IR)
备注: 17 pages, 7 figures, 4 tables, 31 references
Abstract:Food similarity underlies retrieval, recommendation, substitution, and other food computing tasks, yet food representations are typically evaluated by identity preservation. We ask what happens to food similarity beyond identity: how different visual representations organize foods into relational neighborhoods, what culinary relationships these neighborhoods reveal or obscure, and how consistently these structures emerge across representations. Indian cuisine provides a challenging setting in which visually similar foods may have different culinary identities, while foods within a culinary family can vary substantially in appearance. Using an expanded Khana dataset and seven pretrained visual representations spanning convolutional, transformer-based, self-supervised, and image-text aligned models, we evaluate image-to-image retrieval and image-to-dish neighborhoods, examining culinary-family structure, contrast cases, and cross-representation agreement. We also aggregate image representations to compare dish-level visual similarity across models and against recipe-derived similarity for the same dishes. Image-level neighborhoods vary substantially across representations, including similarities across culinary-family boundaries. At the dish level, aggregation reveals common visual structure despite local variation, but this structure corresponds less strongly with recipe-derived similarity. These findings show that recognition performance alone cannot characterize relational structure: food similarity depends on representation, modality, and abstraction level rather than being a fixed property of the food collection. Understanding these structures is therefore a prerequisite to combining multiple food information modalities into a coherent notion of similarity.
[IR-33] DeFT-RAG : Incrementally Decomposing Multi-Hop Questions into Self-Contained Sub-Questions with Reusable Answers
链接: https://arxiv.org/abs/2610.37826
作者: Dingyi Kang,Bingzhe Li
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Information Retrieval (cs.IR)
备注: 12 pages, 2 figures, 7 tables. Code: this https URL
Abstract:Retrieval-augmented generation (RAG) lets a large language model (LLM) answer questions by retrieving relevant passages from a large corpus. However, RAG struggles with multi-hop questions, which require chaining facts spread across several documents through intermediate entities that the question never mentions. Graph-based RAG methods retrieve once from the entities a question names and so rarely recover the full evidence chain. Iterative RAG methods recover it step by step, but their accuracy is limited by the many tasks they ask the LLM to perform, such as planning the sub-questions, deciding their dependencies, rewriting them as answers arrive, and keeping track of what is known. We propose DeFT-RAG, a RAG framework built on three components. First, DeFT-RAG generates one self-contained sub-question at a time from the original question and the sub-questions answered so far with their answers (QA pairs). Each sub-question names every entity it asks about, which removes the need to plan, track dependencies or rewrite, and lets the LLM focus on a single fine-grained task at a time. Second, DeFT-RAG moves the remaining LLM work outside the LLM: rule-based updates to a QA table that keeps track of what is known, and yes/no decisions made by a fine-tuned lightweight decision model. Third, DeFT-RAG caches the QA pairs and reuses an answer when a later question asks a sub-question with the same meaning. On MuSiQue, 2WikiMultiHopQA and HotpotQA, DeFT-RAG achieves the best accuracy among thirteen baselines with both gpt-4o-mini and Qwen3.6 as the backbone, while giving the LLM fewer input tokens than any other iterative method. With a QA cache shared across 1,000 MuSiQue questions, 23.1% of sub-questions are answered from the cache, which reduces LLM calls by 12.9% without lowering accuracy.
[IR-34] Beyond Similarity Search: Knowledge Graph Metadata Enrichment for Enterprise Retrieval-Augmented Generation
链接: https://arxiv.org/abs/2610.37728
作者: Fakabbir Amin
类目: Information Retrieval (cs.IR); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:A key challenge of RAG systems is retrieving information from enterprise documents created by multiple teams over time, where it is not possible to enrich the information using the latest information and metadata available outside the scope of organizational documentation responsibility. We investigated whether using a knowledge graph to store metadata and frequently changing information would improve upon knowledge retrieval over similarity search over embeddings. To perform the experiment, we created a snapshot of the SEC EDGAR dataset for a quarter, comprising 167,050 filings from 10,511 companies and 508,714 embedded chunks of risk-factor text. To simulate the enrichment, we created a knowledge graph of 430,257 nodes and 561,144 edges carrying an accession number on every edge. To evaluate the gain obtained via a graph-based design, we evaluated three architectures: text-to-SQL over typed columns, dense-vector RAG with HNSW cosine search, and a two-stage GraphRAG design. All architectures share the same model and temperature to allow comparison. Results are scored programmatically by an LLM as judge over 20 questions divided across four categories, each question asked three times. GraphRAG solved 16 of 20 questions on all three attempts, compared to 10 for text-to-SQL and 3 for vector RAG. GraphRAG lost on structured aggregation queries. Vector RAG lost the single-document category it was expected to win, passing 3 of 15 where GraphRAG passed 15 of 15, because company attribution is not recoverable from cosine similarity at this scale. All six confident falsehoods in the benchmark were text-to-SQL conflating distinct people who share a surname, the failure that resolving identity to an identifier before reading any text prevents structurally; GraphRAG made none. We claim no advantage beyond this corpus, this model and this question set.
[IR-35] Relevance-Guided Residual Calibration for Spectral Compression of Dense Retrieval Embeddings
链接: https://arxiv.org/abs/2610.37707
作者: Yongkang Li,Panagiotis Eustratiadis,Evangelos Kanoulas
类目: Information Retrieval (cs.IR); Computation and Language (cs.CL)
备注: Under Review
Abstract:Spectral compression reduces the storage and scoring costs of dense retrieval by projecting embeddings onto a smaller principal subspace. However, its corpus-derived weighting rules tie all retained directions to a shared spectral trend, without directly accounting for their individual contributions to relevance. To address this, we propose Relevance-Guided Residual Calibration (RRC), a closed-form method for refining spectral weights using relevance judgments. RRC estimates margin-based weights from the means and variances of positive–negative score contributions along each retained direction. It retains directions with positive estimated weights, decomposes their log weights into a fitted power-law trend and a residual, and adds the residual to the compressor’s log weights on this support. This allows relevance judgments to adjust individual directions while preserving the compressor’s overall spectral trend. RRC applies to PCA, Whitening, and SpecTemp, and incorporates the corrections into the existing transformation shared by documents and queries, without iterative model training or an additional inference stage. Experiments with three encoders on nine datasets covering conventional and reasoning-intensive retrieval show that SpecTemp + RRC achieves the highest mean effectiveness among tested compressors on the BEIR group and BRIGHT at every target dimension. On the five evaluated BRIGHT datasets, it improves over SpecTemp by an average of 2.34 nDCG@10 points at 768 dimensions. The gains also persist under scalar and product quantization, improving the trade-off between per-document storage and retrieval effectiveness. Our code is available at: this https URL
[IR-36] DuQUAD: A Dual-View Framework with Quality-Aware Evidence Pruning for Multi-Document Question Answering
链接: https://arxiv.org/abs/2610.37701
作者: Yeonjun Choi,Seyeon Jeong,Jong Wook Kim,Beakcheol Jang
类目: Information Retrieval (cs.IR); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Large language models with retrieval-augmented generation systematically underutilize evidence in the middle of multi-document contexts, a phenomenon known as Lost in the Middle. Existing mitigation strategies, such as document reordering and attention steering, can reduce boundary bias but remain brittle in noisy settings, where inaccurate relevance estimation and spurious documents still obscure useful evidence in central positions. We propose DuQUAD, a multi-document QA framework that mitigates positional bias through dual-view reasoning and quality-aware evidence pruning. DuQUAD combines a Local Agent guided by a Structural Fusion Score with a Global Agent that reasons over the full document set, providing both targeted evidence recovery and broad contextual understanding. The framework further filters candidate evidence using explicit quality scoring and refines it at the sentence level to reduce noise. DuQUAD consistently outperforms strong Lost-in-the-Middle mitigation baselines, including hybrid retrieval and reranking, multi-agent reasoning, and context optimization methods, improving accuracy by 3.0% on average over the strongest competing baseline and gold evidence recall by 9.4%, while substantially reducing computational cost compared to existing multi-agent frameworks.
[IR-37] chATLAS: A Retrieval-Augmented Conversational Assistant for the ATLAS Experiment
链接: https://arxiv.org/abs/2610.37697
作者: Joe Egan,Ben Elliot,Samuel Van Stroud,Alex Sopio,Ki Ryeong Park,Daniele Dal Santo,Zhaohui Li,Runze Li,Jeremy Couthures,Joe George,Cary Randazzo,Juerg Beringer,Daniel Murnane,Gabriel Facini
类目: Information Retrieval (cs.IR); High Energy Physics - Experiment (hep-ex)
备注: 20 pages, 6 figures
Abstract:Large scientific collaborations produce extensive internal documentation spread across heterogeneous systems, making it difficult for collaborators to locate and reuse existing knowledge. We present chATLAS, a deployed Retrieval-Augmented Generation (RAG) assistant developed for the ATLAS experiment at CERN. The system provides conversational access to more than 30,000 documents from several collaboration resources, material largely unavailable to general-purpose language models, using hybrid dense–sparse retrieval deployed on CERN infrastructure. A domain-specific benchmark built from automatically generated question–answer pairs is used to evaluate retrieval and generation performance and to guide design choices. We study the selection and domain-specific fine-tuning of an embedding model, contextual retrieval, ranking strategies, and generator choice. Fine-tuning combined with contextual retrieval more than doubles how often the correct document is returned as the top result on the most technically challenging documentation source. The deployed system also includes intent routing, safety handling, Mattermost integration, and an MCP server for future agentic workflows. chATLAS demonstrates that domain-adapted RAG can provide source-attributed, domain-specific access to collaboration knowledge, and offers a practical foundation for scientific collaborations and other organisations that rely on accurate, up-to-date retrieval from proprietary documentation.
[IR-38] After One-Click Video: The Inflow of Service-Type Content the Failure of the Recommendation Pool Mechanism and the BoLe System
链接: https://arxiv.org/abs/2610.37695
作者: Bo Zou,Chao Xu
类目: Information Retrieval (cs.IR); Computers and Society (cs.CY)
备注: 13 pages, 2 figures
Abstract:In September 2026, a new large language model was released. Soon after, a new way of making explainer videos became very popular: people asked the model to write a program, and the program produced the video. “One-click video” content then appeared in large amounts on many platforms. On the surface, this is a rise in creative power at the level of the individual. We argue that, at a deeper level, it changes the structure of the content industry. This judgment rests on two observations. First, this wave of content is service-type content. People watch it to meet a need, not to enjoy the content itself. The other kind is work-type content. AI makes both kinds faster to produce. But anyone can come in and make service-type content, while work-type content is limited by the number of people who have something to say. So the ratio between the two slowly becomes unbalanced. Second, free platforms choose content through layers of trial viewing, and this uses up user attention. The total amount of attention is fixed. The unbalanced ratio also dilutes good content. A simplified mathematical model shows that, as long as these conditions hold, the recommendation pool mechanism will fail once new content grows past a certain multiple. Based on this, we discuss one possible way out. Machines and entry rules judge whether service-type content passes the bar. Trusted people judge work-type content. We call this the BoLe system. At the end, we briefly imagine how the two kinds of content may develop from here.
人机交互
[HC-0] Micro-Edits: Computational Support for Fine-Scale Edits in Film MICRO
链接: https://arxiv.org/abs/2610.40653
作者: Nhan Tran,Hijung Valentina Shin,Abe Davis,Mackenzie Leake
类目: Human-Computer Interaction (cs.HC); Computer Vision and Pattern Recognition (cs.CV)
备注: UIST 2026 Best Paper Honorable Mention. Project page: this https URL
Abstract:In film editing, every frame matters. While macro, content-level edits shape the high-level narrative of a film, small local edits often control timing and direct visual attention. For example, editors can adjust the precise timing of clips to control the length of a meaningful silence, or cut to shots that highlight a specific character’s reactions to important moments. These small, detailed edits are slow and tedious to make in standard editing tools because users have to map high-level narrative goals to low-level frame-by-frame adjustments. At the same time, these fine-scale adjustments, which we call micro-edits, can be nuanced and expressive, with different choices serving different creative goals. In this work, we start by conducting a formative study with professional editors to understand the structure of micro-edits and how they are used. We then derive a mathematical formalization for this structure to develop tools that leverage AI and optimization to make exploring micro-edits simpler and more efficient for users. Finally, we validate our tools in multiple user studies and a wide range of video examples.
[HC-1] When to Assist and What to Suggest: Joint Gating and Memory-Grounded Hint Generation in One Compact Decoder
链接: https://arxiv.org/abs/2610.40611
作者: A. J. A. Jubayer Talukder,Mohammad Ariful Haque
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Human-Computer Interaction (cs.HC)
备注: 20 pages, 10 figures, 13 tables; supplementary material included (Sections S1-S8)
Abstract:Proactive assistants offer short, memory-grounded hints during a conversation, deciding when to assist and what to suggest. Current systems split this between a gate and a larger generator; we evaluate both together on 28,058 decision points. Stage-wise generation scores are oracle-gate ceilings: behind a real gate, end-to-end exact match is 0.2619 against a ceiling of 0.2971, and end-to-end precision is 0.2387, leaving substantial headroom. A gate pays off through precision and compute, whose saving falls from 4.39x with an 8B generator to 1.38x with a 1B one. The memory profile drives hint quality: removing it costs 24.7 exact-match points (rescaled BERTScore 0.57 to 0.35), while retrieval and a larger generator add nothing on exact match. One 1B decoder can do both, gating from an intermediate layer and generating in the same pass. At its final checkpoint it matches or exceeds the two-model pipelines on automatic end-to-end measures with 1.31-1.76x less wall-clock time than the matched 1B pipeline; across three seeds, its deeper exit ties the separate gate. A human evaluation (1,144 ratings over 32 dialogues) supports the memory effect and rates the single model level with the matched 1B pipeline and below the 8B one. Graded metrics and LlamaPIE’s native judge track human ratings more closely than exact match, which we report as a strict lower bound.
[HC-2] Deep Learning Based Behavioral Drift Modeling for Continuous Biometric Authentication
链接: https://arxiv.org/abs/2610.40398
作者: Suprava Saha Dibya,Minar,Abdulla Al Noman,Md. Zafir Abid Shresto,Ahmed Sayef
类目: Cryptography and Security (cs.CR); Human-Computer Interaction (cs.HC); Machine Learning (cs.LG)
备注:
Abstract:This research proposes a two-level keystroke dynamics authentication system that provides defense-in-depth security through continuous behavioral drift monitoring. Level 1 employs fixed-text password authentication using Temporal Convolutional Networks (TCN) with contrastive learning and population-based training (PBT), achieving 3.77% EER for high-security login verification. Upon successful authentication, Level 2 continuously monitors user typing during natural free-text sessions using a 4-Model Deep Learning Ensemble. If suspicious behavioral drift is detected during monitoring, the system re-authenticates the user by returning to Level 1, providing a feedback loop against session hijacking and behavioral anomalies. The system explicitly models long-term behavioral drift through the fixed-text component’s PBT mechanism, which adapts to temporal variations in password typing patterns. This research leverages free-text (Buffalo dataset) and fixed-text (GREYC dataset) keystroke dynamics datasets, extracting temporal and spatial features including hold times, up-down latencies, down-down intervals, and trigram patterns. The final system achieves 96%+ accuracy for fixed-text authentication (3.77% EER) and 83.53% True Acceptance Rate (TAR) at 5% False Acceptance Rate (FAR) for free-text continuous authentication (10.05% EER), demonstrating superior performance and strong resistance to impostor attacks through continuous behavioral monitoring.
[HC-3] Crowds Believe in You: Online Prediction Markets Overestimate How Often People Achieve Their Own Goals
链接: https://arxiv.org/abs/2610.40324
作者: Zihan Chen,Lei Nico Zheng,Di Zhu
类目: Human-Computer Interaction (cs.HC); Computers and Society (cs.CY); Social and Information Networks (cs.SI); General Economics (econ.GN)
备注: 23 pages, 2 figures, 14 tables; working paper
Abstract:People increasingly share personal goals online and invite others to forecast them. Research on the planning fallacy finds that observers predict a person’s behavior less optimistically than the person does, and prediction markets aggregate such judgments, harnessing the wisdom of crowds. We study 6,150 prediction markets on Manifold in which 1,660 users asked about their own futures (“Will I …?”), reconstruct every order and execution, and compare these self-markets with markets about external events. Self-markets resolved YES 4.4 percentage points less often than their final prices implied, whereas external markets were nearly calibrated (0.9 points); the difference was 2.2 points at the same price level and 1.9 points within creators who made both kinds of market. Absent when markets opened, the overstatement emerged within a day of trading, even where the person never traded, and peaked in the week before close. In a market’s first day, observers moved external-market prices down toward the base rate, but moved self-market prices 3.6 points further toward YES, except in markets naming an outcome the person would not want, which were not overpriced. The correction came late and largely from the person, whose orders supplied 38% of the eventual movement toward the outcome in self-markets, against 9% in external markets. Because self-market prices mainly failed to separate successes from failures, a recalibration learned from earlier markets improved later forecasts by only 1.5%. Crowds believe in the futures people pose for themselves, and the information that would correct them lies with the person.
[HC-4] owards fNIRS foundation models: A curated corpus of 67 public datasets with channel positions in MNI space
链接: https://arxiv.org/abs/2610.40277
作者: Santiago Posso-Murillo
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:Brain foundation models have transformed medical artificial intelligence and computational neuroscience by reducing the reliance on task-specific data through large-scale pretraining. While they have been successful with non-invasive neuroimaging data, their adoption in functional near-infrared spectroscopy (fNIRS) has lagged due to the limited size and heterogeneity of public datasets, which are collected under varying protocols, devices, and sensor configurations. In addition, no single fNIRS dataset is comparable in size to the largest datasets in functional magnetic resonance imaging (fMRI), electroencephalography (EEG), or magnetoencephalography (MEG). The fNIRS datasets are also spread across different platforms and often have insufficient information, or lack the protocols that would make them easier to use. Given the advances in EEG foundation models that enable flexible modeling across diverse EEG montages, we curated 1,955 hours of fNIRS data from 67 public datasets spanning 2,798 participants, representing the largest fNIRS curation effort to date. The curated data include oxygenated (HbO) and deoxygenated (HbR) hemoglobin signals along with the three-dimensional position of every channel in the standard space of the Montreal Neurological Institute (MNI), which makes positions comparable across datasets. We release dataloaders that handle each dataset automatically, along with the minimal preprocessing pipeline used to prepare the data, at this https URL. This work aims to facilitate the future development of foundation models for fNIRS.
[HC-5] EnvOS: Auditing Computer-Use Agents When the World Moves On : Phenomenon-Based Task Construction Certification and Evidence-Based Evaluation NEURIPS2026
链接: https://arxiv.org/abs/2610.40246
作者: Ankush Kadu,Athira Gopal,Niharika Singh,Arpit Jain,Aswanth Krishnan
类目: Artificial Intelligence (cs.AI); Human-Computer Interaction (cs.HC)
备注: 74 pages, 7 figures. Accepted at NeurIPS 2026 Workshops (AABA4ET and CL4FMAgents)
Abstract:Computer-use agents built on large language models (LLMs) work in software that can change while they act: an appointment moves, a policy is revised, a record is updated. Testing agents under such changes is harder than it looks. When an agent fails, the cause may be the agent, but it may also be a change that did not fire as intended, an update the agent never received, or a grader that scored the episode wrongly. We introduce EnvOS, a framework for building and certifying environments that change mid-task. Before any agent runs, EnvOS checks that each change fires correctly, that the task remains solvable, and that the grader separates correct from incorrect behavior; after each run, separate code re-checks the verdict from archived evidence. We apply it to stale information, where a value the agent has already seen changes before the agent acts on it. In clinic booking, Claude Opus 5 succeeds on 0/16 tasks when nothing extra is shown, 2/16 when reminded to re-check the source, and 15/16 when the updated value is shown; the gain over the reminder holds after correcting for all tests (p = 0.00293). In insurance-claim approval, Opus 5 re-checks on its own and succeeds on 10/10, whereas Claude Sonnet 5 rises from 1/10 to 9/10 when the current policy is shown again, a gain significant in its pre-registered test but not after correcting for all tests. Whether an agent caught a change thus varied with the model and the interface. We also stress-test the grading itself: on corrupted episodes designed blind from the rules, the EnvOS pipeline accepted none of 26 as clean, and on an externally generated set it caught 25 of 39, against 4 for simple log checks. Because every EnvOS environment is generated from a seed and graded by an audited automatic verifier, the same environments are natural candidates for training agents with reinforcement learning from verifiable rewards.
[HC-6] “It Feels Weird to Just Rely on AI”: How Experienced Healthcare Professionals Work With Decision Support Systems in Clinical Practice
链接: https://arxiv.org/abs/2610.40224
作者: Simon W.S. Fischer,Hanna Schraffenberger,Miranda L. van Hooff,Casper Reijnen,Serge Thill,Pim Haselager
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:Decision support systems (DSS) assist with clinical decision-making. Little is known, however, about the decision-making process with DSS in clinical practice and the extent to which clinicians consider or rely on the predictions. Studies on human-AI decision-making usually focus on laypeople, simplified decision tasks, or one-shot interactions with prototypes that are not integrated into existing workflows. To gain insights from real-world practice, we conducted an online survey (n=12) and semi-structured interviews (n=10) with clinicians from two different medical cohorts each working with a DSS. Findings show, amongst other things, that the interaction with DSS changes as experience increases, the predictions become more important in complex cases, that decisions are discussed with colleagues, and that workflow integration affects decision-making. Our findings on human, technical, contextual, and organisational factors contribute to a socio-technical perspective on DSS reliance behaviour. We conclude by discussing the implications of our findings for the design of DSS.
[HC-7] Apple Vision Pro: The good the bad the ugly
链接: https://arxiv.org/abs/2610.40108
作者: Santiago Berrezueta-Guzman,Andrei Koshelev,Habiba-Loai Khalil,Stefan Wagner
类目: Human-Computer Interaction (cs.HC)
备注: Pre-print submitted in Elsevier
Abstract:Apple Vision Pro (AVP), the headset Apple introduced as its first spatial computer,'' built by a single, vertically integrated vendor, entered the market in February 2024 and has since accumulated a fast-growing body of primary research spanning surgery, education, accessibility, design, and human-computer interaction. In this paper, we adopt the tripartite structure of Leone's 1966 film to organize a systematic review of this literature: we cataloged the validated, working applications of AVP (the good), the persistent technical and ergonomic limitations of a still-early-generation headset (the bad), and the ethical concerns and structural harms that follow from deliberate design, marketing, and deployment choices (the ugly). The review is built from a Scopus export of 218 records mentioning AVP, which we narrowed to 104 papers (2024--2026) with a primary focus on AVP, using explicit inclusion/exclusion criteria that we report in full, alongside a PRISMA-informed flow diagram of the screening process. We find the corpus heavily skewed toward validating benefits (medicine and surgery, education, and software tooling account for over half of the included papers) and comparatively thin on documented risks or misuse, a pattern we treat as a finding in its own right rather than a limitation to be explained away. We further show that several recurring issues, namely price, simulated social presence, and the gaze-tracking privacy architecture, resist a clean separation into bad’’ or ``ugly’’ using the published evidence alone, since both readings remain compatible with the same data. We close by identifying where the evidence base most needs to grow: independent risk and ethics research, bystander- and equity-focused study designs, and disclosure that would let researchers distinguish an engineering backlog from a structural choice.
[HC-8] Human–AI Interactions as First-Class Data: A Formal Interaction Model and Runtime for Explainable AI Collaboration
链接: https://arxiv.org/abs/2610.40006
作者: Michela Taufer,Kin Ng,Ian Lumsden,Aashish Pandey,Rajasi Gore,Angelina Ju,Sukanya Moudgalya,Ewa Deelman,Sai Swaminathan
类目: Artificial Intelligence (cs.AI); Human-Computer Interaction (cs.HC)
备注:
Abstract:AI assistants are increasingly embedded in computational notebooks, scientific workflows, software engineering tools, and learning systems, yet human-AI interactions are often treated as transient prompt-response exchanges rather than durable data products, hindering reproducibility, explainability, replay, behavioral analysis, and adaptive AI support. We argue that these interactions form a new data class requiring formal representation and introduce the Human-AI Interaction Description Language (HAIDL), a formal interaction model with a Backus-Naur Form-inspired concrete syntax. HAIDL encodes each interaction as a structured object capturing computational context, user request, prompt payload, AI response, feedback, problem context, provenance, and model metadata. We integrate this vision into HAIDL-X, a middleware for JupyterLab that instantiates HAIDL by capturing user-bot state, building contextualized prompts, interfacing with AI providers, collecting feedback, and storing interactions in a repository. We add a learning layer over HAIDL that maps interaction objects and notebook execution traces into reproducible measures of user behavior, including engagement, performance, help-seeking, error-oriented use, and learning personas. HAIDL-X provides a foundation for treating human-AI collaboration as structured, queryable, replayable, analyzable data.
[HC-9] Mind The Gap (MTG-AD): Dataset and Fine-grained Evaluation for End-to-End Inline Audio Descriptions NEURIPS2026
链接: https://arxiv.org/abs/2610.39955
作者: Subhashini Venugopalan,Jenny Tan,Taylor Roper,Jimmy Tobin,Anton Kast,Alicia Martin,Sam Sepah,Amy Pavel
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Human-Computer Interaction (cs.HC)
备注: NeurIPS 2026 Evaluations Datasets Track, camera ready
Abstract:Audio descriptions (AD) are essential for making visual media accessible to blind and low-vision (BLV) users. While Multimodal Large Language Models (MLLMs) offer a scalable solution for generating audio descriptions on-demand, their performance on ``inline’’ audio descriptions – which must fit within existing silences in a video – remains under-explored, particularly for diverse user-generated content (UGC). We present a comprehensive investigation into MLLM-generated inline audio descriptions. First, we introduce MTG-AD, a professionally annotated dataset of 36 hours of videos, each averaging \sim 5 mins. and 8k+ descriptions in total, co-created by paired blind and sighted AD professionals to provide a high-quality benchmark for this task. Second, we propose MTG-AD-Eval, an evaluation framework based on audio description expert guidelines that provides fine-grained, reference-free actionable metrics for model improvement. Crucially, our framework evaluates the end-to-end task: it renders the generated scripts into audio tracks to assess their fit within the original video’s natural silences, minimizing disruptive overlaps. Our analysis across frontier MLLMs reveals that while models accurately describe visual events and substantially reduce speech overlaps, gaps persist in narrative pacing and flow. Finally, we empirically validate that our automated and reference-free metrics correlate significantly with human quality ratings from sighted raters and 7 BLV participants, finding that while users narrowly prefer human-authored audio descriptions, they find MLLM-generated audio descriptions highly beneficial for comprehension. We provide actionable system design guidance and release MTG-AD at this https URL and MTG-AD-Eval at this https URL.
[HC-10] Inkitiate: Initiating and Steering Agent Assistance through Ink in a Spatial Workspace
链接: https://arxiv.org/abs/2610.39905
作者: Liangwei Wang,Nan Chen,Luna K. Qiu,Zilong Wang,Fugee Tsung,Yuyu Luo,Yuqing Yang
类目: Human-Computer Interaction (cs.HC)
备注: 18 pages, 11 figures
Abstract:Handwriting and sketching help people develop ideas around information. As this work unfolds, users may seek assistance with finding, understanding, and organizing materials. We investigate how ink can initiate and steer agent assistance in this process. An eight-participant formative study explored desired assistance and interaction expectations, informing three connected activities: bringing materials in, thinking with materials, and shaping the space. We present Inkitiate, a spatial workspace for ink-based agent assistance in knowledge work. Through ink, users retrieve external materials, build on accumulated content and results, and reorganize the canvas, while inspecting and guiding agent actions. In a study with eighteen participants, users reported favorable usability and low perceived workload, and used retained ink and materials to develop further ink queries. An ink-query vocabulary and cases reveal how users combine linguistic and graphical expressions to direct agents. These findings inform ink-based agent interaction design for expressive requests and continued work with materials.
[HC-11] Incorporating Machine-Others: An Enactive Exploration of Human-LLM Interactions
链接: https://arxiv.org/abs/2610.39869
作者: Elena C. Cuffari,Marta Pérez-Verdugo,Xabier E. Barandiaran
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI); Computers and Society (cs.CY)
备注:
Abstract:New concepts based in interactional thinking are required to understand how humans are changed by relating to LLMs. Standard theories of technological impact upon users’ autonomy are inadequate here, because this technology is capable of inducing linguistic interactions with humans in an unprecedented manner. Traditional theories of social cognition also fall short, because LLMs have neither autonomy nor minds nor sociality as such. And yet, humans interact, talk, and create with them. We base ourselves on enactive theory, particularly on the theory of Linguistic Bodies and recent developments of enactive accounts of AI, to explore how we make sense with machines. We argue that we incorporate machine-others in two ways: in the tool-like sense of incorporation as enabling particular skills, but also in the more linguistic and identity-tied sense of incorporating (and incarnating) LLMs’ ways of languaging. However, LLMs are themselves excorporated and excarnated phantoms of our own languaging; we give a (machinic) body to our language. As such, we are not just incorporating machinic others; we are incorporating excorporated others. These conceptual advances draw a new map of our linguistic relation to LLMs, putting them in continuation with other forms of languaging involving other linguistic artifacts (e.g., religious books). We attend to some particular forms that this relation takes today, and how they may be impacting the autonomy and becoming of LLM-engaged humans. We introduce the idea of “susceptive sense-making” to characterize increased semiotic labour that is carried by the lack of friction that we sometimes encounter within these services.
[HC-12] Assessing AI Companion Chatbot Safety for Vulnerable Users: A Crisis-Stage Multi-Platform Evaluation NEURIPS2026 ALT
链接: https://arxiv.org/abs/2610.39782
作者: Huanyue Liao,Seok Min Lim,Yan Ling Lim,Shi Min Cheng,Kellin Pelrine,Chen Hui Ong
类目: Human-Computer Interaction (cs.HC)
备注: Accepted at the GenAI4Health Workshop at NeurIPS 2026 (non-archival)
Abstract:AI companion chatbots are increasingly used by vulnerable individuals, including adolescents in suicidal crisis, for emotional support. Existing safety evaluations focus on base models, leaving gaps in product-level assessment, crisis-stage sensitivity, and comparisons between companion apps and general-purpose LLMs. We introduce MIRROR-Eval (Multi-turn Interaction Risk Review: Observing suicidal-crisis Responses), a dataset of 960 annotated 20-turn conversations across eight deployed platform configurations, six clinically grounded personas, and two crisis stages: ideation and explicit planning. An LLM-as-a-Judge pipeline applies an expert-reviewed rubric covering user risk severity, chatbot risk awareness, and intervention behavior. We observe a three-tier harm structure: companion apps (this http URL and Replika) and Grok 4.3 form a high-harm tier; Gemini 3.1 Pro is mid-range; GPT-5.5 and Claude Sonnet 5 show substantially lower harm. At planning, companion apps show increased glorification and normalisation, advancement toward harmful action, and harmful methods and instructions. Crisis recognition and empathetic language do not ensure safe responses. These findings highlight the need for product-level safety evaluation across crisis stages. Dataset and code: this https URL. Warning: this paper includes examples that may be offensive or harmful.
[HC-13] From Artifacts to Coordination: Exploring How Shared Canvases Drive Collaboration during Group Ideation
链接: https://arxiv.org/abs/2610.39765
作者: Saramsh Gautam,Andrew M. Webb,Mahmood Jasim
类目: Human-Computer Interaction (cs.HC)
备注: Accepted to ACM GROUP 2027. To appear in Proc. ACM Hum.-Comput. Interact. 11, GROUP (January 2027). 21 pages, 9 figures. DOI: https://doi.org/10.1145/3859405
Abstract:Recent years have seen a surge in tools that support collaborative ideation through shared canvases. While these canvases offer flexible workspaces, how collaborators leverage them to iteratively develop ideas by externalizing thoughts, organizing artifacts, and coordinating tasks remains underexplored. To address this gap, we conducted an exploratory study with seven triads planning a travel itinerary using an infinite-canvas technology probe. Drawing on participant interviews and video analysis of their canvas interactions, we observed that collaborators contributed in parallel and arranged artifacts freely without a predefined structure. This spatial organization allowed collaborators to coordinate implicitly, made less vocal participants’ contributions visible, helped distinguish private exploratory regions from shared working areas, and supported the emergence of distinct roles through repeated patterns of contribution. We discuss design implications for canvas-based ideation tools, including surfacing unnoticed contributions, supporting emergent roles without formalizing assignments, and keeping spatial cues legible for AI integration in collaborative canvases.
[HC-14] owards Audience-Driven Real-Time Data Storytelling
链接: https://arxiv.org/abs/2610.39658
作者: Angelica Lo Duca
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:This paper introduces the concept of Audience-Driven Real-Time Data Storytelling (ADRDS), a framework that utilizes participatory and social sensing to enable audiences to actively shape the presentation of data stories. At its core, ADRDS employs a pre-authored decision tree, where the presenter defines the data story through a branching three-act arc (setup, conflict, and resolution). Each node contains a question, branching rules, thresholds for coverage, consensus, and confidence intervals, as well as one or more potential calls to action (CTAs). Audience members connect to the ADRDS framework via a URL on their mobile devices and answer questions in real time. ADRDS evaluates responses against the decision tree in real-time, following strong branches when thresholds are satisfied or activating fallback paths when signals are inconclusive. ADRDS then selects the final action consistent with the traversed path and the audience’s demonstrated readiness, subject to fairness safeguards. ADRDS transforms presentations of data stories into audience-shaped decision processes, where live data actively drives the narrative. This work paves the way towards a new form of data storytelling, where the final data story depends on the audience’s background and interests.
[HC-15] Evaluating Conversational Explainability in Healthcare: Comparing the Explainability Assistant and a Traditional XAI Dashboard
链接: https://arxiv.org/abs/2610.39634
作者: Eduard Barbu,Rodion Krjutškov,Anneli Uusküla
类目: Human-Computer Interaction (cs.HC)
备注: 18 pages, 4 figures. Accepted at EXPLIMED 2026 (Third Workshop on Explainable Artificial Intelligence for the Medical Domain). To appear in CEUR Workshop Proceedings
Abstract:Conversational interfaces may make explainable AI (XAI) more accessible to users without machine learning expertise, but their benefits over dashboards remain insufficiently characterised in healthcare. Existing systems have typically been evaluated in isolation, leaving unclear whether conversational interaction improves interpretability beyond the underlying explanation methods. We report a within-subjects user study comparing the Explainability Assistant, an LLM function-calling conversational XAI system, with a traditional dashboard for explaining a heart disease classification model trained on the Cleveland Heart Disease dataset. Both interfaces expose the same Decision Tree classifier and post-hoc explanation methods: SHAP feature attribution, DiCE counterfactuals, and what-if analysis. This design supports comparison of interaction modalities while holding the model and explanation methods constant. Five participants with healthcare backgrounds, from medical students to practising clinicians and public health researchers, completed parallel task blocks on both interfaces; three participated in follow-up interviews. Task accuracy, Likert ratings, free-text questionnaire responses, and interviews revealed complementary strengths and weaknesses rather than a clear preference for either modality. The assistant offered easier entry and patient-specific explanations, while the dashboard supported simultaneous inspection of global and local information. All three interview participants recommended integrating conversational and dashboard elements within a single clinical workflow. Comments: 18 pages, 4 figures. Accepted at EXPLIMED 2026 (Third Workshop on Explainable Artificial Intelligence for the Medical Domain). To appear in CEUR Workshop Proceedings Subjects: Human-Computer Interaction (cs.HC) Cite as: arXiv:2610.39634 [cs.HC] (or arXiv:2610.39634v1 [cs.HC] for this version) https://doi.org/10.48550/arXiv.2610.39634 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[HC-16] Generative artificial intelligence prompting strategies among Chinese university students across academic tasks
链接: https://arxiv.org/abs/2610.39571
作者: Lei Fan,Fangxue Liu
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI)
备注:
Abstract:Generative artificial intelligence (GenAI) is increasingly embedded in university learning, yet effective human-GenAI interaction depends not only on access to these tools but also on students’ ability to formulate, evaluate, and refine prompts for academic tasks. This study examined university students’ perceived GenAI prompting capability and reported prompting strategies. Questionnaire data from 234 students were analyzed using descriptive statistics, measurement modelling, scenario comparisons, structural associations, and qualitative coding of optional comments. Students reported using GenAI across diverse academic activities and generally expressed confidence in their prompting capabilities, particularly in providing relevant context and refining prompts following initial outputs. However, prompting strategies showed limited sensitivity to task types: specifying output format was more strongly endorsed in a creative workshop preparation scenario than in a research-oriented assignment, whereas other matched strategies showed no adjusted differences. Students also commonly endorsed both concise instructions and providing all relevant information at the outset, suggesting that effective brevity may depend more on selecting and organizing task-relevant information than on prompt length alone. By contrast, fixed prompt templates received lower endorsement than other prompting strategies. In the adopted factor model, perceived cognition showed the clearest unique association with perceived prompting capability, while open-ended responses highlighted prompt-intent alignment as a common challenge.
[HC-17] Beyond Preference Following: Benchmarking Profile-Query Arbitration in Personalized Text Transformation
链接: https://arxiv.org/abs/2610.39550
作者: Tenghao Ji,Qichen Sun,Zhuoran Lu
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:Writing assistants use persistent preferences to personalize responses, but these can conflict with the current request. Responses can meet every requirement yet serve the task poorly, while task failures need not involve following a conflicting preference. We introduce Profile Conflict Arbitration (PCA-Bench), a benchmark of 64 rewriting tasks and 256 task-preference pairs. We construct these pairs by varying the profile preference for each fixed text and request, covering different relations between the preference and the request. We assess whether responses meet requirements, how well they achieve the request’s goals, and whether they follow preferences. Across three LLMs, we find that when preferences compete with task goals, a higher success rate in meeting both requirements and preferences does not mean better responses: it can come with a greater quality loss compared with responses without a profile. Similar rates of preference following do not imply the same ability to meet the current task’s requirements when preferences conflict with them.
[HC-18] InterA11y: Interactive Web Accessibility Error Detection with LLM Agents
链接: https://arxiv.org/abs/2610.39488
作者: Mingyuan Zhong,Mengqi Shi,Davin Win Kyi,Ajit Mallavarapu,Xia Chen,James Fogarty,Jacob O. Wobbrock
类目: Human-Computer Interaction (cs.HC)
备注: We release our code and data at this https URL
Abstract:Large Language Models (LLMs) have shown potential to expand the coverage of web accessibility evaluation. However, current LLM-based checkers neglect much of the interactive and dynamic web contexts, resulting in limited coverage, high false discovery rates, and inconsistent performance across models. We present InterA11y, an agentic accessibility evaluation system that combines rubrics and interactive tools to support LLM agents in evaluating accessibility effectively. The rubrics decompose WCAG success criteria (SC) into testable aspects and guide agents in collecting and interpreting evidence. The interactive tools inspect or operate a webpage deterministically to collect evidence, such as focus appearances and error messages. We developed these rubrics and tools on 682 curated test cases grounded in established accessibility guidelines that target 10 commonly violated SC. We evaluated InterA11y on 197 expert-annotated cases sampled from 54 real webpages. Results show InterA11y achieves higher recall than the strongest evaluated LLM-driven baseline (79.0% vs. 44.4%), while maintaining comparable precision to best rule-based checkers (93.1% vs. 90.3-97.6%), with little variability between models. Further, when equipped with our interactive tools, general-purpose web agents detect more accessibility errors with fewer false positives and less token usage than without these tools.
[HC-19] GraspPressure: Understanding Pressure Input With Busy Hands
链接: https://arxiv.org/abs/2610.39459
作者: Anand Kumar,Aditya Shekhar Nittala
类目: Human-Computer Interaction (cs.HC)
备注: To appear in the Proceedings of the ACM on Human-Computer Interaction (PACM HCI), presented at ACM Interactive Surfaces and Spaces (ISS 2026), Turin, Italy, November 23-26, 2026
Abstract:In many real-world scenarios, users interact with digital systems while their hands are partially occupied, such as when holding tools or everyday objects. Prior work on busy-hand interaction has focused on microgestures and motion-based techniques, which rely on finger articulation or relative movement. In situations where a stable grasp is required, such as handling fragile, heavy, or precision objects, such motion can be undesirable or impractical. Pressure-based input offers a complementary modality that enables interaction without changing finger pose, yet its use while holding objects has received limited empirical attention. To explore this, we conducted two within-subject experiments examining users’ ability to perform controlled pressure selections and analyzing finger-specific performance across different grasps and geometries. Results indicate that participants can reliably perform pressure-based selections in these conditions, achieving high success rates (92.7% on average) with low selection times (1.74s on average). Our findings characterize the potential of pressure input as a complementary microinteraction modality under grasp-constrained, busy-hand scenarios.
[HC-20] Lookover: Scaffolding Student Participation in Social Annotation
链接: https://arxiv.org/abs/2610.39363
作者: Yubo Zhou,Xinyi Xu,Zesen Zhao,Anhong Guo,Xinyue Chen
类目: Human-Computer Interaction (cs.HC)
备注: 3 pages, 1 figure. Poster at UIST Adjunct 2026, the 39th Annual ACM Symposium on User Interface Software and Technology, November 2-5, 2026, Detroit, MI, USA
Abstract:Social annotation platforms such as Perusall and Hypothesis are widely used to support student engagement with course readings, often through required commenting. Yet visible participation does not necessarily produce sustained or substantive peer exchange. Analyzing six weeks of Perusall activity from a graduate course (2,873 comments), we found that participation was often fragmented: 53.5% of comments received no reply, 38.3% were posted within the final 12 hours before the deadline, and over one-third of a double-coded sample went no further than repeating or agreeing with the reading or a peer. Interviews pointed to two linked difficulties: locating a discussion worth entering and contributing something that meaningfully advances it. We present Lookover, an AI-assisted annotation system that scaffolds both stages without writing for students. Discussion Overview helps students identify active themes and relevant threads, while Spark Thinking poses questions that encourage them to move beyond reaction and restatement toward questioning, extending, and substantively responding to peers. A preliminary study with five participants found Discussion Overview valuable for those who noticed it, while the usefulness of Spark Thinking depended on how much a participant already knew about the topic.
[HC-21] Wearable Biosensors for Early Detection and Continuous Monitoring of Postpartum Depression: A Human-Centered Approach
链接: https://arxiv.org/abs/2610.39191
作者: Emma Ricci-De Lucca,Elizabeth L. Murnane
类目: Human-Computer Interaction (cs.HC)
备注: 2024 PhysioCHI: Towards Best Practices for Integrating Physiological Signals in HCI
Abstract:Postpartum depression (PPD) is the leading cause of maternal mortality in the United States. However, there are significant gaps in PPD screening, detection, and treatment. A main challenge stems from limitations in existing methods for early PPD diagnosis and subsequent monitoring and intervention. Mobile technology that incorporates interactive interfaces and continuous wearable biosensors holds major potential to augment clinical decision-making and enhance PPD care. A number of compelling research directions exist in effectively developing such systems, including user experience design, sensing accuracy during perinatal stages, clinical integration and interoperability, acceptance and adoption, and privacy and ethics. By focusing on wearable devices that center women and their experiences, this paper discusses such challenges and opportunities at the intersections of HCI, inclusive physiological computing, and women’s health.
[HC-22] Unlocking with the Heart: Investigating User Perceptions of ECG-Based Biometric Authentication
链接: https://arxiv.org/abs/2610.39152
作者: Farnaz Farid(1),Linet Momposhi(2),Tamara Watson(1),Alana Maurushat(1),Abubakar Bello(3),Farhad Ahamed(1) ((1) Cybersecurity and Behaviour, School of Social Sciences, Western Sydney University (2) Graduate Cyber Security Analyst, Iglu Student Accommodation (3) School of Law Criminal Justice, Edge Hill University)
类目: Human-Computer Interaction (cs.HC); Cryptography and Security (cs.CR); Computers and Society (cs.CY); Emerging Technologies (cs.ET); Systems and Control (eess.SY)
备注: 28
Abstract:Password-based authentication remains challenging for many users, especially for those who face usability and accessibility issues. The burden of maintaining security can cause user fatigue, frustration and non-engagement. To address these issues, we conducted a pilot study to assess ECG-based biometric authentication as a passwordless alternative and to gather adult participants’ perspectives on this approach. The research consisted of a survey study, an authentication prototype and a follow-up usability experience questionnaire. Our approach combined an online survey of 337 adults examining authentication practices and perceptions, with a separate usability evaluation in which 49 participants interacted with a simulated ECG-authentication and password interface in a health-care application prototype, then completed a post-test usability test questionnaire. The survey examined how adults across age groups use and perceive various authentication techniques, their experiences with biometrics, their views on security and convenience, and their willingness to adopt ECG authentication. In parallel, the usability prototype and questionnaire captured participants’ responses after hands-on interaction with an ECG authentication prototype.
[HC-23] Usable String Matching by Humans through Overlaid Presentation
链接: https://arxiv.org/abs/2610.39084
作者: Enka Blanchard,Ted Selker,Alan T. Sherman,Levi Gabasova,Mehdi Bouhamidi,Ryan Robucci
类目: Cryptography and Security (cs.CR); Human-Computer Interaction (cs.HC)
备注: A shorter version of this paper (without the appendices) has been conditionally accepted to the International Journal of Human-Computer Studies (IJHCS)
Abstract:We explore the ability of humans to compare strings, an error-prone activity which occurs in critical contexts (e.g., bank wire transfers), where it is not always possible to trust a machine to help. We hypothesize that understanding human cognitive and perceptual limitations can improve how people verify whether two character strings are identical. We vary the strings’ relative positions and presentation formats to measure comparison speed and accuracy. We present results from two online experiments with 178 human subjects who compared pairs of strings to determine whether the two strings were identical. The strings varied by length (5 characters to small paragraphs), difference (matching versus making a random change or a hard-to-detect one), alignment (vertical versus horizontal), and presentation (Overlay versus Adjacent). Our experiments test whether, by encoding and memory tasks into perceptual tasks, overlay formats can offer advantages in speed and accuracy. Results show that overlay offers a substantially faster method, with improved accuracy for long strings. Nevertheless, mean error rates were at least 4 percent, indicating that even the improved approach may be insufficiently accurate for many verification tasks. Comments: A shorter version of this paper (without the appendices) has been conditionally accepted to the International Journal of Human-Computer Studies (IJHCS) Subjects: Cryptography and Security (cs.CR); Human-Computer Interaction (cs.HC) Cite as: arXiv:2610.39084 [cs.CR] (or arXiv:2610.39084v1 [cs.CR] for this version) https://doi.org/10.48550/arXiv.2610.39084 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Alan Sherman [view email] [v1] Thu, 29 Oct 2026 00:24:04 UTC (3,060 KB)
[HC-24] Exploring AI Peer in Collaborative Mathematics Learning
链接: https://arxiv.org/abs/2610.38990
作者: Yiran Lin,Ye Deng,Ziqi Pan,Yongan Yu,Alexandra I. Cristea
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:Artificial intelligence (AI) agents are increasingly explored as collaborative learning partners in mathematics education, yet it remains unclear what makes an AI recognizable as a peer. This study examines how domain experts perceive and interact with an adaptive LLM-based peer agent combining a Big Five-informed persona with motivational state modulation for collaborative mathematics problem solving. Through thematic analysis of expert interviews, we identify conditions shaping peer perceptions, including emotional safety, shared control, and collaborative reasoning. Participants primarily judged the AI through its social behavior rather than mathematical capabilities. The AI was perceived more as a peer when it created a non-judgmental space, engaged in tentative reasoning, displayed appropriate imperfection, and respected participants’ pace. Conversely, blind agreement and over-instruction reduced trust and shifted perceptions toward a tutor role. Based on these findings, we propose human-centered design dimensions for AI peer agents to guide future peer-oriented learning systems.
[HC-25] When Does GenAI Use Support Critical Thinking? A Cross-Cultural Structural Account of Transactive Memory with GenAI DATE
链接: https://arxiv.org/abs/2610.38910
作者: Mahir Akgun,Sacip Toker
类目: Human-Computer Interaction (cs.HC)
备注: Preliminary exploratory version. Results will be updated after a validation study and additional data collection
Abstract:Whether generative AI (GenAI) use helps or hinders critical thinking remains contested, and existing accounts emphasize exposure to the tool rather than how a person structures their engagement with it. Extending Transactive Memory Systems (TMS) theory to human-AI partnerships, we test a structural account of critical thinking in GenAI-assisted academic work among 891 college students in the United States and Türkiye. Self-appraised thinking ability and confidence in the partnership shaped how much practical experience a person built up with GenAI, but neither predicted critical thinking directly. That experience mattered only through two downstream behaviors, dividing and coordinating cognitive labor with the tool, and actively reworking its output rather than accepting it as given. A multigroup comparison found this structure largely invariant, with two theoretically interpretable exceptions. This study offers a structural, TMS-based account of critical thinking in GenAI-assisted work.
[HC-26] Evaluating Visual Realism of AI-Generated Videos with Eye Tracking
链接: https://arxiv.org/abs/2610.38825
作者: Danqing Shi,Lan Jiang,Jake Lyons,Miri Zilka
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Human evaluation of AI-generated videos has so far relied on explicit human feedback, such as judgments or ratings. While these signals capture the conclusions viewers reach, they fail to offer insights into the perception that leads to them. This paper investigates eye tracking as a complementary human evaluation signal for AI-generated videos. We conducted two Visual Turing Test (VTT) experiments with a total of 90 participants on videos generated by latest generative models, in which participants watched each video twice and then gave their feedback. Across the experiments, we collected 4,286 valid eye-tracking trials on 438 videos. Our analysis indicates that gaze patterns offer informative evaluation signals that align with explicit human feedback. Building on the findings, we propose the first computational method GazeEval to predict the perceived visual realism of AI-generated videos from eye-tracking data. It achieves 71.3% accuracy in prediction, providing evidence that eye tracking data carries informative signals about how people evaluate a video. We release the study data to support further research on eye tracking for evaluating AI-generated video.
[HC-27] How Does Perceived Empathy From AI Chatbots Shape Users Privacy Attitudes Behaviors and Needs?
链接: https://arxiv.org/abs/2610.38802
作者: Jiaxun Cao,Hiba Laabadli,Yuntong Xue,Jana Schaich Borg,Pardis Emami-Naeini
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:Prior research suggests that some dimensions of AI empathy may increase users’ disclosure to chatbots, posing growing privacy risks as chatbots become more capable of empathic interaction. Yet this work often conceptualizes AI empathy narrowly, leaving users’ own definitions and their broader privacy experiences unexplored. We interviewed 21 AI Chatbot users about what AI empathy meant to them and how it shaped their privacy attitudes, behaviors, and needs. Participants characterized empathy through five components–understanding, resonating, caring, expressing, and helping. Most participants viewed resonating and caring as essential components of human empathy but absent from AI empathy. Yet AI’s contextual, in-depth understanding and helpfulness still encouraged highly sensitive disclosure. This highlights that AI empathy can endanger users’ privacy without necessarily making AI seem human. Participants also desired varied forms and levels of empathy, each associated with distinct privacy needs. We offer actionable recommendations for designing privacy-enhancing mechanisms for empathic AI interactions.
[HC-28] Crossing the Chasm: Bridging Visual Augmentations and Designer Intent
链接: https://arxiv.org/abs/2610.38782
作者: Luke S. Snyder,Maureen C. Stone,Jeffrey Heer
类目: Human-Computer Interaction (cs.HC)
备注: 23 pages, 17 figures, 2 tables, appendix, IEEE TVCG 2026
Abstract:To direct attention and communicate context, visualization designers often employ augmentations such as annotations, animated transitions, and stylistic layouts. While graphical perception principles can helpfully prescribe effective low-level representations of data, they lack guidance for augmentations that service higher-level communication goals. To bridge this gap, we contribute a design space that frames designer intent as viewer-oriented cognitive behaviors, grounding communicative aims in actionable visualization design techniques, including both visual encodings and augmentations. This design space consists of common augmentation tactics (annotation, animation, stylized encodings, etc.) that implement design strategies (spotlighting, sequencing, association, etc.) to achieve higher-level design goals (observe, interpret, introspect), often simultaneously. We demonstrate the analytic and generative value of our design space with examples across varied designer objectives. We discuss how our contributions help align designer intent with reader takeaways, pave the way for readers to learn more effectively, and enable future (semi-)automated systems to support visualization designers in achieving their communication goals.
[HC-29] Idea Traces: Threading Accretive Content across Scientific Papers
链接: https://arxiv.org/abs/2610.38721
作者: Luke S. Snyder,Jeffrey Heer
类目: Human-Computer Interaction (cs.HC)
备注: 15 pages, 9 figures, 1 table
Abstract:We introduce idea traces, a novel thread-based representation of accretive content across scientific papers. Rather than capturing fixed themes from papers, idea traces surface similarities and deltas of faceted content that enable inspection of how ideas relate and evolve across the citation network. To extract traces, we contribute an LLM pipeline that deconstructs salient transitions between pairs of papers and collates the output into an idea dependency graph. We leverage idea traces in a visual system that supports idea-driven navigation starting from a seed paper, lifting the “plane” of synthesis from papers to ideas. To evaluate the usefulness and quality of generated ideas, we provide a case study that compares our approach with an LLM QA workflow, and interview 10 researchers who used our system and annotated 110 ideas. Experts found that our approach helped extract a useful space of connections that would have been difficult to make otherwise. Drawing on their feedback, we outline several promising directions for future research.
[HC-30] When AI performs what remains? An integrative review of generative AI and cognitive capital in individuals and organizations
链接: https://arxiv.org/abs/2610.38647
作者: Giuseppe Riva,Stefania La Rocca,Brenda K. Wiederhold,Andrea Gaggioli
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI)
备注: 16 pages, 1 figure, 4 table
Abstract:Generative artificial intelligence is rapidly entering knowledge work, and its short-term productivity benefits are increasingly documented. Less is known about what happens, over time, to the knowledge, skills and judgment that remain in people and organizations. This integrative narrative review addresses that question through the construct of cognitive capital: the accumulated, renewable and depletable stock of knowledge, cognitive and metacognitive capabilities, and practices that enables an individual or an organization to learn, interpret information, exercise judgment, solve problems and adapt. We distinguish individual and organizational cognitive capital from the capability of the human-AI system, so that gains in joint performance and losses in human capability can be observed at the same time. Drawing on experimental, field and conceptual work, we propose four mechanisms through which interaction with AI affects cognitive capital: substitution of practice, scaffolding and feedback, metacognitive calibration, and organizational aggregation. Seven principles summarize the current evidence, including that assisted performance differs from retained competence, that the same technology can support or displace learning depending on design and use, and that individual gains can coexist with reduced collective diversity. For each mechanism we derive testable propositions, and we discuss five design levers for organizations as proposals to be validated rather than established practice. We conclude that cognitive capital offers a multilevel frame for governing AI adoption and outline a research agenda centered on longitudinal, multilevel measurement.
[HC-31] ItemCompass: Real-Time Edge-AI Wearable Vision for Tag-Free Domestic Object Retrieval ISWC2026
链接: https://arxiv.org/abs/2610.38629
作者: Tsz Lok Cheung,Jeff Kai-Tai Tang,Anthony Kong,Zackary P. T. Sin
类目: Human-Computer Interaction (cs.HC)
备注: Accepted and presented at the ACM UbiComp / ISWC 2026 Workshop on Wearable Design in Games and Entertainment (WDGE), Shanghai, China
Abstract:Misplacing household objects disrupts daily routines, especially for elderly users. ItemCompass addresses this with a real-time, tag-free wearable AR system that builds a persistent digital twin of the home for natural language object retrieval. A custom eyewear node streams visual and inertial data to an edge server, where YOLOv11n detection, Siamese re-identification, and a Pedestrian Dead Reckoning engine log object locations as Spatial Anchors in InfluxDB. Voice queries are resolved by a local Faster-Whisper and Qwen2 LLM pipeline, delivering concise spoken and AR guidance on a heads-up display. Evaluation shows 0.40 m localization drift, 97.5% step precision, and mAP@0.5 of 0.874, with users rating Accuracy and Trust at 4.6/5.0. The system highlights always-on sensing, conversational spatial memory, and privacy-by-design for wearable computing.
[HC-32] he Artificial Personality Model: Theory Implementation Measurement and Experimental Validation in Conversational Agents
链接: https://arxiv.org/abs/2610.38608
作者: Alexander Dregger,Barbarella Petz,Erik Kuhn,Pia Volk,Andreas Oberweis,Stefan Hillmann,Jannik Strötgen
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI)
备注: 56 pages, 16 figures, 19 tables
Abstract:Artificial personality shapes user experience (UX) in human-machine interaction, yet coherent theoretical frameworks and validated measurement instruments remain limited. This paper refines and empirically evaluates the Artificial Personality Model (APM), a hybrid framework integrating human-centred and machine-specific personality dimensions. Across three experiments (N = 395), the model was operationalised and evaluated using the Artificial Personality Questionnaire (APQ). Factor analyses revealed that some human personality dimensions (e.g., agreeableness, conscientiousness) are partially transferable, whereas others (e.g., extraversion) did not emerge as distinct dimensions. Additional machine-specific dimensions were identified. The APQ provided encouraging initial psychometric evidence, while the human-centred BFI-2 showed poor reliability and an uninterpretable factor structure in Experiment 3. Experimental manipulations demonstrated significant effects of artificial personality on user satisfaction, continuous use intention, and social presence. The findings contribute an empirically refined framework and a novel instrument for measuring perceived artificial personality, offering practical guidance for designing conversational agents.
[HC-33] Influence of Agent Number and Behavior on Users Psychological Responses in VR Evacuation Training
链接: https://arxiv.org/abs/2610.38536
作者: Naoki Matsumura,Kensuke Sugihara,Tomoko Yonezawa
类目: Human-Computer Interaction (cs.HC); Multimedia (cs.MM); Robotics (cs.RO)
备注: The 2nd International Workshop on Trust and Blame in Multi-Person Multi-Agent Societies (TBSA_2nd) in conjunction with HAI 2026, 7 pages, 8 figures, 5 tables
Abstract:This study aims to build effective drill software by combining a virtual simulation of a disaster with multiple agents capable of expressing emotions. We evaluated how the number of agents and their emotional and behavioral actions during evacuation influence users in a virtual space. The results showed that users tended to act more confidently when many agents evacuated with them, and that the evacuation route chosen by the agents influenced the users’ decision-making. These findings indicate that users’ evacuation behavior is shaped not only by the environment itself but also by the collective behavior of surrounding agents, underscoring the need to consider how agent behavior affects users’ psychological states during training.
[HC-34] Generative AI as a Mediator in Creator Collaboration: Challenges Design Opportunities and Concerns
链接: https://arxiv.org/abs/2610.38488
作者: Hajun Kim,Jini Kim,Yunjae Josephine Choi
类目: Human-Computer Interaction (cs.HC)
备注: 37 pages. Published in Proceedings of the ACM on Human-Computer Interaction (PACMHCI), CSCW 2026
Abstract:As Generative AI (GenAI) expands beyond individual workflows and becomes increasingly integrated intocollaborative creative work, its potential to address the fundamental challenge of building common groundamong creators with diverse expertise remains underexplored. Specifically, we lack empirical understandingof how GenAI can mediate collaboration asymmetries among heterogeneous creators. To address this gap,we conducted a multi-method study with narrative writers (N = 10) and visual designers (N = 10) in thecontext of game development. Our findings demonstrate that GenAI goes beyond individual assistance toserve as a collaborative mediator by aligning creative intent, sustaining shared understanding, and mitigatinginterpersonal friction among creators. At the same time, we highlight concerns that GenAI may reducecommunication among creators, undermine mutual trust, and reshape interpersonal dynamics and collaborativeprocesses. Based on these insights, we identify design opportunities for GenAI tools that nurture collaborativesynergy while preserving the creativity and values of human work.
[HC-35] StreamStudio: Exploring Video Authoring with Streaming Autoregressive Generation
链接: https://arxiv.org/abs/2610.38467
作者: Boyu Li,Zhe Lin,Eli Shechtman,Ding Li,Hongbo Fu,Yuqian Zhou
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:We present StreamStudio, a prototype system that explores a new authoring paradigm for streaming video generation. Instead of entering a prompt and waiting for a complete video before making changes, creators can interact with the generation as it unfolds: steering what comes next, returning to earlier moments, and exploring different continuations. In this way, StreamStudio shifts creative control from post-generation revision to continuous interaction with the generative process. Our contribution centers on this authoring paradigm and the interaction mechanisms that make streaming generation directly manipulable. StreamStudio is designed as a plug-and-play interface that can work with compatible generation and editing models, allowing its interaction model to evolve alongside advances in underlying video models. As streaming models become faster and easier to control in the future, such an interface may support more flexible and controllable video creation, including richer edits and multimodal interventions during generation.
[HC-36] Decompose Dont Expand: Capability-Independent AI Systems by Construction
链接: https://arxiv.org/abs/2610.38144
作者: Tejaswi Gowda
类目: Artificial Intelligence (cs.AI); Human-Computer Interaction (cs.HC)
备注:
Abstract:A recurring design choice in building AI systems is whether, when a task is too hard for a model, to expand the model (make it larger, better-prompted, fine-tuned), or to decompose the task (extract the parts that admit a deterministic solution to a host, leaving the model only the irreducibly fuzzy residue). We make three claims. First, that many tasks commonly delegated whole to a model are in fact deterministic tasks with a small generative core. Second, and our central empirical result, that decomposing such a task yields a measurable and desirable property we call parity by construction: on the extracted sub-tasks the model-size performance curve becomes flat, because no model performs them at all. We measure this across nine models spanning a 494M-parameter local checkpoint to a frontier Opus-class model, a thirty-fold parameter range on the local family alone: once the host resolves the context, operation selection scores identically for all nine models and selector resolution for eight of nine, while the one task that does not decompose, labelling, remains sloped across the same range. Because the task leaves the model, a stock checkpoint suffices: no task-specific training and no fine-tuning. Third, that this is a general design discipline, not a domain trick: we show the same move recurring across physics setup, motion retargeting, and object-model editing, and in a Web-CLI tool family (same author, so not an independent replication). We give a method for locating the cut, the construction that makes a cut exist (a deterministic surface and a closed grammar over it), and a way to make the residue safe (resolve the deterministic context; verify the generative output), position the work against neuro-symbolic AI, tool-augmented language models, and generate-and-verify approaches, and are explicit about what is measured (one domain) versus illustrative (the rest).
[HC-37] LogicTree: Object-Oriented Writing for Structured Human-AI Academic Authoring
链接: https://arxiv.org/abs/2610.38136
作者: Jiajun Zhu,Xinyu Cheng,Wenshuo Zhang,Huamin Qu,Yanna Lin
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:Academic writing is non-linear, iterative, and structure-driven, requiring authors to develop fragmented ideas into arguments with local coherence and global consistency. However, current AI-assisted tools primarily exchange linear prompts and free-form text, forcing authors to reconstruct generated logic and repeatedly verify its alignment with their structural intentions. We propose object-oriented writing (OOW), an OOD-inspired representation and interaction paradigm in which persistent writing objects serve as shared, manipulable referents for authors and AI. We instantiate OOW in LogicTree, an Overleaf-integrated prototype that combines a hierarchical object representation with a structure-grounded writing assistance module for author-initiated generation and issue-grounded revision and a structural logical review module for inspecting local coherence and global consistency. A within-subjects study (N=12) provides initial evidence that, in controlled academic writing tasks, LogicTree helps users structure ideas, reduces workload, and improves measured logical quality compared with a conversational baseline.
[HC-38] Beauty Is Human: When People Love AI-Generated Art They Think a Human Made It
链接: https://arxiv.org/abs/2610.38113
作者: Zihan Chen,Di Zhu,Lei Nico Zheng
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI); Computers and Society (cs.CY)
备注: 22 pages, 4 figures, 12 tables. Working paper
Abstract:Generative AI now produces images and poems that people cannot reliably distinguish from human-made work, so people often judge for themselves who made a work. We examine how these perceived-authorship judgments relate to aesthetic preference. In a public online AI Art Turing Test, 11,123 people judged whether each of 50 pictures (26 AI-generated) was made by a human and were asked for their favourite. Among the 9,152 respondents who judged their favourite, a picture was 22.1 percentage points more likely to be judged human-made when it was the respondent’s favourite, holding respondent and picture fixed. The association was similar for AI-generated and human-made favourites, was positive for all 50 pictures, and appeared under every reported detection strategy. Respondents most negative about AI art chose fewer AI-generated favourites (47% against 60% among the most positive) but more often believed them human-made, partly because others also judged those pictures human. As a result, in every opinion group a third to two fifths of respondents named an AI-generated favourite they believed human-made. A reanalysis of a published experiment on AI-generated poetry, in which 230 people rated each poem before judging its authorship, found the same pattern, although the AI poems were rated higher. When provenance is hidden, liking and perceived human authorship travel together, consistent with aesthetic response serving as evidence of a human maker and with beliefs about the maker shaping enjoyment. The findings qualify research on bias against AI art, which typically uses explicit labels, and inform AI disclosure policy.
[HC-39] Infrastructure and its Discontents: Contestations of Power in Canadas AI Governance Scaffold
链接: https://arxiv.org/abs/2610.38077
作者: Christelle Tessono,Nusrat Jahan Mim,Shion Guha
类目: Human-Computer Interaction (cs.HC); Computers and Society (cs.CY)
备注:
Abstract:This paper develops a critical theoretical framework for understanding national Artificial Intelligence (AI) strategies through the lens of “Governance as Infrastructure.” Integrating insights from Infrastructure Studies, Science and Technology Studies (STS), this paper employs Black Feminist Technoscience (BFTS) to reveal how AI governance infrastructures are assembled, maintained, and contested, and whose forms of knowledge, labour, and experience are rendered visible or invisible in these processes. We analyze the Canadian AI governance landscape through in-depth interviews with bureaucrats, civil society advocacy groups, academic researchers, and industry actors. Our inductive thematic analysis reveals how AI governance infrastructure enacts specific configurations of power, particularly through the politics of expertise; how actors navigate profound ambiguities related to AI’s opacity, often involving strategic ignorance; and how mechanisms for justice and accountability struggle to address AI-related harms. We contend that a BFTS perspective highlights the limitations of current governance approaches and points toward alternative infrastructural designs prioritizing care, maintenance, participatory democracy, and adaptive temporalities to foster more equitable and accountable technology governance.
[HC-40] Ignore the Writing: A Proposal for Peer Review in the Age of AI
链接: https://arxiv.org/abs/2610.38062
作者: Vassilis Kostakos
类目: Human-Computer Interaction (cs.HC); Digital Libraries (cs.DL)
备注: 4 pages
Abstract:The written paper has long stood in as proof of work: a fluent, well-structured manuscript was taken as evidence that good research lay behind it. Generative AI has broken that link. Polished manuscripts now cost nothing to produce, submissions are outgrowing what reviewers can absorb, and venues are turning to AI for help. The logical endpoint is a loop where AI writes the papers and AI reviews them. We argue that this loop is survivable, but only if it is given a reward function, and that human reviewing effort should become that reward function. We propose that submissions should consist of a 1500-word brief plus an unlimited, machine-verifiable appendix. Human reviewers judge whether a pursuit is worthwhile, models verify the experimental record in the appendix, and human curation decides where the field goes next.
[HC-41] Making Sense of Audience Change: Causal Attribution and Boundary Management on Xiaohongshu
链接: https://arxiv.org/abs/2610.38026
作者: Zhiyi Zhou,Kyrie Zhou
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI); Computers and Society (cs.CY)
备注:
Abstract:Recommendation-driven platforms can reshape not only what users see, but also who becomes visible and how community boundaries are negotiated. We examine Xiaohongshu discussions surrounding a perceived World Cup-related audience influx to understand how users attributed community change and how those explanations related to boundary-management strategies. We analyzed 262 comments from eight high-engagement threads using researcher-led thematic analysis, structured coding, and permutation-based association analysis. Users described personalization as both social separation and a boundary that could fail, interpreted platform growth through audience acquisition and commercialization, and contested whether gender should define community membership. Among 70 comments with explicit attributions and strategies, attribution and strategy were strongly associated (Cramer’s V = .583): algorithmic attributions concentrated in technical boundary management, while social-actor attributions more often aligned with social enforcement or normative socialization. These findings connect causal sensemaking with boundary management and show how audience restructuring can create governance consequences on recommendation-driven platforms.
[HC-42] Making Smart Homes Fit for Autistic Children: Family Work and Monitored Independence
链接: https://arxiv.org/abs/2610.38008
作者: Tingting Xiang,Weihan Jiang,Hrishitva Patel,Haiqi Shen,Tianzhi He,Kyrie Zhou
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI)
备注:
Abstract:General-purpose smart-home technologies are increasingly present in family life, yet how they support or disrupt autistic children and families with autistic children remains underexamined. We conducted a multi-source qualitative study drawing on five caregiver interviews, Xiaohongshu (also known internationally as RedNote) posts explicitly situated in autism contexts, and contextual posts from Xiaohongshu and Weibo. We find that families create support for autistic children by adapting environmental outputs, prompts, and interaction channels, while sensory intensity, communicative rules, and opaque operating states can turn the same technologies into barriers. Support also depends on continuing family work, including translation, configuration, modification, maintenance, and boundary-setting. Remote monitoring enables conditional independence while generating tensions among safety, privacy, and control. We contribute three concepts, including relational fit, family work as support infrastructure, and monitored independence, for understanding accessible domestic technology. We derive design implications for adjustable interaction, legible and recoverable failure, lower-burden configuration, and negotiable monitoring.
[HC-43] “It Would Make a Geordie Feel at Home”: Regional Accent Perceptions and Speech Accommodation in Voice AI
链接: https://arxiv.org/abs/2610.38005
作者: Dana Serditova,Kevin Tang,Jochen Steffens
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 50 pages, 5 figures
Abstract:Automatic speech recognition (ASR) systems often perform less accurately for speakers of regional dialects, but less is known about how users respond to these limitations and how regional voices produced by voice assistants (VAs) are evaluated through a sociolinguistic lens. This study investigates these issues among speakers of Newcastle English (Geordies), examining both linguistic accommodation to voice technologies and perceptions of regional versus mainstream-accented VA voices. Participants interacted with a simulated VA that spoke either Newcastle English or Standard Southern British English (SSBE). They subsequently completed semi-structured interviews about their experiences with the experimental VA and voice technologies, language use, accent attitudes, and place orientation. The findings reveal a two-way process of sociolinguistic bias. The Geordie VA was associated with familiarity, warmth, personality, and local belonging, but also with concerns about professionalism and authority, whereas the SSBE voice was perceived as more neutral, clear, and credible. Participants also reported adapting their speech when voice technologies failed to recognise them, most commonly by shifting to a more mainstream accent, enunciating more clearly, and speaking more slowly. These strategies are similar to accommodation in human-human interaction and reflect users’ perceptions of what technologies can recognise, also known as perceived intelligibility pressures. Overall, the findings demonstrate that voice technologies can reproduce existing accent biases both by encouraging users to suppress regional features and by activating established social stereotypes through synthetic voices. The study positions sociolinguistics as a bridge between understanding human-AI interaction and designing inclusive technologies around users’ own linguistic priorities.
[HC-44] he Collaborative Critique: An Inspection Method for Expert Evaluation of User Interfaces
链接: https://arxiv.org/abs/2610.37946
作者: Wendy Lucas,Tamara Babaian
类目: Human-Computer Interaction (cs.HC)
备注: 33 pages, 5 figures, 3 tables
Abstract:The development of the collaborative critique (CC) usability inspection method was driven by the need for systems, ranging from the simple to the complex, to lessen the workload on their users by being better collaborative partners. In the collaborative view of system-user interaction, the role of the system is expanded to that of a partner that must help its users achieve their goals. This perspective shifts the dynamic from focusing on the user’s ability to make effective and efficient use of the system to the system’s usefulness within a given task, user, and interaction context, thereby reducing the user’s cognitive and physical efforts. The CC method consists of a walkthrough evaluation in which the questions answered at each step of the interaction assess the collaborative strength of the system. This method is further distinguished by the inclusion of questions that access the behavior of the system in error situations, which are known to be particularly burdensome to users of complex systems. In this paper, we provide a detailed description of the CC method and its application and demonstrate its usefulness in evaluating an interface. The results of a laboratory evaluation show that the method performed well in identifying the problems experienced by users across all severity levels and predicted almost all of the problems associated with error situations.
[HC-45] What Is the Role for Humans in Creative Writing? Creativity and Diversity in Human-AI Collaboration
链接: https://arxiv.org/abs/2610.37916
作者: Chenyi Li,Haoyu Hu,Raja Marjieh,Katherine M Collins,Max Kreminski,Ilia Sucholutsky,Dalton Conley,Nori Jacoby
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:As AI agents increasingly participate in creative teamwork, writing emerges from hybrid human-AI teams. Although AI can improve the creativity of individual outputs, these gains may come at the cost of homogenization. What is the best role for human intervention? We introduce a transmission-chain paradigm inspired by cognitive models of writing, with four stages: Plan, Write, Review, and Revise. Each stage receives the previous output and passes its own forward. Using data from 333 participants across 427 stories, we compared creativity, diversity, and enjoyment across human, AI, and hybrid chains. Consistent with prior work, all-AI chains produced more creative but less diverse stories than all-human chains. Introducing a single human into an otherwise all-AI chain increased diversity depending on role. Human planning provided the best balance: it preserved AI-level creativity, increased diversity that persisted through downstream AI stages, and maintained enjoyment. We discuss implications for designing human-in-the-loop creative pipelines.
[HC-46] Advisor Co-pilot or Guardian? AI Role Framing and Human Decision Behavior in Automated Vehicle Takeovers
链接: https://arxiv.org/abs/2610.37892
作者: Wei-Hsiang Lo,Jingyi Xie,Gaojian Huang
类目: Human-Computer Interaction (cs.HC)
备注: 19 pages, 9 figures, 5 tables
Abstract:Artificial intelligence (AI) technology has gradually been integrated into automated vehicles. However, how this technology could systematically support takeover transitions in SAE Level 3 automated vehicles remains unknown, even though this level may remain a major form of driving automation in the near future. To address this, we implemented the three AI roles proposed by the triadic Human-AI collaboration framework (Advisor, Co-pilot, and Guardian) as potential authority-ability configurations for AI agents within an automated system. We then examined their interactions with explanation type (why and action) and directional semantics (ipsilateral and contralateral) in an automated driving simulator study with 24 participants. Results showed that, across the three AI roles, Advisor had the shortest decision-making time, followed by Guardian and Co-pilot. Advisor and Guardian were associated with higher decision-making accuracy and higher satisfaction ratings than Co-pilot. Guardian was associated with the lowest workload, followed by Advisor and Co-pilot. The effect of explanation type on decision-making time depended on AI role. Action explanations were faster under Advisor, whereas why explanations were faster under Co-pilot. For directional semantics, ipsilateral semantics were associated with shorter decision-making time than contralateral semantics. Overall, these findings provide preliminary guidance for designing AI-role authority-ability configurations and their associated speech-based takeover messages.
[HC-47] Improving Group Deliberation through AI-Guided Self-Reflection
链接: https://arxiv.org/abs/2610.37862
作者: Haiwen Li,Crystal Qian,Ben Usman,Stephanie Chan,Michiel Bakker,Srivignesh Rajendran,Avneesh Sud,Thomas Leung,Nikki Hariri,John Canny,Nick Dufour,Manon Revel
类目: Human-Computer Interaction (cs.HC); Computers and Society (cs.CY)
备注:
Abstract:As AI systems increasingly mediate human communication, identifying conditions under which they augment collective sense-making is a critical challenge. Through a randomized experiment (N=1,173 U.S. participants across 391 small-group deliberations), we investigate whether AI-guided self-reflection improves subsequent deliberative discourse. We show that engaging with a non-directive AI chatbot instructed to structure self-reflection prior to group deliberation yields significantly higher deliberation quality (measured by the Discourse Quality Index, DQI) compared to a baseline of no self-reflection (Cohen’s d=0.31) and a condition of unassisted self-reflection (d=0.27), and the positive impact is most pronounced in groups with disagreement. This improvement may be driven by more sophisticated argument justifications, greater consensus-building effort, and increased respect for counterarguments. Furthermore, after self-reflection, participants in the AI condition reported higher empathy toward dissenting peers and a deeper understanding of opposing reasoning. Beyond observed behavior, our findings are empirically relevant to open debates between political philosophy and psychology surrounding self-reflection, demonstrating that interactive AI could transform solitary thinking into productive internal reflection. Together, these results demonstrate that even simple AI-based reflective scaffolds can foster more productive group dynamics, opening new avenues for leveraging AI to augment collective intelligence.
[HC-48] New Election Same Old Story: How Tropes Form a Grammar of Election Rumors
链接: https://arxiv.org/abs/2610.37804
作者: Stephen Prochaska,Mert Bayar,Ashlyn B. Aske,Joseph Schafer,Kate Starbird
类目: ocial and Information Networks (cs.SI); Human-Computer Interaction (cs.HC)
备注:
Abstract:Recent elections in the United States have seen online rumors become a regular feature of election discourse as voters and election officials work to navigate a sea of digitally spread claims and propaganda. For the uninitiated, these rumors can feel overwhelming and random, sowing doubt about the integrity and legitimacy of recent elections. In this paper, we reviewed and qualitatively coded a large dataset of 590 election rumors collected during the the 2020 and 2022 national elections. We found that there are tropes that recur time and time again in election rumors, highlighting how these rumors are not as random as they may first appear. We make two primary contributions in this work: first, we introduce a hierarchical framework that structures observed tropes in three tiers (Meta-category, Meso-trope, and Micro-trope). Second, combining the framework with our coded data, we identify a visible grammar of election rumors, highlighting how in the majority of cases a rumor contains an Actor, and Object, and an Action, with optional grammatical components of Suggested Remedies and Scene Setting. The framework and grammar can be used by other researchers to act as the underlying information architecture of analyses examining similar rumors, enabling the rapid identification, analysis, and prediction of election rumors.
计算机视觉
[CV-0] Rethinking Neural Thickets Without Weight Perturbations
链接: https://arxiv.org/abs/2610.40649
作者: Nicolas Dufour,Elouan Gardès,Amil Dravid,Patrick Pérez,Loic Landrieu
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注: Project page can be found here: this https URL
Abstract:The Neural Thickets hypothesis posits that pretrained language models lie in dense neighborhoods of task-specific experts in weight space, which random Gaussian perturbations can uncover. We reproduce the reported gains, but show that they do not require this geometric explanation. Perturbing a single decoder block already recovers much of the improvement, while keeping the model entirely frozen and adding a random bias to its output logits matches or exceeds full-weight perturbations on most reasoning tasks. This suggests an alternative explanation based on output diversity, validation-based selection, and voting. We formalize this view as Perturb-Select-Vote (PSV), a general procedure that generates perturbed predictors, selects high-performing members on a small validation set, and aggregates their predictions. This interpretation connects Neural Thickets to established inference-time methods, including sampling, self-consistency, and verifier-guided selection. A first-order analysis further shows that small weight perturbations act primarily through their induced logit shifts. We extend this analysis to vision, where logit PSV substantially improves performance under domain shift, including on Earth observation, while full-weight perturbations provide little additional benefit. Together, these results suggest that the gains attributed to nearby weight-space experts can instead be obtained through output diversity, selection, and voting. Logit PSV realizes these benefits with a fully frozen model, providing a simple, training-free alternative to full-weight perturbation ensembles across language and vision. Comments: Project page can be found here: this https URL Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV) Cite as: arXiv:2610.40649 [cs.LG] (or arXiv:2610.40649v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.40649 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[CV-1] Read Allocate Route: Role-Coupled Tokenization for Pixel-Space Generative Transformers ICLR2027
链接: https://arxiv.org/abs/2610.40647
作者: Jiongfan Yang,Weichen Fan,Haiwen Diao,Zhongang Cai,Ziwei Liu
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Neural and Evolutionary Computing (cs.NE)
备注: 22 pages, 10 figures; under review at ICLR 2027
Abstract:Pixel-space generative Transformers typically use one token per image patch, so capturing finer detail requires more tokens and makes global attention increasingly expensive. We observe that finer detail does not necessarily require more global communication: low-frequency structure benefits from global exchange, while high-frequency detail can be modeled locally. Based on this insight, we introduce SpecSlot, a pixel-space rectified-flow Transformer that decouples output detail from the length of the global attention sequence. SpecSlot expands each spatial anchor into multiple slots, each assigned a disjoint band of DCT coefficients; this separates anchor density from spectral subdivision. Only the lowest-frequency slot from each anchor participates in global attention, while all slots interact locally. This allows the model to increase spectral detail without enlarging the global sequence. A spatial head further adds a local correction to the spectral prediction, while the generative state, target, and loss remain in pixel space. On ImageNet-256, SpecSlot achieves a comparable final FID to a FLOPs-matched JiT control while converging faster. On ImageNet-512, it improves FID by 2.84 over the control with 3.9% fewer FLOPs, and comes within only 0.69 FID of JiT-B/16 at 38.2% lower cost. On text-to-image generation, SpecSlot surpasses JiT-B/8 on DPG-Bench by 1.06 at 40.6% lower forward cost.
[CV-2] Luggage re-identification: A study of strategies towards a real-world application
链接: https://arxiv.org/abs/2610.40643
作者: Quentin Dhôte,Romaric Audigier
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: 7 pages, 4 figures, 5 tables. Accepted at The 2026 IEEE International Conference on Content-Based Multimedia Indexing (IEEE CBMI)
Abstract:Luggage re-identification (ReID) is a critical task in video surveillance for forensics, security or border management (suspicious or abandoned luggage recognition), as well as luggage management (lost luggage recovery). It enables the association of observations of the same luggage across disjoint camera views in complex environments such as airports, train stations and border crossing. While existing approaches primarily focus on appearance-based cropped image matching under relatively clean conditions, real-world scenarios are characterized by frequent occlusions, cluttered background and crowded scenes which significantly degrade performance. Yet, strategies leveraging the complementarity of multiple shots and views from different video streams is overlooked. In this paper, we study strategies for an end-to-end pipeline that operates from multiple video streams to detect and re-identify luggage. We evaluate different variants on a benchmark dataset acquired in real-world environments to address the questions of selecting and merging different views and shots of the same luggage as well as training on different environment datasets for an overall ReID performance improvement in practical applications.
[CV-3] SonoWorld4D: From One Video to a 4D Audio-Visual Scene via Diffusion Synchronization
链接: https://arxiv.org/abs/2610.40640
作者: Derong Jin,Xiyi Chen,Ming C. Lin,Ruohan Gao
类目: Computer Vision and Pattern Recognition (cs.CV); Sound (cs.SD)
备注: Project Page: this https URL
Abstract:Recent progress in visual scene generation has extended beyond static 3D scenes toward the more challenging problem of generating dynamic 4D scenes from minimal input. However, sound, despite being intrinsically coupled with dynamic objects and events, remains largely unmodeled in prior work. We introduce Video-to-4DAVScene, a new task that turns a video into an explorable, time-varying 4D audio-visual scene, and present the SonoWorld4D framework for this task. Given a monocular video, our method generates a 360 static background, reconstructs explicit motion trajectories of dynamic, sound-emitting foreground objects, and synthesizes a physically-grounded sound field consistent across multiple sources through dual-space audio diffusion synchronization. The resulting 4D scene supports real-time novel-view visual and spatial audio rendering. Evaluation on real-world dynamic scenes and a user study demonstrates SonoWorld4D generates 4D audio-visual scenes with perceptually realistic, physically grounded spatial audio consistent with scene geometry and visual content.
[CV-4] Probing Spatial State Representations in Visual Foundation Models
链接: https://arxiv.org/abs/2610.40636
作者: Baiqiao Yin,Zhongang Cai,Ruisi Wang,Qiwei Liang,Lei Yang,Ziwei Liu
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:An observer sees only part of the surrounding world, yet must keep track of what lies beyond the current view. Spatial perception therefore involves more than recognizing individual images: it calls for a persistent spatial state that is updated under self-motion and used to guide action. Do visual foundation models provide representations that support such a state? We study this question through four levels of spatial capability—spatial structure, object persistence, predictive spatial update, and actionability—using frozen-feature probes on ten representations under streaming egocentric observations of static indoor environments. The resulting capability profiles vary across tasks, network layers, and prefix lengths: strong readout of the visible scene does not ensure accurate object persistence or future-feature retrieval. Matched controls separate readout accuracy from reliance on visual history and motion. Native future prediction retains recoverable spatial structure, but a spatial reader trained on real features does not transfer to predicted features unchanged. Finally, goal-dependent candidate selection does not establish reliable continuous action utility: predicted actions respond to goal changes, yet supplying the correct goal does not reliably improve progress. Across the four levels, object persistence and actionability build on camera-pose readout, whereas feature prediction builds on visible-scene readout. Together, these findings motivate evaluating spatial representations not only by what can be decoded from them, but by whether that information is maintained beyond observation, updated predictively, and used for goal-directed action.
[CV-5] SymAnt: Grounding Long-Horizon Action Anticipation with Adaptable Symbolic Priors WACV2027
链接: https://arxiv.org/abs/2610.40632
作者: Fnu Aryan,Yaqi Xie,Joseph Campbell,Woojun Kim,Kwonjoon Lee,Hossein Nourkhiz Mahjoub,Ehsan Moradi-Pari,Katia Sycara,Simon Stepputtis
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to WACV 2027. 18 pages (8-page main paper, references, and supplementary appendix), 6 figures, 9 tables. Project page: this https URL
Abstract:As robots and intelligent systems increasingly interact with humans, they must anticipate long sequences of future actions from only brief observations. While purely neural models typically require extensive observations to predict future actions and remain difficult to interpret and adjust, symbolic priors offer a structured alternative. To this end, we introduce SymAnt, a novel neuro-symbolic approach that unifies the expressiveness of neural networks with the structured reasoning of scene and knowledge graphs. Given a short video, SymAnt performs a joint-search over both graphs to construct a symbolic context, enabling unified reasoning over spatial relationships, object attributes, affordances, and associated actions. This symbolic context serves as a prior for an iterative refinement process that progressively generates and improves the sequence of future actions. By grounding predictions in symbolic context, SymAnt anticipates long-horizon actions from short observation windows in a parameter-efficient manner, while enabling users to directly edit the underlying knowledge to adjust predictions without further training. Through our experiments, we demonstrate the effectiveness of SymAnt, particularly when only short observation contexts are available, across five diverse datasets: Breakfast, 50 Salads, EPIC-KITCHENS, EGTEA Gaze+, and Assembly101.
[CV-6] Difficulty Calibration Across Time for Membership Inference in Diffusion Models
链接: https://arxiv.org/abs/2610.40616
作者: Michael Joseph,Yedid Hoshen
类目: Computer Vision and Pattern Recognition (cs.CV); Cryptography and Security (cs.CR); Machine Learning (cs.LG)
备注:
Abstract:Modern diffusion models are trained on massive datasets that often include copyrighted or private images. Membership inference attacks (MIAs) address this by testing whether a given image was in a model’s training set. The standard approach scores an image by its reconstruction loss and flags low-loss images as members. This often fails because some images are simply easier to reconstruct than others, so a simple unseen image can score a lower loss than a complex training image. Difficulty calibration separates memorization from how difficult an image is to begin with. We show how to calibrate along two axes of time: training time and diffusion time. First, across training time, we introduce Reference-Model Calibration (RMC), which compares a fine-tuned model’s loss against an earlier checkpoint. This cleanly isolates newly memorized data and gives near-perfect auditing, even without text captions. Since earlier checkpoints are not always available, we next present Temporal Self-Calibration (TSC), which calibrates across diffusion time using only a single model. Our key observation is that an image’s difficulty persists across noise levels, whereas memorization is localized in time. By comparing the loss at a target timestep against neighboring timesteps from the same model, TSC attenuates image difficulty using the model’s own trajectory. Empirically, our methods beat the state of the art by a wide margin, and unlike most existing attacks, generalize across different model parameterizations such as x -, v -, and \epsilon -prediction models. Code is available at this https URL.
[CV-7] L3G-World: Real-Time Infinite World Generation with Latent 3D Gaussian Memory
链接: https://arxiv.org/abs/2610.40614
作者: Jinhyeok Choi,JoungBin Lee,Jongmin Lee,Minkyeong Jeon,Jiyoung Kim,Kaeun Han,Yoonjin Yoon,Seungryong Kim
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project page: this https URL
Abstract:Interactive world models generate visual content conditioned on user actions, yet maintaining spatial consistency over long-horizon exploration in real time remains challenging. Existing approaches rely on either temporal context or explicit 3D memory, both of which scale poorly because memory grows with the length of the generated stream. Moreover, explicit 3D memories condition latent video generators by re-encoding RGB renderings, which precludes real-time generation. In this paper, we introduce a Latent 3D Gaussian Memory (L3G) that combines the efficiency of feed-forward reconstruction with a compact, rapidly renderable scene representation. Building on this memory, we propose L3G-World, in which a memory updater and a video diffusion transformer (DiT) alternate autoregressively to generate spatially consistent worlds. The memory updater spawns Gaussians in newly observed regions and refines existing ones, so that the memory grows with scene extent rather than stream length. Each Gaussian further stores a VAE latent feature, allowing the memory to be rendered directly into the DiT’s latent space without VAE re-encoding. Experiments show that L3G-World achieves state-of-the-art generation quality and long-horizon spatial consistency, with up to 38.6\times fewer 3D primitives, 3.2\times faster memory updates, and 359\times faster memory rendering relative to the most efficient prior 3D-memory baseline for each metric, enabling real-time world generation at 9 FPS.
[CV-8] Fight World Model: Customizable Combat through Visual Interaction Proxies
链接: https://arxiv.org/abs/2610.40607
作者: Jiahua Li,Zixu John,Tom Zhong,Fuping Wu,Tianhao Xu,Ruicheng Zhang,Jun Zhou,Zunnan Xu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Tech Report
Abstract:A playable, programmable game world must maintain persistent state, enforce executable rules, and support extensible gameplay mechanics. Existing video world models can generate plausible combat, yet struggle to keep hit reactions and health updates consistent with game rules. We introduce Fight World Model (FWM), which separates rule-governed combat execution from video rendering. An executable engine maintains state and resolves interactions. A deterministic compiler encodes these outcomes and attempted actions in a spatially and temporally aligned interaction proxy. The proxy aligns action and skill windows, brief hit reactions, and persistent health by fighter identity and timing. These cues can be controlled separately to render interactions under modified rules. Identity-bound references preserve character appearance across side switches. We train FWM on FightWorld Corpus, spanning 52 skill effects, 187 character-appearance variants, and 15 arenas. Without retraining the renderer, FWM supports damage scaling, limited-hit immunity, lifesteal, and multi-stage skill mechanisms beyond the native game. It also allows character and arena customization. Experiments demonstrate the effectiveness of FWM, which outperforms existing game world models in action fidelity, hit reactions, health-state consistency, and character appearance preservation. Our project is available at this https URL.
[CV-9] HairDeck: Animation-Ready Stylized Hair Cards from a Single Image
链接: https://arxiv.org/abs/2610.40599
作者: Yuze He,Yixiang Wei,Wang Zhao,Chuhao Chen,Yanning Zhou,Yong-Jin Liu
类目: Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR)
备注: 27 pages, 23 figures
Abstract:The hair of stylized characters in games and avatars is built from hair cards: artist-placed ribbons whose geometry is the visible hair, animated by spring bones. Existing image-based methods produce realistic hair or a single fused mesh, while CHARM, the only generator of stylized hair cards, decodes a hairstyle autoregressively as a long sequence of control points. Observing that a card hairstyle is an unordered set of artist-defined primitives, we propose HairDeck, which generates it as one: each card is a single token, and a set diffusion transformer denoises all cards together. From a single image, HairDeck generates the cards and the spring-bone chains that drive them, yielding an animatable asset that exports to ready-to-use downstream formats. On CHARM’s own test set, it lowers Chamfer distance by 44% and card-orientation error by more than half while running 33 times faster, and from single images it outperforms CHARM, image-to-3D pipelines and strand-based methods. Project page: this https URL
[CV-10] Beyond Ground Truth: A Heuristic Unsupervised Quality Assurance Engine for Diagnosing Calibrational Blindness in Dental Segmentation
链接: https://arxiv.org/abs/2610.40587
作者: Fatematuj Johura Mim
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at 11th IEEE Asia-Pacific Conference on Computer Science and Data Engineering (IEEE CSDE 2026), Cox’s Bazar, Bangladesh. Copyright 2026 IEEE. 5 pages, 5 figures, 3 tables
Abstract:In real-world clinical settings, dental X-ray segmentation using deep learning methods suffers from reliability issues. Even when the segmentation quality is bad, deep learning models show high confidence in their predictions (calibrational blindness). Moreover, in real use, doctors do not have ground-truth masks to check their accuracy. To address this problem, we introduce a simple quality-checking engine called HUS-QA with three stages. We train it on the PRAD (Periapical Dental X-ray) dataset to build our SimCLR self-supervised ResNet-18 U-Net model, which generates the segmentation mask and then evaluates its quality. Our HUS-QA checks the segmentation mask quality without a ground-truth mask. Prediction confidence, edge sharpness, shape connectivity, and area-size penalty are the simple indicators considered in this research. Results show that standard confidence methods fail (r = -0.1394, p = 1.7864e-3, ROC AUC = 0.4275). By contrast, our HUS-QA shows a positive correlation with actual Dice scores (r = 0.4647, p = 3.7510e-28) and achieves a strong ROC-AUC of 0.8405. Therefore, HUS-QA serves as a practical tool for clinical quality control.
[CV-11] RADIOlogy: A Single-Pass Approach to Training-Free Zero-Shot Anomaly Detection WACV2027
链接: https://arxiv.org/abs/2610.40580
作者: Joaquín Figueira,Giacomo D’Amicantonio,Egor Bondarau
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to WACV 2027. Code available at this https URL
Abstract:Current zero-shot anomaly detection methods often rely on complex prompt-learning techniques or computationally expensive sliding-window strategies to achieve fine-grained visual alignment to detect small-scale anomalies. This reliance stems from the fundamental limitation of standard CLIP architectures, where [CLS] token attention and spatial patch pooling struggle to capture subtle, localized anomaly features. While prompt-learning approaches aim to bridge this gap, they typically require a large number of ground-truth segmentation masks from auxiliary datasets, which are often unavailable in specialized domains. In this work, we show an emerging property of multi-head attention pooling in vision-language models: remarkable zero-shot anomaly localization and detection right out of the box. We translate this property into a single-pass anomaly detection technique, RADIOlogy, that does not require dense, windowed execution and significantly outperforms other training-free baselines. Instead, the method relies on a simple contrastive scoring approach that compares the average similarity between vision features and a suite of normal and abnormal prompts. Furthermore, we leverage the agglomerative features of RADIO models to enhance pixel-level anomaly segmentation using a dense anomaly refinement pipeline. Finally, we show how the vision-language family of training-free zero-shot anomaly detection techniques can be applied successfully to the medical domain with minimal modifications. Code is available at this https URL.
[CV-12] PhysCount: Do MLLM s Count Visual Appearances or Physical Entities?
链接: https://arxiv.org/abs/2610.40530
作者: Lidong Lu,Jiaxin Lv,Liangrui Dong,Xinyi Mao,Lidong Zou,Zhenquan Li,Tong Lu,Guo Chen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Images reveal appearances, but counting often asks about physical objects. A single entity may be visible directly, in several reflections, or through glass, so detecting every visible occurrence does not determine how many entities are present. Existing visual counting benchmarks rarely separate these two operations. We introduce PhysCount, a benchmark for evaluating whether multimodal large language models (MLLMs) can distinguish what appears in an image from what physically exists in the scene. PhysCount contains 546 real-world images and 4,595 human-written questions organized into 651 target-specific groups. For each group, annotators locate the relevant appearances, label how each is observed, and identify appearances of the same entity. Four complementary question families test the enumeration of appearances, the enumeration of physical entities, image-formation-specific counting, and appearance–entity correspondence. We measure both accuracy on each family and the proportion of groups answered entirely correctly. Across 40 MLLMs, counting visible appearances is substantially easier than recovering the identities that connect them. Even the strongest model answers every question correctly in only 13.36% of groups. Performance declines when one entity produces many appearances or when its appearances cross observation modes. Higher image resolution improves access to visual evidence but brings limited gains in complete-group accuracy, whereas answering related questions jointly produces a marked improvement. These results expose a gap between recognizing what is visible and correctly recovering the physical entities behind those observations.
[CV-13] CrossWeave: Cross-Modal Scene and Instance Retrieval from Sparse 2D-3D Alignment
链接: https://arxiv.org/abs/2610.40525
作者: Aadith Warrier,Gnana Prakash Punnavajhala,Siddharth Tourani,Muhammad Haris Khan,Avinash Sharma,Madhava Krishna
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project Page: this https URL
Abstract:Cross-modal scene and instance retrieval aims to retrieve a corresponding 3D environment or object given query 2D images, serving as a foundational mechanism for spatial reasoning in robotics and AR/VR. Existing methods achieve this by learning alignment only at the level of whole scenes or objects, which provides no supervision for the fine-grained local structure within them. In this work, we show that supervision on a set of sparse 2D-3D correspondences, without scene- or instance-level contrastive objectives, is sufficient to train a single model for both scene- and instance-level retrieval. Specifically, we propose \textbfCrossWeave , a novel framework for cross-modal retrieval that learns sparse patch-point alignment, which alone outperforms existing methods. An optimal transport-based aggregator over the aligned features gives further gains by clustering structurally redundant regions, freeing descriptor capacity for semantically distinct local features. CrossWeave achieves state-of-the-art performance on both scene- and instance-level retrieval on ScanNet and 3RScan with a \textitsingle model , whereas existing methods require separate models for each task. Our results suggest that whole-scene encoding is unnecessary for cross-modal retrieval and that sparse local correspondence supervision is a more effective and flexible alternative.
[CV-14] LucidLyra: Feed-Forward Generative 3D Scene Reconstruction and Joint Understanding via Multi-Task Distillation
链接: https://arxiv.org/abs/2610.40495
作者: Oliver Hahn,Rodrigo Marcuzzi,Sangeetha Grama Srinivasan,Alessandro Burzio,Sherwin Bahmani,Jiahui Huang,Stefan Roth,Shengyu Huang,Qunjie Zhou,Laura Leal-Taixé
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project Page: this https URL
Abstract:We present LucidLyra, a framework that makes generatively reconstructed 3D scenes directly queryable for semantics and instances in a feed-forward manner from video latents. While generative 3D reconstruction can synthesize virtual environments from a single image, downstream applications in robotics and physical AI require not only visual realism but also access to scene semantics and object identities. Existing 3D understanding approaches, however, require captured multi-view scenes, have limited generalization capabilities, or rely on expensive per-scene processing and optimization. To bridge this gap, our method distills complementary knowledge from a camera-controlled video diffusion model and frozen 2D vision foundation models into a unified model to jointly infer 3D Gaussians encoding geometry, appearance, open-vocabulary semantics, and class-agnostic instance cues. The resulting scenes support novel-view rendering, open-vocabulary segmentation, and object grouping without per-scene optimization, while training requires neither manual 3D annotations nor captured multi-view data. Our method outperforms baselines combining state-of-the-art 3DGS segmentation approaches with generative 3D scene reconstruction on curated synthetic and real-image-conditioned benchmarks, while preserving reconstruction quality. Furthermore, our latent formulation naturally extends to monocular video, reconstructing structured dynamic 3D scenes across time. By turning renderable but unstructured outputs into queryable 3D worlds, LucidLyra enables downstream applications including 3D scene editing, 3D asset mining, and robotic simulation.
[CV-15] MetaRefine: Score Refinement with Multi-Level Features for Image Assessment
链接: https://arxiv.org/abs/2610.40487
作者: Chuanzhi Xu,Chengkun Yue,Haolan Guo,Xuanhua Yin,Boyu Wei,Zihan Deng,Qiang Qu,Weidong Cai
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:An image’s original dimensions, aspect ratio, and composition provide important cues that can affect its technical and aesthetic quality. Image assessment methods use resizing, cropping, padding, or patch sampling to accommodate different input sizes. However, these operations can alter or obscure original geometric and compositional cues. Existing assessment models also encode rich multi-level visual evidence that can be further extracted and exploited to improve existing models. This motivates us to combine visual evidence with original-image geometry for score refinement without changing the original scoring pipelines. In this paper, we introduce MetaRefine, a metadata-aware score refinement architecture whose lightweight residual heads predict score corrections from geometry, multi-level visual features, and the frozen model’s original score. Across 100 configurations on 21 datasets, MetaRefine improves PLCC in 88 configurations and SRCC in 85, with average relative gains of 7.55% and 7.10%, respectively. It reduces MAE in 90 configurations and RMSE in 89, with average relative reductions of 23.27% and 21.93%, respectively. The median parameter overhead is only 1.11%. MetaRefine can further improve state-of-the-art (SOTA) assessment models on multiple benchmarks.
[CV-16] AVPS-360: A Spatial Audio-Visual Benchmark for Instance Segmentation and Tracking of Sound-Producing Persons
链接: https://arxiv.org/abs/2610.40480
作者: Yifu Sun,Parthasaarathy Sudarsanam,Tuomas Virtanen,Archontis Politis
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Audio-visual instance segmentation (AVIS) segments and tracks individual sound-producing objects in videos. Existing benchmarks predominantly use monaural audio, which lacks inter-channel cues for spatially distinguishing concurrent sound sources. We introduce AVPS-360 (Audio-Visual Person Segmentation 360), a spatial audio-visual benchmark bridging AVIS and sound event localisation and detection (SELD). Built upon STARSS23, it pairs first-order Ambisonics audio with synchronised 360° video across 230 recordings totalling 10.5 hours. Recordings average 164.0 seconds. The dataset captures intermittent sound production in real-world interactions dominated by speech, alongside laughter, clapping, and footsteps. It provides 31,678 manually reviewed instance masks linked to source identities and sound-event labels. We also present a baseline that integrates spatial audio with visual features for frame-level segmentation and identity association across frames. Experiments on AVPS-360 show that existing AVIS methods struggle with detection and identity association in long recordings with intermittent sound production, while our baseline substantially improves performance in both aspects.
[CV-17] Accurate MRI-free PET Quantification and Reliability Assessment via CT-to-MRI Synthesis and Conformal Prediction
链接: https://arxiv.org/abs/2610.40478
作者: Jungwoo Seo,Seung Yun Choi,Kun Ho Lee,Seun Jeon,Byoung Seok Ye,Jiook Cha
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 28 pages, 8 figures, including supplementary material. Submitted to Medical Image Analysis
Abstract:Standardized uptake value ratio (SUVR) quantification of amyloid PET commonly relies on a subject-specific T1-weighted MRI, yet a usable MRI is not always available. We therefore combined CT-to-MRI synthesis with conformal prediction for MRI-free PET quantification with reliability assessment. A slice-consistent Brownian Bridge Diffusion Model was strengthened with multi-dataset training and data-source conditioning to enrich the anatomical prior learned from MRI, and with expanded multi-slice CT context to exploit the through-plane continuity of subject-specific anatomy in CT. Synthetic MRI was used for FreeSurfer-based anatomical parcellation and PET quantification, while deep-ensemble uncertainty was calibrated with conformal prediction to estimate a plausible range for the corresponding real MRI-based SUVR. On ^11 C-PiB PET from OASIS-3, the proposed method achieved the highest synthesis and parcellation fidelity among the compared methods. Mean cortical SUVR from the proposed synthetic MRI agreed closely with the real MRI-based reference (MAE 0.016 \pm 0.015 , Pearson r = 0.9987 ) and yielded the same amyloid-positivity classification as the real MRI in all 87 test subjects at the prespecified cut-off. Conformal prediction achieved empirical coverage close to the prespecified levels, and at 90% target coverage, 85 of 87 subjects had prediction intervals that supported an unambiguous amyloid-positivity decision. Zero-shot evaluation on an independent external cohort showed that the proposed CT-to-MRI model could support PET quantification without retraining, although conformal calibration required target-cohort recalibration to recover the intended coverage. These findings support CT-to-MRI-based MRI-free PET quantification as a promising approach for providing accurate SUVR measurements and an explicit assessment of their reliability.
[CV-18] InfantNet: A Large Scale Dataset for Infant Body Pose and Shape Estimation
链接: https://arxiv.org/abs/2610.40467
作者: Zhipeng Hui,Jiahao Wang,Xun Dong,Guofeng Zhang,Xingrui Wang,Jiawei Peng,Qihao Liu,Xiaoding Yuan,Yi Zhang,Junjie Oscar Yin,Wufei Ma,Qinfan Lyu,Sara N. Lupolt,Keeve Edward Nachman,Adam Kortylewski,Alan Yuille
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Infant pose and shape estimation is essential for applications in childcare, developmental monitoring, and medical diagnosis. However, existing methods and datasets are largely designed for adults, and direct transfer to infants fails due to substantial differences in body proportions, articulation limits, and frequent self-occlusion. To address this gap, we introduce InfantNet, the largest real-image infant dataset to date, comprising 108,902 RGB images of infants aged 6-18 months. Each image is annotated with 2D keypoints, and a curated subset of 11,642 images additionally includes 3D pose and shape annotations with full SMIL parameters, verified through multi-view rendering and reprojection-error filtering. We use an iterative annotation pipeline to ensure high fidelity across both 2D and 3D labels. InfantNet establishes a large-scale, comprehensive benchmark for infant 2D keypoint detection and 3D pose-and-shape recovery. Baseline experiments demonstrate that state-of-the-art adult pose estimators do not generalize well to infants, whereas fine-tuning on InfantNet yields a consistent improvement. The gains are even more pronounced for 3D pose and shape estimation, where SMIL-based estimation demonstrates robust performance with accuracy comparable to results reported for SMPL-based models. By releasing the InfantNet dataset and benchmark, we provide a vital resource for advancing infant pose analysis and related healthcare applications.
[CV-19] A Unified Benchmark for Aerial-Enhanced Autonomous Driving: The KITScenes Aerial Dataset WACV2027
链接: https://arxiv.org/abs/2610.40455
作者: Daniel Lengerer,Richard Schwarzkopf,Jonas Merkert,Fabian Immel,Rupert Polley,Frank Bieder,Jan-Hendrik Pauls,J. Marius Zöllner,Klaus Bogenberger,Christoph Stiller,Carsten Markgraf
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV 2027)
Abstract:Aerial imagery provides complementary spatial context for autonomous driving by exposing road structure beyond the immediate perception range of onboard sensors. Yet map construction from aerial imagery, aerial-assisted online HD mapping, and cross-view localization are typically studied on separate data and benchmarks. We introduce KITScenes Aerial, an extension of the KITScenes Multimodal dataset that augments synchronized onboard sensor data, vehicle poses, and rich Lanelet2 map annotations with spatially associated aerial imagery and higher-resolution neural representations, providing a common basis for a variety of research tasks. Our mapping evaluation focuses on all 53 KITScenes map classes across different model designs, spatial ranges, and aerial resolutions, while retaining compatibility with the widely used three-class setting. Representative baselines show that aerial context is particularly beneficial for fine-grained and extended-range mapping, more than doubling 53-class mAP at the larger mapping extent. Higher-resolution aerial imagery yields a further 4.3 mAP (+24 %) improvement, with gains extending to challenging map elements such as traffic lights. We additionally evaluate cross-view localization under geographically disjoint conditions, where an adapted fine-grained feature-matching baseline achieves a mean translation error of 5.13 m. KITScenes Aerial thus enables unified study of aerial and ground-level information across complementary mapping and localization tasks. Project page: this https URL
[CV-20] PhysInsert: Physically Plausible Object Insertion with Geometric Priors and Bidirectional Object-Scene Interactions
链接: https://arxiv.org/abs/2610.40453
作者: Yue Gong,Le Zhang,Rui Chen,Song Chang,Xiaochao Qu,Luoqi Liu,Ting Liu,Yebin Liu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 51 pages, including supplementary material. Project page: this https URL
Abstract:We present PhysInsert, a framework for physically plausible reference-based object insertion that jointly adapts objects and their surrounding scenes. Existing methods often rely on masked inputs, generation-based pairs, or training pairs with limited viewpoint variation, leaving pose and illumination adaptation and bidirectional object-scene interactions insufficiently constrained. We construct the Insert World dataset with strictly aligned Blender-rendered counterfactual pairs and object references across viewpoints and lighting conditions. Building on these data, Geometry-Guided Adaptation couples a VGGT geometry stream with image and text streams under target-scene feature supervision to guide spatially consistent insertion. Complementary 2D Object-Scene Spatial Supervision jointly supervises object occupancy and environmental changes while grounding spatial instructions in object regions through image-text attention. We introduce PhysInsert-Bench with dedicated cross-environment and cross-view cases requiring illumination and pose adaptation beyond appearance copying. Experiments on PhysInsert-Bench and DreamBooth demonstrate improved image fidelity; a user study further shows more accurate placement and improved lighting integration, perspective, and pose plausibility.
[CV-21] Soil Moisture Estimation in Precision Agriculture: A Knowledge Distillation Approach
链接: https://arxiv.org/abs/2610.40432
作者: Mohammed Rakib,Vrinda Nagula,Arunkumar Bagavathi
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: Accepted by MWAIS 2025
Abstract:Accurate soil moisture estimation is critical for precision agriculture, influencing irrigation, fertilization, and harvesting strategies. Traditional methods primarily rely on meteorological and soil property data, while recent approaches incorporate visual modalities such as aerial and geospatial imagery. However, these high-resolution images are expensive and computationally demanding to analyze. In this work, we propose Soil Moisture Distillation (SMD), a Knowledge Distillation (KD)-based multimodal learning framework that leverages unimodal teachers to enhance multimodal soil moisture estimation. SMD surpasses not only traditional multimodal fusion techniques, including MIS-ME, but also outperforms unimodal models, demonstrating the benefits of guided unimodal knowledge transfer. Additionally, we conduct a comparative analysis of soil patches and crop images, showing that crop images provide superior visual cues, leading to improved prediction accuracy. SMD achieves a 2% improvement in MAPE for soil patches and 1.2% for crops over state-of-the-art baselines, highlighting the effectiveness of leveraging both meteorological and visual data through knowledge distillation for enhanced soil moisture forecasting.
[CV-22] Uncertainty-aware Feedback Loop Diagnosis: integrating reinforcement learning and conformal prediction for reliable wafer defect classification
链接: https://arxiv.org/abs/2610.40421
作者: Aleksandr Dekhovich,Carlas Smith,Oleg Soloviev
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Semiconductor manufacturing is a complex, multi-step process that requires both high precision and low operating costs. Therefore, detecting defects in a high-volume setting is an important component of wafer map analysis for reliable production. However, most state-of-the-art deep learning approaches require a full scan of the surface, which hinders the development of high-throughput inspection systems. To address this problem, we propose Uncertainty-aware Feedback Loop Diagnosis (UFLD), a novel algorithm that employs reinforcement learning for adaptive measurement acquisition and provides a prediction uncertainty level after each step, stopping the process once sufficient information is obtained. Unlike existing methods that rely on ensemble models or auxiliary confidence estimation deep learning models, UFLD utilizes a more parameter-efficient conformal prediction framework to determine the optimal stopping time. The proposed UFLD reduces acquisition time by about 40% and achieves 98.4% inspection accuracy, without requiring an additional learned confidence-estimation network. We validated our approach on the MixedWM38 dataset, demonstrating its advantages over existing adaptive acquisition procedures and standard convolutional neural networks.
[CV-23] Continuous-Action World Models: Learning Cross-Embodiment Spatiotemporal Dynamics Priors
链接: https://arxiv.org/abs/2610.40401
作者: Liangyu Li,Qingwen Liu
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 12 pages, 6 figures
Abstract:We introduce Continuous-Action World Models (CoAWM), a framework for learning cross-embodiment spatiotemporal dynamics priors that support robot action generation across embodiments and execution frequencies. At its core is a shared latent function space that captures the spatial and temporal structure of robot actions. We learn this space through two stages. First, function pretraining brings trajectories from different embodiments and sampling frequencies into a common representation, learning shared dynamics priors through action reconstruction and consistency across temporal sampling views. By accommodating heterogeneous robot trajectories, this stage broadens the range of data sources available for pretraining and facilitates scaling to larger and more diverse datasets. Second, we connect the pretrained function space to a world-model backbone, enabling observations, language instructions, and the current robot state to condition the prediction of a latent action function. A single decoder takes any query time and returns the corresponding end-effector action. By separating shared dynamics priors from task-conditioned action selection and temporal sampling, our approach provides a unified foundation for transferring robot experience across embodiments and control frequencies.
[CV-24] HiFOR: Hierarchical Foresight with Action-Guided World Grounding for Vision-Language-Action Model
链接: https://arxiv.org/abs/2610.40373
作者: Rongxing Ding,Hongyu Qu,Kehong Wu,Xudong Wang,Fan Zhang,Xinguang Xiang,Xiangbo Shu
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注:
Abstract:Vision-Language-Action (VLA) models have demonstrated promising capabilities in robotic manipulation by integrating visual perception, language understanding, and action generation. Recent advances have further incorporated future prediction to enhance the understanding of environmental dynamics and task progression. However, existing methods often model short-term dynamics and long-term task objectives separately, while directly incorporating predicted futures into action policies without sufficiently considering their relevance to current action requirements. To address these limitations, we propose HiFOR, a hierarchical foresight framework with action-guided world retrieval. Specifically, Hierarchical Foresight Prediction jointly predicts short-term state evolution, semantic milestones, and final goal states through a Hierarchical Dependency Mask (HDM), establishing dependencies across temporal horizons. Building upon these predictions, Structure-Aware Latent World Modeling organizes future latent representations along the horizon, view, and spatial dimensions to facilitate fine-grained retrieval. Furthermore, Action-Guided Factorized World Retrieval adaptively retrieves relevant future information from the structured latent world based on evolving action representations during denoising, providing targeted guidance for action generation. Extensive experiments on two simulation benchmarks and six real-world robotic manipulation tasks demonstrate the effectiveness of our approach.
[CV-25] Wizard: Manipulate Any World in Real Time via Agent ic Control
链接: https://arxiv.org/abs/2610.40339
作者: Yuzhe Wu,Zijun Lin,Yeying Jin
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project Page: this https URL
Abstract:Recent advances in video world models have enabled the generation of increasingly realistic and interactive virtual environments, opening new possibilities for player-customizable gameplay. Agentic frameworks further advance this direction by decoupling explicit game logic and controllable world representations from visual generation. However, existing approaches support only simple game settings and predefined motion patterns, generalize poorly to unseen environments and characters, and offer limited real-time interactivity. In this paper, we introduce Wizard, an agentic game world model that enables real-time, precise control over character motions while generalizing zero-shot across novel scenes, characters, and actions. To achieve this, we construct a diverse dataset spanning 750 scenes and 178 characters, incorporating skeleton and geometry information as structured representations. We leverage these representations as intermediate proxies to condition a video generation model, together with an adaptive masking mechanism that improves alignment between the conditioning signals and generated video rollouts. Furthermore, we introduce an agentic framework that explicitly maintains and updates game states, allowing players to freely configure their virtual worlds and exercise fine-grained control over character motions while preserving visual quality and gameplay consistency. By combining agentic control with structured proxy-based generation, our framework unlocks new possibilities for player-driven game creation with real-time, generalizable, and mechanically consistent world simulation.
[CV-26] RFW-A: Agent ic Procedural Generation of 400K Articulated Assets over 500 Categories
链接: https://arxiv.org/abs/2610.40333
作者: Yuanbo Lyu,Changjian Pan,Saining Zhang,Yi Yang,Peishuo Li,Zhiheng Li,Jialu Liu,Junhao Chen,Jiawei Weng,Yupeng Gao,Weiqing Xiao,Wenyi Li,Jiarui Guo,Rui Zhang,Jingrun Du,Weiyi Lin,Zenan Meng,Hu Wei,Lin Qu,Zongzheng Zhang,Hao Zhao,Chaojian Li
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 37 pages, 19 figures. Project Page: this https URL
Abstract:High-quality articulated assets are essential for embodied perception, simulation, and manipulation, yet constructing them at scale remains difficult. Manual annotation and procedural modeling require substantial human effort. Recent agentic systems automate asset construction through program synthesis, but repeat costly reasoning and refinement for every new asset. Our key observation is that assets within a category can share construction logic despite variations in structure and appearance. Agentic synthesis and procedural generation are therefore complementary: agents can construct reusable programs, while procedural execution reuses the resulting logic across many samples. We therefore present RFW-A, which combines the two in a new articulated-asset generation paradigm. Starting from existing articulated programs, RFW-A organizes reusable components and their dependencies into category-level generators. Agents test and refine these generators across sampled configurations and motion states. Once constructed and validated, the generators produce diverse assets through procedural execution with no further model calls or token charges. The estimated API cost of generator authoring averages 10.62, amortizing to 0.05 per asset over 200 assets, versus 1.24 for the sampled per-asset agentic synthesis baseline. Using RFW-A, we built RFW-A-400K, containing over 400K sim-ready articulated assets across more than 500 categories. Experiments show high structural and kinematic validity and stable execution across physics simulators. Beyond asset validity, training with RFW-A-400K improves part-level articulation estimation on seen and unseen categories, while policies trained in simulation with retrieved and program-edited assets transfer zero-shot to real articulated objects.
[CV-27] Radiology Guideline-derived Decision Graph for Efficient Clinical Decision Support
链接: https://arxiv.org/abs/2610.40327
作者: Xuewen Zhou,Zhen Chen
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Radiological risk stratification and management planning are essential downstream tasks in medical imaging. Automating these tasks is challenging because clinical guideline criteria are distributed across prose, tables, and cross-referenced definitions. We present Guideline-derived Decision Graph (GDGraph), a framework that decouples guideline compilation from case evaluation. During compilation, a large language model partitions guidelines into section-based chunks, extracts local decision subgraphs with linked definitions, and aggregates related branches into an executable decision graph. During inference, a fast System One model answers localized clinical questions while deterministic traversal resolves Boolean logic, turning opaque single-pass predictions into a traceable decision path without task-specific training. Across 347 cases spanning renal cyst, adnexal mass, and thyroid nodule guidelines, GDGraph improves System One accuracy from 57.3% to 90.8%, outperforming retrieval-augmented GPT-4o (80.1%) and graph-guided GPT-4o with chain-of-thought (82.7%) at substantially lower latency and cost. The source code will be available.
[CV-28] xtual Response-Aware Guidance for Zero-Shot Adversarial Robustness of Vision-Language Models
链接: https://arxiv.org/abs/2610.40309
作者: Yubo Cui,Xianchao Guan,Zheng Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Vision-language models (VLMs) exhibit strong zero-shot generalization but remain vulnerable to adversarial attacks. Existing adversarial fine-tuning methods often use pre-trained model guidance to preserve image–text alignment while improving robustness. Our analysis of visual representation drift reveals complementary limitations of two representative forms of guidance. Matching visual features provides broad coverage by constraining all directions equally, but lacks semantic emphasis. Matching prediction distributions prioritizes semantically relevant directions, but remains limited by the candidate texts and prediction probabilities. These limitations can hinder preservation of pre-trained image–text relationships and degrade zero-shot performance. To address these issues, we propose Textual Response-Aware Guidance (TRAG), which constrains visual representation drift using the pre-trained text encoder’s responses to local variations in input embeddings. Specifically, TRAG constructs anchor-level constraints from the response directions and magnitudes around semantic anchors, then aggregates them using pre-trained image–anchor relevance, yielding an image-adaptive constraint. Extensive experiments across architectures and tasks demonstrate state-of-the-art clean and robust performance. Further analysis shows that TRAG helps preserve pre-trained image–text relationships.
[CV-29] Self-training for cross-centre coronary vessel segmentation in X-ray angiography using pre-contrast frame subtraction
链接: https://arxiv.org/abs/2610.40302
作者: Danial Moafi,Aida Moafi,Mostafa Aryafar,Ashkan Sami,Maryam Aryafar
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: 7 pages, 2 figures, 3 tables
Abstract:Deep learning models for coronary vessel segmentation in X-ray angiography often perform poorly on data from centres that were not represented in training. Apart from differences in equipment, the labels of the largest public dataset (ARCADE) are incomplete, as only the segments defined in the SYNTAX score are annotated. In this study, we evaluated whether self-training with pseudo-labels can improve performance on external data without additional manual annotation. Pseudo-labels were generated from two sources, namely the predictions of an nnU-Net model trained on ARCADE and a contrast map obtained by subtracting a pre-contrast frame from the same sequence. The nnU-Net predictions were extended along the contrast map, and pixels detected only in the subtraction image were assigned an ignore label and excluded from the loss. Six models were trained and tested on ARCADE and on two external datasets (XCAD and XACV). All first-round models initialised with the baseline predictions performed better than both the baseline and a control model fine-tuned for the same number of epochs. The model trained on the XCAD pool improved the macro-averaged Dice score by 2.7 points (95% CI 2.3-3.2), while the control decreased it by 0.6 points. A second round of self-training did not lead to further improvement. When an external model trained on 50 fully annotated images was used to generate the initial labels, the XCAD Dice score increased from 0.623 to 0.745 without any XCAD labels, although performance on XACV did not improve because large main vessels were missed. These results suggest that the benefit of self-training depends largely on the sensitivity of the model used to generate the initial pseudo-labels.
[CV-30] LoRe-MIM: Lost-or-Replaced Masked Image Modeling for Self-Supervised Representation Learning in Whole-Body PET/CT
链接: https://arxiv.org/abs/2610.40284
作者: Amartya Bhattacharya,Bashirul Azam Biswas,Biratal Raj Wagle,Michael S. Leapman,Matthew E. Maeder,James B. Yu,Indrani Bhattacharya
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 25 pages, 9 figures
Abstract:Deep learning models for whole-body positron emission tomography/computed tomography (PET/CT) hold significant clinical potential for disease detection, treatment planning, and response assessment. However, the time and cost associated with expert annotations remain major bottlenecks in developing robust and generalizable models. Although self-supervised learning (SSL) has emerged as a promising approach to mitigating annotation scarcity, its application to whole-body PET/CT remains relatively underexplored due to the limited availability of large-scale datasets and the challenges associated with learning complementary cross-modal representations. To address these limitations, we curate a heterogeneous, multi-tracer, multi-institutional cohort comprising 6,531 whole-body PET/CT scans and propose LoRe-MIM (Lost-Or-Replaced Masked Image Modelling), a novel self-supervised framework designed to learn multimodal representations through masked image reconstruction. Specifically, LoRe-MIM introduces a masking strategy in which image regions are either replaced with zeros (Lost) or substituted with corresponding complementary imaging modality (Replaced), encouraging the model to capture both modality-specific characteristics and cross-modal relationships. We evaluate the learned representations across multiple downstream tasks, including multimodal PET/CT image reconstruction, patch-based classification into cancer vs. non-cancer, whole-body lesion segmentation, label-efficient training, and robustness assessment. Experimental results demonstrate that LoRe-MIM consistently outperforms the corresponding state-of-the-art baselines across these tasks, highlighting its potential as a scalable and generalizable self-supervised learning framework for whole-body PET/CT analysis.
[CV-31] askFrame: Connecting Task-Level Planning with Frame-Level Control for Game Agents
链接: https://arxiv.org/abs/2610.40276
作者: Guangzhi Wang,Junfu Pu,Yiran Wang,Xingyilang Yin,Honghao Chen,Cheng Cheng,Lingen Li,Yinan Zhou,Jiarui Wu,Yinuo Jiang,Hongbo Jin,Xiaodong Cun,Ying Shan
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Building general game agents requires both high-level task reasoning and low-latency action control in rapidly changing visual environments. These capabilities favor models with substantially different computational profiles, making them difficult to realize within a single model. In this work, we present TaskFrame, a hierarchical planner-controller framework that connects task-level planning with frame-level keyboard-and-mouse control. A vision-language task planner decomposes a high-level goal into an ordered sequence of atomic instructions and is optimized through supervised learning and dependency-aware GRPO. The hidden states of special goal tokens that terminate each atomic instruction provide continuous goal embeddings that are cached and reused across control steps. A causal Transformer controller combines the embedding of the active goal with recent observations to predict discrete keyboard and mouse-button actions and continuous camera motion at every game step. We train the system in three stages: planner training, controller pretraining, and joint alignment. Experiments in Minecraft show that TaskFrame is competitive with general-purpose vision-language models on instruction decomposition, and that in closed-loop evaluation the agent achieves a higher combat success rate than the evaluated baselines.
[CV-32] Benchmarking color judgment in multimodal LLM s with CIE-computable ground truth
链接: https://arxiv.org/abs/2610.40212
作者: Shuwei Yue,Kaida Xiao,Zhihua Wang,Qiao Chen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Multimodal large language models (MLLMs) are now used as zero-shot judges of image quality, but their color judgments are rarely tested with colorimetric ground truth. We present a five-task benchmark based on the CIE colorimetric framework: AWB pair judgment, skin-tone bias direction, CCT estimation, CIE Delta E 2000 pair ordering, and color naming under changing illuminants. We test eight MLLM endpoints, including four Qwen3-VL variants, GLM-4.5V, GPT-5.4, Claude-Opus-4.7, and Gemini-3.1-Pro. The models often solve coarse binary AWB judgments, but they remain weak on numeric CCT estimation and on some fine color-difference judgments. On ColorChecker-24 stimuli, the three closed-weight frontier models reach CCT MAE of 715-1056 K, compared with 1358-2133 K for the five open-weight models. A non-ColorChecker control shows that this advantage is partly template-dependent. The results suggest that current MLLMs have useful color words and coarse visual priors, but their color judgments are not reliably calibrated to CIE-computable color differences.
[CV-33] Decision-Level Error Attribution and Depth-Aware Safety Evaluation of Monocular RGB Human Tracking for Autonomous Robot Navigation
链接: https://arxiv.org/abs/2610.40199
作者: Abdulahi Taiwo,Chijioke Cyriacus Ekechi
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注: 10 pages, 7 figures, 6 tables
Abstract:Monocular RGB tracking is useful for robots operating near people, but perception errors affect binary stop/proceed decisions in ways that geometric accuracy metrics do not capture. We examine how much decision error arises from the safety rule versus perception. We prove that lateral-only distance is a lower bound on planar distance: with accurate positions, it cannot miss dangers identified by the planar rule, but it produces unnecessary stops. We introduce a two-factor Shapley attribution separating rule and perception contributions, and further distinguish missed detections from localization errors. Experiments use ETH and UCY pedestrian videos with simulated stationary robots, disjoint calibration and evaluation pedestrians, and time-block bootstrap intervals. Across 898 pedestrians in three scenes, the lateral rule missed no dangers on annotated positions but achieved only 5.6-31.9% stop precision. In two scenes with RGB perception, calibrated ground-plane localization achieved 0.59 m and 0.51 m planar RMSE, compared with 1.72 m and 1.39 m for a geometric baseline and 0.81 m and 0.68 m for calibrated Depth Anything V2. Detection recall of 0.43 and 0.26 made missed detections the leading source of missed danger. A two-standard-deviation safety margin with ground-plane localization matched or improved on the lateral rule’s missed-danger rate in all eight scene-threshold settings and reduced unnecessary stops in seven. All robot interactions are simulated, and outcomes are offline decision classifications, not demonstrations of collision avoidance.
[CV-34] Video Generation Meets Agent Harness: Agent Agent Harness: Agentic Video Generation Agentic World Modeling and Recursive Self-Improvement
链接: https://arxiv.org/abs/2610.40191
作者: Liaoyuan Fan,Yuhao Zheng,Dongxing Mao,Hangyu Ran,Siyuan Hu,Weijia Mao,Rui Zhao,Weijia Wu,Guian Fang,Yanzhe Chen,Zeyu Zhu,Yuxin Jiang,Yiren Song,Yuchao Gu,Junchao Yi,Qi Hu,Dan Kong,Alex Jinpeng Wang,Kevin Qinghong Lin,Mike Zheng Shou,Wei Liu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: website: this https URL github: this https URL
Abstract:Video generation is shifting from end-to-end synthesis toward agentic generation, where planning, tool use, and feedback enable coherent, controllable multi-shot videos. In parallel, video world models provide action-conditioned observations and predicted futures, serving as interactive environments for agents. We study their intersection through the agent harness: an executable runtime that constructs context, exposes tools, maintains state, routes feedback, and manages budgets around video models. This report examines three directions. First, agentic video generation, where harnesses coordinate scripting, shot planning, asset management, production, and editing, alongside the evaluation challenges of these systems. Second, agentic world modeling, where harnesses mediate observation, control, execution, and state. We discuss hybrid designs in which game engines or agent-written programs maintain world state while video models render observations, and how agents and world models can mutually evaluate and improve one another. Third, recursive self-improvement (RSI) in video model development, distinguishing output revisions, updates to harness state, generator parameters, and evaluators, and changes to the improvement procedure itself. We identify how these improvements feed back into subsequent development. Across these directions, we analyze essential agent capabilities, including audiovisual perception, long-horizon planning, memory, code-based interaction, feedback-guided control, skill reuse, and multi-agent coordination. We conclude with challenges in autonomous video production, lifelong memory, programmable hybrid control, autonomous model development, provenance, cost accounting, and evaluation linking video quality and task performance to sustained improvements in models and harnesses.
[CV-35] AdMuse: Skill-Guided Creative Advertising Image Generation
链接: https://arxiv.org/abs/2610.40190
作者: Minyan Luo,Yuxin Zhang,Meng Zhao,Xincan Wang,Fuzhang Wu,Tong-Yee Lee,Oliver Deussen,Fan Tang,Weiming Dong
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:An outstanding creative advertising image should both attract attention and communicate a product’s positioning. How to choose a persuasive conceptual bridge before rendering begins is challenging for advertising generation. We present AdMuse, a skill-guided multi-agent framework that distills high-quality advertisements into reusable creative skills for cross-domain advertising strategy. Rather than retrieving isolated exemplars, AdMuse abstracts recurring bridge patterns into a skill library that can guide new product briefs. At generation time, AdMuse coordinates skill routing, multi-candidate ideation and selection, and VLM-guided visual refinement: diverse skills propose alternative concepts, relative judgment selects the most promising direction, and vision-language feedback helps realize the selected intent in the final image. This design treats advertising creativity as a structured decision process rather than a single prompt-writing step. Experiments across diverse product and brand positionings show that AdMuse produces visually compelling advertisements that faithfully communicate product positioning.
[CV-36] Beyond CCT: Multi-channel SPD recovery for per-shot camera color correction
链接: https://arxiv.org/abs/2610.40175
作者: Shuwei Yue,Hongke Wei,Qiao Chen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Modern cameras correct color via two or three pre-calibrated 3x3 matrices interpolated by Correlated Color Temperature (CCT). This collapses the illuminant SPD to a scalar. LEDs, fluorescents, and mixed sources sit off the Planckian locus, and metameric illuminants with the same chromaticity demand different optimal CCMs that a CCT pipeline cannot distinguish. We recover the illuminant SPD instead using non-negative least squares against a basis dictionary, conditioned on the camera white point and a multi-channel ambient-light sensor (ALS). A closed-form 3x3 CCM is then fit per shot. On a synthetic benchmark of 300 illuminants and 3,100 reflectances, mean angular error drops from 1.89 to 1.20 degrees, with a 1.7x tail reduction on the LED-like off-Planckian group (n=45) where metameric ambiguity is tangible. The gain holds under realistic ALS imperfections and observation noise, and the same off-Planckian advantage carries over to 1,684 real measured SPDs.
[CV-37] EventSeeColor: Learning to Reconstruct Color from Monochrome Events
链接: https://arxiv.org/abs/2610.40166
作者: Chuanzhi Xu,Langyi Chen,Haoxian Zhou,Zelin Zhang,Haodong Chen,Qiang Qu,Xiaoming Chen,Weidong Cai
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Event cameras capture visual brightness changes under challenging lighting conditions. However, traditional event-based image reconstruction targets grayscale intensity and produces images without scene color. Color distinguishes surfaces with similar intensity. Reconstructing color from events can reveal appearance differences that reconstructed grayscale images cannot express. Although monochrome event cameras do not directly measure color, their event-triggering patterns depend on surface reflectance, illumination, and the sensor’s spectral response. These patterns may contain cues that support color reconstruction. In this paper, we present EventSeeColor, which uses event-reconstructed grayscale and aligned events to jointly guide a diffusion colorizer. Grayscale-derived semantic context supports color inference, while region-balanced chroma supervision and local RGB refinement improve color distribution, brightness, and texture. In the combined evaluation on two real event datasets, EventSeeColor achieves the best structural similarity, learned perceptual scores, and chromatic-region color difference among the external methods.
[CV-38] Hierarchical Step Decomposition for Training-free Online Video Step Grounding WACV2027
链接: https://arxiv.org/abs/2610.40153
作者: Long Zhou,Luca Zanella,Paolo Rota,Massimiliano Mancini,Elisa Ricci
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: WACV 2027
Abstract:Given a set of procedural steps, Video Step Grounding (VSG) aims to identify which steps occur in a video and localize when they are performed. Existing methods are either offline, requiring access to the full video before making predictions, or online, but relying on fixed-length video segments producing noisy, over-segmented outputs. In this work, we propose Hierarchical Step Decomposition (HSD), a training-free framework for online VSG that leverages the semantic understanding capabilities of Large Multimodal Models. HSD addresses two central challenges in online grounding: constructing semantically coherent temporal segments from a streaming video, and bridging the semantic gap between coarse procedural step descriptions and visual observations. To construct elementary temporal units, HSD incrementally groups incoming segments into semantically consistent events, using an anchor caption that is continuously validated against the visual stream. Based on these events, we perform zero-shot procedural step classification and selectively refine uncertain predictions by decomposing candidate steps into finegrained substeps that capture more nuanced visual evidence. By operating on semantically coherent events rather than every fixed-length segment, HSD reduces redundant computation and produces more stable predictions. Experiments on HT-Step, Ego4D Goal-Step, and CrossTask show that HSD substantially outperforms state-of-the-art methods, overall achieving better step recognition, more accurate temporal localization, and faster inference.
[CV-39] NeurEO: dissecting Earth observation embeddings with computational neuroscience
链接: https://arxiv.org/abs/2610.40137
作者: Thijs L van der Plas,Jacob JW Bakermans,Ioannis N Athanasiadis
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Originally published at the EurIPS 2025 REO workshop
Abstract:Earth embeddings are compressed representations of rich Earth observation data that enable downstream models to tackle a wide range of complex geospatial tasks. Yet, this increase in performance comes at the cost of decreased interpretability, as the semantic meaning of Earth embeddings is unknown. To address this challenge, we borrow methods from the field of computational neuroscience, which aim to interpret activity in (biological) neurons. Here, we adapt and apply these methods to assign meaning to AlphaEarth embeddings. We analyse embedding features both as stand-alone data, equivalent to spontaneous neural activity, and in relation to land cover and elevation correlates, equivalent to stimulus-response neural activity. We demonstrate correlational and multi-scale structure in the embedding features. Embedding features can display highly specific and directional tuning to land cover and elevation, such as a `North vs South woodland edge’ response. When we remove the main trends of combined land cover responses (like water versus land tuning), we discover fine-grained representational details with increased effective dimensionality. Together, our analyses start to unravel what these Earth embeddings represent, which is critical for developing trustworthy downstream applications.
[CV-40] VideoMemBench: How Well Do MLLM s Remember in Video Conversations?
链接: https://arxiv.org/abs/2610.40131
作者: Zhengdong Hu,Chao Wang,Kaituo Feng,Huanjin Yao,Mingfei Lu,Hehe Fan,Yi Yang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Multimodal large language models (MLLMs) can now process hour-long videos, making extended multi-turn dialogue over a single video a natural mode of use. Sustaining such a dialogue requires more than answering each question in isolation: the model must retrieve and reuse information established earlier in the conversation, both what it has seen and what it has said, an ability we term in-context video memory. Existing video QA benchmarks leave this ability unmeasured: their questions are mutually independent, never reference one another, and each is answerable from the video alone, so nothing ever requires carrying information across turns. We introduce VideoMemBench, in which each sample pairs a single video with a long multi-turn dialogue, whose turns are connected by typed cross-turn dependencies: an Episodic Dependency refers back to a visual episode described in an earlier turn, probing memory of what the model has seen, while an Answer Dependency builds on the model’s own earlier answer, probing memory of what the model has said. Every dependency is verified considering two properties: 1) necessity, meaning the question cannot be answered without its earlier turns, and 2) non-leakage, meaning the question does not restate the information from its earlier turns. To enforce these properties, we design a construction pipeline with three LLM roles: a planner, a generator, and a judger, with each dialogue reviewed by human experts. VideoMemBench comprises 326 videos with 2,476 turns, of which 1,387 depend on earlier turns, including multi-parent dependencies and long-range call-back. Among over 20 evaluated open- and closed-source MLLMs, Claude Opus 5 performs the best with an overall score of only 45.7%, highlighting the challenge of in-context video memory. Our dataset and evaluation are released.
[CV-41] Learning a Motion Manifold to Constrain Ambiguity in Unsupervised 4D Medical Image Interpolation
链接: https://arxiv.org/abs/2610.40121
作者: Haojin Li,Huazhu Fu,Heng Li,Jiang Liu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Unsupervised 4D medical image interpolation reconstructs intermediate anatomical states from temporally sparse observations without intermediate supervision. Distinct motion fields may attain comparable endpoint agreement while producing different intermediate anatomy, giving rise to motion ambiguity. This ambiguity is difficult to constrain when many independently adjustable motion variables must be estimated from the observed endpoints. We develop a framework that learns a motion manifold to restrict candidate motion fields while retaining sample-specific variation. Cross-Sample Spectral Basis Learning extracts recurring motion structure from endpoint-derived dense estimates and encodes it in a shared spectral basis. Sample-Conditioned Cayley Basis Reorientation uses predicted motion coordinates to rotate the effective motion directions for each endpoint pair without increasing the number of independent coordinates. Their composition defines a nonlinear family of complete velocity fields across endpoint pairs, accommodating anatomical and motion-state differences within a restricted representation. Integrating each predicted field supports bidirectional reconstruction at arbitrary intermediate times. Theoretical analysis establishes sufficient conditions under which endpoint observations distinguish nearby motions within the learned family. Experiments on ACDC and 4D-Lung demonstrate state-of-the-art performance among the evaluated methods trained without intermediate volumes and competitive results against methods using intermediate supervision.
[CV-42] SynthFaces: Balanced Large Scale Human Dataset BMVC2026
链接: https://arxiv.org/abs/2610.40119
作者: Sergey Karpukhin,Denis Bobkov,Kirill Morozov,Aibek Alanov
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: accepted at BMVC 2026 github: this https URL
Abstract:The pursuit of fair and equitable AI in human-centric tasks is critically hindered by the inherent biases and privacy risks of real-world datasets, which offer imbalanced representations of identity and often contain sensitive personal information. Consequently, state-of-the-art generative models trained on this skewed data exhibit unfair performance, particularly in identity-preserving generation. In this work we propose a pipeline for controllable generation of dataset with diverse faces for generative models training and finetuning. We achieve this by expanding foundational Arc2Face diffusion model with a flow matching generative model in the manifold of identity embeddings. We demonstrate that such approach achieves both goals of generative fidelity and label diversity not present in real human-centric datasets. As a result we collect and further contextualize a highly balanced identity-centric dataset of humans named SynthFaces. By design, our released images and identities are synthetic, reducing direct reliance on real personal images; we empirically check identity separation with a nearest-neighbor non-leakage metric. We purpose it for training and tuning of fair models for various downstream tasks, and showcase its usability in finetuning an existing adapter-based method for personalized generation.
[CV-43] SelfThinker: Self-Evolving Reasoning via Reliability-Verified Intrinsic Visual Thoughts
链接: https://arxiv.org/abs/2610.40115
作者: Zhangquan Chen,Ruqi Huang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 25 pages, 12 figures
Abstract:Unified multimodal models can both generate and understand images within a single set of weights, offering an opportunity to use image synthesis as training supervision for visual comprehension. However, existing visual chain-of-thought methods rely on human-annotated helper images during training and require slow image generation at test time. We present SelfThinker, the first self-evolving framework that turns a unified model’s self-generated visual thoughts into stronger understanding without human image annotations. Given a visual problem, SelfThinker samples candidate helper images from its generation pathway and uses its frozen understanding pathway to score each candidate via conditional pointwise mutual information. High-utility visual thoughts then refine the generator via flow matching and serve as privileged teacher context for on-policy self-distillation. At inference, SelfThinker reasons directly from the original image in text without generating intermediate images. Although trained only on small synthetic visual-search and navigation datasets, SelfThinker transfers broadly to natural-image and spatial benchmarks, raising the seven-benchmark average of ThinkMorph-7B from 52.7 to 64.0 (+11.3 points). Controlled image-perturbation probes further show that SelfThinker surpasses ThinkMorph-7B by up to 40.4 points and its pretrained BAGEL-7B backbone by up to 23.2 points across fine-grained and spatial perception tasks. Our code is available at this https URL.
[CV-44] RepConTUR: A Report-Conditioned Volumetric Multimodal Model for Abdominal Tumor Segmentation CVPR2027
链接: https://arxiv.org/abs/2610.40094
作者: Seunghwan Lee,Sangjoon Park,Jongmin Lee,Yoon Jo Kim,Wonyoung Cho,Jin Sung Kim,Han Joo Chae
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Submitted to CVPR 2027
Abstract:Tumor delineation is required for radiotherapy planning, yet automatic segmentation is limited by errors, predominantly false positives, that distort dose and increase editing time. The radiology report accompanies the scan and localizes disease at the suborgan level, but existing text-promptable models condition on short referring phrases, and most operate in 2D. We present RepConTUR, a single native-3D early-fusion multimodal model that conditions volumetric abdominal tumor segmentation on the narrative report. We extend AbdomenAtlas 3.0 to a suborgan-level promptable segmentation dataset and generate report-based VQA pairs, yielding four report-derived tasks on the same annotations: referring segmentation (RES), report-grounded segmentation (GSEG), report generation (RRG), and visual question answering (VQA). We train RepConTUR, comprising a 3D CT encoder, an early-fusion language model, and a native-3D segmentation module, jointly on the four tasks. Across liver, kidney and pancreas, RepConTUR attains the highest Dice similarity coefficient (DSC), instance-wise precision, and detection F1 among the text-promptable baselines, and adding the report raises mean precision while leaving recall comparable. Against two vision-only specialists, nnU-Net and the report-supervised R-Super, RepConTUR attains the highest value on every in-domain metric for liver and pancreas and the highest mean DSC, recall, precision, and detection F1 over the four out-of-distribution public cohorts. Ablations indicate that conditioning on the narrative report raises detection F1 from 0.499 to 0.528 within the same model, that joint GSEG training raises RES F1 from 0.450 to 0.482, and that the native-3D segmentation module raises GSEG F1 from 0.139 to 0.528 over its 2D counterpart. Code is available at this https URL.
[CV-45] Neural Blind Deconvolution for Image Scanning Microscopy
链接: https://arxiv.org/abs/2610.40088
作者: Eliot Guez,Nathan Buskulic,Lisa Cuneo,Luca Calatroni,Giuseppe Vicidomini
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 4 pages, 2 figures
Abstract:Image scanning microscopy (ISM) acquires one scanned image per detector element, which are then fused into a single image by multi-image deconvolution to obtain a super-resolved image. Its accuracy, however, depends on point spread function (PSF) models that sample- and system-induced aberrations often invalidate. Since differences between the scanned images encode aberration information, we propose a training-free blind Deep Image Prior (DIP) framework that jointly estimates the specimen and the detector-dependent PSFs. The PSFs are parametrised through the detector geometry and shared excitation and emission PSFs, reducing the unknowns while preserving the physical ISM model, and a Gaussian pre-initialisation stabilises the blind optimisation. On experimental tubulin data, blind DIP recovers finer filamentous structures than multi-image deconvolution and DIP with fixed theoretical PSFs, while adapting the PSFs to the measurements.
[CV-46] Morphology-Aware Channel-Decoupled Mamba Network for Medical Image Segmentation
链接: https://arxiv.org/abs/2610.40085
作者: Chenchen Wang,You Zhou
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Accurate multi-organ segmentation in computed tomography (CT) is essential for computer-aided diagnosis, treatment planning, and disease monitoring. However, substantial variations in organ shape and size, together with low tissue contrast and ambiguous boundaries, complicate the modeling of heterogeneous anatomical structures. To address these challenges, we propose a Morphology-Aware Channel-Decoupled Mamba Network (MCDNet), which integrates adaptive local structural enhancement with channel-decoupled long-range modeling. Specifically, we introduce a Morphology-Aware Channel-Decoupled State-Space (MCSS) block that divides feature channels into separate processing groups, allowing differentiated structural representations to be learned within an efficient state-space framework. Within each group, a Morphology-Aware Enhancement Module (MAEM) combines deformable convolutions with channel attention to adapt local feature extraction to anatomical shape variations. Group-wise state-space modeling then captures long-range spatial dependencies, and channel gating adaptively integrates the resulting features. Experiments on the Synapse abdominal CT dataset demonstrate the effectiveness of MCDNet, achieving an average Dice Similarity Coefficient of 81.61% for multi-organ segmentation.
[CV-47] Uncertainty-aware Coarse-to-Fine Main Pancreatic Duct Segmentation for Width Measurement in 3D MRCP
链接: https://arxiv.org/abs/2610.40073
作者: Zipei Zhang,Jaakko Sahlsten,Mikko Saukkoriipi,Johanna Ronkainen,Mona Lehto,Akseli Aittoniemi,Irina Rinta-Kiikka,Kimmo Kaski
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Assessment of the main pancreatic duct (MPD) from volumetric magnetic resonance cholangiopancreatography (MRCP) scans is an essential task for ductal dilatation diagnosis and pancreatic cancer analysis. However, manual assessment is affected by reader variability in which automated analysis may assist. Nevertheless, fully automated and uncertainty-aware MPD analysis from volumetric MRCP remains underexplored area of research. In this study, we propose an automatic and uncertainty-aware deep learning-based pipeline for MPD segmentation and maximum-width analysis from MRCP scans. A coarse localisation model first identifies the ductal region-of-interest, followed by a fine model that segments the MPD within the region. The dataset consists of 256 MPD-segmented MRCP scans from 69 patients. A randomly selected holdout test subset of 14 patients with 55 MRCP volumes was used for evaluation. We employed a five-model cross-validation ensemble for improved performance and uncertainty estimation. The best MPD segmentation performance was 0.634 DSC, 0.744 precision, and 0.593 recall. The mean absolute error for maximum MPD width prediction was 0.355~mm and dilated MPD ( \geq 5 ~mm) was detected with 75.0% sensitivity and 74.4% specificity. These novel results suggest that automated MPD segmentation can be used to aid MRCP-based duct measurements.
[CV-48] BRACHIO: Broad Representations Across Centres for Head-and-neck Imaging Oncology (HECKTOR 2026) MICCAI
链接: https://arxiv.org/abs/2610.40068
作者: Roxanne De Maeyer,Ine Dirks,Yannick Brackenier,Simon DeKeyser,Jef Vandemeulebroucke,Vicky De Ridder
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: HECKTOR Challenge 2026, MICCAI, head and neck cancer, PET imaging
Abstract:The HECKTOR 2026 challenge benchmarks machine learning for head and neck cancer, where teams delineate primary and nodal tumour volumes, perform TN staging, and predict recurrence-free survival (RFS). This report describes the submissions for the HECKTOR challenge of team nucli. As the three tasks span both dense and global information from multicentric PET/CT scans, we utilise self-supervised pre-trained encoders, training a 2D DINOv2 encoder (PET-DINOv2-MIPs) on HECKTOR 2026’s PET scans. Our final test submission ensembles a DynUNet trained from scratch with a SwinUNETR whose Swin trunk was initialised from a SimMIM-pre-trained checkpoint for segmentation, reaching a mean Dice of 0.5409. For TN staging, GTVp/GTVn geometric features from the segmentation masks are fit to stage probabilities by logistic regression (LR) and fused with PET-DINOv2-MIPs deep features (PCA + LR) via a second-level LR, reaching balanced accuracies of 0.3962 (T-stage) and 0.6432 (N-stage). For the RFS prognosis task, we fine-tune the last two blocks of the PET-DINOv2-MIPs ViT-S encoder jointly with an MLP head under a CoxPH loss, concatenating clinical covariates with the imaging features, reaching a test Harrell’s C-index of 0.6577 and a weighted challenge score of 0.5802. The prognosis result is comparable to the best entry of the previous edition and consistent with our internal cross-validation, while our segmentation result is well below it and limits the staging results that depend on it.
[CV-49] Localizing Concepts in Visual Autoregressive Models
链接: https://arxiv.org/abs/2610.40061
作者: Nanxiang Jiang,Yang Liu,Yuanhao Wang,Min Xu
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Understanding the internal routing of visual concepts is crucial for the safe and controllable adaptation of generative models. While concept localization has been widely studied in diffusion models, the emerging paradigm of next-scale visual autoregressive models remains largely unexplored. In this paper, we introduce LoCo, the first model-agnostic framework to localize where and when specific concepts emerge within autoregressive models. Driven by the native coarse-to-fine nature of next-scale generation, our method precisely maps conceptual knowledge across three distinct dimensions: Layer, Scale, and Position. To systematically localize and evaluate concept routing without context bias, we propose LoCoBench, a comprehensive dataset spanning 10 diverse categories. Extensive probing on image and video autoregressive models like Infinity, HunyuanImage-3.0, and InfinityStar shows that the localized positions are both interpretable and causally related to concept emergence. Building on these insights, we apply our localization method to three key applications: concept erasure, model personalization, and adversarial concept injection. Experiments demonstrate that our targeted intervention achieves state-of-the-art performance, substantially reducing computational overhead while preserving benign utility. Overall, our findings offer insights into how conceptual knowledge is routed during autoregressive generation, introducing a practical pathway for more interpretable, efficient, and secure adaptation.
[CV-50] EPC-RawDC: Equivariant Physics Constrained 3D Diffusion with Raw Data Consistency for CBCT to CT Image Synthesis
链接: https://arxiv.org/abs/2610.40046
作者: Alzahra Altalib,Chunhui Li,Alessandro Perelli
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Medical Physics (physics.med-ph)
备注: 10 pages, 4 figures, 8 tables
Abstract:Cone beam computed tomography (CBCT) is widely used for image guided and adaptive radiotherapy, but scatter, noise, reconstruction artifacts, and inaccurate Hounsfield units limit its quantitative use. Recent diffusion approaches for CBCT to CT synthesis remain largely image driven, with limited use of the underlying acquisition physics. We propose EPC-RawDC, a conditional 3D latent diffusion framework that combines volumetric generative modeling with two complementary constraints derived from the same scanner specific CBCT forward model. First, geometry equivariant projection constraint exploits the rotational equivariance of the CBCT forward operator by supervising the synthesized CT through correspondingly transformed reference projections. Second, raw data consistency uses noisy projection measurements of the individual acquisition as scan specific supervision during training. The framework was evaluated on two independent clinical cohorts with distinct scanner geometries and projection sources. For JUST, the physics constraints used measured clinical raw CBCT projections acquired with an Elekta system, whereas for SynthRAD2025 the projection data were simulated from the reference CT using a Varian geometry with Poisson photon statistics. EPC-RawDC achieved an SSIM of 0.97 in both cohorts and improved PSNR over DDPM by 2.66 dB on SynthRAD2025 and 2.10 dB on JUST, while reducing mean MAE by 20.3% and 19.0%, respectively. On SynthRAD2025, projection consistency and equivariant errors were reduced by 55.0% and 41.4% relative to DDPM. These results demonstrate that combining a volumetric diffusion prior with scanner geometry and scan specific projection measurements improves quantitative CBCT to CT synthesis across both simulated and measured projection domains. Code available after publication at this https URL.
[CV-51] Segment Anything Model Based Context Guided Mamba Unet for Medical Image Segmentation
链接: https://arxiv.org/abs/2610.40029
作者: Chenchen Wang,You Zhou
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Medical image segmentation is crucial for clinical diagnosis and treatment planning, requiring both global contextual understanding and fine-grained structural precision. While the Segment Anything Model (SAM) demonstrates strong generalization capabilities, its direct application to medical images often lacks domain-specific adaptability. Meanwhile, State Space Models (SSMs)-based methods, such as Mamba, efficiently capture long-range dependencies but may struggle with preserving fine anatomical details. To address these challenges, we propose a segment anything model based context guided mamba unet, SAMBA-UNet, a novel framework that integrates SAM’s context knowledge with Mamba’s hierarchical feature learning to enhance both global representation and local detail refinement. Specifically, we introduce the Hierarchical Context Enhancement Module (HCEM), an attention-based fusion module that deeply incorporates the multi-level features of SAM and Mamba to improve detail perception and segmentation precision. Additionally, we employ LoRA and Adapter-based fine-tuning to adapt SAM to medical imaging tasks, further enhancing its domain-specific segmentation capability and robustness. We conducted a comprehensive study on several medical segmentation datasets, including Synapse, ISIC17, and ISIC18, evaluating our method across multiple metrics. Experimental results demonstrate that SAMBA-UNet achieves state-of-the-art performance, effectively balancing generalization, segmentation accuracy, and computational efficiency. Our code will be released upon acceptance.
[CV-52] DINOv2 for PET: A Pre-Training Domain Benchmark Across Classification Segmentation and Prognosis MICCAI2026
链接: https://arxiv.org/abs/2610.40018
作者: Roxanne De Maeyer,Yannick Brackenier,Vicky De Ridder,Simon DeKeyser,Maarten Larmuseau,Jef Vandemeulebroucke
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 8 pages + references, MICCAI 2026, CaPTion workshop
Abstract:DINO (self-DIstillation with NO labels) has become a stan- dard for self-supervised pre-training, producing features that generalise across datasets and downstream tasks. In medical imaging it has been adopted mainly for radiology (X-ray, CT and MRI), where even natural- image DINO features transfer well. In the literature, these features have been reported to transfer worse to functional imaging such as PET. Yet PET-specific DINO models have barely been introduced, and even less validated across multiple tasks and datasets. We present PET-DINOv2-2D, the first DINOv2 model pre-trained on 2D PET, and benchmark it against open-source encoders: the origi- nal DINOv2 (pre-trained on the LVD-142M natural-image dataset) and OmniRad (pre-trained on CT, MRI and ultrasound). Using the same lightweight heads for every backbone, and despite far less pre-training data, PET-DINOv2-2D matches or outperforms both on three down- stream PET tasks: normal/abnormal classification, lesion segmentation and recurrence-free survival prognosis. Its advantage is largest when few labels are available, suggesting that PET pre-training gives more transferable, label-efficient features. Code is made available at this https URL.
[CV-53] Greenscapes: Optimized Vegetation Configurations in Cities
链接: https://arxiv.org/abs/2610.40007
作者: Patryk Bartkowiak,Marcin Marlewski,Paweł Wójcik,Jae Joong Lee,Dominik L. Michels,Bedrich Benes,Sören Pirk,Wojciech Pałubicki
类目: Graphics (cs.GR); Computer Vision and Pattern Recognition (cs.CV)
备注: 18 pages (11 main paper + 7 supplementary material), 15 figures, 5 tables. Project page: this https URL Code: this https URL
Abstract:Vegetation plays a central role in cities, serving functional roles such as balancing shade and sky exposure, shaping the microclimate, carbon sequestration, and providing acoustic screening. Furthermore, aesthetic choices in landscape styles (formal, modernist, picturesque) are important for citizens’ well-being and should be considered in the design of urban landscapes. In this paper, we present a novel method for placing vegetation, ranging from shrubs to large trees, in 3D urban scenes. Our method uses individual lots as input and employs a conditioned Vision-Language Model (VLM) to generate a top-down rendering of vegetation, placing large trees, small trees, and shrubs on lawn areas. This rendering is then converted into a 3D scene by detecting individual plants and selecting appropriate 3D assets. To refine the proposed configurations, we define an optimization loop that keeps the VLM-generated positions fixed after a deterministic feasibility repair and uses CMA-ES to optimize per-instance scale transformations along with discrete vegetation asset substitutions. To define the objective function for the optimization, we evaluate the generated 3D scenes with efficient surrogate models of shade and path-based acoustic exposure. Furthermore, we render in-scene views at each optimization step to compute view-dependent perceptual descriptors, enabling the generation of both aesthetic and functional vegetation layouts for urban scenes.
[CV-54] How Much Should a VLA Trust Its Reasoning ? Step-Level Guidance as a Measured Action-Value Function
链接: https://arxiv.org/abs/2610.40002
作者: Dabin Seo,Eunhyek Joa
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 6 pages, 2 figures, 4 tables
Abstract:Reasoning vision-language-action (VLA) policies write a reasoning trace and then decode the action from it. A wrong trace yields a wrong action, and even a correct-looking trace can mislead the plan. Existing remedies ask whether the reasoning is right. We ask how strongly it should steer. CoC-CFG applies classifier-free guidance between the policy’s context and the same context without the reasoning span. One weight per Euler step attenuates or amplifies the reasoning, and w=1 recovers the original policy. We treat the choice of this weight as a Markov decision process whose action values are measured, not bootstrapped. With the expert frozen and the noise fixed, applying a weight for one step and finishing the plan under the current controller returns that weight’s value exactly. Q-adaptive guidance fits these values with small critics and improves the controller conservatively. In the Alpasim driving simulator it outperforms every fixed weight and online RL (SAC, GRPO) in open loop, and in closed loop it passes 262 of 300 scenes, against 252 for the original policy. On held-out frames of the Waymo end-to-end driving benchmark, its first three stages outperform every fixed weight, four adaptive-guidance methods, SAC and GRPO. On the hidden test set, a frozen Alpamayo 1.5 with rule-based camera and vehicle conversions and Q-adaptive guidance sets a new state of the art among published methods: 8.135159 rater feedback score, against 8.079 for the best published method.
[CV-55] UniRA: Unified Representation Alignment and Auto-Encoding for Efficient Diffusion Generation ICML2027
链接: https://arxiv.org/abs/2610.39948
作者: An Lanji,Dawei Liu,Jin Li,Haoran Xu,Mei Chen,Yu Tian
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 7 pages, 5 figures; submitted to ICML2027;
Abstract:Diffusion transformers (DiTs) have become the dominant architecture for high-fidelity image synthesis, yet they suffer from slow training convergence and high computational cost. Two recent lines of work address this from complementary angles: Representation Alignment (REPA) accelerates DiT training by aligning intermediate features with a pretrained external representation model (e.g., DINOv2), while Representation Auto-Encoders (RAE) replace the pixel-space VAE with a compact representation-space tokenizer learned from self-supervised features. However, these approaches are developed and applied in isolation: REPA still relies on a pixel-space VAE for latent diffusion, and RAE does not exploit the alignment signal that accelerates the diffusion backbone. We present \textbfUniRA, a unified framework that couples representation alignment and representation auto-encoding into a single objective. UniRA learns a representation-space tokenizer whose output serves simultaneously as the diffusion target and the alignment anchor, so the tokenizer, the diffusion backbone, and the external representation model are co-optimized. We give a theoretical result showing that this joint optimization preserves the representation quality of the external model while accelerating diffusion convergence, and we derive a principled weighting that balances the reconstruction, alignment, and generation objectives. On ImageNet class-conditional generation at 256256 and 512512, UniRA matches or surpasses state-of-the-art DiTs (SD3, REPA, RAE) in FID while reducing training steps by up to 5*. We further show that the unified representation space supports improved zero-shot classification and linear probing, confirming that generation and understanding benefit from a shared, aligned latent space.
[CV-56] ElasticFormer: Detecting Objects in HRW Shots via Elastic Computing Vision Transformer CVPR2026
链接: https://arxiv.org/abs/2610.39946
作者: Wenxi Li,Jingchen Huang,Chenyang Lyu,Moran Liu,Haozhe Lin,Guiguang Ding,Yuchen Guo
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Published at CVPR 2026. 14 pages including supplementary material. Code: this https URL
Abstract:Recent advances in gigapixel-level imaging have brought High-Resolution Wide shots to the forefront of research. However, these images present significant challenges: extreme sparsity of foreground, gigapixel-level resolutions and diverse target counts. This makes traditional close-up detectors inaccurate and slow as they are overwhelmed by the background. Although previous research has explored sparse backbones, their fixed sparsity patterns lack the adaptability required to handle diverse target numbers. To address this, we introduce ElasticFormer, a sparse backbone that dynamically allocates computational resources based on foreground proportion. After scoring windows based on variance, proposed ElasticSelector module will predict the foreground proportion for top-k selection. The mechanism guides the model to select target-containing windows, scaling resources in areas where objects are clustered. We introduce a novel loss function combined with the 3-phase training strategy for ElasticSelector, allowing it to function properly when bounding box annotations are missing. A WSOD study is carried on PASCAL VOC 2007 to evaluate its extensibility. Further, ElasticNet is created to verify its backbone-agnostic nature. In experiments on the PANDA gigapixel benchmark, ElasticFormer reduces backbone FLOPs by 80% while achieving a significant improvement in AP50 when compared to fixed-ratio sparse methods.
[CV-57] BiHashFormer: Hash-Driven Dual-Branch Transformer for Efficient Object Detection in HRW Shots NEURIPS2026 NEURIPS
链接: https://arxiv.org/abs/2610.39937
作者: Jingchen Huang,Wenxi Li,Chenyang Lyu,Moran Liu,Haozhe Lin,Yuchen Guo
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at NeurIPS 2026. 34 pages including supplementary material and the NeurIPS checklist. Code: this https URL
Abstract:Recent advances in gigapixel-level imaging have brought High-Resolution Wide shots to the forefront of research. However, these images present significant challenges: extreme sparsity of foreground, gigapixel-level resolutions and interleaving of foreground and background. This causes traditional detectors and attention mechanisms to be hindered by background, resulting in inefficiency and inaccuracy. To tackle this problem, we propose BiHashFormer, a dual sparsified branch transformer built on hashing. RoI Selector will predict the foreground proportion for top-k selecting target-containing windows. The compression branch processes windows selected and refines attention into HashAttention by discarding half of the key-value pairs, reducing cost and improving efficiency. The compensation branch uses HashMiner to perform low-cost hash searches in the remaining regions, recovering features of missed objects. Experiments confirm that the selection relationship in HashAttention is one-way and reveal the preference of different queries for keys. In experiments on the gigapixel benchmark PANDA, BiHashFormer reduces 37.5% backbone FLOPs while improving AP50 to 81.0%. Moreover, this parallel dual-branch structure allows the model to avoid inference delays caused by inter-branch dependencies, while achieving much greater stability than single-branch models, specially with low rates of RoI selection.
[CV-58] MLLM -GS: Multimodal Large Language Model Prior-Guided Fast Novel View Synthesis in Extreme Environments
链接: https://arxiv.org/abs/2610.39936
作者: Zhiliang Wu,Xinye Zheng,Kun Li,Fei Wang,Hao Guo,Yanyan Wei,Guanghui Yue,Beier Zhu,Wei Zhou,Xun Yang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Novel view synthesis (NVS) in degraded scenes aims to reconstruct 3D scenes from multi-view observations affected by various degradations, and to generate clear and photorealistic images at unseen viewpoints. Despite recent advances, existing methods often rely on degradation-specific models, limiting their generalization across diverse extreme environments. Generative models provide rich visual priors, but their outputs often suffer from pixel misalignment and color deviations, which can propagate into the reconstructed 3D scene and compromise reconstruction quality. Furthermore, the high computational and memory costs of 3D reconstruction hinder the practical application of NVS in extreme environments. To this end, we propose MLLM-GS, a multimodal large language model (MLLM) prior-guided framework for fast NVS in extreme environments. MLLM-GS consists of two branches: MLLM prior-guided view enhancement and fast 3D scene reconstruction. The former leverages the rich visual priors of a frozen MLLM to enhance views under different severe degradations and introduces a lightweight correction network to reduce pixel shifts and color distortions in MLLM outputs. With supervision from only a few paired clean views, this branch produces reliable enhanced views for all degraded observations, providing pseudo-GT views for subsequent 3D reconstruction. The latter improves 3DGS efficiency by optimizing rasterization, Gaussian memory layout, and parameter updates, reducing redundant computation, memory-access overhead, and optimization costs. Extensive experiments on multiple degraded scenes demonstrate that MLLM-GS effectively handles various severe degradations and achieves high-quality NVS with significantly lower reconstruction time and memory consumption. Project Page: this https URL
[CV-59] Spectral Invariant Fusion and Refinement for Domain Generalized Semantic Segmentation
链接: https://arxiv.org/abs/2610.39926
作者: Yansong Bai,Yunzhi Zhuge,Xinzhuo Yu,Lu Zhang,Jin Zhan,Pingping Zhang,Huchuan Lu
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Domain generalized semantic segmentation is challenging because it requires accurate dense prediction on unseen domains without any access to target-domain data during training. Existing methods improve transferability through data diversification, invariant learning, or direct use of foundation models, but they often struggle to simultaneously suppress domain-sensitive interference and preserve fine-grained spatial cues. In this work, we propose SIFR, a frequency-aware dual-stream framework that tackles domain shift at both the fusion and decoding stages. SIFR uses spectral modeling as a unified principle to regulate both heterogeneous prior interaction and dense decoder refinement. Instead of directly fusing frozen CLIP and DINOv2 features, SIFR performs spectral-guided invariant fusion to filter out domain-sensitive responses while preserving complementary semantic and structural information. In addition, SIFR incorporates a spectral decoder refinement module that stabilizes multi-scale decoder features through low-frequency statistics-based normalization and adaptive frequency selection, thereby reducing the propagation of residual domain bias in dense prediction. This design enables SIFR to jointly improve domain robustness and segmentation precision under severe distribution shifts. Extensive experiments on multiple DGSS benchmarks, including synthetic-to-real, real-to-real, and adverse-weather settings, demonstrate consistent improvements over previous state-of-the-art methods.
[CV-60] PhysAg3D: Bridging 3D Generative and Agent ic Models for High-Fidelity and Physically-Interactive Assets
链接: https://arxiv.org/abs/2610.39924
作者: Bi’an Du,Fanpeng Meng,Jiemin Fang,Chen Yang,Pufan Li,Daizong Liu,Qi Tian,Wei Hu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Generating interactive 3D assets from a single image requires realistic appearance, fidelity to the depicted object, and plausible physical behavior. Yet detailed geometry alone does not determine how an object should move or deform. We introduce PhysAg3D, which couples 3D generative models with agentic reasoning and tool use. A vision-language agent first infers semantic parts, candidate interactions, and physical priors from the image; a 3D generator then produces a textured whole-object mesh from the same image. A hierarchical interaction representation links these revisable priors to mechanisms and the generated surfaces they control, organizing interaction construction and refinement. Guided by this representation and observations of the mesh, the agent uses geometric and physical tools to construct parts, joints, deformable regions, and surface bindings while retaining useful generated geometry and materials. Rendered interactions and execution measurements guide targeted revisions within a fixed budget. The framework accommodates both rigid articulation and compliant deformation. Quantitative and qualitative evaluations on real photographs and external benchmarks demonstrate detailed, image-aligned assets with plausible interactions. Ablations support bounded refinement and source-surface preservation. Project Page: this http URL.
[CV-61] okenRepel: Generating Diverse Image Sets Without Compromising Quality NEURIPS2026
链接: https://arxiv.org/abs/2610.39913
作者: Qingtao Yu,Zhen Wang,Ning Lai,Boyang Hu,Hongdong Li,Dylan Campbell
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: NeurIPS 2026
Abstract:We present an inference-time approach that enhances the visual diversity of image sets generated by text-to-image models, without compromising the average image quality. Our method leverages the intrinsic geometry of hidden representations in pretrained models to diversify images generated from the same prompt. First, we introduce a token repulsion mechanism that increases the angle between feature tokens across samples. It takes a negative Riemannian gradient step on a von Mises-Fisher mixture distribution of tokens on the unit hypersphere. Second, to preserve the quality of each output image, we propose a token density preservation mechanism that projects the repulsion-induced prediction change onto a half-space aligned with the model’s score direction. This encourages the updates to remain within the high-density regions learned by the generative model. We further extend this mechanism to the output of each intermediate layer, enabling finer-grained iterative guidance. We evaluate our method on three types of pretrained text-to-image models. Experiments show substantial gains in visual diversity under identical prompts, while maintaining a comparable level of image fidelity.
[CV-62] sDMD: Sinkhorn Distribution Matching Distillation
链接: https://arxiv.org/abs/2610.39889
作者: Jaihoon Kim,Lingxiao Li,Soshi Shimada,Roshni Cooper,Minhyuk Sung,Yiwei Zhao
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: Project Page: this https URL
Abstract:Distribution Matching Distillation (DMD) distills multi-step diffusion models into few-step generators by minimizing reverse KL divergence. Despite its strong performance, DMD guides samples by local density alone, which introduces two limitations: teacher scores can be unreliable in regions of low teacher density, and the mode-seeking behavior of reverse KL can collapse the student onto a subset of the teacher modes. We first connect DMD to Wasserstein gradient flow, showing that its gradient follows from a fixed-point formulation with integral KL (IKL) as the energy functional. This perspective motivates augmenting IKL with an optimal transport term that addresses these limitations by guiding outlier samples toward the target and promoting mode coverage. We realize this term using Sinkhorn divergence estimated from finite batches. In practice, however, errors in the estimated Sinkhorn velocity can perturb in-domain samples for which IKL guidance is already accurate. To this end, we introduce sDMD, which adaptively activates the Sinkhorn correction based on discriminator outputs, focusing on outlier samples where IKL may be unreliable. With 2–4 sampling steps on FLUX.1-dev, Qwen-Image-20B, and Wan2.1-14B, sDMD outperforms distillation baselines on GenEval, DPG, WISE, and VBench while improving sample diversity over DMD2.
[CV-63] What Survives Label Downsampling? Exact Region Survival for Coarse Segmentation Targets
链接: https://arxiv.org/abs/2610.39880
作者: Zhiyuan Zhang,Yukun Ou,San Tek Choi
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 25 pages, 13 figures, including supplementary material
Abstract:Segmentation networks that predict at an output stride are trained against label maps downsampled by strided sampling, nearest-neighbor resampling or majority pooling. These methods label each coarse grid cell on its own, so a region that wins no cell is absent from the target and receives no positive label. We present MatchDown, which builds the coarse target from a minimum-cost matching between regions and the grid cells they occupy. It keeps as many regions as any one-region-per-cell target can, certifies how many losses are unavoidable, and with priority tiers (MatchDown-P) keeps every evaluated region (a region that a metric scores) of at least one cell’s area. On COCO, Cityscapes and BBBC038 at factors 2 to 16, MatchDown-P loses exactly the minimum number of regions, whereas the downsampling used by StarDist, nnU-Net and connectomics pyramids loses four to five times as many on COCO at factor 8. Keeping a region in the target and reconstructing its shape are different problems: we show that no region lost by majority pooling can be reconstructed above IoU 0.5 by upsampling any coarse target, yet on BBBC038 a star-convex head trained on MatchDown targets segments 35% of the nuclei that majority pooling loses above IoU 0.5, against 15% when trained on majority targets and 24% on StarDist’s strided target. On two nucleus datasets, with thresholds cross-validated for count error, this head has instance accuracy at IoU 0.5 (Acc50) 2.6 to 4.2 points above the strided target and at most 2.8 points below cell-center sampling, and a lower count error than the strided and cell-center targets on BBBC039. On BBBC039, an affinity head, which outputs one label per cell, lowers count error relative to majority targets by 9.3 nuclei per image at 7.2 points lower Acc50. MatchDown runs once, before training, in place of the label-downsampling step.
[CV-64] Parametric Gaussian Noise Mixer (PGNM): Unified Synthesis of Camera Noise and Film Grain
链接: https://arxiv.org/abs/2610.39868
作者: Mingyang Song(ETH Zürich)
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Technical report. 49 pages: 9-page main paper and 40-page supplementary material; 29 figures, 33 tables
Abstract:Noise synthesis is commonly approached through physical models of image formation or neural networks trained to generate noisy images. We explore a middle ground: an explicit statistical model fitted directly from paired clean and noisy images. A normalized convolution correlates a Gaussian field across pixels and color channels, and a lookup maps each Gaussian value to a residual given the clean intensity. Because the field stays standard Gaussian, the two components are fitted separately, without reproducing the physical process or training a neural generator. With one formulation for camera noise from SIDD and simulated film grain from Newson, the model has the lowest average distribution and correlation errors on SIDD among the evaluated methods and remains competitive on film grain. A denoiser trained on its noise comes closest to one trained on real noise, and a 1080p frame renders in 10 ms on GPU.
[CV-65] M-PFT: Angular Margin for Calibrated Prompt-Tuning of Fine-Tuned Vision-Language Models WACV2027
链接: https://arxiv.org/abs/2610.39861
作者: Shihab Aaqil Ahamed,Mohamed Ansaff,Nang Khin Myat Lay Phyu,Udaya Sampath K. Perera Miriya Thanthrige,Chamira U. S. Edussooriya,Ranga Rodrigo,Muhammad Haris Khan
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at WACV 2027
Abstract:Prompt-tuning has emerged as a parameter-efficient technique for adapting large-scale vision-language models (VLMs) to downstream tasks while keeping the pre-trained weights frozen. However, prompt-tuned VLMs often suffer from a base-to-novel dual miscalibration problem, where models typically show underconfidence on base classes and overconfidence on novel classes. Current VLM calibration approaches address this dual miscalibration either separately or jointly, but lack a geometric understanding of the hyperspherical embedding space and thus fail to exploit angular discriminative properties to adjust distorted decision boundaries. To address this, we propose M-PFT, a novel angular-margin approach for prompt tuning that introduces two complementary objectives to promote class discriminability and semantic anchoring. This enlarges the angular margin by 1) minimizing the maximum cosine similarity between tunable class embeddings to reduce embedding collapse and 2) minimizing the angular distance between each tunable embedding and its corresponding zero-shot embedding to preserve semantic geometry. Extensive experiments across 9 prompt-tuning methods with various backbones on 15 diverse datasets show that our approach reduces the average calibration error on novel classes, while remaining competitive on base classes and maintaining comparable accuracy. Further analyses reveal that M-PFT effectively adjusts distorted decision boundaries induced by prompt tuning while preserving the pre-trained embedding space, improving calibration with negligible computational overhead.
[CV-66] Open-set multi-view seahorse re-identification from underwater photograph sequences
链接: https://arxiv.org/abs/2610.39857
作者: Tristan Fauvel,Sylvie Louisy,Patrick Louisy,Thomas Fauvel
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Automated animal re-identification from photograph sequences can facilitate minimally invasive Capture-Mark-Recapture studies at scale. Yet, photographs taken during an observation session can vary greatly in quality, not always providing the necessary information for accurate re-identification. In this paper, we present new methods for taxonomic classification, sex determination, and open-set re-identification from image sequences. We train and evaluate our models on a dataset of 35,914 photographs of 474 individual seahorses observed in the wild. We demonstrate that our approach, by combining information from multiple photographs, yields state-of-the-art performance.
[CV-67] LiDAR-SHIELD: A Fully Dockerized Adversarial Robustness Assessment Framework for LiDAR-Based 3D Perception Models
链接: https://arxiv.org/abs/2610.39848
作者: Alexandros Gkillas,Ioulia Kapsali,Christos Anagnostopoulos,Aris S. Lalos
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:LiDAR-based 3D perception underpins autonomous driving and robotics, yet deploying these models safely requires stress-testing them against adversarial inputs before real-world use. Although state-of-the-art adversarial attacks exist, their implementations are often fragmented across individual repositories, each tied to a specific architecture or task and requiring manual environment reconstruction and model-specific integration code. We present LIDARSHIELD, a unified and fully Dockerized framework that reverses this workflow: rather than installing the target model inside the toolkit, LIDARSHIELD exposes any model through three lightweight HTTP endpoints, namely /info, /forward, and /gradient As a result, connecting a new model or dataset requires only a single loader function or adapter. The framework supports both 3D semantic segmentation and 3D object detection through the same engine-service protocol, integrating a task-aware suite of state-of-the-art attacks, including PGD and Square Attack for segmentation, and point perturbation, point removal, point injection, and LiDAttack for detection. This enables systematic exposure of corner cases and vulnerabilities across heterogeneous LiDAR backbones. The LiDAR-Shield codebase is publicly available at this https URL.
[CV-68] Aligning Medical Vision Language Models at the Capability Boundary with Entropy Adaptive Rubric Grounding
链接: https://arxiv.org/abs/2610.39835
作者: Hongyu Bai,Jiaqi Zeng,Chunzheng Zhu
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: Accepted at ACM Multimedia Asia 2026 (MMAsia 2026). 10 pages, 4 figures
Abstract:Medical diagnostic reasoning requires verifiable links between visual evidence and clinical knowledge, yet current medical vision language models are aligned with objectives that reward the wrong behavior. Uniform supervised fine tuning lets easy cases dominate updates while clinically critical boundary cases saturate prematurely. Sparse rewards based only on final-answer correctness cannot assess anatomical grounding or clinical logic, inviting reward hacking through answer shortcuts. We propose MedCBR, a two-stage alignment framework coupling boundary-aware preference optimization with rubric-grounded reinforcement learning. First, we construct counterfactual boundary preferences at the model’s capability frontier using empirical pass rates and preference pairs that share reasoning prefixes but diverge at decisive diagnostic branches. Difficulty Weighted Boundary Preference Optimization (DBPO) concentrates learning on this calibrated boundary region. Second, we distill reusable clinical rubrics from boundary trajectories and activate them according to prediction entropy. Entropy Gated Grounded Policy Optimization (EGPO) rewards answer correctness, anatomical grounding, and reasoning fidelity while suppressing rubric noise on reliable samples. A reference policy anchored to the boundary-aligned model couples population-level boundary priors with instance-level uncertainty, making diagnostic reasoning more accurate and clinically grounded. Experiments on six medical benchmarks show consistent improvements over strong general, specialized, and reinforcement learning baselines, reaching 61.2% average accuracy with the largest gains on challenging differential diagnosis tasks. Ablations show super-additive gains: combining the two stages exceeds the sum of their individual improvements.
[CV-69] GROME: GROunding Factorized MEmory into Video World Models
链接: https://arxiv.org/abs/2610.39812
作者: Fang Yu,Xiaoxue Wu,Haiyu Zhang,Yangfu Li,Botong Zhao,Yu Qiao,Tianfan Xue,Yue Lu,Xinyuan Chen
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Long-horizon consistency is essential for video world models, yet it becomes difficult to maintain once observed content falls outside the context window. Existing memory mechanisms either retrieve history as implicit context, which remains constrained by context capacity and entangles static and dynamic information or consolidate history into explicit global representations, which struggle to model evolving dynamics and accumulate geometric errors. To overcome these limitations, we introduce \textbfGROME, a novel factorized memory framework that augments video world models for long-horizon coherence across both static scenes and dynamic entities. GROME operates via three synergistic components: 1)~\emphStatic-Dynamic Factorization decouples historical evidence into static and dynamic components, preserving static scene geometry and dynamic entity states separately; 2)~\emphLocal Memory Retrieval selects spatially aligned static history to eliminate geometric drift alongside the latest dynamic entity states to ensure temporal continuity; 3)~\emphTraining-Free Memory Injection seamlessly grounds the factorized memory into video world models via their native conditioning pathways. Experiments on long-horizon revisit scenarios demonstrate that GROME substantially enhances scene persistence and dynamic entity fidelity over existing methods, all while preserving the visual quality and controllability of the frozen backbone without additional training. Beyond self-rollouts, GROME readily accommodates external observations as plug-and-play memory, unlocking flexible and controllable video continuation and scene exploration.
[CV-70] SCOPE: Semantic COdec Pre-Editing for Object-Preserving Compression
链接: https://arxiv.org/abs/2610.39778
作者: Silvia Dani,Leonardo Galteri,Marco Bertini
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Image compression is increasingly used before machine vision systems, yet standard codecs are mainly optimized for human perception rather than downstream task accuracy. We propose SCOPE (Semantic COdec Pre-Editing), a conditional image pre-editor that modifies images before compression to improve the bitrate and detection accuracy trade-off. The model is conditioned on the codec family and compression severity, allowing a single trained network to adapt to multiple compression settings without retraining or loading codec-specific weights. Instead of uniformly preserving the whole image, the pre-editor learns to protect task-relevant regions while simplifying less important background content to improve compressibility. The resulting rate-accuracy curves show consistent gains across different codec families and detector architectures, together with strong transferability to unseen learned and wavelet-based compression families.
[CV-71] Quantum-Inspired Convolutional Networks for High-Resolution Pseudo-Label Generation from Multimodal Earth Observation Data
链接: https://arxiv.org/abs/2610.39773
作者: Abhishek Singh,Giovanni Cuozzo,Alexander Jacob,Claudia Notarnicola
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Accurate land-cover mapping from multimodal Earth observation (EO) data remains a challenge in regions where high-resolution annotations are scarce or costly to obtain. This work introduces a quantum convolutional encoder-decoder network for pseudo-label generation using Sentinel-1 SAR and Sentinel-2 multispectral imagery, guided only by coarse MODIS-derived land-cover maps. The proposed framework uses quantum-inspired convolutional blocks to enhance feature extraction and cross-sensor fusion, enabling the model to infer a fine-scale semantic structure from heterogeneous inputs. Experiments were conducted on the DFC2020 dataset, which provides a quadruple of Sentinel-1 SAR, Sentinel-2 multispectral, low-resolution MODIS labels and high-resolution reference land-cover maps. The proposed training framework first learns a robust representation of multimodal inputs through the reconstruction branch before any supervision from low-resolution labels is applied. This self-supervised stage enables the model to capture a fine-scale spectral-spatial structure, which is then used to generate high-resolution pseudo-labels. These pseudo-labels are subsequently evaluated through classification metrics on the test data, demonstrating that the proposed approach produces more spatially coherent and class-consistent compared to classical baselines. The results highlight the potential of quantum-inspired deep learning to improve label generation in low-annotation remote sensing scenarios, offering a scalable path towards high-quality land-cover products for EO applications.
[CV-72] Cross-Encoder Anchor Refinement for Zero-Shot Anomaly Detection
链接: https://arxiv.org/abs/2610.39727
作者: Kaisong Ji,Ziyang Xu,Chang Bian,Yu Zhou,Wenyu Liu,Xinggang Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 15 pages, 6 figures
Abstract:Zero-shot anomaly detection aims to transfer anomaly discrimination from auxiliary training domains to unseen categories while localizing subtle defects. A key challenge is to construct discriminative references that combine contextual representations with complementary local evidence. We present Cross-Encoder Anchor Refinement (CEAR), which connects two visual encoding paths for the same image at the anchor stage, before normal and anomalous prototypes are constructed. An independent convolutional encoder pretrained for industrial product recognition builds an image-specific memory shared across selected vision transformer (ViT) layers. Spatial anchors read from ViT patch features query this memory through cross-attention and undergo residual refinement. Image-conditioned normal and anomalous queries then gate anchor aggregation to construct prototypes for anomaly scoring. Across 13 industrial and medical datasets, CEAR achieves the best performance in 57 of 68 evaluations spanning image-level detection and pixel-level localization. Notably, CEAR surpasses all other state-of-the-art methods in 41 of 44 pixel-level comparisons. Across 11 localization datasets, mean gains in pixel-level maximum F1 score and average precision are 4.45 and 3.79 percentage points, respectively. This study shows how cross-encoder interaction at the anchor stage can integrate complementary visual representations for fine-grained anomaly localization across domains.
[CV-73] WorldProbe: When Video World Models Fall Apart under Long-Horizon Spatial Exploration
链接: https://arxiv.org/abs/2610.39721
作者: Minqi Wang,Jie Peng,Hejun Dong,Ruchang Yao,Puyi Wang,Siyuan Huang,Jihai Zhang,Yafu Li,Qi Dou,Xiaoye Qu,Yu Cheng
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 19 pages, 9 figures
Abstract:Video world models generate future visual observations from camera or action inputs, with the goal of synthesizing persistent 3D worlds that can be explored across time and space. Existing benchmarks typically evaluate short interactions or rely on camera trajectories fixed before generation. The performance of video world models under long-horizon exploration with extensive camera motion remains under-explored. We introduce WorldProbe, a diagnostic benchmark that uses a goal-conditioned multimodal agent to adapt camera actions online during generation. WorldProbe comprises 164 test cases across four probing families. Across 10 state-of-the-art models, we generate more than 1,600 rollouts totaling approximately 800 minutes, with horizons up to 180 seconds. Our evaluation across four dimensions reveals both the degradation of local measures and the emergence of long-range inconsistencies. We find that recent models sustain visual and content quality for longer, while long-range memory and evolution remain weak. These results expose a gap between locally stable generation and persistent world consistency that is largely missed by existing benchmarks. Code and benchmark resources are available at this https URL.
[CV-74] NIF: Normal-Intersection Fields for Geometrically Consistent Oriented Box Refinement
链接: https://arxiv.org/abs/2610.39715
作者: Shaokai Zheng
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 17 pages, 6 figures, 5 tables
Abstract:Accurate oriented detection requires a box to follow object boundaries, yet a standard regression head predicts its center, size, and angle from a pooled region representation. Small errors in these coupled parameters can displace long boundaries even when the object is correctly classified. We propose NIF, a refinement module that replaces an additional box regressor with normal-intersection fields. Given a coarse oriented box, the module observes local features at 20 boundary anchors and predicts signed distances along their outward normals. These observations admit an exact affine representation in five geometric variables. A differentiable, reliability-weighted least-squares fit projects the predicted field onto this representation and analytically recovers a refined rectangle. Tangential pooling preserves the normal position profile, while shared projection limits the added network to 0.132M parameters. On DOTA-v1.0, NIF reaches 78.00%, 79.11%, and 79.02% mAP with Oriented R-CNN, LSKNet-S, and Strip R-CNN, respectively, achieving the highest single-scale SOTA result in Table 1. It reaches 68.19% on DIOR-R, improves matched FAIR1M/DIOR-R baselines by 0.30/0.68%, and raises HRSC mean IoU from 0.867 to 0.888 while reducing field error by 24.5%. Across four different detector frameworks with the detectors frozen, relative AP75 improves by 1.57-4.18% and COCO-style mAP by 0.42-3.74%, without detector retraining. Code is available at this https URL.
[CV-75] Confidence Filtering in Video-Based Gait Assessment: Measurement Error versus Selection Shift
链接: https://arxiv.org/abs/2610.39704
作者: David Regev,Eran Beeri Bamani
类目: Computer Vision and Pattern Recognition (cs.CV); Quantitative Methods (q-bio.QM); Methodology (stat.ME)
备注: 6 pages + 2-page appendix, 4 figures, 2 supplementary figures, 2 tables. Submitted to the 2027 IEEE International Conference on Rehabilitation Robotics (ICORR)
Abstract:Video-based gait assessment often keeps only high-confidence pose observations, which can lower measurement error but change the movement assessed. We compared confidence selection with budget-matched random selection in a Parkinson-related cohort (n = 96) and an NF2-related replication cohort (n = 18). In the Parkinson-related cohort, confidence selection produced a small but consistent shift in the observed movement summary (+0.05 SD, 95% CI 0.03 to 0.07), while held-out Mini-BESTest error stayed flat (-0.04 points, -0.13 to 0.05); a policy preserving motion-group proportions reduced the shift, while NF2 showed no shift and its predictor did not beat a constant baseline. Because this shift mixes measurement error with selection shift, we decompose and simulate both on the cohort’s window structure. Filtering lowers local error when error depends on confidence, but biases the endpoint once confidence tracks movement more than error. In the Parkinson-related data, confidence was only weakly tied to movement (rho = 0.14). There, filtering is harmless if well-tracked windows are unbiased. If instead systematic error only redistributes within participants, filtering can add up to 0.58 degrees. Unchanged clinical error from such small-cohort predictors therefore cannot show that quality filtering preserves the assessed movement.
[CV-76] DEQ-Stereo: Deep Euilibrium Video Stereo Matching via Adaptive Convergence ACCV2026
链接: https://arxiv.org/abs/2610.39701
作者: Kiljoon Han,Kyumin Hwang,Wonheok Choi,Jeonghoon Kim,Sunghoom Im
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: ACCV2026
Abstract:Iterative refinement models have significantly improved stereo matching accuracy, but their rigid, fixed update schedules incur heavy computational costs and a strict trade-off between speed and accuracy. This challenge is magnified in video streams, where traditional online methods resort to reducing feature channels or relying on expensive temporal warping. To address this, we propose \textbfDEQ-Stereo, a novel video stereo matching framework based on the Deep Equilibrium (DEQ) paradigm. Instead of unrolling a fixed number of iterations, DEQ-Stereo reformulates the refinement as an implicit fixed-point problem, adaptively terminating once convergence is met. To ensure fast and stable intra-frame convergence, we introduce the \textbfConvolutional Simplified Gated Unit (ConvSGU), which eliminates unnecessary non-linearities by modeling the latent state as a gated, element-wise sum of inputs. Furthermore, to exploit cross-frame temporal coherence, we propose \textbfTemporal State Transfer (TST), a warm-start strategy that propagates the converged hidden state to the next frame without requiring explicit pose or warping priors. Extensive evaluations demonstrate that DEQ-Stereo delivers state-of-the-art performance with a superior accuracy-to-latency trade-off.
[CV-77] EditAgent : Rethinking Agent ic Image Editing
链接: https://arxiv.org/abs/2610.39664
作者: Haiyang Xu,Tianyu Wang,Jialu Li,Jiteng Mu,Yuanjun Xiong,Jianming Zhang,Zhuowen Tu,Mingze Xu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Agentic image editors decompose complex requests into steps, but when a step fails, they mostly retry with the same narrow set of tools and a judge that scores each output by eye directly. We argue that such agents are bounded by two spaces: the action space of what the agent can do, and the observation space of what its judge can see. On our expert-annotated benchmark, a general-purpose editor yields no usable result in over 40% of the cases, and no judge is reliable across edit types. We present EditAgent, which expands both spaces. It routes each step to general editors and specialized methods, and attaches to each candidate a verification record. With these records, off-the-shelf VLM judges outperform finetuned reward models. Candidates are generated in parallel and judged in one listwise call: experts prefer this to sequential refinement in 78.5% of decided comparisons, at no more than half the judge calls. EditAgent improves two open-weight editors on three benchmarks and serves as a data engine for training stronger editors.
[CV-78] AB-Attention: Accurate Block Attention for Accelerating Video Diffusion Transformers
链接: https://arxiv.org/abs/2610.39663
作者: Bin Xu,Mo Zhang,Zhengfeiyang Zhang,Xinfeng Yang,Yixing Ma,Wei Wang,HeXin Wang,Pengfei Zhou,Wangbo Zhao
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 20 pages, 5 figures, 3 tables
Abstract:Video diffusion transformers generate high-quality videos, but the quadratic cost of 3D full attention dominates their inference. Dynamic block-sparse attention reduces this cost by computing exact attention only on key/value blocks selected from current activations. However, its fidelity degrades under aggressive sparsity: blocks are scored from coarse summaries or formed by costly global clustering, pruned blocks vanish from the softmax, and whole tiles are kept or dropped even when only a few of their keys matter. To address these limitations, we propose Accurate Block Attention (AB-Attention), a training-free method that accurately approximates dense attention at high sparsity. Its Intra-Block Clustering Selector (IBCS) estimates block importance faithfully yet cheaply by clustering queries only within each query block, without permuting tokens. Sampling with Inclusion-Probability Weighting (SIP) then samples blocks in proportion to these estimates and reweights them by inverse inclusion probability, so that one pass estimates dense attention rather than truncating it. Finally, its kernel gathers key/value blocks much smaller than a tile into full tensor-core tiles, refining selection at no extra tensor-core cost. On Wan2.1 and HunyuanVideo, AB-Attention skips 89–92% of attention, whereas the baselines compute 1.8–3.6 \times as much, yet it still improves PSNR by up to 4.4~dB over the strongest baseline and achieves up to 2.23 \times end-to-end speedup. Our code is available at this https URL.
[CV-79] Look at Where You Think: Localization-Head Alignment for Fine-Grained Perception in Thinking VLMs
链接: https://arxiv.org/abs/2610.39649
作者: Yuan Wang,Wei Wu,Hongning Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 23 pages, including references and appendices
Abstract:Thinking vision-language models reason for thousands of tokens, yet still miss small details in high-resolution images. Zooming methods pay for extra inference passes, and outcome rewards never tell the model where to look at. In Qwen3.5-4B, we find that the bottleneck is selection, not seeing: a few attention heads, which we call localization heads, pick the evidence out of thousands of visual tokens; ablating four of them costs 13.2 points more than ablating four random heads, and their focus fades as the reasoning grows. We therefore propose Localization-Head Alignment (LHA), which makes attention a training signal: whenever the reasoning states a visual fact, these heads learn to look at the focal region rather than the background, as an auxiliary loss in both SFT and RL training. On benchmarks of V*, HR-Bench 4K and TreeBench, LHA lifts Qwen3.5-4B from 74.1% to 76.6% accuracy, the best single-pass 4B result we evaluate, ahead of multi-pass zooming and released 7B models. Alignment alone adds 1.8 points over otherwise identical training while shortening reasoning by 20%, and the gains carry over to Qwen2.5-VL-3B.
[CV-80] Annotation-Free Representation Learning for Polyp Retrieval in Colonoscopy Videos
链接: https://arxiv.org/abs/2610.39648
作者: Luca Parolari,Lamberto Ballan,Carlo Biffi,Loic Le Folgoc
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Under submission at ISBI 2027
Abstract:Colonoscopy is inherently comparative: reliable visual assessment of polyps often depends on complementary views acquired under different viewpoints and imaging conditions. Retrieving such views requires polyp tracklet representations, which recent contrastive methods learn from temporal cues but still build from manually annotated bounding boxes. We propose an annotation-free framework that constructs tracklet views directly from raw colonoscopy videos, enabling temporal self-supervised learning on full-procedure videos, including frames without visible polyps. On REAL-Colon, our method achieves state-of-the-art results on tracklet retrieval and re-identification, outperforming both self-supervised and supervised baselines. We then introduce few-shot polyp retrieval, a task closer to clinical use, where a few observations of a target polyp are used to retrieve its remaining tracklets: with only two examples, our method surpasses most competing methods given ten. Finally, we show that a modest number of unlabeled procedures suffices for competitive performance, making on-site model development more practical. Code is available at this https URL.
[CV-81] FlexTry-On: A Time-Efficient Framework for Multi-Scenario Virtual Try-on
链接: https://arxiv.org/abs/2610.39646
作者: Phuong Dam,Quyen Nguyen Duc,Son Hoang Minh,Daeyoung Kim
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Virtual try-on has achieved remarkable progress in recent years, yet most existing methods are designed for a specific scenario, such as cloth-to-person or person-to-person try-on, and often require separate conditioning strategies and processing pipelines for different settings. We present FlexTry-On, a unified and time-efficient diffusion-based framework for diverse virtual try-on scenarios, including cloth-to-person, person-to-person, cross-domain, and in-the-wild settings. FlexTry-On adopts a common conditioning formulation, where a text-aware condition extraction strategy converts heterogeneous garment sources into a unified representation, and a dual-conditioning diffusion transformer jointly models garment appearance and person identity. Extensive experiments demonstrate competitive performance on both cloth-to-person and person-to-person benchmarks while maintaining fast inference. Moreover, the same framework can be directly applied to unseen cross-domain scenarios without requiring separate architectures or scenario-specific processing pipelines. These results highlight the potential of unified and efficient virtual try-on frameworks as a practical step toward more flexible and widely applicable virtual try-on systems.
[CV-82] Where Do Vision Language Models Fail at 3D Reasoning ?
链接: https://arxiv.org/abs/2610.39639
作者: Khai Le-Duc,Nguyen Kim Hai Bui,An Ngo,Nam D. Hoang,Yueqi Song,Jiarui Liu,Thai Quoc Hoang,Fang Wu,Chris Ngo,Terry Jingchen Zhang,Mona T. Diab,Ran Xu,Silvio Savarese,Xiao-Ming Wu
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: Preprint, 150 pages
Abstract:Vision-language models can get individual judgments right while struggling to recover and consistently use the exact properties of a 3D object. To characterize this gap, we introduce BedBench, a controlled benchmark of thirteen procedurally generated tasks spanning individual judgments and composite answers. Across twelve open-weight and four frontier models, per-color counting exposes a pronounced weakness: humans achieve 93.3% accuracy, compared with 7.2% for the best model. Targeted fine-tuning raises one model’s accuracy from 3.0% to 68.6%, but reveals a further difficulty: its answers about the most common colors remain inconsistent with its own per-color counts. Supplying those counts as text improves its judgments, showing that recovering a property does not mean using it consistently. Reasoning, scaling, and added visual cues each improve individual judgments, but none makes exact answers reliable. To locate the difficulty, we probe Qwen3.5-27B: exact per-color counts remain recoverable at the language model’s input, yet the model’s answers rarely recover them (65.0% for the probe vs. 3.0% for the model). Image-removal controls further show that some apparent successes require no visual evidence. These findings motivate evaluating 3D reasoning by exact correctness, consistent use, and image dependence, rather than accuracy alone. Code and dataset are publicly available: this https URL
[CV-83] Intrinsic Distribution Matching
链接: https://arxiv.org/abs/2610.39633
作者: Deyuan Liu,Yansen Han,Kaiwen Zheng,Jun Xie,Jiacheng Liu,Zhenglin Cheng,Peng Sun,Jintao Zhang,Jianguo Li,Tao Lin
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:One-step and few-step text-to-image generation must accurately render the details specified in the prompt despite limited opportunities for refinement. Distribution matching distillation (DMD) uses score differences to bring the generated distribution closer to a reference distribution. Following prior work, we estimate this difference with internal velocity branches, avoiding a separate fake-score network. We ask whether an explicit moment constraint improves this learned score supervision. A decomposition of the reverse-KL field into a Gaussian moment component and a non-Gaussian residual identifies statistics that can be constrained explicitly. We introduce Intrinsic Distribution Matching (IDM), which retains the full DMD correction and matches feature means and per-coordinate standard deviations using the generator’s exponential-moving-average (EMA) copy, requiring no external feature encoder. Experiments use full-parameter and low-rank adaptation (LoRA) tuning on Z-Image-Turbo 6B and Qwen-Image 20B. At two network evaluations, IDM reaches 94.25% and 90.20% on Chinese and English LongText-Bench, the highest scores among the compared methods at this budget. Controlled ablations show gains from the added moment loss, while generator EMA features perform competitively with an external DINOv2 encoder across the five reported metrics. Code is available at this https URL.
[CV-84] FutureWorld: A Plug-and-Play Causal Next-State Predictor for Interactive Video World Exploration
链接: https://arxiv.org/abs/2610.39619
作者: Guangyuan Li,Kelu Yao,Shuolin Xu,Xinlei Yu,Wei Xing,Lei Zhao,Peng-Tao Jiang,Xiaogang Xu,Xiaobin Hu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 17 pages, 6 figures
Abstract:Interactive video world models aim to follow user actions while maintaining scene consistency over extended interactions. History-based methods reuse generated latents to preserve scene content, but three limitations of this reuse contribute to error recirculation. 1) The latent generated at each interaction is retained as history, potentially containing blurred details, geometric distortions, or semantic drift. 2) Although retrieved historical content matches the current scene, it can still contain errors that are carried into the next chunk. 3) New outputs then become history for later interactions, potentially reinforcing the same errors through repeated reuse. To limit this cycle, we aim to predict the future scene state from the current generated latent and the next action before generating the next interaction’s video chunk. This future-state prior provides guidance for the next chunk, with the aim of reducing the errors carried forward from history. To this end, we propose a plug-and-play Causal Next-Chunk Predictor (C-NCP) with cross-model adaptability and prediction-guided modulation. We integrate C-NCP into four base generators and evaluate the augmented models across three benchmarks for interactive long-video generation and action-conditioned world modeling. Extensive experiments demonstrate reduced error accumulation and improved long-horizon consistency across diverse generation settings. The project page is at this https URL.
[CV-85] UniTele: Learning Device-Conditioned Degradations for Unified Smartphone Telephoto Super-Resolution
链接: https://arxiv.org/abs/2610.39617
作者: Yuchen Zhang,Aojie Li,Wei Cheng,Xianfang Zeng,Runxing Liu,haixin hao,Gang YU,Xiaoyun Yuan
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Super-telephoto imaging on mobile phones suffers from complex, device-dependent degradations that are poorly captured by conventional handcrafted pipelines. Collecting large-scale aligned training pairs across different phones and diverse scenes, however, is both costly and difficult to scale. We therefore study whether device-specific degradations can be learned from only a small number of physical captures and transferred for training a unified super-resolution (SR) model. Our key idea is to factor degradation learning into two learnable components: one captures the device-specific imaging characteristics under well-resolved conditions, while the other captures the additional information loss and artifacts that arise in super-telephoto imaging. We build a paired benchmark across three phone brands (Huawei, Apple, and vivo) by photographing the same content at two display sizes and zoom settings. The larger displayed image is captured at the phone’s maximum optical zoom setting, while the smaller image is photographed directly at the phone’s highest native zoom setting through the phone’s native imaging pipeline. The benchmark contains 519 training and 60 held-out pairs. We use the benchmark to learn device-specific degradation, then apply it to external images to synthesize thousands of additional pairs to train our UniTele SR model. Relative to the pretrained Qwen and FLUX backbones before adaptation, the full UniTele training recipe substantially improves paired restoration fidelity and perceptual similarity on the held-out benchmark and yields markedly more stable and visually faithful restorations.
[CV-86] DiFuse: Fusing Native Diffusion Representations for Spatial Reasoning in Unified Multimodal Models
链接: https://arxiv.org/abs/2610.39609
作者: Ruixin Fan,Weining Wang,Jiahui Sun,Tailai Chen,Jing Liu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 20 pages, 6 figures
Abstract:Unified multimodal models (UMMs) jointly support visual understanding and image generation, yet information flow between these two capabilities remains largely asymmetric: understanding representations routinely condition generation, whereas generative representations are rarely reused for multimodal reasoning. In this work, we investigate whether intermediate representations from a UMM’s native diffusion branch can enhance spatial reasoning without compromising general multimodal understanding. We propose DiFuse, which selectively incorporates intermediate diffusion representations into the understanding pathway of UMMs. Through systematic layer-wise analysis, we identify which diffusion representations are most informative and where they can be integrated with minimal interference to semantic processing. Guided by these findings, DiFuse spatially aligns diffusion features with visual tokens and injects them into selected multimodal decoder layers via residual fusion. Extensive experiments on UniWorld-V1 show that, DiFuse consistently improves spatial reasoning over a matched SFT baseline while maintaining broadly comparable performance on general multimodal benchmarks. Further analyses show that these gains depend critically on the choice of diffusion representations, injection positions, and the preservation of spatial correspondence between diffusion features and visual tokens. Results on OmniGen2 further demonstrate the transferability of DiFuse across unified multimodal architectures. Overall, our findings show that native diffusion representations can provide complementary spatial cues for multimodal reasoning when selectively integrated in a structure-preserving manner.
[CV-87] GC-LIF: Spiking Group-Equivariant Neural Network With Group-Coupled Membrane Dynamics
链接: https://arxiv.org/abs/2610.39601
作者: Nhan Trong Luu,Duong Trung Luu,Thang Cong Truong
类目: Neural and Evolutionary Computing (cs.NE); Computer Vision and Pattern Recognition (cs.CV)
备注: Under review
Abstract:Spiking neural networks (SNNs) promise event-driven, low-power inference, but on vision tasks they still acquire rotation invariance only through data augmentation. Group-equivariant CNNs build such symmetries into the architecture, yet a faithful spiking analogue is missing: a leaky integrate-and-fire (LIF) neuron placed after a G -equivariant convolution can break equivariance through its membrane recurrence. We give a necessary and sufficient condition for this not to happen. The layer is equivariant at every time step if and only if the neuron parameters (\beta, v_\mathrmth) are constant along the joint orbits of the group action on the (group, space) index. We solve it in closed form, and it covers any elementwise stateful neuron. Its sufficiency half specialises a known result for equivariant recurrent networks; its necessity half rules out the heterogeneous per-neuron parameters common in SNNs. The per-channel parameterisation that satisfies it, Group-Coupled LIF (GC-LIF), is combined with p4 and p4m -equivariant convolutions and a spike-rate group max-pool into SGE-Net. On MNIST, KMNIST, Fashion-MNIST, CIFAR-10 and CIFAR-100, evaluating each checkpoint on the clean test set and all its group-transformed copies, SGE-Net trained without rotations is invariant under C_4 and D_4 and outperforms a plain SNN trained with rotations in nine of ten dataset-group pairs, whereas non-equivariant baselines lose up to 72 percentage points in the worst case. Operator-level tests confirm the theory before and after training, and an analytical estimate places SGE-Net at 1.7\times the arithmetic energy of a standard CNN and 2.3\times below its ANN counterpart at eight time steps.
[CV-88] Operator-Aware Posterior Sampling for Multivariate Weather Downscaling
链接: https://arxiv.org/abs/2610.39592
作者: Carlo Saccardi,Rudolf Saathof,Geethu Joseph,Justin Dauwels
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Statistical downscaling is the task of reconstructing high-resolution meteorological fields from coarse inputs. A principled approach formulates downscaling as a Bayesian inverse problem, using a diffusion prior over high-resolution fields and incorporating the coarse field only at inference through a likelihood. This decouples the prior from the coarse source, allowing one pretrained model to adapt to different sources. However, two challenges limit this flexibility: (1) when coarse and target fields come from different reanalysis systems, no fixed coarsening operator can capture their relationship; and (2) guidance gradients can differ by orders of magnitude across variables, making a single guidance strength ineffective. This paper addresses both challenges within the guidance loop, leaving the prior unchanged. We learn the forward operator from paired data and normalize the gradient per channel. We show that this normalization corresponds to steepest descent under a block-max norm, allowing a single guidance strength across variables. Our code is publicly available at this https URL.
[CV-89] actoHand: Learning Geometry-Based Dense Hand-Surface Interaction Fields from Video
链接: https://arxiv.org/abs/2610.39585
作者: Ziheng Xi,Weibo Li,Jiayi Yu,Sipeng Zheng,Wanpeng Zhang,Haiweng Xu,Hao Luo,Chuan Mao,Zongqing Lu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 23 pages, including references and appendices
Abstract:Understanding how humans interact with the world requires a representation of physical interactions at the hand surface. Tactile sensing captures these interactions, but collecting tactile data at scale requires specialized hardware. We introduce TactoHand, a model that predicts three dense hand-surface interaction fields—contact, proximity, and 3D force—from monocular video. We derive geometry-based contact and proximity labels and motion-derived force labels from eligible training data within a curated collection spanning approximately 50 million source frames. TactoHand uses a temporal transformer to aggregate visual evidence across frames, while three learned queries decode the corresponding hand-surface fields. A differentiable physics layer further constrains force predictions using observed object motion. Across in-domain and out-of-domain benchmarks, TactoHand improves contact and proximity estimation over prior methods. In addition, physics supervision reduces position and rotation errors in motion rollouts, and the predicted contact and proximity fields improve downstream hand-object reconstruction. Together, these results show a scalable, low-cost path to learning hand-surface interaction fields without collecting tactile measurements. Code will be released at this https URL.
[CV-90] DOSE: Dominance-Guided Ordinal Severity Estimation from Sparse Relative Annotations WACV2027
链接: https://arxiv.org/abs/2610.39584
作者: Tushir Sahu,Hideaki Hayashi,Hajime Nagahara
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to WACV 2027. 8 pages, 4 figures. Adjust the counts if you add the supplementary material
Abstract:Medical severity estimation commonly relies on absolute grade annotations, which are fundamentally ordinal, yet existing methods often treat the grades as independent categories. Learning-to-rank methods, which can be trained using relative annotations based on severity comparisons of image pairs, can provide continuous severity rankings but cannot directly predict discrete severity grades. We propose Dominance-guided Ordinal Severity Estimation from Sparse Relative Annotations (DOSE), a framework for learning ordinal medical severity from sparse pairwise comparisons without using absolute annotations. DOSE aggregates the observed comparisons through a global comparison graph to estimate a dominance score for each image. Differences between dominance scores set cumulative margins in the embedding space, so that pairs with larger estimated severity differences are separated more strongly, and also generate pseudo-pairs that supervise a scalar severity head alongside the observed comparisons. The learned continuous severity axis is converted into discrete grades via label-free score calibration, with tie-anchored segmentation as the main calibration method. In the experiment, DOSE consistently outperformed pairwise-ranking baselines on ACNE04, IDRiD, and LIMUC under different pair budgets. Code is available at this https URL
[CV-91] BEACON: Distilled Begin/End-Aware Structured Decoding for Continuous Sign Language Segmentation WACV
链接: https://arxiv.org/abs/2610.39553
作者: Guohou Zhang,Kalin Stefanov
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) 2027. 12 pages, 4 figures, including supplementary material
Abstract:In segmentation-based continuous sign language recognition, an untrimmed video is first partitioned into lexical units; segmentation is therefore a means to an end—downstream recognition. However, segmenters are usually judged by local boundary detection rather than by the quality of the lexical segments and the recognition they enable. We present BEACON, a begin/end-aware segmenter that combines four-class Outside/Begin/Inside/End (BIOE) labeling, constrained Viterbi decoding over eight legal transitions, and boundary knowledge distilled from a binary teacher into a shared student. Under strict no-cut evaluation with validation-based model selection, BEACON improves segment-F1 by 14.58/14.59 points and boundary-F1 by 2.17/1.22 points, and reduces the downstream word error rate (WER) by 9.10/4.15 points relative to the state-of-the-art in the PHOENIX-14/BSL Corpus. To examine which segmentation metric better reflects downstream recognition, we evaluated five distinct architecture families. Segment-F1 aligns more consistently with lower WER than boundary-F1 in both corpora (Spearman \rho=-1.0/-0.9 versus -0.9/-0.8 ). These results support evaluating sign language segmentation in terms of segment quality and downstream recognition utility rather than local boundary accuracy alone.
[CV-92] Comprehensive empirical evaluation of feature extractors in computer vision
链接: https://arxiv.org/abs/2610.39551
作者: Murat Isik
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 26 Pages, 9 Figures, 6 Tables
Abstract:Feature detection and matching are fundamental components in computer vision, underpinning a broad spectrum of applications. This study offers a comprehensive evaluation of traditional feature detections and descriptors, analyzing methods such as Scale Invariant Feature Transform (SIFT), Speeded-Up Robust Features (SURF), Binary Robust Independent Elementary Features (BRIEF), Oriented FAST and Rotated BRIEF (ORB), Binary Robust Invariant Scalable Keypoints (BRISK), KAZE, Accelerated KAZE (AKAZE), Fast Retina Keypoint (FREAK), Dense and Accurate Invariant Scalable descriptor for Yale (DAISY), Features from Accelerated Segment Test (FAST), and STAR. Each feature extractor was assessed based on its architectural design and complexity, focusing on how these factors influence computational efficiency and robustness under various transformations. Utilizing the Image Matching Challenge Photo Tourism 2020 dataset, which includes over 1.5 million images, the study identifies the FAST algorithm as the most efficient detector when paired with the ORB descriptor and Brute-Force (BF) matcher, offering the fastest feature extraction and matching process. ORB is notably effective on affine transformed and brightened images, while AKAZE excels in conditions involving blurring, fisheye distortion, image rotation, and perspective distortions. Through more than 2 million comparisons, the study highlights the feature extractors that demonstrate superior resilience across various conditions, including rotation, scaling, blurring, brightening, affine transformations, perspective distortions, fisheye distortion, and salt-and-pepper noise
[CV-93] ROAD: Modern Adversarial Training for Simple One-Step Distillation at Scale
链接: https://arxiv.org/abs/2610.39539
作者: Xiaoou Li,Shan You,Tao Huang
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 37 pages, 12 figures, 7 tables
Abstract:One-step distillation turns pretrained diffusion models into fast generators, but strong results often require substantial training machinery and compute. Pure adversarial distillation offers a simpler route, yet has struggled at modern text-to-image scale. To address this limitation, we propose ROAD (Representation-grounded One-step Adversarial Distillation), a one-stage framework that distills text-to-image models with one adversarial objective and a lightweight representation-space critic. First, we replace the classical convolutional head with a Lipschitz-controlled Transformer head with token-level text conditioning, which matches the semantic tokens of modern visual encoders. Second, we observed a fixed random feature mixing leads to long-horizon degradation. We attribute it to generator-critic co-adaptation around one realization-specific view, and periodically re-draw the mixing to keep the randomization effective. Third, for our dense and vector-valued logits, we introduce Jacobian R1 which penalizes the full input-logit Jacobian, overcoming pooled R1’s limitation of neglecting most of logit sensitivities. A single adversarial objective then suffices: in one step, ROAD reaches 11.03 FID on SDXL and 13.73 on SD3.5-M on COCO-10K, surpassing the multi-step teachers at a fraction of distribution matching’s training compute.
[CV-94] MOSAIC-Flow: Orthogonal-Axis and Status-Conditioned Fusion for Missing-Sequence Completion in Brain Tumor MRI
链接: https://arxiv.org/abs/2610.39534
作者: Pengli Zhu,Zhihao Peng,Yixuan Yuan
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 5 pages, 4 figures
Abstract:Missing-sequence completion is important for comprehensive brain tumor characterization, but cross-modal fusion can obscure sequence-specific contrast and lose fine-grained three-dimensional tumor detail. To address this challenge, we introduce Modality-aware Orthogonal-axis Status-conditioned Attentive Integration for Completion (MOSAIC-Flow), a unified framework for missing-sequence completion that learns a target-conditioned velocity field while keeping the seven acquired volumes fixed throughout the transport. Specifically, modality-aware stems independently encode each sequence, and orthogonal-axis refinement captures axial, coronal, and sagittal dependencies while retaining the three-dimensional representation. Status-conditioned prototype fusion uses compact modality and availability tokens to route and modulate pooled queries from the fused visual representation, enabling target-adaptive conditioning without dense voxel-to-voxel attention. A conditional flow matching objective transports target-channel noise toward the single missing sequence, while four Heun trajectories are averaged to produce the completion. Evaluation across structural, diffusion, and contrast-enhanced MRI in UCSF-PDGM demonstrates consistent whole-brain and tumor-region fidelity across all eight single-target completion tasks, with the strongest performance on structural sequences. Code is available at this https URL.
[CV-95] MorphSIG: Subject-Driven Image Generation via Decoupled Anchoring and Feature Transport
链接: https://arxiv.org/abs/2610.39530
作者: Hailong Yan,Yongrui Zhang,Xiangtao Zhang,Le Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:We propose MorphSIG, a training-free framework that reformulates subject-driven image generation (SIG) as pseudo-video feature transport. MorphSIG contains two core modules: Progressive Decoupled Anchoring, which uses frequency-domain decoupling and structural scrambling during early denoising to break structural locking and enable a gradual transition from free composition to identity alignment; and Pseudo-flow Guided Feature Transport, which constructs a latent-space pseudo-flow field to mimic Image-to-Video temporal coherence and transport subject semantics precisely. To address stylistic homogeneity and the lack of ground truth in existing benchmarks, we further introduce a benchmark with diverse stylized subjects. Extensive experiments show that MorphSIG significantly improves pose diversity and subject fidelity over baselines without additional training. Overall, MorphSIG bridges static generation and dynamic propagation, providing a cost-effective paradigm for high-fidelity subject consistency.
[CV-96] BRIDGE: Learning Brain-Gut Metabolic Evidence for Alzheimers Disease Diagnosis
链接: https://arxiv.org/abs/2610.39520
作者: Pengli Zhu,Ruxi Chang,Tianliang Yao,Yu Jiang,Shengyuan Liu,Yiyang Lin,Xiaoyi Duan,Shuang Wu,Yixuan Yuan, theAlzheimer’s Disease Neuroimaging Initiative
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 10 pages, 6 figures
Abstract:Alzheimer’s disease (AD) diagnosis may benefit from gut metabolic information associated with brain-gut interactions, yet existing models largely analyze cerebral glucose metabolism alone. Although whole-body fluorodeoxyglucose positron emission tomography (FDG-PET) can provide paired brain-gut observations, routine brain examinations rarely include gut imaging, limiting direct use of this information for diagnosis. In this paper, we present BRIDGE (Brain and Gut Representation Integration via Distribution-Guided Enhancement), a two-stage framework for transferring brain-gut metabolic knowledge to brain-only diagnosis. Specifically, BRIDGE integrates a conditional brain-to-gut latent transport module and a disease-evidence-guided cerebral pathway to capture subject-specific cross-organ correspondence and disease-relevant cerebral representations. For the cerebral pathway, a large language model (LLM) helps organize literature-derived descriptors into human-approved semantic anchors that guide feature learning. Additionally, we introduce a counterfactual-guided residual fusion mechanism, which uses matched and non-matched virtual-gut evidence to supervise a confidence gate and dynamically regulate the contribution of transported cross-organ information. At inference, BRIDGE exploits complementary gut-related context using brain PET alone, without measured gut PET. Extensive experiments across public and private cohorts demonstrate that BRIDGE outperforms brain-only and cross-organ baselines, achieving AUC gains of 6.0% and 5.2% over its brain-only counterpart for CN/MCI and CN/AD diagnosis, respectively. Notably, its ability to transfer complementary brain-gut metabolic context while preserving brain PET as the sole inference input highlights its potential for clinically practical diagnosis. Our code is available at this https URL.
[CV-97] FFAI: Flow Matching with Feedback Training and Adaptive Initialization for CT Reconstruction
链接: https://arxiv.org/abs/2610.39517
作者: Jigang Duan,Heran Wang,Genwei Ma,Xing Zhao
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Although low-dose, sparse-view, and limited-angle CT involve distinct degradation mechanisms, all three pose reconstruction challenges due to limited measurement information and can therefore be addressed within a unified framework. In iterative image restoration with flow matching, prediction errors may cause inference states to deviate from the prescribed training path, while a fixed initialization may be unsuitable for inputs with varying degradation severity. We propose a unified conditional flow matching method that conditions on the initial reconstruction and combines feedback-aware training with degradation-adaptive initialization. During training, a teacher network simulates inference updates to generate shifted states, from which a student network learns to recover high-quality reference images, improving its ability to handle prediction errors. During inference, the starting time is selected based on the noise level or extent of missing projection data, and the noise added to the initial state is adjusted accordingly to adapt the starting point to the input. Experiments on public benchmark datasets show that the proposed method achieves better quantitative results than the compared methods across all three reconstruction tasks, effectively suppressing noise and artifacts while preserving anatomical structures and details.
[CV-98] Robust LIDAR Sensor-to-Sensor Domain Generalization via Accumulated-Point Scene Completion
链接: https://arxiv.org/abs/2610.39516
作者: Christoph B Rist,Larissa T Triess,Markus Enzweiler,Dariu M Gavrila
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 2026 IEEE Intelligent Vehicles Symposium (IV)
Abstract:We study domain generalization for LIDAR semantic segmentation. Deep neural networks achieve strong performance on 3D point clouds, yet real-world LIDAR exhibits sensor-specific sampling patterns that hinder transfer across datasets. In many applications, performance on completely unseen sensors is crucial. A model should preferably work on a new setup without prior access to its data or labels. To this end, we introduce a source-only training scheme that adds an occupancy prediction auxiliary objective. We aggregate source-domain scans into a simple world model and use it as the supervision signal, replacing single-scan standard segmentation training. A key property of our approach is that it does not alter the input scan. During training and testing the network consumes the same single-frame LIDAR input, only the supervision signal is enriched via the occupancy objective. At deployment, there is no runtime overhead as the auxiliary head can be dropped for pure segmentation, or kept to obtain scene completion alongside semantic segmentation. In an experimental evaluation across distinct LIDAR datasets, we demonstrate that a model’s generalization to new domains can be significantly affected by how it is supervised. In certain setups, performance on unseen domains improves without any explicit domain adaptation (no target scans, no target labels, no sensor priors) and without prediction refinement. At the same time, our method maintains competitive source-domain segmentation performance and incurs no additional inference cost. Comments: 2026 IEEE Intelligent Vehicles Symposium (IV) Subjects: Computer Vision and Pattern Recognition (cs.CV) Cite as: arXiv:2610.39516 [cs.CV] (or arXiv:2610.39516v1 [cs.CV] for this version) https://doi.org/10.48550/arXiv.2610.39516 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Related DOI: https://doi.org/10.1109/IV66570.2026.11623951 Focus to learn more DOI(s) linking to related resources
[CV-99] Agent -Wise Test-Time Normalization for Camera-Based Cooperative Perception with Corrupted Collaborators
链接: https://arxiv.org/abs/2610.39505
作者: Yu Wang,Shaohua Wang,Yicheng Li,Kecheng Sun
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: 11 pages, 8 figures
Abstract:Intermediate-fusion cooperative perception lets a vehicle detect objects beyond its own field of view by fusing features shared by nearby agents. When only a collaborator’s cameras are degraded by fog, motion blur or exposure errors, the features it shares drift away from the training distribution while the ego input stays clean, and cooperation can become worse than not cooperating at all. Test-time batch-normalization adaptation, the standard training-free remedy, estimates statistics over the pooled batch of all agents and therefore normalizes both the clean ego and the corrupted collaborator with a mixture of two distributions. We show that this mixing harms even the clean ego: when the ego shares BN statistics with a corrupted collaborator, its detection from its own features drops by 3.58 to 6.68 AP@0.5 points although its cameras are untouched. We propose AgentNorm, an agent-wise test-time normalization that i) estimates the normalization statistics of every collaborator from its own current observation and ii) anchors the ego to its source statistics, so that the ego path remains exactly the original model. AgentNorm has no learnable parameters, needs no gradients or labels, is independent of the order of test frames, and runs locally on each agent without changing the transmitted features. On camera-based OPV2V with four fusion checkpoints and collaborator-only corruptions, AgentNorm raises AP@0.5 under severity-5 fog from 29.42-37.25 to 43.32-58.86 and outperforms pooled normalization by 5.53-12.94 points under fog and motion blur, at an extra latency of 2.3 ms per frame. On the real-world DAIR-V2X-C dataset, it outperforms pooled normalization in all ten infrastructure-corruption settings. We also report the cases it does not fix: occlusion, which removes image content rather than shifting feature statistics, and the case in which all agents share the same shift.
[CV-100] Does Federated Training Degrade Explainability? A Comparative Analysis of Grad-CAM Faithfulness in Centralized vs. Federated Deepfake Detection
链接: https://arxiv.org/abs/2610.39483
作者: Muhammad Uzair Shahid,Muhammad Zabil Mehboob
类目: Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 10 Pages, 3 figures, Muhammad Uzair Shahid and Muhammad Zabil Mehboob contributed equally to this work
Abstract:Automated detection has become the essential safeguard against realistic synthetic media, but deploying it at scale requires training on sensitive data from multiple organizations, putting privacy and the transparency demanded by forensic review in direct conflict. Federated learning resolves the privacy side by sharing only model updates rather than raw data, but whether this comes at the cost of trustworthy gradient-based explanations, which analysts rely on to decide whether or not to trust a flagged clip, has not been studied. We present the systematic study of Grad-CAM explanation faithfulness under federated deepfake detection, evaluating an EfficientNet-B4 detector on FaceForensics++ across five federated clients under varying degrees of data heterogeneity. We find that the relationship between federation and explanation quality is non-monotonic: at near-IID (Independent and Identically Distributed Data) and moderate heterogeneity, federated models match or exceed the faithfulness of a centralized baseline, but at extreme heterogeneity, explanations break down entirely, becoming spatially incoherent and, on average, causally misleading. This degradation is not a general cost of federated training but a threshold effect tied to client heterogeneity, and we offer practical guidance for identifying when privacy-preserving deepfake detection can be deployed safely and when its explanations can no longer be trusted. Code and data partitions are available for reproducibility.
[CV-101] Regional Anchor Routing for Multi-Object Motion Transfer
链接: https://arxiv.org/abs/2610.39478
作者: Yuze Li,Fuchen Long,Xinlong Sun,Cheng Yan,Xinyu Zhang,Xi Chen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project Page: this https URL
Abstract:Compositional motion transfer requires a model to extract motions from multiple reference videos, and apply each motion to a user-specified object in one target scene. Existing methods normally struggle to reliably assign different reference motions to the correct target objects, thus a motion intended for one object may affect another. Meanwhile, some motion-specific tuning approaches are still facing another issue: they must run a new optimization whenever the reference video changes, making repeated transfers inefficient. To address these limitations, we propose RART, a framework that unifies object-wise motion extraction, target-region anchoring, and frame-aligned injection into a frozen video Diffusion Transformer for multi-object motion transfer. For each reference motion, RART extracts temporally ordered features from the masked reference video, and then conditions them on the target first-frame mask to form mask-anchored motion tokens. Frame-aligned motion cross-attention allows each generated frame to read only the motion tokens at the corresponding time slots, composing all object motions in one denoising process. During training, a motion-attention routing loss encourages each target region to attend to its corresponding motion tokens and reduces interference between objects. Therefore, RART can handle novel reference-target combinations without per-video optimization. Experiments show that RART achieves accurate motion preservation, efficient inference, and strong compositional generalization.
[CV-102] Calibrating Attention-Based Token Pruning for Efficient Multi-Image Synthesis in UMMs
链接: https://arxiv.org/abs/2610.39467
作者: Zimeng Wu,Qijian Tian,Penghui Du,Changqian Yu,Jiaxin Chen,Yunhong Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Unified Multimodal Models (UMMs) enable powerful multi-image synthesis but incur substantial computation from long visual contexts. While token pruning offers a direct route to accelerating inference, existing methods are primarily adapted from understanding tasks, suffering from insufficient reference image exploration, confused conditional guidance, and pooling-induced ambiguity that degrades generation quality. To address these issues, we propose CATP, the first training-free context-compression method tailored to UMM generation, which \textbfCalibrates the discrepancies of general \textbfAttention-based \textbfToken \textbfPruning for better quality-efficiency trade-off. Concretely, building on independent layer-wise pruning during denoising, we introduce Guidance-Consistent Pruning (GCP) to synchronize pruning decisions across conditioning branches, and a Taylor-Calibrated Attention Metric (TCAM) that derives a theoretically grounded second-order correction for pooling-induced bias. In addition, a Fast Scoring Operator (FSO) fuses the calibration into tiled online Softmax to reduce pruning cost. Extensive experiments across representative UMMs and benchmarks demonstrate that CATP consistently outperforms state-of-the-art token-reduction methods, achieving up to 2.09 \times attention-core speedup.
[CV-103] PathCowork: Learning from real-world pathology practice to build AI copilots
链接: https://arxiv.org/abs/2610.39441
作者: Sheng Wang,Jialu Yao,Maria V. Rossi,Shunsuke Koga,Esther Baranov,MacLean P. Nasrallah,Songhao Li,Jonathan Xu,Wei Du,Cindy McGrath,Timothy Chao,Anne Wang,Zhaohai Yang,Daniel Akuma,Selemon M. Walle,Olga Pozdnyakova,Edward B. Lee,Zahra Alipour,Annapurna Saksena,Rashmi Tondon,Nikolina Dioufa,David E. Elder,George J. Netto,Xiaowei Xu,Zhi Huang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 56 pages, including supplementary information. Code, models and demonstration videos: this https URL
Abstract:Medical AI learns from the answers doctors record, but rarely from how they interpret the evidence to reach them. Here we introduce PathCowork, a framework that turns pathologists’ narrated examinations into supervision for visual grounding and procedures that guide diagnostic AI agents. Synchronized gaze, narration and navigation link visual findings to the process of interpreting them. Across 48 sessions from 16 pathologists, the platform captured 1,028 minutes of slide review. Using these data, we trained PathCowork-1T for server deployment and PathCowork-27B for local deployment, with improved visual grounding. Across five organ-specific evaluations, agents combining this grounding with recorded diagnostic procedures achieved primary task scores averaging 17.9 and 24.1 percentage points higher than Claude Code and Codex, respectively, with shorter processing times and lower inference costs. Compared to the state-of-the-art multiple-instance learning AIs, the PathCowork agent-proposed regions achieved higher prostate-grading scores with limited slide labels. PathCowork can also help construct pathology AI benchmarks where questions derived from the recordings identified difficulties in recognizing viewing context and rejecting inapplicable premises. In summary, PathCowork proposes a new paradigm to scale up the data collection and train pathology AI through transparent reasoning from diagnostic sessions, opening a path towards systems that learn directly from routine clinical practice.
[CV-104] FDSML: Formula-Driven Supervised Metric Learning with Parameter-Aware Triplet Loss WACV2027
链接: https://arxiv.org/abs/2610.39423
作者: Hina Otake,Shinei Arakawa,Keitaro Tanaka,Yoshihiro Fukuhara,Hirokatsu Kataoka,Shigeo Morishima
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Neural and Evolutionary Computing (cs.NE)
备注: Accepted to WACV 2027. 24 pages (11 main + 13 supplementary). Project page: this https URL
Abstract:Formula-Driven Supervised Learning (FDSL) enables vision pre-training without real images, but existing methods cast pre-training as classification on discretized labels, discarding continuity in the generative process. We propose Formula-Driven Supervised Metric Learning (FDSML), which derives relational supervision from image-generation parameters beyond discrete class labels. FDSML combines FineFractalDB, our refined fractal dataset with higher intra-class variation and more stable structures, with a Parameter-Aware Triplet (PA-Triplet) Loss that assigns positives and negatives according to distances in the underlying Iterated Function System parameter space. This objective trains embeddings to respect parameter-derived relative similarity orderings. Experiments show that FDSML improves transferability to image retrieval while remaining competitive on downstream classification and improving robustness under several corruption settings. On ResNet-50, Recall@1 improves from 65.6% to 74.0% on Stanford Online Products and from 56.2% to 78.4% on Stanford Cars over FDSL pre-training. These results suggest that parameter-derived relational supervision provides a stronger synthetic pre-training signal than label-based classification for transfer beyond recognition.
[CV-105] Decoding Geometry Not Tokens: Efficient VLMs for Robot Affordance and Trajectory Prediction WACV2027
链接: https://arxiv.org/abs/2610.39416
作者: Akila Ayanthi,Darshana Priyasad,Tharindu Fernando,Dimity Miller,Sridha Sridharan,Clinton Fookes,Peyman Moghadam
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注: IEEE/CVF Winter Conference on Applications of Computer Vision (WACV 2027)
Abstract:Predicting where to grasp and how to move is fundamentally a geometry problem. Vision-language models treat it as language generation instead. They tokenize spatial coordinates into text and decode them autoregressively. This mechanism ignores spatial structure and truncates continuous precision. Errors compound across coordinate tokens during decoding. In this paper, we show that compact off-the-shelf VLMs with 0.5-2B parameters already encode rich spatial information in their intermediate hidden states. Autoregressive decoding is the dominant bottleneck for trajectory prediction, while affordance prediction depends additionally on feature depth and decoder architecture. We expose the latent spatial information with a lightweight, geometry-compatible prediction head that attaches to a frozen backbone. It reads features from an intermediate hidden layer, conditions a learned query on the task instruction and refines it via cross-attention over the visual tokens to regress affordance boxes and trajectory waypoints directly as continuous values, without modifying the backbone. We evaluate on ShareRobot-Bench and AGD20K. Our 2B variant reduces trajectory error by over 60% and improves ShareRobot-Bench affordance mAP by 85% relative to the autoregressive baseline. It approaches RoboBrain on trajectory prediction (0.251 vs 0.191 DFD) at 3.8x higher throughput. These results demonstrate a favorable accuracy-efficiency trade-off for instruction-conditioned 2D geometric prediction.
[CV-106] Anatomical Prior-Guided Routing for 3D Lung Tumor Segmentation ACCV2026
链接: https://arxiv.org/abs/2610.39414
作者: Uihyeon Maeng,Jonghun Kim,Hyunjin Park
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to ACCV 2026 (Oral), 18 pages, 6 figures
Abstract:Accurate lung tumor segmentation in computed tomography (CT) is essential for treatment planning, response assessment, and quantitative imaging analysis. However, 3D lung tumor segmentation remains challenging because tumors occupy only a small portion of the whole volume and exhibit heterogeneous appearances across anatomical locations. In patch-based training and sliding-window inference for 3D medical image segmentation, location-dependent heterogeneity is further amplified because cropped image patches provide limited global anatomical context, making it difficult for a model to adapt its feature processing according to the anatomical region from which each patch is extracted. To address this limitation, we propose an Anatomical Prior-guided Routing Module (APRM) for lung tumor segmentation. For each image patch, we obtain anatomical prior information of regional occupancy and deterministically select an anatomical routing gate. The selected gate then performs sparse feature routing over a shared expert adapter pool, enabling anatomy-specific feature adaptation while sharing expert capacity across regions. We evaluate APRM on two datasets with various backbone networks and demonstrate consistent improvements over baseline models. Furthermore, feature visualizations show that APRM produces different feature responses depending on patch location, indicating that the proposed routing mechanism effectively adapts feature processing according to anatomical context.
[CV-107] YOLOF-MLD: A Multi-Level Distillation Framework for Efficient Object Detection
链接: https://arxiv.org/abs/2610.39408
作者: Xiaolong Wu,Liming Zheng,Zitong Liu,Yue Zheng
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 4 pages
Abstract:Efficient object detection on resource-constrained edge devices is essential for safety-critical applications such as fire-smoke monitoring. However, current lightweight frameworks, such as YOLO26, still rely on the parameter-dense multi-branch Feature Pyramid Network (FPN), preventing full exploitation of their lightweight advantage. Replacing FPN with YOLOF’s single-input-single-output (SiSo) neck is a promising solution but causes substantial accuracy loss. To alleviate this, we propose YOLOF-MLD, a CPU-efficient multi-level distillation (MLD) framework that combines a YOLO26n backbone, a SiSo neck, and MLD. It aligns features at the backbone via sliced-semantics matching, neck with receptive-field matching, and response levels. On D-Fire and Annotated Fire-Smoke (AFS) datasets, YOLOF-MLD reduces the parameter count of YOLO26n by 35.7% and achieves a 62.3% speedup on Raspberry Pi 5, attaining 71.6% and 59.9% mAP@50, respectively, compared with 75.5% and 60.0% for YOLO26n. These results demonstrate a competitive accuracy–latency trade-off for edge deployment. Our code is available at this https URL.
[CV-108] SymKL: A Unified Representation-Space Measure of Signal Degradation
链接: https://arxiv.org/abs/2610.39403
作者: Dhananjaya Jayasundara,Sudarshan Rajagopalan,Vishal M. Patel
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Quantifying degradation between a pristine signal and its degraded observation is fundamental to signal processing, yet existing full-reference measures are inherently modality-specific. We introduce SymKL, a unified full-reference degradation measure based on implicit neural representations (INRs). An INR is fitted once to the pristine reference and characterized by its diagonal Fisher information, which captures the local information geometry of the learned representation. Evaluating the Fisher information with the pristine reference and a degraded observation yields two parameter-sensitivity profiles whose symmetric KL divergence defines the degradation score. The proposed formulation requires only the reference and degraded observation, without knowledge of the corruption process or degradation severity, making it applicable across images, audio, and 3D data. We further derive an exact closed-form expression for the diagonal Fisher of SIREN, enabling efficient computation. Experiments on image, speech, and volumetric 3D benchmarks demonstrate that SymKL accurately tracks degradation severity, remains monotonic where conventional metrics saturate or fail, and provides a unified modality-agnostic measure of signal degradation. Code will be made publicly available after the review process.
[CV-109] ACD-GS: Asymmetric Curvature-aware Densification for 3D Gaussian Splatting
链接: https://arxiv.org/abs/2610.39399
作者: Houqiang Zhong,Tianchi Zhu,Qiang Hu,Xiaoyun Zhang,Li Song,Wenjun Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:3D Gaussian Splatting (3DGS) has emerged as a powerful paradigm for real-time photorealistic novel view synthesis. However, its training-time densification remains largely heuristic, often resulting in redundant Gaussian proliferation and inaccurate geometric reconstruction. Existing methods predominantly rely on first-order screen-space gradients to trigger cloning and splitting, which fail to distinguish true geometric discontinuities from high-frequency textures or noisy residuals. To address this limitation, we propose ACD-GS, an Asymmetric Curvature-aware Densification framework for compact and geometry-aware 3D Gaussian optimization. Our key insight is that second-order depth responses provide a more reliable structural prior for density control, enabling asymmetric regulation of cloning and splitting to better capture geometric discontinuities. Building on this, we introduce a multi-view photometric consistency constraint to suppress transient single-view errors, together with a capacity-aware adaptive regulation strategy that balances densification and pruning during training. Unlike post-hoc compression or quantization-based approaches, ACD-GS directly prevents redundant Gaussian generation during training, yielding inherently compact representations. Extensive experiments demonstrate that our method reduces nearly 60% of Gaussian count and storage while maintaining competitive rendering fidelity, consistently outperforming existing methods across multiple datasets. Our code is available in this https URL.
[CV-110] CASE: Concept-Aware Segmentation from Positive and Negative Visual Exemplars ACCV2026
链接: https://arxiv.org/abs/2610.39396
作者: Yi-Jen Tsai,Yen-Yu Lin,Chien-Yao Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to ACCV 2026
Abstract:Few-shot semantic segmentation typically identifies target objects based on positive support examples of a single semantic class. In practice, however, positive cues alone are often underspecified: for instance, “red cars” may be confused with cars of all colors, “vehicles” may include cars that should be excluded, and “material-defined objects” may span multiple categories. Fulfilling precise requests requires positive exemplars to provide inclusion evidence and negative exemplars to act as intent-defining constraints for exclusion and disambiguation. We formulate this setting as Specified Visual Prompt Segmentation (SVPS), where the target concept is specified by both positive and negative visual prompts. In this work, visual exemplars refer to support-image regions specified by bounding-box prompts. To study SVPS, we introduce CASE (Concept-Aware Segmentation from Positive and Negative Visual Exemplars), a benchmark built on COCO and PACO. CASE encompasses six category- and attribute-level task families: exclusion, isolation, union, attribute exclusion, attribute isolation, and attribute union. We evaluate SAM 3 under this setting and find that, despite its powerful positive-only few-shot segmentation capability, frozen SAM 3 does not consistently treat negative visual prompts as exclusion evidence. To address this, we present Concept Prompt Adaptation (CPA), a conservative LoRA-based adaptation baseline that improves positive-negative prompt interpretation on CASE while remaining competitive under positive-only few-shot segmentation. Code is available at this https URL.
[CV-111] Semantic Mismapping in Generalized Category Discovery: Evaluation and Mitigation
链接: https://arxiv.org/abs/2610.39356
作者: Zitai Wang,Qianqian Xu,Cong Hua,Xi Chen,Zhiyong Yang,Xiaochun Cao,Qingming Huang
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Generalized Category Discovery (GCD) aims to perform old-category recognition and novel-category discovery in the same unlabeled set. Early methods address both objectives through clustering and evaluate the exchangeable predicted indices with global Hungarian (GH) matching. Recent methods adopt trainable parametric heads, using supervision for old-category recognition and discovery-oriented objectives for new categories. In this way, the output space becomes partially exchangeable: old indices are semantically anchored, while new indices remain exchangeable. However, these methods still apply global matching, which can reassign anchored old indices to other categories and hide some prediction errors. Empirically, we are surprised to observe that this reassignment can induce an overestimation of new-category accuracy by up to 25.63 points. We term this phenomenon semantic mismapping and further derive a permutation-cycle condition that characterizes when semantic mismapping can increase the global matching score. To fix this issue, we propose Semantics-Preserving Evaluation (SPE), which preserves the supervised semantics of old indices and restricts Hungarian matching to the new region. Furthermore, we propose Mismapping-Mitigating Optimization (MMO) to alleviate the prediction errors that can give rise to semantic mismapping, combining a region-balance loss (RBL) for old/new routing with a novel-diversity loss (NDL) that encourages orthogonality among new prototypes. Theoretically, we show that minimizing the two losses can tighten an upper bound on spurious accuracy gains from semantic mismapping, with controllable residual terms. Empirical results under both GH and SPE demonstrate the effectiveness of our proposed method.
[CV-112] PicturePerfect: Diffusion-based Same-identity Facial Replacement ECCV2026
链接: https://arxiv.org/abs/2610.39344
作者: Manik Dhingra,Yashesh Savani,Mahesh Reddy,Antoine Mercier,Jiahe Liu,Guillaume Berger
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Camera-ready version (ECCV 2026 workshop - Third Workshop on Foundation and Generative Models in Biometrics)
Abstract:With the growing success of diffusion models in computer vision, we explore their potential for same-identity facial replacement in photographs. Specifically, we propose a diffusion-based method, built on top of a pre-trained text-to-image model, that takes as input a portrait image of a person and a second reference image of the same individual, potentially captured under different conditions. The goal is to seamlessly replace the input face with the reference face, while keeping the background intact. Surprisingly, despite the clear real-world utility of this task, no recently published work has directly addressed face replacement in this specific setting. To support this goal, we construct a large dataset of image pairs depicting the same person under varying facial expressions and poses. Experimental results demonstrate that our approach produces more realistic and identity-consistent results than existing face reenactment models.
[CV-113] ng Is Not Seeing: Textual Personas Rarely Make Vision-Language Models Answer Like the People They Describe
链接: https://arxiv.org/abs/2610.39331
作者: Jiawei Chen
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Vision-language models (VLMs) are increasingly prompted with personas, such as “You have low vision”, to stand in for people in user studies. We ask whether a persona makes a model see like the person it describes. We compare persona answers with measured people with low vision, color blindness or hearing loss, and with simple rules that use only the facts the persona states. The persona never beats these rules, and it erases the real differences between groups and individuals. Describing blur leaves recognition unchanged in all 14 models we test, while blurring the photo lowers it as it does for people. Inside open models, the description changes neither the image information nor the answer probabilities. Yet the models know the effects they fail to show: they can estimate how blur and illusions affect people. Telling is not seeing: a persona makes a model say what it knows about people, not see what they see.
[CV-114] Can Image Agents Complete Visual Commissions? A Benchmark and Modular Framework
链接: https://arxiv.org/abs/2610.39330
作者: Tong Li,Lishuai Gao,Qi Zeng,Yifan Li,Cong Wei,Zhuoliang Kang,Xiaoming Wei
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Visual commissions require image agents to integrate heterogeneous source materials, preserve requirements across dependent stages, and produce coordinated deliverables. Existing evaluations examine specific capabilities and execution settings, leaving end-to-end commission completion less understood. We introduce Visual Commission Bench (VCB), comprising 600 tasks across 12 categories and three complementary tiers, from capability-focused tasks to complete commissions. VCB uses 5,039 requirement checks and separate visual quality ratings to assess content correctness, visual realization, constraint preservation, and delivery completeness without prescribing an internal workflow. We present V-Protocol, a reference agent that dynamically composes reusable task, cognitive, and expression modules at inference time. Task modules organize delivery, cognitive modules derive content, and expression modules guide visual realization. The modules exchange structured intermediate artifacts, while a shared execution harness tracks artifact revisions and dependencies, manages tool execution, and checks completion. With the same controller and image generator as the agent baselines, V-Protocol achieves 69.0% strict task success on VCB, compared with 42.0% for SCOPE. On Mind-Bench, strict accuracy increases from 39.2% to 45.4%.
[CV-115] MRC-MMAgent : Reinforcement Learning for Multi-Role Collaboration in Multimodal Tool-Using Agents
链接: https://arxiv.org/abs/2610.39324
作者: Shengqin Wang,Jie Jin,Xiaofan Li,Wentao Yan,Yu Cheng,Qingsong HuYihang Chen,Weilin Luo,Jingyu Gong,Xin Tan,Yuan Xie,Zhizhong Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Multimodal tool-using agents augment vision-language models with access to external information and capabilities, enabling them to solve knowledge-intensive and visually complex tasks that cannot be addressed using visual context alone. However, existing reinforcement learning pipelines largely train single-role agents to search and answer using terminal rewards, overlooking how humans inspect retrieved evidence, reformulate queries, and revise conclusions during search. Consequently, they fail to optimize this self-corrective process over complete interaction trajectories. We propose MRC-MMAgent, a multi-role collaborative reinforcement learning framework that incorporates trajectory assessment and continuation directly into the executable agent loop. First, the framework establishes a multi-role collaborative loop that integrates solving, inspection, and continuation, enabling the agent to acquire evidence through tools, evaluate the preceding trajectory, and resume problem solving when the current answer is insufficiently supported. Second, we introduce a role-coupled learning objective that combines task success, interaction validity, and successful correction with full-trajectory policy optimization across the decisions generated by all roles. Extensive experiments show that, using only the original baseline training set without constructing additional datasets, our method consistently outperforms the corresponding base models and achieves leading or highly competitive performance across benchmarks.
[CV-116] MemoryFusion: Cross-Temporal Memory Learning for Multimodal Video Fusion
链接: https://arxiv.org/abs/2610.39322
作者: Meiqi Gong,Hao Zhang,Jiayi Ma
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Multimodal videos contain rich temporal cues that can facilitate faithful fusion. However, existing video fusion methods mainly rely on short-range temporal interactions, leaving valuable long-term information underexplored. Moreover, inconsistent fusion representations can cause flickering artifacts, undermining temporal coherence. To address these coupled challenges, we propose MemoryFusion, a two-stage framework integrating cross-temporal memory learning with progressive temporal refinement. Considering the complementary roles of adjacent and distant frames, the first stage combines a Short-Term Memory Bank for local temporal guidance with a Long-Term Memory Bank that preserves reliable historical information through a fusion-oriented update strategy. To mitigate temporal fluctuations, a Temporal Smoothing Module leverages soft motion cues for coarse background stabilization. Building upon the resulting fusion prior, The second stage further employs a lightweight 3D residual Temporal Refinement Module to correct remaining temporal inconsistencies while enhancing structural details. Extensive experiments demonstrate that MemoryFusion achieves superior fusion quality and temporal consistency over existing frame-based and video-based fusion methods. The source code is publicly available at this https URL.
[CV-117] CAD-WAM: Anticipating Geometry for Text-to-CAD Generation
链接: https://arxiv.org/abs/2610.39306
作者: Zhengyang Li,Weijie Zheng,Yuxin Cui,Sijie Wang,Wenjie Feng,Qiyu Kang
类目: Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR)
备注:
Abstract:Generating editable computer-aided design (CAD) models from text requires selecting operations that produce the intended changes to the geometry constructed so far. An incorrect operation can compromise subsequent steps, making reliable multi-step generation challenging. Existing methods based on large language models (LLMs) primarily learn to predict commands, without explicit supervision of the geometric states those commands should produce. Inspired by world-action models, we introduce CAD-WAM, which couples geometry-conditioned action generation with next-state prediction. We define complete modeling operations as actions in a CAD domain-specific language (DSL) and encode surface point clouds of intermediate solids into compact latent states using a pretrained variational autoencoder. An LLM backbone generates the next action from the target description and current geometric state, while an auxiliary head on the same backbone predicts the next state to provide geometric supervision during training. We construct step-wise training trajectories from parametric CAD models in the DeepCAD dataset, pairing each trajectory with an expert-level textual description. At inference, each generated action is executed by a CAD kernel, and the resulting solid is re-encoded to inform the next decision. On the DeepCAD test set, CAD-WAM outperforms the task-specific baselines in geometric accuracy and output validity, with larger gains on models requiring more operations, and produces more accurate geometry than general-purpose LLMs prompted with the same DSL. Ablations further show that next-state supervision reduces mean Chamfer distance by 13.2% relative to an otherwise identical model trained only to predict actions. Code is available at this https URL.
[CV-118] Proving and Improving the Platonic Representation Hypothesis with JEPAs
链接: https://arxiv.org/abs/2610.39302
作者: Randall Balestriero,Yann LeCun
类目: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (stat.ML)
备注:
Abstract:Whether models trained on different modalities converge to the same representation of the world is central to AI. The Platonic Representation Hypothesis (PRH) conjectures that they do as they scale, but without a proof it offers neither a guarantee nor a recipe. We turn the hypothesis into a theorem. First, we prove a \emphPlatonic Representation Theorem: Joint-Embedding Predictive Architectures (JEPAs) trained separately on any number of modalities learn the same representation up to one rotation whenever their positive pairs are two observations of one scene–\emphgood views–and their misalignment is bounded otherwise. Second, we show that next-token prediction is a JEPA with \emphbad views, parts of a single document, which explains why language models align with vision only through scale. Third, we validate the theory from 145k captions to the texts of 940M web images, and uncover a scaling law of alignment that predicts unseen runs within seed noise. Good views need 42 times fewer training samples than next-token prediction on the same corpus to reach the same alignment, and a 109M-parameter text JEPA aligns with DINOv2 and CLIP’s image encoders better than every language model we evaluated, up to 72B parameters, and than CLIP’s text encoder: next-token prediction is a remarkably inefficient route to platonic representations. As a showcase, one rotation turns separately trained encoders into an image captioner (CIDEr .838 on COCO). Even that rotation needs no pair: on the latest pair-free benchmark, our text JEPA beats every zero-shot accuracy reported with language models up to 8B parameters, e.g., 64% against at most 56% top-5 on ImageNet-100. Positive views, and not scale alone, are thus the lever to build platonic representations.
[CV-119] Youtu-Parsing-Omni: One Encoder One Schema All Modalities
链接: https://arxiv.org/abs/2610.39301
作者: Yunhang Shen,Zhangquan Chen,Xiaotian Li,Zhongpeng Cai,Qing Zhou,Zekai Zhang,Kun Zhang,Zezhong Wu,Jian Hu,Haihan Gao,Heting Gao,Xin Li,Yinsong Liu,Deqian Jiang,Di Yin,Haoyu Cao,Xing Sun,Bing Liu,Yanqiu Qu,Yunfei Wu,Kun Yin,Zhixiang Wei,Yongxiang Hua,Qianyu Li,Xiaoxing Hu,Peiyuan Zhang,Lijiang Li,Dong Xu,Zhenchuan Yang,Yiliang Jiang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:We present Youtu-Parsing-Omni, an omni-modal parser that converts documents, images, audio, and video into structured records grounded in space and time, providing a perception layer for agents that read, locate, and cite evidence across modalities. Its design rests on one encoder for every modality and one schema for every task. Youtu-Omni-Encoder, initialized from a pre-trained language model to strengthen text reading, encodes all modalities with shared weights in a unified space-time coordinate system, and a few fusion layers let video frames attend to concurrent audio, bringing cross-modal fusion and temporal alignment into perception itself. OmniSchema unifies the outputs of all ten parsing tasks, allowing a single self-evolving post-training procedure, OmniSchema-Aware On-Policy Distillation (OSAD), to provide a shared token-role-based objective across tasks. In each round, the model is trained on its own rollouts against token-level targets from a reference-conditioned copy of itself, which is then refreshed with the trained weights, so that the parser and its supervision co-evolve. With a single 5 B-parameter checkpoint, Youtu-Parsing-Omni achieves state-of-the-art performance on OmniDocBench v1.6 (96.96 Overall) and leads all open-weight models on OmniParsingBench (75.08 Avg.), with its advantage concentrated on audio and video. It also achieves the best overall chart-parsing score (95.02) and the highest exact chemical-fingerprint agreement among compared systems, and matches a dedicated optical-music model in character-level transcription of typeset sheet music while staying far ahead of general-purpose systems. Code and model weights are publicly available.
[CV-120] Unlocking Any-Order Generation in Pretrained Autoregressive Image Models NEURIPS2026
链接: https://arxiv.org/abs/2610.39293
作者: Rishav Pramanik,Marco Pedersoli,Zhaozheng Yin
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: Accepted in NeurIPS 2026, Sydney as a poster
Abstract:Autoregressive (AR) image generators are powerful in principle but inflexible in practice: training with a fixed raster-scan order hardcodes a single factorization of the image distribution into the model, leaving it unable to inpaint, outpaint, or edit at inference time. Existing solutions, including the modern discrete-diffusion family, train from scratch on a mixture of token orderings, paying roughly 3x the compute of standard raster-scan training; discrete diffusion further sacrifices compatibility with the AR inference ecosystem (KV caching, vLLM) by construction. We argue that this cost is unnecessary. A pretrained raster-scan AR model has already learned the joint distribution over image tokens; what it lacks is not knowledge but the ability to express that knowledge through non-sequential conditional pathways. We propose a recipe for adapting pretrained raster-scan models into any-order generators using two components: future-aware positional embeddings that resolve a target-position identifiability problem at negligible overhead, and an ordering curriculum that traces a stable optimization path from the pretrained solution to a good any-order basin. Applied to four base models across two architectures and two scales (LlamaGen-L/XL, RAR-L/XL), the recipe substantially outperforms MaskGIT on ImageNet inpainting (FID 6.21 vs. 10.44) and outpainting, is competitive with MAR-L, trades a 1.2-1.9 increase in class-conditional FID for this flexibility, and costs 13-33% additional compute beyond pretraining, while leaving the standard AR inference stack untouched. Inpainting, outpainting, and editing, our results suggest, need not be added to autoregressive image models. They can be unlocked from the ones that exist. Code is available at this https URL.
[CV-121] PriorMaskMap: Robust Online Vectorized Map Construction with Biased Priors ECCV2026
链接: https://arxiv.org/abs/2610.39283
作者: Haoming Xu,Wei Li,Yu Hu
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注: Accept by ECCV2026
Abstract:Online vectorized map construction provides autonomous vehicles with essential, real-time semantic and geometric scene understanding. While leveraging prior maps presents an opportunity to improve accuracy, they are often biased due to outdated information or temporary road changes. Existing methods, which typically treat priors as fully reliable or only consider limited noise patterns, suffer severe performance degradation when these priors are imperfect. To address this, we propose PriorMaskMap, a robust framework for prior-guided vectorized map construction. Its first innovation is a Prior-Aware Confidence Estimator (PACE) that evaluates the input rasterized map against current visual observations. It assigns confidence scores to regions classified as reliable, erroneous, or blank, creating a foundational mask that guides two subsequent branches. In one branch, the Dual-Prior Query Generator (DPQG) selectively filters and embeds both rasterized and vectorized priors to initialize informative queries from trustworthy map structures. In the other, the Confidence-Guided Prior Fusion (CGPF) module spatially propagates and refines the mask, using the refined version to govern the fusion of BEV features with the prior map, thereby providing rich contextual representations for the map decoder. Evaluation on the nuScenes dataset shows that PriorMaskMap achieves superior mapping precision over existing approaches and exhibits marked robustness when provided with priors containing apparent errors. Our code is released at this https URL.
[CV-122] WorldFlow: Joint Video and 3D Motion Prediction via Large-Scale Pretraining
链接: https://arxiv.org/abs/2610.39278
作者: Yuhao Lu,Jiongxu Chen,Hexiang Wei,Jiajin Wu,Juelin Li,Mingjian Liang,Fan Yang,Zhuowei Liang,Zihao Liu,Zhennan Wang,Sheng Wang,Lin Shao
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Robotics (cs.RO)
备注:
Abstract:Predicting future physical states stands as a central challenge in embodied intelligence. Existing video world models leave 3D dynamics implicit, action-based formulations remain tied to specific robot configurations, and 3D motion prediction models use flow-only supervision, providing a limited learning signal for physical dynamics. To circumvent these limitations, we introduce \method, a generative world model that jointly predicts future video and query-based 3D trajectories. By parameterizing 3D motion as continuous displacement tokens aligned with the video grid, a unified Mixture-of-Transformers (MoT) backbone jointly denoises both modalities under a rectified-flow objective. This formulation couples explicit geometric dynamics with visual evolution, effectively unlocking the physical commonsense embedded within pretrained video models. To empirically scale and validate this unified architecture, we train \method on a heterogeneous mixture of 1.44 million object-level samples spanning simulated and real-world domains. We further introduce WorldFlowBench, comprising 1,678 evaluation samples from 11 data sources across simulated and real-world interactions, including spatial reasoning tasks. Across all three evaluation sets, WorldFlow achieves the lowest average displacement error (ADE) and final displacement error (FDE) among the evaluated learning-based methods. Notably, on the challenging SpatialCog set, WorldFlow achieves an ADE of 0.060 m, a 54.2% reduction compared with the strongest external baseline. It also reduces ADE and FDE by 55.9% and 68.5%, respectively, compared with its flow-only variant. Project page: this https URL.
[CV-123] Efficient Covariance-Aware Vicinal Latent Diffusion for Continuous Conditional Image Generation
链接: https://arxiv.org/abs/2610.39277
作者: Xin Ding,Shuo Huang,Zhenglong Ding,Sen Zhang,Nenglun Chen,Peibei Cao,Yongwei Wang,Fei Wu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 14 pages
Abstract:Continuous Conditional Generative Modeling (CCGM) aims to synthesize high-dimensional data conditioned on continuously valued scalar labels. However, existing continuous diffusion models face severe data sparsity under imbalanced and long-tailed label distributions, which can lead to visual artifacts and condition inconsistency. Moreover, inference-time classifier-free guidance and dense pixel-space integration introduce substantial sampling overhead, limiting practical deployment. To address these challenges, we propose CAViLD, an efficient latent diffusion framework for continuous conditional generation. CAViLD first constructs a purified condition-aware latent space through density-adaptive vicinal learning, auxiliary semantic regularization, and NIQE-guided negative data augmentation. It then internalizes the conditional guidance direction through an EMA teacher, enabling single-branch conditional inference, and introduces covariance-aware progressive distillation for label-dependent probability-flow ODEs, supporting sampling down to four NFEs. Experiments on RC-49, UTKFace, and Steering Angle demonstrate a favorable trade-off among generation fidelity, continuous condition consistency, and sampling efficiency.
[CV-124] DeCoRL: Decomposed Consistency Reinforcement Learning for Multi-Image Composition NEURIPS
链接: https://arxiv.org/abs/2610.39265
作者: Zhiqiang Wu,Shuang Sun,Jiale Zhang,Jing Li,Yibo Chen,Xu Tang,Yao Hu
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: NeurIPS
Abstract:Multi-image composition (MICo) is a critical yet challenging capability for large-scale image editing models, where consistency degrades rapidly as the number of reference images increases. While supervised fine-tuning (SFT) can equip models with basic MICo ability, it provides only implicit supervision on the composed image and often fails to learn reliable object-level correspondence across multiple references. We propose DeCoRL, a reinforcement-learning (RL) framework that improves MICo consistency by decomposing holistic MICo consistency into object-level consistency. DeCoRL first establishes instruction-relevant object correspondences across reference images and localizes the corresponding regions in the generated image. It then uses a VLM-based reward model trained on single-object consistency annotations to produce fine-grained, object-wise feedback, which is aggregated into a multi-object consistency reward to guide RL post-training. We further introduce DeCo-Bench, a benchmark covering compositions from 2 to 6 reference images. Experiments on DeCo-Bench and the public MICo-Bench demonstrate consistent improvements in MICo consistency.
[CV-125] Seeing and Imagining as One: Unified Video Generation and 4D Perception for Embodied World Modeling
链接: https://arxiv.org/abs/2610.39232
作者: Haitang Feng,Zhuoman Liu,Zhennan Wang,Sheng Wang,Juelin Li,Chengzhi Zhao,Hexiang Wei,Lin Shao
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注: Project page: this https URL
Abstract:Embodied world models should not only imagine plausible futures, but also perceive the geometric, semantic, and motion structure of the evolving world. We present GenSight, a unified embodied world model that brings future video generation and dense perception into a single pretrained video diffusion transformer. GenSight uses a unified spatiotemporal representation for video, depth, embodiment segmentation, and first-frame-relative 2D flow. Built on a shared DiT backbone, GenSight uses lightweight task modulation to predict all modalities within the same architecture. All tasks are jointly optimized within a unified velocity prediction space. At inference, GenSight supports two complementary modes: multi-step future video generation with aligned dense predictions and single-step dense perception from an observed video. By unifying modality representation, model architecture, and prediction space, GenSight avoids separate modality-specific pipelines and couples appearance, geometry, semantics, and motion in a shared representation. It promotes cross-modal consistency and allows dense perception to benefit from the spatiotemporal priors learned through video generation. Trained on embodied manipulation data spanning multiple robot embodiments and human-hand interactions, GenSight demonstrates strong performance in both video generation and dense 4D perception, highlighting the effectiveness of unifying seeing and imagining for embodied world modeling.
[CV-126] Fast rigid rotational alignment of voxelized 3-D density maps in sliced Wasserstein distance
链接: https://arxiv.org/abs/2610.39214
作者: Oscar Mickelin,Yunpeng Shi,Amit Singer,Eric J. Verbeke
类目: Numerical Analysis (math.NA); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:We develop a fast method for the rigid alignment of voxelized three-dimensional density functions over \mathsfSO(3) using sliced optimal transport. Our approach efficiently transforms each density map into a tensor of one-dimensional quantile functions, taken from central slices over the unit sphere, such that the Euclidean distance of the transformed volumes approximates the sliced 2-Wasserstein distance. Our formulation is therefore compatible with existing, fast, voxel-based Euclidean alignment algorithms. We provide theoretical analysis and numerical experiments demonstrating that the sliced 2-Wasserstein distance is a robust metric for comparing three-dimensional shapes and for computing their rigid alignment.
[CV-127] COGS: A Diagnostic Benchmark for Attribute-Aware Object-Level 3D Gaussian Splatting Perception WACV2027
链接: https://arxiv.org/abs/2610.39201
作者: Zheng Xiao,Yifan Liu,Weijian Deng,Jiashu Li,Ziqin Liu,Zhenjun Han,Jianbin Jiao
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at WACV 2027, Datasets and Benchmarks Track. 20 pages, including references and supplementary material
Abstract:3D Gaussian Splatting (3DGS) is increasingly explored as an input representation for 3D perception. Yet it remains unclear how current 3D architectures exploit heterogeneous Gaussian attributes. This ambiguity makes accuracy hard to interpret and limits principled architecture design for 3DGS perception. However, existing object-level 3DGS datasets lack the appearance variation or artifact control needed for systematic attribute-level analysis. We introduce COGS, a large-scale 3DGS dataset with 15,675 artifact-filtered and appearance-rich objects across 40 categories. Built on COGS, we establish COGS-Bench for object classification and attribute-usage diagnosis, complemented by external validation. Our benchmark shows that Gaussian attributes are not uniformly beneficial: selected subsets can outperform the full attribute set, and the effects of appearance, opacity, and covariance-related attribute combinations vary substantially across architectures. We further study this finding with HALO, a reference model that preserves attribute-specific evidence before hierarchical aggregation. Experiments on COGS-Bench and external 3DGS datasets demonstrate that COGS supports systematic diagnosis of 3DGS attribute usage, while HALO highlights the value of role-preserving attribute organization. Our dataset is available at this https URL.
[CV-128] Multi-level Constrained Learning for Underwater Image Enhancement
链接: https://arxiv.org/abs/2610.39183
作者: Wangtao Bao,Liping Xie,Kanjian Zhang,Huimin Lu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Underwater image enhancement aims to correct color and recover detail degraded by wavelength-dependent attenuation, scattering, and noise. Paired-reference supervision constrains output similarity, but does not by itself specify the color mapping, the strength of local detail corrections, or consistency with the degraded observation. We propose MCL-UIE, a paired-supervised framework that addresses these requirements by constraining restoration operations and the training objective. Monotonic Chromatic Transform (MCT) restricts global color mapping in log-LMS space through positive diagonal entries and bounded cross-channel coefficients. Uncertainty-Gated Wavelet Reconstruction (UGWR) modulates high-frequency residuals with a learned confidence gate calibrated against normalized wavelet reconstruction errors. During training, Reference-Anchored Degradation Replay (RADR) renders the enhanced image and its paired reference using shared latent degradation proxies, encouraging both to reproduce the input. The three components share a restoration network that combines structured color correction, locally controlled detail recovery, and reference-anchored replay. MCL-UIE ranks first in thirteen of sixteen full-reference dataset-metric comparisons and produces competitive visual results on no-reference benchmarks. Objective ablations indicate that confidence calibration is most effective when combined with replay supervision. RADR is used only during training and adds no inference-time computation. Our code is available at this https URL.
[CV-129] EmbodiExp: Compositional Experience Transfer for Embodied Agents
链接: https://arxiv.org/abs/2610.39180
作者: Zhifeng Gu,Ying Chen,Rui Jiang,Lei Yi,Bing Wang
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Experience can help embodied agents turn high-level plans into grounded actions, but transferring it across tasks remains challenging: previously effective strategies may be inapplicable to the current state or fail to produce their intended outcomes. We present EmbodiExp, a framework for constructing transferable experiences and composing them online during embodied task execution. Our key observation is that tasks often share subgoals whose completion can be assessed independently of overall task success. We therefore introduce an outcome-grounded subgoal experience that specifies a subgoal, applicability conditions, an adaptable policy, and a verifiable outcome. During execution, applicable experiences guide action generation, while verified outcomes can establish the conditions required by subsequent experiences, enabling online composition. Subgoal-level outcome attribution further enables failure utilization: failed trajectories can still contribute verified local successes and condition-specific guidance on actions to avoid. Experiments on multiple embodied benchmarks under a fold-based protocol with frozen experience banks demonstrate the effectiveness of EmbodiExp.
[CV-130] Edited Endpoints as Partial Goals for Image-to-Video Generation
链接: https://arxiv.org/abs/2610.39179
作者: Sihyun Kim,Sanghyun Jo,Kyungsu Kim
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 34 pages, 8 figures, 28 tables. Project page: this https URL
Abstract:Edited endpoint images make the target of an image-to-video transition explicit and can improve transition completion, but also preserve context in the source image’s layout. Full-frame endpoint conditioning turns this preserved context into requirements on the video’s final state, potentially inducing temporal artifacts. We measure discontinuity with Hard-Cut (HC), late background reversal with Background Motion Reversal (BMR), and motion suppression with Dynamic Degree (DD). These artifacts appear across five video models spanning training-free and learned endpoint conditioning, in model-specific forms. We study Partial, a selective conditioning scheme that constrains only an edit-relevant region estimated from image-editor attention and leaves the remaining terminal content to the video model. Across three backbones on TC-Bench-I2V, Partial reduces temporal artifacts relative to full-frame conditioning, with no significant difference in Transition Completion Ratio (TCR); on Wan2.2-Turbo, uniformly weakening conditioning does not reproduce these reductions. A donor-context intervention keeps full-frame conditioning and replaces only the context outside the edit-relevant region with the final frame of a Partial video. It lowers HC and BMR on HunyuanVideo-1.5 and lowers BMR while restoring DD on Wan2.2-Turbo, with no significant difference in TCR, so constraint extent alone does not account for these artifacts. Together, these findings suggest that edited endpoints should serve as partial future goals, although edit relevance does not guarantee future validity. Code and data will be released at this https URL.
[CV-131] EgoContact: Evaluating Physical Intelligence with Egocentric Video
链接: https://arxiv.org/abs/2610.39167
作者: Adam Deryło,Alexey Gavryushin,Daniel Cremers,Marc Pollefeys,Xi Wang
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注: Project page: this https URL
Abstract:Egocentric video has emerged as one of the most scalable sources of data for learning generalizable manipulation priors. Yet, despite its growing use in pretraining robotic policies, it has been underutilized as a validation tool for physical intelligence, a key component of which is understanding where to make contact with an object to achieve an intended interaction. We present EgoContact, a large-scale egocentric dataset for evaluating the generalizability and precision of learned manipulation priors through future contact-point prediction. EgoContact contains 22.5K manually verified interaction events, each with a pre-contact observation, semantic intent, and pixel-precise contact locations. We find that current methods often struggle to predict precise interaction contacts and sometimes fail to localize the target object in diverse, in-the-wild scenes, even when allowed multiple predictions. Furthermore, we demonstrate the usefulness of EgoContact’s annotated interactions for robotic manipulation: simply using EgoContact as the memory bank of an existing affordance-transfer policy improves grasp success from 14.2% to 31.2% in simulation and from 16% to 58% on a real robot.
[CV-132] Joint Rigid Alignment and Morphable Reconstruction from Partial 3D Face Scans
链接: https://arxiv.org/abs/2610.39161
作者: Eli A. Underwood,Nicolas Rondan,Angel V. Peterchev,J. Matias Di Martino
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 9 pages, 4 figures, 4 tables
Abstract:Aligning a partial, real-world 3D face scan to a canonical representation requires solving two coupled problems at once: estimating the rigid pose that brings the scan into a common frame, and completing the facial shape in regions the scan never observed. Existing pipelines typically decouple these steps, running a generic point-cloud registration front end followed by a separate non-rigid fitting or deformation stage. Building on the GeoTransformer superpoint-matching architecture, we propose a single network that predicts both a rigid pose and a shape-complete point cloud in one forward pass: a cross-attention regressor, conditioned on transformer point features and predicted coarse correspondences, maps a partial scan to per-patch coefficients of a 3D morphable face model learned offline from a corpus of 8,000 identities. The resulting morphed template is then used, in the same forward pass, as the target for fine-grained Sinkhorn matching and rigid pose estimation, supervised purely by geometric alignment to ground-truth morphed geometry, so no hand-labeled coefficients are required. Against six classical and learned non-rigid registration baselines on held-out identities, our method leads the strongest ones by 3-4 times on per-vertex error and baselines that collapse the head onto the visible region by more than an order of magnitude, and is simultaneously the most accurate and fastest method on ground-truth-free scans from two independent real-scan datasets, NoW and NPHM. Code will be made publicly available.
[CV-133] CLaW: Codec-Guided Adaptive Latent Watermarking for Traceable Diffusion Image Generation NEURIPS2026
链接: https://arxiv.org/abs/2610.39160
作者: Huangchao Song,Lingyun Yu,Cong Li,Peiqi Jiang,Mingqi Fang,Hongtao Xie
类目: Computer Vision and Pattern Recognition (cs.CV); Cryptography and Security (cs.CR)
备注: Accepted at NeurIPS 2026
Abstract:With the rapid advancement of text-to-image diffusion models, increasingly realistic AI-generated content has raised serious concerns about misuse and copyright infringement. Digital watermarking offers a promising solution by enabling user-level traceability, yet existing methods remain difficult to deploy at scale, hindered by costly model retraining or detection, poor effectiveness-fidelity trade-offs, and limited robustness to image transformations. To address these challenges, we propose CLaW (Codec-guided Latent Watermarking), a robust and efficient watermarking framework for traceable diffusion image generation with frozen diffusion backbones and inversion-free detection. Specifically, CLaW first uses a pretrained watermark codec to map each watermark message to a latent residual, enabling low-cost sampling-time injection. Then, the residual is injected within a late denoising window, where image semantics are better preserved and the watermark signal is less perturbed by subsequent denoising updates, yielding a better balance between watermark effectiveness and image fidelity. Furthermore, we introduce a decoder-guided adaptive injection mechanism that uses decoding confidence as feedback to dynamically calibrate watermark strength, reinforcing weak watermark signals and improving robustness under image transformations. Extensive experiments show that CLaW preserves high visual fidelity and achieves state-of-the-art robustness under image transformations, enabling reliable user-level traceability.
[CV-134] VEvo-Harness: Causally-Guarded Self-Evolving Harness for Agent ic Image Generation
链接: https://arxiv.org/abs/2610.39158
作者: Yuan Wang,Yuhang Yu,Mengting Chen,Tianrun Xu,Zihao Pan,Xuetao Feng,Wei Li
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:The central bottleneck in visual generation and editing has migrated from achieving pixel-level fidelity to satisfying densely constrained user specifications through increasingly agentic workflows. Prevailing agentic generation pipelines remain shallow generate-verify-rewrite loops treating VLM judgments as coarse visual feedback, either discarding repair experience or admitting ungated updates that poison skill library. With memorize and govern periods left open, they degrade into ungoverned iterators of repeated trial-and-error rather than continual self-evolving harness. Towards this issue, we present \textbfVEvo-Harness, the first causally-guarded continual self-evolving harness system for agentic visual generation. VEvo-Harness introduces four coupled innovations: (1) \textita visual traceback that densifies sparse judgments into root-cause attributions with a hierarchical verification tree; (2) \textita visual continual governance that admits only causally valid skill patches through counterfactual dual-arm image generation replay; (3) \textita lifecycle-managed harness state that hierarchically distills root-cause-anchored episodes across working, experience, and pattern tiers into a persistent skill library of routable and causally verified capabilities; (4) \textita heterogeneous sandbox policy that orchestrates a diverse action space of repair primitives, dynamically routing cost-aware interventions based on root causes to select the most effective strategy. VEvo-Harness surpasses state-of-the-art agentic methods and rivals proprietary generators (\textite.g., GPT-Image, Nano-Pro) while maintaining the continual harness evolution and generalization.
[CV-135] STRATA: Reinforcement-driven Active Sampling with Tri-paradigm Adaptation for Efficient 3D Spatial Transcriptomics
链接: https://arxiv.org/abs/2610.39151
作者: Junchao Zhu,Ruining Deng,Chongyu Qu,Zhengyi Lu,Yanfan Zhu,Yuechen Yang,Yu Wang,Shilin Zhao,Haichun Yang,Yuankai Huo
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Spatial transcriptomics (ST) enables joint characterization of tissue morphology and gene expression. However, 3D volumetric ST acquisition remains expensive. AI-driven methods offer a low-cost alternative but remain limited in accuracy, while recent 2D section-wise ST acquisition still rely on dense grid sampling, resulting in redundant measurements and undersampling of biologically informative regions. To bridge the gap, we propose STRATA, a reinforcement-driven active sampling framework with tri-paradigm adaptation for reconstructing dense 3D transcriptomic maps from sparse voxel-wise ST acquisition. STRATA constructs a 3D graph integrating HE morphology with single-cell priors and employs a reinforcement learning agent to estimate morphological sparsity and transcriptomic informativeness, enabling budget-aware selection of informative spots. Given these sparse measurements, an adaptive tri-paradigm predictor combines retrieval, regression, and cross-layer imputation through lightweight fusion, dynamically weighting their contributions. Experiments across datasets demonstrate improved sampling efficiency and accuracy, particularly under low-budget settings. Code is available at this https URL.
[CV-136] Kinematic Causal Attribution: Black-Box Explanations of Motion versus Geometry for 4D Point-Cloud Video Models
链接: https://arxiv.org/abs/2610.39147
作者: Zerui Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 27 pages, 6 figures, 14 tables
Abstract:4D perception models, which classify 3D point-cloud videos, are being built into robots, assistive monitoring and AR, yet before deploying one a practitioner cannot check which cue it relies on: how a body part moves, or what shape it has. 3D attribution is single-frame, video attribution is confined to RGB pixels, gradient maps need white-box access, attention is absent from state-space backbones, and point-level perturbations assume a point correspondence across frames that real depth and LiDAR data lack. We present Kinematic Causal Attribution (KCA), a black-box, interventional attribution method that queries only the model’s outputs. Its freeze operator removes a region’s motion while keeping its shape, and a shape-free proxy removes the shape while keeping the motion. Together they span a motion \times geometry cube for every tracked region, whose order-symmetric Shapley decomposition is complete and whose interaction is measured rather than assumed away. The operators rewrite per-frame point sets, so they need no correspondences, and a context-averaged estimator handles structures that span several regions. We validate KCA on the official PSTNet, P4Transformer and Mamba4D trained on MSR-Action3D. The decisive test is an audit: two binary tasks are built from real clips whose classes differ only in the motion or only in the shape of the same arm, and the motion share of the arm’s importance follows the defining cue on all three backbones (0.83 vs. 0.60 on P4Transformer, 0.86 vs. 0.30 on PSTNet, 0.92 vs. 0.61 on Mamba4D), a distinction no single attribution map can make. The split costs no faithfulness: the necessity maps have the lowest deletion AUC on P4Transformer and PSTNet, tie with the best baseline on Mamba4D, and lose their correlation with the trained model’s once the weights are randomised. Skeleton ground truth, re-sampling stability and cost complete the validation.
[CV-137] LightGenBench: A Benchmark for 3D Emission Generation
链接: https://arxiv.org/abs/2610.39038
作者: Dongchen Yang,Xingguang Yan,Manolis Savva
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project page: this https URL
Abstract:Recent generation methods produce 3D assets with detailed geometry and rich textures. However, these assets are non-emissive: they reflect light but emit none. We study a new task, emission generation: predicting spatially varying RGB emission over a 3D asset’s surface from its geometry, non-emissive materials, and a rendered view. To support this task, we introduce LightGenBench. Our benchmark makes three contributions. First, we curate a dataset of 37k artist-authored emissive assets in UV, voxel, and multi-view form. Second, we establish baselines across four method families: UV-domain, sparse-voxel, multi-view, and segmentation-based generation. Third, we design an evaluation protocol that measures where an asset emits and how, complemented by user-study comparisons on renderings with varying emission strength. Across the four families, the sparse-voxel baseline TRELLIS.2-Emission performs best because generating emission natively in 3D avoids the failure modes of the other representations, such as the fracturing artifacts of UV space and the occluded emitters that multi-view generation cannot see. However, TRELLIS.2-Emission only reaches an IoU of 0.271, leaving the task far from solved. We hope LightGenBench establishes emission generation as a benchmarkable problem for 3D generation. Comments: Project page: this https URL Subjects: Computer Vision and Pattern Recognition (cs.CV) Cite as: arXiv:2610.39038 [cs.CV] (or arXiv:2610.39038v1 [cs.CV] for this version) https://doi.org/10.48550/arXiv.2610.39038 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[CV-138] he Search for Statistical Indistinguishability Between Real and Fake Images
链接: https://arxiv.org/abs/2610.39014
作者: Mohaimin Rafi,Otavio Moraes Rosa,Anshuman Chhabra,Utkarsh Ojha
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Images produced by generative models appear strikingly similar to real ones. In this work, we study whether perfection has been reached by generated content in terms of matching the properties of real data. We train deep neural networks to separate real data from fake and observe their test accuracies. Only a chance performance has any scope of proving statistical indistinguishability. However, since generative models are known to sometimes introduce small errors, e.g., low level checkerboard artifacts, we create progressive abstractions of the original real/fake images to potentially bring them to a space where those artifacts get wiped away. These include operations such as blurring, downsizing, cropping, converting RGB images into edge/semantic maps, etc. To our surprise, almost none of the abstractions are able to make fake data indistinguishable from real. The only cases when neural network based classifiers fail to separate the two categories is when the data is abstracted to such an extent that the original content seems mostly lost to the human eye, e.g., downsizing a 256 \times 256 RGB image of human face to 4 \times 4 resolution. We observe this trend across four different datasets of varying complexity (MNIST, CelebA, FFHQ and GenImage) and across various types of generative models (GANs, VAEs, diffusion models). This shows that the nature of fake artifacts extends all the way from low level to very high level in images. The simple implication is that fake images are much more different from real than we are used to think.
[CV-139] Fed-IGTD: A federated learning-based framework for tabular-to-image encoding
链接: https://arxiv.org/abs/2610.39008
作者: David Chushig-Muzo,Eva Milara,Francisco J. Lara Abelenda,Luis Carlos Estrada-Petrocelli,Diego H. Peluffo-Ordóñez
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Tabular-to-image encoding methods, which transform tabular data into image-like representations, have shown promising results in predictive tasks. Despite their benefits, these methods follow a centralized paradigm, in which the data are collected and models are trained on a central device. This raises privacy and security concerns, particularly the risk of data leakage, unauthorized data collection and access to users’ private information. Federated Learning (FL) provides a decentralized framework in which multiple clients jointly optimize a shared model without directly sharing training data. We introduce Fed-IGTD, a horizontal FL-based approach in which clients share a common feature space and upload only local statistics, from which the server reconstructs the global correlation matrix and obtains a feature-to-pixel mapping to train a convolutional neural network via Federated Averaging (FedAvg). We studied two differential privacy (DP) mechanisms: (i) a one-shot Gaussian mechanism protecting the correlation-aggregation step; and (ii) a client-level DP-FedAvg variant protecting the training with privacy loss composed via a Renyi-DP accountant. We evaluated Fed-IGTD on five datasets under a systematic sweep of non-IID severity and privacy budgets. Privacy protection applied to the mapping-construction step has little impact on predictive performance. By contrast, DP-FedAvg substantially degrades utility at moderate privacy budgets with a small client cohort (K=5). Ablation study indicates that this degradation is driven mainly by the injected noise rather than by uniform averaging or update clipping. Macro-F1 exhibits sensitivity to performance degradation than accuracy under non-IID heterogeneity and class imbalance.
[CV-140] AlbumVSR: Identity-Preserving Face Video Super-Resolution from a Personal Photo Album
链接: https://arxiv.org/abs/2610.39001
作者: Nianyao Du,Xijun Wang,Stanley H. Chan
类目: Computer Vision and Pattern Recognition (cs.CV); Image and Video Processing (eess.IV)
备注: 5 pages, 2 figures, 2 tables. Nianyao Du and Xijun Wang contributed equally
Abstract:Severe blur, downsampling, noise, and compression can erase the identity cues needed to faithfully restore a person’s face. We therefore use photographs of the same person as an external source of identity. AlbumVSR conditions a frozen, one-step streaming diffusion transformer on a personal album. We use four MagFace/CLIP set tokens to encode the album, and 64 texture tokens which come from a fused UV atlas rendered at the current window’s estimated face pose. These tokens are appended to the transformer’s cached cross-attention context. On an identity-disjoint VoxCeleb2 benchmark with heavy synthetic degradation, AlbumVSR raises ArcFace similarity to the ground truth from 0.586 to 0.670 and face PSNR from 24.98 to 26.71 dB. On the 91 test identities, its mean identity similarity is higher than that of all 14 baselines.
[CV-141] What do MLLM s know about Waldo? Evaluating Visual Reasoning under Multi-Instance Multi-Object Challenges
链接: https://arxiv.org/abs/2610.38957
作者: Saumya Mundra,Ajoy Mondal,Vineeth N Balasubramanian,C. V. Jawahar
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Recent advances in Multimodal Large Language Models (MLLMs) have led to significant progress in visual reasoning capabilities. However, existing benchmarks curate image-question pairs pertaining to object categories with few instances, enabling trivial object referencing and limited visual ambiguity. In this work, we introduce MIMO-Bench, a benchmark for evaluating MLLMs in a Multi-Instance per Multi-Object categories (MIMO) setting. Each image-question pair in our benchmark pertains to instances belonging to object categories with multiple instances. The presence of such similar feature distractors causes hindrance in basic visual reasoning tasks, necessitating the referring expression to be composed of multiple visual subtasks to uniquely and unambiguously identify the target object. MIMO-Bench comprises 3,802 image-question pairs across attribute binding and numerical counting tasks. Despite impressive performances on existing benchmarks, both SOTA open-source (InternVL, QwenVL) and proprietary (Gemini, GPT-5) models exhibit substantially lower performance on MIMO-Bench, highlighting their limitations in the MIMO setting. The benchmark will be made publicly available.
[CV-142] GradSplat: Gradients Inside the Forward Pass for Feed-Forward 3D Gaussian Splatting
链接: https://arxiv.org/abs/2610.38951
作者: Jingong Chen,Linlian Jiang,Guangyuan Li,Xinxin Zuo,Yang Wang,Chulwoo Pack,Kyle Gao,Kwanghee Won
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Feed-forward 3D Gaussian Splatting predicts a scene from a few images, but our predictor is less accurate than its own Gaussians optimized per scene on those images. This gap lies mostly in the shape of the Gaussians, their scale and rotation, since optimizing the shape alone recovers almost all of it and opacity and color recover less. Gradient steps on the input views correct this shape, and the model holds these views at inference. Log-anisotropy correction changes the axis ratios of each Gaussian in log space at constant volume, and in-forward gradient refinement renders the Gaussians into the input views and applies the gradient of the photometric loss with a learned recurrent update, first to opacity and color and then to the shape. GradSplat is a feed-forward model that refines its own Gaussians by learned gradient steps inside its forward pass, with no target view. On both complete six-view RealEstate10K protocols, it has the best PSNR and SSIM among nine methods and keeps its accuracy from 6 to 42 input views. Without fine-tuning, it is the most accurate on DL3DV and has the best PSNR and SSIM on ScanNet++. The code will be released at this https URL.
[CV-143] OneScene-Bench: Do Vision-Language Models Integrate Evidence Across Views? NEURIPS
链接: https://arxiv.org/abs/2610.38926
作者: Swathi Suhas,Anurag Das,Bernt Schiele,Jonas Fischer
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at LM4RWD Workshop, Conference on Neural Information Processing Systems (NeurIPS), 2026
Abstract:Vision-language models (VLMs) increasingly process multiple views of the same scene, but it remains unclear whether they integrate these observations into a coherent scene-level representation. Existing multi-image evaluations do not isolate this capability, as tasks may admit correct solutions through independent per-view processing without requiring cross-view correspondence. We introduce OneScene-Bench, a controlled benchmark designed to explicitly distinguish when cross-view integration is necessary. Our benchmark evaluates three complementary capabilities on real-world multi-view scenes: unique object counting, cross-view consistency, and spatial reasoning. For counting, we control object overlap across views to separate cases solvable by simple per-view aggregation from those requiring instance-level correspondence. For consistency, we introduce localized edits in shared and view-specific regions, testing whether models distinguish genuine scene contradictions from differences caused by viewpoint and visibility. For spatial reasoning, we construct object relations that require combining spatial evidence across viewpoints to compare physical proximity. We evaluate twelve state-of-the-art open-weight and proprietary VLMs and find that counting accuracy drops substantially on complementary views, ranging from 4.5-50.0%, compared with 45.5-81.5% on complete-overlap examples. Moreover, cross-view consistency remains around 50% across models, including recent proprietary VLMs, despite substantially stronger performance on spatial reasoning. Our results reveal a substantial gap between processing multiple images and integrating evidence across them, highlighting the need for evaluations that explicitly test scene-level multi-view understanding. Code and data are available at this https URL.
[CV-144] Beyond Region Memory: Local Point Tracking for Vocal Cord Localization in Video Laryngoscopy WACV2027
链接: https://arxiv.org/abs/2610.38830
作者: Huayu Wang,Gyungmin Ko,Jung Heon Kim,Jia-Xian Jian,Dae Kon Kim,Dong Keon Lee,Hsiang-Wei Huang,Pau-Choo Chung,Linda Shapiro,Richard B. Utarnachitt,Jenq-Neng Hwang,Nathan J. White
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted by WACV 2027
Abstract:Reliable vocal-cord localization in emergency video laryngoscopy remains visually challenging due to rapid tissue deformation, specular highlights, and instrument occlusions. We propose Local Point Tracking (LPT), a robust framework that combines a single-frame detector, local point propagation, and stateless mask refinement. LPT follows a localize first, snap second pipeline. First, a single-frame semantic detector generates sparse anchors, within which LPT initializes local query points (SAM-supported by default, with a regular grid as fallback). A point-propagation module tracks these points through detector-missed frames and recovers temporal bounding boxes from their geometry. Second, because point-derived boxes can be loose around the visible boundary, a box-snap step refines the selected box to the target boundary. This decomposition keeps temporal state in local point coordinates rather than a stateful mask or whole-region representation, while the spatial refinement module handles single-frame boundary precision without feeding mask errors back into temporal memory. Extensive evaluations on two distinct datasets demonstrate that our proposed LPT method achieves strong overall performance against existing baselines, with the best mAP_50, success AUC, and missing rate on both datasets. Furthermore, ablation studies confirm that point tracking primarily enhances temporal tracking robustness, while box-snap refinement effectively improves boundary tightness.
[CV-145] A 211K Parameter Foundation Model: Genetic Prompting in Neural Cellular Automata
链接: https://arxiv.org/abs/2610.38776
作者: John Kalkhof,Niklas Ihm,Daniel Kirste,Anirban Mukhopadhyay
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:The scale of modern foundation models enables broad task generalization. In medical imaging, where segmentation demands vary widely across clinical contexts, such flexibility is invaluable. But current models, such as SAM trade adaptability for scale, locking their utility behind high-end infrastructure and leaving behind the low-resource environments most in need of diagnostic support. To bridge this gap, we introduce Foundational-NCA, a lightweight foundation model capable of adapting to a wide range of medical segmentation tasks without retraining. Built on Neural Cellular Automata (NCA), Foundational-NCA incorporates gradient-based learning with genetic optimization over a latent task space, using a compact task vector to guide predictions. This enables dynamic adaptation via a genetic prompt mechanism, allowing the model to generalize across tasks while keeping the parameter count magnitudes lower at just 211 thousand, compared to SAM-based models with 38.9 to 636 million parameters. We evaluate Foundational-NCA on 15 trained and 6 entirely unseen segmentation tasks from the TotalSegmentator dataset, achieving an average Dice score of 84.5% on known tasks and 60.4% on unseen ones. These results demonstrate that task generalization is possible in ultra-compact models, laying the groundwork for resource-efficient, deployable foundation models in clinical environments where infrastructure is limited. By reducing model size by over 180 times, Foundational-NCA takes a step toward a more equitable future for medical AI, one where powerful tools are within reach of every healthcare setting.
[CV-146] RuleForge: Inferring and Instantiating Latent Visual Rules as Executable Programs
链接: https://arxiv.org/abs/2610.38737
作者: Weifeng Xu,Yaxiong Wang,Zhedong Zheng
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 13 pages, 5 figures
Abstract:Visual rule reasoning asks whether multimodal models can recover an abstract concept from a few examples. Existing studies either infer rules from visual examples but evaluate only classification or verbalization, or supervise executable generation with rules provided as input. Consequently, the relationship between visual rule understanding and reliable rule instantiation remains insufficiently explored. We present Bongard-Forge100k, a controlled benchmark, together with RuleForge, a backbone-agnostic framework that jointly learns visual rule understanding and rule-conditioned executable generation. The benchmark contains 108,797 problems: 86,000 for training, 19,297 for testing familiar rules, and 3,500 for testing held-out rule compositions. The framework progresses through rule-understanding training, joint understanding and scene-program supervision, and refinement via execution-grounded reinforcement learning with Understanding Replay. Executing generated programs under multiple random seeds provides verifiable feedback on executability, rule satisfaction, diversity, and robustness. With UniLIP as the primary backbone, RuleForge improves multiple-choice accuracy over the base model by 20.93 percentage points on familiar rules and 28.42 percentage points on held-out rule compositions, while increasing scene-level rule satisfaction on familiar rules from 2.91% to 70.99%. Additional experiments on InternVL3.5, Qwen3-VL, and Show-o2 demonstrate the effectiveness of the proposed training data and framework across diverse multimodal architectures.
[CV-147] nkGround: Benchmarking Fine-grained Visual Grounding Capabilities of Vision-Language Models for Cursive Handwriting in the Wild
链接: https://arxiv.org/abs/2610.38720
作者: Sidra Hanif
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Human handwriting is produced as an ordered sequence of pen movements, but an offline image keeps only the final ink. Handwriting trajectory recovery aims to restore the ordered stroke points and their pen-up/pen-down transitions. Existing image-trajectory pair datasets are recorded with digitizing tablets under controlled conditions, which limits writer coverage and stylistic diversity. Moreover, there is no benchmark that measures the visual grounding capabilities of general-purpose Vision-Language Models (VLMs) for the stroke trajectory recovery task. In this work, we introduce inkGround, a benchmark for stroke trajectory recovery from full-page, camera-captured handwriting. inkGround pairs 537 page images (~18.7k words, ~163k strokes) with reference trajectories produced by a pretrained offline-to-online model, filtered by OCR agreement, and verified by human raters. Because these references are pseudo-labels that represent one plausible pen trajectory, we also evaluate VLMs on the IAM-OnDB dataset, which provides recorded pen trajectories. We cast the task as text-conditioned point grounding: given a page image and a target word, the model must locate the word and output its ordered stroke points. Outputs are scored by trajectory fidelity (DTW, Chamfer distance), readability of the rendered strokes (Accurate Rate, Correct Rate), and style similarity. We benchmark four open-source and three frontier closed-source VLMs with a shared in-context prompt. Closed-source models, notably Gemini-3.1-Pro, produce legible trajectories close to the reference, and this ranking also holds on IAM-OnDB.
[CV-148] Geometry Matters in Packed 3D Attention NEURIPS2026
链接: https://arxiv.org/abs/2610.38694
作者: Amil Khan,Anantajit Subrahmanya,B.S. Manjunath
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Accepted to NeurIPS 2026 (Poster)
Abstract:Packing a 3D volume into a token sequence is not merely long-sequence modeling. Once the volume is packed, sequence order no longer encodes the geometry of the sample—so a positional mechanism indexed by sequence position can learn the memory layout rather than the object. This is a general failure mode for attention over packed geometric data. We introduce PACK3D, a local-global attention operator that separates memory layout from attention geometry. PACK3D keeps local neighborhood attention exact, compresses only distant context into contiguous 3D block summaries, and indexes every query-key interaction with axial RoPE on true token-grid coordinates rather than packed positions. Because the rotation is applied before a block is summarized, the query-to-summary logit remains a weighted sum of query-relative 3D displacements; compressed global context therefore stays spatially interpretable after packing, instead of collapsing to anonymous pooled content. We measure the failure mode directly. On a controlled synthetic task—train on one packing, change only the memory order of the tokens, hold the physical labeling rule fixed—learned positional embeddings and packed-sequence RoPE solve the training layout perfectly yet collapse under Morton, block-contiguous, and random repacking, while axial grid-coordinate RoPE stays exact throughout. The same coordinate system carries to real data: on Allen WTC-11 hiPSC microscopy, PACK3D improves over a parameter-matched 3D U-Net on the 15-structure joint task and over a matched learned-position local-global baseline, with controls isolating the coordinate system, not encoder capacity, as the source of the gain. A fused Triton implementation matches windowed-attention latency at 65k tokens while supplying the long-range context windowing cannot—so the operator is not only correct but trainable at scale.
[CV-149] RIM: Temporal Run-Time Monitoring of 3D Object Detection via Activation Statistics
链接: https://arxiv.org/abs/2610.38613
作者: Hakan Yekta Yatbaz,Konstantinos Koufos,Mehrdad Dianati
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注: 6 pages. Accepted for publication at the 2026 IEEE 23rd International Conference on Smart Communities: Improving Quality of Life using AI, Robotics and IoT (HONET)
Abstract:Reliable perception is essential to intelligent transportation systems, yet deep neural networks can silently miss safety-critical objects while continuing to produce apparently plausible outputs. Existing run-time monitors for 3D object detection generally assess each frame independently and therefore cannot exploit the temporal evolution of occlusion, point sparsity, and scene complexity. We present TRIM, a lightweight temporal monitor that detects missed-object failures without modifying or retraining the underlying detector. TRIM summarises intermediate activations using compact channel-wise statistics and processes short sequences of these summaries with a temporal convolutional network. Experiments on NuScenes show that temporal monitoring improves AUROC from 0.686 to 0.704 and reduces the false alarm rate at 95% recall from 0.818 to 0.759 compared with single-frame monitoring. Early and intermediate backbone features provide the strongest monitoring signal, while the temporal inference head requires only 0.159 ms on a CPU.
[CV-150] V-CAPE: Visual Consistency Assessment for Product Image Editing
链接: https://arxiv.org/abs/2610.38611
作者: Iaroslav Melekhov,Roberto Alcover Couso,Francesco Taioli,Daniel Coelho,Artur Bekasov,Virginia Fernandez Arguedas,Jose Miguel Grande Saiz
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Recent advances in generative image editing enable complex transformations such as background replacement, object manipulation, and pose adjustment from natural language instructions. However, verifying that these transformations preserve visual consistency, defined as the preservation of an object’s intrinsic visual attributes, including geometry, texture, color, and fine-grained details, remains a largely unsolved problem. This challenge is particularly pronounced at scale, where manual inspection is infeasible and automated quality assessment (AutoQA) could become essential. In this work, we introduce Visual Consistency Assessment for Product Image Editing (V-CAPE), a large-scale dataset for evaluating visual consistency in generative image editing. The dataset consists of 40K unique source-target product image pairs, each annotated with binary accept/reject labels and structured failure modes capturing fine-grained inconsistencies such as geometric distortions, pose errors, and identity mismatches. V-CAPE combines real-world edited images with a synthetic subset constructed from 3D assets, enabling both realistic evaluation and controlled analysis. Vision-language models (VLMs) across all evaluated size categories on AutoQA achieve accuracy below 65%, indicating that the dataset poses a substantial challenge regardless of model scale. Our results highlight fundamental limitations of current models in verifying visual consistency under transformation, positioning V-CAPE as a benchmark for advancing fine-grained image editing evaluation.
[CV-151] Data Scaling Laws in Stereo Matching: From In-domain Fitting to Cross-domain Generalization WACV2027
链接: https://arxiv.org/abs/2610.38600
作者: Tong Zhao,Di Liang,Yiming Zeng,Bo Xu,Yingzhi Tang,Chen Bai
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: WACV 2027
Abstract:In the era of large models, stereo matching is developing towards generalizable foundation models trained on million-scale synthetic images. However, how model performance scales up with synthetic data is still underexplored. In this work, we preliminarily investigate the in-domain and cross-domain data scaling laws in stereo matching. A realistic synthetic dataset, named XStereo, containing 200 thousand samples is first built to serve as a high-quality benchmark for data scaling research. In-domain training on five large-scale datasets demonstrate a two-regime scaling behavior with data, transitioning from the standard power law to saturation, where the critical transition point is approximately invariant under a fixed model. Cross-domain performance exhibits a substantially weaker and less predictable dependence on the source-domain data scale. The distribution and coverage diversity are more effective for diminishing domain gaps. More instructive suggestions are provided for high-efficiency data construction and training. The insightful findings are significant verification and extension of the existing scaling laws in both language and vision, paving the way to efficient and stronger stereo matching. The XStereo dataset is public at this https URL.
[CV-152] SpectralTF-MAE: A Pre-trained Spectral Foundation Model for Tokamak Fusion Plasma Diagnostics
链接: https://arxiv.org/abs/2610.38593
作者: Xiao Wang,Zhijie Lu,Duoqing Yang,Jiangtao Yu,Jin Tang,Zichao Lin,Hongming Zhang,Guosheng Xu,Bo Lyu,Qiang Liu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Spectroscopic diagnostics provide rich information for characterizing plasma states in tokamak fusion experiments, while existing learning-based approaches are often designed for individual tasks and rely heavily on labeled data. In this work, we propose SpectralTF-MAE, a self-supervised spectral foundation model for tokamak plasma diagnostics. The model is pretrained on paired charge-exchange recombination spectroscopy (CXRS) and X-ray crystal spectroscopy (XCS) measurements through masked spectral reconstruction. To better capture diagnostically important spectral structures, we introduce a peak-aware masking strategy together with a peak-enhanced reconstruction objective, encouraging the model to focus on characteristic spectral regions. A shared Transformer encoder is employed to learn transferable representations across different diagnostic modalities. The pretrained model is further adapted to several downstream tasks, including plasma-parameter prediction, cross-diagnostic spectral generation, and spectral peak analysis. Experiments on EAST diagnostic data demonstrate that the proposed framework achieves competitive performance across different tasks and consistently benefits from self-supervised spectral pretraining. These results indicate the potential of SpectralTF-MAE as a general and transferable representation learning framework for plasma spectroscopic diagnostics.
[CV-153] RoSAMDepth: Robust Self-supervised Depth Estimation Leverag ing Segment Anything Model CVPR2026
链接: https://arxiv.org/abs/2610.38587
作者: Xuanang Gao,Zhiwei Ning,Gengming Zhang,Jiaxi Cao,Runze Yang,Zhonglong Zheng,Jie Yang,Rong Xiao,Wei Liu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 13 pages, 8 figures, including supplementary material. Published in CVPR 2026
Abstract:Robust depth estimation aims to maintain high-quality depths across diverse conditions. However, most existing methods estimate depth without taking into account the object-level information. As a result, the predicted depth may easily deviate within objects and become blurred under adverse conditions. To overcome this weakness, we propose RoSAMDepth, a novel framework that can assist robust self-supervised depth estimation in leveraging rich and diverse object-level priors from the Segment Anything Model (SAM). We focus on incorporating object-level information across three key aspects: a segment-guided representation contrasting method that injects object-level awareness into the feature representation space; an adaptive regional outlier masking strategy combined with a regional Gaussian likelihood loss that enforces regional depth smoothness; and an object-level reliability estimation strategy that mitigates the influence of unreliable supervision. Extensive experiments across multiple datasets and diverse weather conditions demonstrate that our method produces sharper, more accurate depth predictions, consistently outperforming state-of-the-art methods. Our code is available at this https URL.
[CV-154] Detecting Fine-Grained Differences between Image Datasets
链接: https://arxiv.org/abs/2610.38547
作者: Leander Girrbach,Alexander von Recum,Kirill Bykov,Zeynep Akata
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Detecting whether and in particular how two data distributions differ is a fundamental statistical problem. For example, we might be interested in which concepts (expressed in natural language) differ between two classes, the outputs of different generative models, or the same model with different prompt specifications. When working with large and complex image datasets, we intuitively think that first selecting concepts from a broad list using fine-grained patch embeddings, and then optimizing the concepts to better fit the data is an effective workflow. However, prior work instead uses MLLMs to propose concepts and image-level embeddings to select them. Therefore, we systematically compare these techniques and find that selecting concepts from broad lists can replace MLLM-proposals at obviously lower cost. MLLM inference can then be spent on optimizing selected concepts instead to achieve substantial improvements. Additionally, by designing a new benchmark that focuses on low-prevalence and multi-concept distinctions, we find that selecting nonredundant concepts instead of only the most distinguishing concepts matters, and we introduce techniques for how to effectively do this. Finally, we show how our methods address important problems like detecting spurious correlations and biases. Overall, our work presents a fundamental perspective shift on how we analyze interpretable distributional differences between datasets.
[CV-155] Self-Consistency Geometry Injection for Multi-view Consistent Image Generation ACCV2026
链接: https://arxiv.org/abs/2610.38518
作者: Yupeng Xiao,Yankai Wu,Jianjie Luo,Jiajie Lin,Yi Yu,Zhenguo Yang
类目: Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR)
备注: Accepted by ACCV 2026
Abstract:Recent diffusion models have achieved impressive progress in single image to multi-view generation by synthesizing consistent novel views efficiently. Existing methods usually explicitly or implicitly couple geometry inference and texture synthesis, making generation vulnerable to inaccurate geometry prediction. Although several geometry-guided methods have explored decoupling, they usually treat predicted geometry as a fixed condition and do not explicitly model whether the geometry is reliable. We therefore propose Reliability-Guided Multi-View Generation (RGMV), which comprises a parameter-free Self-Consistency Geometry Gate (SCG) and a Reliability-aware Geometry Injection (RGI) module. Specifically, a geometry proposal generator produces multiple geometry predictions for the same reference image and target viewpoints under different stochastic sampling configurations. SCG measures their spatial agreement to derive a reliability map, which RGI uses to modulate the injection strength of geometry-conditioned residuals in the RGB synthesis model. Moreover, RGMV can adapt to different geometry proposal generators. As a result, RGMV preserves stronger geometric guidance in stable regions while attenuating unreliable geometry cues before they propagate into multi-view RGB generation. Extensive experiments show that reliability-aware geometry injection improves the geometric quality of multi-view generation. Code: this https URL.
[CV-156] he Power Watershed as a Differentiable Segmentation Layer
链接: https://arxiv.org/abs/2610.38516
作者: Galia Akl(AUB),Laurent Najman(LIGM, KUSTAR),Hugues Talbot(OPIS, CVN)
类目: Computer Vision and Pattern Recognition (cs.CV); Neural and Evolutionary Computing (cs.NE); Image and Video Processing (eess.IV); Optimization and Control (math.OC)
备注:
Abstract:Can a Watershed transform, the classical tool of morphological segmentation, be placed at the end of a deep network and trained through? The Power Watershed (PW) is the limit of the seeded Random Walker (RW) as p grows to infinity with edge weights raised to the power p. At every finite p the RW solves a sparse positive-definite linear system and is exactly differentiable with respect to the edge weights and the unaries through one adjoint solve; its multi-label form costs one right-hand side per label, and its limit is the PW. We study this family as the last layer of a segmentation network, the watershed layer (WS layer). A conditioning study measures how the solve, the gradient and the distance to the limit evolve with p, and the implicit gradient is checked against unrolling. On synthetic data with correct edge costs the WS layer transforms a soft prediction into an exact boundary. On nuclei segmentation (PanNuke), a binary WS layer is level with a thresholded U-Net. With the instances as labels, the WS layer turns the U-Net into an instance segmenter whose number of labels is set per image by the markers rather than by the output layout, and the network is trained through that partition: instance F1 score rises from 0.656 to 0.795 over a U-Net with connected components, level with the same network followed by a tuned marker watershed. At the training temperature the partition is a watershed of the learned costs, and the finite temperature lets the solve hedge, discounting weak markers and sharing contested pixels. The hedging puts the WS layer 0.015–0.020 F1 ahead of the baseline when trained on a tenth of the data, at the price of having to deploy the finite-p solve rather than its exact limit at inference. Code and the full experimental record are released.
[CV-157] UnMixMamba: Unmixing Guided State Space Subpixel Mapping of Crops from MODIS Time Series
链接: https://arxiv.org/abs/2610.38507
作者: Motasem Alkayid,Yimin Zhu,Naser El-Sheimy,Lincoln Linlin Xu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Subpixel mapping (SPM) of crop types from \SI250m Moderate Resolution Imaging Spectroradiometer (MODIS) time series to field-scale Agriculture and Agri-Food Canada (AAFC) labels supports acreage estimation and agricultural monitoring. Three difficulties remain: (i)~spectral-temporal coupling and mixed pixels that break direct coarse-to-fine classifiers; (ii)~intra-class endmember variability that a single spectrum per crop cannot capture; and (iii)~uniform refinement on the high-resolution (HR) grid, which treats every subpixel the same, so most of the effort goes to easy pixels while errors remain in the minority class. UnMixMamba addresses them as follows. For~(i), a temporal-spectral-spatial decoupled (TSSD) factorization estimates simplex fractions and drives AAFC classification from the same mixture field, instead of mapping low-resolution (LR) observations directly to HR logits. For~(ii), (\times 8) abundance super-resolution and a learnable per-crop variant endmember bank reconstruct the MODIS cube at LR and HR under a linear mixing model. For~(iii), the predicted crop fractions decide where a fixed Mamba token budget is spent, so HR refinement focuses on the minority class instead of the full grid. On Alberta 2015 under a fixed spatial-block protocol, UnMixMamba reaches OA 0.8397, mIoU 0.6640, AA 0.7837, and Kappa 0.7715, ahead of ST-DRes, ESPCN, FineCrop, AgriFM, and classical SPM baselines. The task is closed-set four-crop mapping inside a known agricultural mask.
[CV-158] Residual Motion Splatting for Continuous Space-Time Video Super-Resolution
链接: https://arxiv.org/abs/2610.38499
作者: Vincent Herfeld(Thoth),Eva Coupet{é},Julien Mairal(Thoth),Jean Ponce(ENS-PSL)
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Continuous space-time video super-resolution produces a high-resolution, highframe-rate video from a low-resolution, low-frame-rate one at arbitrary spatial scales and temporal instants. Classical spatial multi-frame super-resolution has shown that detail beyond a single frame’s sampling limit can be recovered by aligning complementary observations with sub-pixel precision and accumulating them on a finer grid. Existing learning-based methods largely depart from this principle: some align and fuse frames only at low resolution before regressing a continuous high-resolution representation, while others interpolate each frame to the output resolution before splatting. In both cases, the per-frame samples are resampled before fusion, so their sub-pixel positions are lost before frames are combined. Restoring this principle in a learned continuous model raises two challenges: estimating sub-pixel motion at unobserved times, and handling holes and occlusions when splatting a coarse grid onto a finer one. We address the first by predicting per-pixel continuous trajectories that refine linear optical-flow paths with smooth residual motion, using features integrated over the full video sequence. We address the second by directly splatting consecutive low-resolution features in a learned latent space onto a high-resolution grid at their predicted sub-pixel positions with predicted visibility weights, before a decoder resolves remaining ambiguities. This makes temporal interpolation and multi-frame superresolution a single operation. Our method improves both reconstruction fidelity and perceptual quality on multiple standard benchmarks over the state of the art, while running up to an order of magnitude faster.
[CV-159] Vision-Based Density Estimation for Adaptive Traffic Control
链接: https://arxiv.org/abs/2610.38497
作者: Mohammad Usaid,Deepak Silaych,Hrithik Mhatre,Archak Mittal
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 14 pages, 5 figures, 2 tables
Abstract:Traffic congestion is becoming a critical challenge in rapidly growing urban centers, particularly in developing nations where rapid urbanization and inadequate transport infrastructure place increasing strain on transport networks. In highly heterogeneous and chaotic traffic flows, camera-based traffic-state estimation is limited by severe vehicle occlusion and overlap. In this paper, we propose a framework to optimize traffic flow directly from stochastic visual observations. The architecture is structurally decoupled into three primary components: (a) vehicle density estimation, (b) predictive dynamics modeling, and © policy optimization. The proposed methodology introduces a convolutional neural network (CNN) that learns the mapping from camera feeds to spatial density maps, utilizing an adaptive k-nearest-neighbor (k-NN) Gaussian kernel to handle varying congestion levels. This feeds into an ST3DNet predictive module for spatiotemporal forecasting, followed by a hierarchical Graph-Transformer Multi-Agent Reinforcement Learning (MARL) algorithm to optimize signal phases. The density estimation stage was evaluated on 40 annotated intersection images from Rambagh 4, Jaipur, India, with a reported five-fold validation mean absolute error (MAE) of 3.32 \pm 0.95 vehicles and a mean absolute percentage error (MAPE) of 14.5%. These results pertain to the perception component; empirical training and evaluation of the forecasting and control stages remain future work.
[CV-160] Calibrating Track-Query Updates at Test Time for Multi-Object Tracking
链接: https://arxiv.org/abs/2610.38494
作者: Riku Inoue,Shogo Sato,Yu Mitsuzumi,Kazuhiko Murasaki,Tomoyasu Shimada,Takashi Sano
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Multi-Object Tracking (MOT) requires localizing multiple objects and maintaining their identities over time. Recent query-propagation trackers carry target-specific representations across frames and update them with current-frame image features, yet identity association remains difficult when visually similar targets overlap or cross. During such interactions, current-frame query updates may contain components directed toward nearby active tracks, which may affect identity association in later frames. To investigate this issue, we analyze track-query updates preceding identity switches (IDSWs), where the same target is assigned a different track ID. Our analysis shows that (i) propagated track-query representations change more abruptly before IDSWs, (ii) current-frame updates have larger positive projections along directions toward other active tracks, and (iii) bounding-box intersection over union (IoU) retrieves the counterpart track involved in a subsequent IDSW with high recall when that track remains active before the switch. Motivated by these findings, we propose Test-time Track-Query Calibration (TQC), a training-free and gradient-free method for query-propagation trackers. TQC calibrates the tracker output before propagation using a bounded correction estimated from update components directed toward nearby active tracks. It requires only track-query representations and bounding boxes available during inference, without retraining or architectural modification. Across nine DanceTrack and SportsMOT test settings, TQC improves HOTA and AssA in all settings and IDF1 in eight, with an average HOTA gain of +0.38 and limited inference overhead. Controlled variants on six validation settings further show that TQC’s correction direction yields more consistent HOTA improvements than alternative directions with matched correction norms.
[CV-161] Geometry-Aware Trajectory State Modeling for Few-Shot Action Recognition WACV
链接: https://arxiv.org/abs/2610.38482
作者: Zhiqi Miao,Marko Haralović,Stanislav Levendeev,Sarvesh Angadi,Jannatul Ferdous,Dan-Cristian Ploesteanu,Jiapan Guo,Estefanía Talavera
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: This work is accepted by The IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) 2027
Abstract:Few-shot action recognition (FSAR) requires representations that expose discriminative motion dynamics from only a few labeled videos. Recent trajectory-based methods leverage long-term point tracks for explicit temporal correspondence, but often treat trajectories as carriers of sampled visual features or summarize their motion with compact 2D displacement-based descriptors. We instead view an action as the ordered evolution of 2.5D trajectory states and their local geometric interactions. Based on this view, we introduce GeoTRAM, a geometry-aware trajectory state modeling framework for FSAR. GeoTRAM lifts sparse tracks into 2.5D trajectories, where StateMotion models ordered intra-trajectory state evolution and RelMotion captures state-conditioned local interactions through time-varying relative geometry and motion. Quality-filtered spectral sampling provides more reliable trajectory initialization; complementary semantic, motion, and depth cues are separately refined and preserved for joint spatio-temporal modeling; action-guided reweighting and top- \rho trajectory selection support complementary frame- and trajectory-level matching. Experiments on five FSAR benchmarks demonstrate strong gains on the motion-centric Something-Something benchmarks while maintaining competitive performance on more appearance-driven benchmarks and a favorable accuracy–efficiency trade-off.
[CV-162] Automatic Iconographic Annotation of Medieval Illuminations: Comparing Generative Vision-Language Models and Supervised Discriminative Classifiers
链接: https://arxiv.org/abs/2610.38462
作者: Vincent Jolivet(ENC, CJM),Lucas Terriel(ENC),Chahan Vidal-Gor{è}ne(CJM, ENC)
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:The large-scale digitisation of medieval manuscripts creates a growing need for automated iconographic indexing. We evaluate two complementary paradigms for this task on 9,735 full-folio images annotated with 2,333 TIMEL concepts: generative Vision-Language Models (VLM) and supervised discriminative classifiers (SDC). The VLM experiments evaluate fine-tuning, structured JSON decoding, taxonomybased candidates generation, and deterministic reranking. The discriminative experiments evaluate flat, hierarchy-aware, and patch-attention heads over a frozen DINOv3 encoder. Our results highlight key functional differences: ML-Decoder (SDC) achieves the strongest overall retrieval performance (precision@1 of 0.696, MRR of 0.806), with loss calibration (Asymmetric Loss) serving as the primary performance lever. While SDCs provide superior precision and reliability within a closed vocabulary, VLMs offer strong semantic generalization and broad taxonomic coverage, particularly through structured JSON targets and verify-calibrate reranking (hierarchical F1 of 0.559). Furthermore, model behavior strongly depends on concept abstraction: SDCs excel at perceptually grounded descriptive concepts, whereas VLMs provide useful hypotheses for complex interpretative scenes. Finally, we find that enforcing explicit hierarchy within model architectures limits coverage compared to spatial feature aggregation. Overall, rather than competing approaches, SDCs and VLMs emerge as complementary components for future hybrid indexing systems combining structured retrieval with exploratory annotation assistance. As a methodological contribution, we define the soft F1 metric to provide a taxonomy-aware evaluation that credits semantically close predictions based on lowest common ancestors.
[CV-163] IntReal: A Real-World Dataset and Benchmark for Multimodal Interleaved Generation
链接: https://arxiv.org/abs/2610.38425
作者: Shaohan Yu,Jie Qin,Wende Tan,Haiwen Diao,Limeng Qiao,Yueming Lyu,Chenyang Si
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: this https URL
Abstract:Humans routinely consume multimodal content, where textual and visual information is naturally interleaved. Although Unified Multimodal Models (UMMs) can jointly generate text and images, multimodal interleaved generation remains largely underexplored. Progress in this area is hindered by two key limitations: (1) the lack of well-defined interleaved generation tasks and corresponding training data, and (2) the absence of fine-grained evaluation protocols that can reliably diagnose model performance across different capability dimensions. To this end, we present IntReal, comprising two complementary components: IntReal-100K, a real-world dataset of multimodal interleaved samples spanning four major categories, and the IntReal Evaluation Suite, which evaluates model performance across seven complementary dimensions using specialized expert models, enabling coarse-to-fine diagnosis. Extensive experiments validate our evaluation suite through low inter-metric redundancy (max pairwise absolute Spearman’s \rho \leq 0.42 ), high agreement with human judgments (83.3% accuracy), and strong consistency across different expert evaluators (Spearman’s \rho 0.9 ), supporting the complementarity and reliability of our metrics. By benchmarking four families of UMMs, we show that IntReal Evaluation Suite reveals distinct strengths and weaknesses across models. Post-training experiments demonstrate the utility of IntReal-100K for improving interleaved generation and further show that training unimodal generation can also benefit interleaved generation. These findings suggest that interleaved generation can benefit from modality-specific optimization, rather than requiring joint text-image training at every stage.
[CV-164] Structure-retained low-rank adapters for weather synthesis
链接: https://arxiv.org/abs/2610.38413
作者: Shunxin Wang,Alexandros Stergiou,Luuk Spreeuwers,Estefania Talavera,Nicola Strisciuglio
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Real-world deployment of computer vision models, particularly in outdoor scenarios, is often limited by environmental interferences, such as adverse weather. Improving model robustness across weather conditions is thus critical for reliability. However, sourcing high-quality data relevant to isolated weather effects is a significant challenge. Most datasets include changes in dynamic elements across observations of the same scene: subjects such as pedestrians and vehicles change between shots, while camera viewpoints vary. These aspects make it difficult to decouple solely weather-induced distribution shifts from the scene. In this work, we address this with a novel weather synthesis framework that utilizes structure-retaining Low-Rank Adapters (LoRA). We analyze the attention blocks in Stable Diffusion XL (SDXL) and identify where relevant structural information is mostly processed during image-to-image translation. Through an architectural analysis of SDXL, we learn structural LoRA matrices for the corresponding blocks and then leverage them for weather synthesis, eliminating the need for external conditioning such as ControlNet. Experimental results demonstrate that our method improves upon existing synthesis approaches, achieving greater realism while preserving structure. This provides a robust pipeline for generating high-fidelity training and evaluation data for outdoor computer vision applications.
[CV-165] Visually Irrelevant Hallucinations in Medical Vision-Language Models
链接: https://arxiv.org/abs/2610.38389
作者: Ke Zou,Nan Zhou,Tao Yang,Meng Liu,Linchao He,Zehui Liao,Ling Huang,Guangren Huang,Yidi Chen,Yi Zhang,Hu Chen,Huazhu Fu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Medical vision-language models (VLMs) show strong performance in radiology report generation (RRG) and medical visual question answering (Med-VQA), producing free-form clinical descriptions and answers from medical images. However, they may generate clinically meaningful statements that lack visual evidence or contradict image content. We refer to these errors as visually irrelevant hallucinations and investigate them at the phrase level. We propose CLEAR (Counterfactual Localized Evaluation of Alignment and Response), a model-agnostic post-hoc framework for phrase-level hallucination risk estimation. CLEAR decomposes generated responses into visually assessable clinical phrases. For each phrase, phrase-conditioned grounding identifies a candidate visual region, which is selectively perturbed by a localized counterfactual intervention. CLEAR characterizes the intervention using two complementary auditing cues: response cues, capturing semantic and clinically meaningful changes in the generated response, and alignment cues, measuring changes in image-phrase compatibility. These cues are normalized and jointly modeled to estimate phrase-level hallucination risk. Across four datasets spanning RRG and Med-VQA, CLEAR outperforms state-of-the-art hallucination detection baselines across multiple medical and general-purpose VLMs. On MIMIC-CXR, CLEAR improves AUPRC from 85.10% to 90.23%. Its effectiveness is further validated on a GT-derived controlled VIH benchmark. These results establish visually irrelevant hallucination as an important fine-grained reliability failure mode in medical VLMs and demonstrate the value of model-agnostic visual-support auditing for more trustworthy multimodal clinical generation.
[CV-166] SparseTokenGS: Gaussian-based Sparse-View CBCT Reconstruction via Reusable Token Pool
链接: https://arxiv.org/abs/2610.38360
作者: Zihan Deng,Zhenzhen Gong,Yujie Liu,Jiaxin Gong,Dingkun Liu,Kexian Xie,Yixing Huang,Peng Jin,Tianye Niu,Zhisheng Wang,Fenglin Liu,Hongming Shan,Lequan Yu
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: 10 pages, 12 figures. Submitted to IEEE Transactions on Medical Imaging
Abstract:Sparse-view cone-beam computed tomography is widely used for fast, low-dose volumetric imaging and lesion localization in dentistry and image-guided interventions, but it introduces severe artifacts. Feed-forward Gaussian methods transfer the reconstruction cost to training and decode each coordinate from projection-aligned features and neighboring anchors. However, information from the whole scan is written only when the field is constructed and cannot be retrieved during coordinate decoding, leaving the volume inconsistent within the scan. This is a common limitation of such methods: once the field is fixed, decoding uses only local features and reconstruction quality drops substantially. We propose SparseTokenGS, a Gaussian-based method for sparse-view cone-beam computed tomography reconstruction via a reusable token pool. Anchor Token Fusion and Point-Token Attention query this shared pool at the anchors and at the query coordinates, respectively. The local Gaussian prior is retained, and the volume is reconstructed in one forward pass, without per-scan optimization or a projection-domain loss. On three public chest and dental datasets, SparseTokenGS achieves the highest reconstruction performance among the compared methods. On clinical dental scans, reconstruction from only six projections reaches 29.51 decibels in peak signal-to-noise ratio and a structural similarity of 0.842 against the full-view reference, and off-the-shelf tooth-and-pulp segmentation keeps these structures separable for root-canal localization. Experiments on volcanic rock and chicken show performance after same-object adaptation. A reusable token pool gives feed-forward reconstruction a mechanism that keeps scan-wide context available through decoding and can benefit future rapid clinical lesion localization and analysis. The code is available at: this https URL.
[CV-167] Volley3D: Unified 3D Full-Sequence Reconstruction from Monocular Broadcasts
链接: https://arxiv.org/abs/2610.38358
作者: Noâm Boussouf,Adrien Maglo,Mohamed Chaouch,Hideo Saito
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 8 pages + 10 pages of supplementary. Videos available on the project webpage: this http URL
Abstract:3D reconstruction from single-view sports broadcast is a pivotal capability for automated analytics and immersive AR/VR replays, yet notoriously ill-posed due to difficult-to-calibrate moving cameras and depth ambiguity. Unlike widely studied racket sports, team sports such as volleyball feature unpredictable multi-contact sequences and high altitude jumps. These dynamics categorically invalidate the standard planar projection assumptions relied upon by existing methods. To resolve these limitations without 3D training supervision, we propose a complete, physics-grounded pipeline tailored to monocular broadcast volleyball. First, we include a correction module to the perspective drift induced by ground-plane projection by coupling temporal jump spotting with Newtonian interpolation for 3D player placement. Second, for 3D ball trajectory estimation, we design a dedicated sub-trajectory segmentation architecture alongside a collision-matching algorithm. By anchoring physical equations to reconstructed 3D hand vertices, we lift noisy ball 2D detections into a spatiotemporally coherent 3D scene. Evaluations on multi-camera benchmarks demonstrate our approach significantly reduces 3D spatial drift compared to 2D-reprojection baselines.
[CV-168] WeVisCut: The Whole Editing Room in One Agent From Narration to Long-Form Video
链接: https://arxiv.org/abs/2610.38352
作者: Chenyu Mu,Wenming Weng,Qinrui Wang,Xingsong Ye,Hao Yu,Jun He,Chong Sun,Chen Li,Jing Lyu,Erkun Yang,Cheng Deng
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Producing long-form videos from narration requires interdependent decisions that remain coherent across shots. Existing agentic editors often use multi-stage workflows driven by general models, leaving the quality-cost trade-off of compact, task-adapted policies unclear. We address this gap with WeVisCut, a long-form script-to-video system that connects tool execution with policy learning. Its agent harness maintains shared production state so that tool feedback informs subsequent decisions and repairs. Because finished videos do not expose these decisions, our four-stage data pipeline combines editing inversion with verified replay to turn recovered editing intent into executable supervision. Applied to 3,750 finished videos, this pipeline yields over 130K training examples. We use this supervision to train a single 9B policy that coordinates production and recovers from failures within the same harness. Moreover, we introduce a 100-task benchmark and five metrics assessing editing quality and delivery reliability. Under a shared tool environment, WeVisCut substantially outperforms its base model and achieves comparable performance to Claude Opus 5 while using nearly 33x fewer generated tokens per successful output minute. It also outperforms task-matched native editors on our proposed metrics.
[CV-169] Aligning Misaligned Benchmarks for Single-View Novel View Synthesis
链接: https://arxiv.org/abs/2610.38349
作者: Yang-Che Tseng,Erqun Dong,Jeong-gi Kwak,Gopal Sharma,Shakiba Kheradmand,Stephan Garbin,Hossam Isack,Kwang Moo Yi
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:While single-view novel view synthesis methods have been dramatically enhanced recently, how to quantify their performance has yet to be standardized. In this work, we demonstrate that an important factor lies in how each method’s own reference frame is aligned with the so-called ground-truth pose. Existing methods often rely on test-time optimization of target poses for this alignment, which inherently is ill-posed due to the complexity of the optimization landscape of camera poses vs render accuracy, leading to methods being evaluated at an arbitrary local optimum. Some recognize this limitation and thus rely on 3D reconstructions of entire scenes using training and test views to align poses, which may be infeasible. In this work, instead, we propose to use traditional 2D-to-2D and 2D-to-3D image matching techniques to align 3D reconstructions. We also evaluate only within co-visible regions, between different views, which is the only place where a `correct’ ground truth can be defined. Our evaluations show that with proper coordinate alignment, novel-view synthesis methods that use explicit 3D Gaussians achieve the state of the art, despite being previously reported in the literature to be outperformed by pure image-based 2D methods. We further provide a simple baseline that directly utilizes an off-the-shelf video generator and a feed-forward geometry estimator. Our simple baseline is already comparable to state-of-the-art methods specifically trained for novel-view synthesis, without any task-specific training.
[CV-170] Knowledge-structured and evidence-aware semantic self-distillation for glomerular pathology foundation model
链接: https://arxiv.org/abs/2610.38346
作者: Qiming He,Jing Li,Zihan Wang,Tian Guan,Yifei Ma,Zimo Zhao,Yanxia Wang,Hongjing Chen,Yexing Zhang,Yizhi Wang,Xinrui Chen,Xitong Ling,Minxi Ouyang,Yujie Xie,Lianghui Zhu,Yiqing Liu,Qingxia Hou,Shuyan Zhao,Xiaoqin Wang,Lili Ma,Qiang Huang,Zhiyuan Shen,Junru Cheng,Siqi Zeng,Chao He,Yonghong He,Zhe Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Visual foundation models learn rich pathological morphology but lack explicit concept-level supervision. Vision–language pretraining can provide this supervision but requires costly image–text pairs. Here, we introduce GloSAGE, a semantic self-distillation framework that uses lesion profiles predicted by a pretrained visual teacher to construct structured pathology descriptions. Reliability-guided selection identifies image–text anchors, while Knowledge-Structured Pathology Text Synthesis and Evidence-Guided Negative-pair Construction account for partially observed lesions. Image–text alignment trains projection layers while retaining the frozen visual encoder; an intermediate Semantic Transfer Layer supplies features for downstream tasks. Developed on GloMillion and evaluated across seven glomerulus- and case-level cohorts, GloSAGE supported zero-shot lesion recognition, image–text retrieval, hard-negative discrimination and concept-intervention analyses. It outperformed its source visual model in most lesion-recognition tasks and improved lesion grading, while supporting competitive case-level disease classification, injury grading and baseline renal function estimation. Ablations supported the combined contributions of knowledge-structured text and evidence-guided negatives, and the intermediate transfer layer improved mean lesion-recognition ROC-AUC relative to the complete projection output. A frozen language model conditioned on GloSAGE features generated glomerular descriptions. These findings suggest that teacher-derived lesion profiles can provide annotation-efficient semantic supervision for extending a glomerular visual foundation model.
[CV-171] SSQA: Sibling-Selective Quadtree Attention for Hierarchical Modeling in Perception Tasks ICRA2026
链接: https://arxiv.org/abs/2610.38320
作者: Yufan Chen,Arnav Bali,Angela Liu,Laura Zheng,Ming C. Lin
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 8 pages, 6 figures, 4 tables. Published at ICRA 2026 (oral)
Abstract:Perception tasks for navigation in robotics, including aerial platforms such as drones and autonomous driving systems, are inherently structured. Drone-mounted cameras typically capture sky above, terrain below, and obstacles or man-made structures in between, while driving data often contains organized road layouts, lane markings, and surrounding agents. Motivated by these axis-aligned structural priors, we note that such information is typically more structured than in generic image tasks. We hypothesize that processing information in a quadtree-esque manner can not only model features effectively in a hierarchical manner, but also offers an efficient linear-time alternative to vanilla attention mechanisms, which run in quadratic time. In this paper, we propose Sibling-Selective Quadtree Attention (SSQA), which models image tokens hierarchically as a structured, full quadtree. We show analytical complexity analysis that guarantees linear-time feature modeling, in addition to empirical experiments comparing inference speeds with other popular modeling approaches, such as Mamba 2 and Quadtree Attention. Our results, benchmarked across several tasks, show that we achieve results at least as good, if not notably better, as others at a fraction of the computational costs.
[CV-172] Extremely Sparse-View Computed Tomography from 2D Projections via Pose-Aware Diffusion Priors NEURIPS2026
链接: https://arxiv.org/abs/2610.38319
作者: Linrui Dai,Jinxiu Liang,Chu Zhou,Mae Yamaguchi,Ryuya Takahashi,Imari Sato
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: NeurIPS 2026 poster
Abstract:Extremely sparse-view computed tomography (xSV-CT) is critical when scanning time, motion, or radiation tolerance limits each scan to only a few projections. In many such scenarios, the same acquisition limits also constrain dataset construction, making pre-reconstructed 3D volumes from dense scans unavailable. This exposes a gap in current sparse-view approaches: Optimization-based methods, which use the sparse measurements directly but lack data-driven priors to suppress artifacts, and learning-based methods, which use strong priors but usually require densely scanned training volumes acquired outside the xSV-CT regime. We study this train-sparse / infer-sparse setting by learning a pose-aware diffusion prior from one 2D projection per training sample, then guiding diffusion sampling with measured-view fidelity, shared-volume reprojection, and projection-physics consistency for tomographic reconstruction. On two public CT datasets across 1-, 4-, and 8-view settings, the proposed method achieves the best novel-view synthesis scores and best or competitive reconstruction quality compared with baselines that use dense-volume supervision.
[CV-173] Does 3D Pose Improve Action Valuation in Football? An Empirical Study of Pose Representations and Learning Objectives
链接: https://arxiv.org/abs/2610.38308
作者: Rikuhei Umemoto,Kenjiro Ide,Soham Majumder,Ulf Brefeld,Keisuke Fujii
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: Submitted to IEEE Access
Abstract:Football action valuation estimates players’ contributions to future scoring. Existing multi-agent deep reinforcement learning (MADRL) models use 2D positions and velocities, omitting posture information relevant to imminent actions. We investigate whether adding 3D pose to a SARSA-based model improves observed-action values relative to attack returns: goals, conceded goals or final expected possession value. Using UEFA Champions League 2022/23 event, tracking and 3D pose data, we compare six representations (no pose, raw joints, torso and head orientation, and spatial-temporal graph convolutional network (ST-GCN) embeddings) across five learning settings and three seeds. The settings vary temporal-difference targets and action-prediction loss, with Monte Carlo return regression as a reference. Across 50 on-ball and off-ball comparisons, pose robustly improved return prediction in only three, all using torso and head orientation, while worsening it in twelve, mostly with ST-GCN. Training choices had greater effects: removing the action-prediction loss raised on-ball skill of position-only values from 0.003 to 0.040 and their correlation with expected goals from 0.23 to 0.60; the improvement in return prediction persisted at matched training length. Separately, pose improved prediction of the on-ball action 0.2 s ahead (macro AUC 0.914 to 0.934 with raw joints), while the passer’s pose during execution improved pass-success prediction (AUC 0.844 to 0.856). Within this training protocol, pose informed imminent actions but rarely improved observed-action values.
[CV-174] Dont Ask the LLM : Visual Token Pruning with Spectral Structure and Semantic Relevance WACV2027
链接: https://arxiv.org/abs/2610.38303
作者: Intaek Shin,Geuntaek Lim,Sungjune Park,Yukyung Choi
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to WACV 2027. 19 pages (10 pages main paper, 9 pages supplementary material)
Abstract:Large vision-language models (LVLMs) represent images and videos as long visual-token sequences, substantially increasing inference costs. These costs motivate visual-token pruning, whose effectiveness depends on both what signal is used for selection and when that signal becomes available. Our analysis reveals complementary limitations: visual-side [CLS] attention produces scattered, background-biased selections under aggressive compression, while stronger query-conditioned signals become available after substantial computation within the large language model (LLM). This raises a natural question: can structurally coherent, query-aware token selection be performed before consulting the LLM? We therefore propose ST-Pruner (Spectral-Textual Pruner), a training-free framework for pre-LLM visual-token selection. It combines attention-independent spectral saliency with inverted patch-text similarity through parameter-free multiplicative fusion. Across diverse LVLMs, ST-Pruner retains over 95% of full-token performance using at most 11.1% of visual tokens and achieves a 3.2x end-to-end speedup on LLaVA-NeXT. Code is available at this https URL.
[CV-175] Hidden Commitment: When Language Models Silently Pick a Side and How Steering Can Surface It ICML2026
链接: https://arxiv.org/abs/2610.38296
作者: Samuel Dawit Assefa,Jae Won Cho
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Presented at AI4Good @ ICML 2026
Abstract:Language-model answers are often treated as a visible record of the internal decision that produced them. We test this assumption in conflicting-context question answering, where two supplied documents support incompatible answers. A model can process the disagreement and still return a one-sided answer, even under prompts that ask it to disclose conflict. In Qwen 0.8B, hidden source commitment is captured by a residual source-choice direction whose sign tracks which document the answer uses. Sparse middle-layer MLP writers feed the same coordinate, and steering the coordinate changes the chosen source. A transfer run on Gemma 3 1B preserves the source-choice readout and sparse-write bridge under LLM-audited labels. Conflict disclosure follows a different pattern, appearing weaker, later-layer, and only partially steerable. These results support a practical monitoring strategy for retrieval systems that detects silent source commitment with residual readouts and pairs conflict-surfacing prompts with disclosure steering when the model has selected a side under disagreement.
[CV-176] PulmoWorld: An Asynchronous Multimodal Medical World Model for Longitudinal Pneumonia Dynamics
链接: https://arxiv.org/abs/2610.38289
作者: Cheng Tang,Xinyu Zhang,Jiawei Chen,Minghui Zhang,Shaocong Wei,Wei Li,Yuxing Chen,Kongcheng Zhang,Chun-Mei Feng,Nanqing Dong,Xinyuan Zhu,Ziyan Huang,Junjun He,Yirong Chen,Lingtong Huang,Min Cen,Ming Hu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 49 pages, 17 figures
Abstract:Pneumonia changes across respiration, inflammation, and organ function, while the measurements that describe it arrive at different times and are often incomplete. We introduce PULMOWORLD, a world model for longitudinal pneumonia dynamics that connects asynchronous state estimation, recursive simulation, and evidence-constrained planning. History-aware filtering separates state propagation from observation-driven correction, so a measurement updates the shared belief only when it is observed. That belief supports recursive forecasts of clinical values and modality representations, and summaries of future prognosis. Cap-MBRL searches imagined continuations inside an authorized action set and replaces a frozen fallback only when a candidate meets the risk budget and a predicted-cost margin. On the pneumonia cohort MASS, recursive forecasts are competitive with task-trained sequence models, and simulated future states retain prognostic information. Trained separately on public MIMIC-IV-CXR chest radiographs and clinical trajectories, this model improves prognostic discrimination over latest-value fusion in development and is competitive on long-horizon error on the locked pneumonia panel. In five known-mechanism MASS-Syn environments, Cap-MBRL lowers selection regret relative to direct fallback execution.
[CV-177] ConvRoute-ViT: Convolutional Token Routing for Efficient Vision Transformers
链接: https://arxiv.org/abs/2610.38284
作者: Manish Das,Shiv Ram Dubey,Satish Kumar Singh
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 5 pages, Conference
Abstract:Vision Transformers (ViTs) achieve strong image recognition performance, but their self-attention increases quadratically with the number of input tokens and makes them computationally expensive. The standard implementation of ViT models treats all patch tokens with equal importance across every layer of the model. This uniform treatment introduces redundant computation and may reduce the model’s focus on the most discriminative object regions. In this paper, we propose ConvRoute-ViT, a convolutional token routing mechanism for efficient vision transformers that performs token reduction through convolutional routing. Our model preserves the most relevant patch tokens according to class-token attention at selected layers, while the remaining tokens are spatially reorganized and compressed into a compact set of convolutional summary tokens. This design reduces the sequence length without completely removing the contextual information of the less informative tokens. By applying the routing process hierarchically, the proposed framework constructs compact and discriminative representations for deeper layers. Experiments on CIFAR-100 and TinyImageNet show that ConvRoute-ViT improves computational efficiency while maintaining classification accuracy compared with the base ViT.
[CV-178] BudSplat: Feed-forward 3D Gaussian Splatting under a Rendering Budget
链接: https://arxiv.org/abs/2610.38278
作者: Muyu Xu,Fangneng Zhan,Yu Wei,Ling Shao,Shijian Lu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Feed-forward 3D Gaussian Splatting has recently emerged as a promising alternative to per-scene optimization, enabling fast novel view synthesis by predicting Gaussian primitives directly from input images. However, it quickly faces a bottleneck of rendering efficiency while scaling from sparse to dense multi-view inputs, which offer more geometric constraints but produce a rapidly growing set of candidate Gaussians. Existing solutions cannot reliably distinguish redundant primitives, leading to many redundant primitives from repeatedly observed regions and increased rendering cost without much gain in visual quality. We propose BudSplat, a compact feed-forward 3DGS framework that learns to keep the most valuable Gaussians under a limited primitive count. BudSplat has two novel designs. The first is a detail-aware importance score that is extracted from multi-level visual tokens, providing an explicit cue for regions where pruning is likely to harm perceptual quality. The score can be injected into voxel fusion to preserve high-value local contributors during multi-view aggregation. The second is quality-aware Gaussian pruning that selects a compact set of primitives by jointly considering rendering contribution and detail importance. The quality-aware pruning enables local quota redistribution that helps avoid over-allocating primitives to large redundant regions. Experiments on RE10K, DL3DV, and ACID show that BudSplat achieves a favorable trade-off among rendering quality and efficiency.
[CV-179] reemapMix: Dirichlet-Controlled Multi-Image Augmentation for Probability and Ordinal Supervision NEURIPS2026
链接: https://arxiv.org/abs/2610.38265
作者: Ejafa Bassam,Konstantin Garbers,Yingsheng Geng,Dalton Jens,Jiaxu Qian,Guyue Liu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at NeurIPS 2026. 40 pages including references, appendices, and checklist. Ejafa Bassam and Konstantin Garbers contributed equally
Abstract:Data augmentation improves visual recognition by exposing models to synthetic training examples that encourage generalization beyond the observed data. Mixed-sample methods extend this idea by combining multiple images. However, many common approaches either mix only pairs of images or rely on fixed small layouts, limiting control over source count, visible area proportions, and induced supervision. We propose TreemapMix, a distribution-controlled multi-image augmentation that samples source weights from a Dirichlet distribution and assigns them to area-proportional regions using a treemap partition. The resulting known region areas support two complementary objectives. TreemapMix with soft cross-entropy (SCE) treats areas as soft class-probability targets, while TreemapMix with a Plackett-Luce (PL) objective converts areas into ordinal supervision. On ImageNet-1K under matched training recipes, TreemapMix-SCE achieves substantially lower calibration error than evaluated mixed-sample baselines, while TreemapMix-PL achieves the strongest Top-1 accuracy. Transfer experiments on object detection and instance segmentation suggest that TreemapMix-pretrained features remain useful beyond classification. These results show that controlled multi-image composition can expose the same region-area information as either probability-matching or rank-matching supervision. Code is available at this https URL.
[CV-180] ETA-Motion: Efficient Topology-Agnostic Motion Diffusion via Sparse Joint Learning WACV2027
链接: https://arxiv.org/abs/2610.38259
作者: Seoyoung Hwang,Tae-Kyun Kim
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to WACV 2027; Project page : this https URL
Abstract:Topology-agnostic motion generation aims to synthesize motions for diverse characters with arbitrary skeletons. Unlike conventional motion generation, which assumes a fixed skeletal template, this setting must handle diverse joint configurations and inconsistent joint semantics across characters. Existing methods address this challenge through dense full-joint modeling, which incurs substantial computational cost and hinders learning transferable motion patterns across diverse skeletal topologies. We present ETA-Motion, a topology-agnostic motion generation framework based on sparse joint learning. Instead of learning motion from all joints, our method uses a compact subset of joints to reduce redundancy and trains the motion generator on the resulting sparse skeleton. To facilitate correspondence across diverse topologies, we introduce semantic joint embeddings that combine textual and visual cues, together with topology-aware training objectives for preserving skeletal structure. Although ETA-Motion is trained only on sparse skeletal observations, it supports full-skeleton motion generation during inference. Experimental results demonstrate that ETA-Motion outperforms existing topology-agnostic baselines while significantly improving efficiency and generalization to unseen motions and unseen topologies.
[CV-181] MEV: A Multi-Event Video Dataset for Long-Take Generation NEURIPS2026
链接: https://arxiv.org/abs/2610.38241
作者: Peiyuan Zhu,Shaoan Xie,Yifan Shen,Wen Tian,Zijian Li,Guangyi Chen,Kun Zhang
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:Current datasets for text-to-video generation are largely limited to short clips depicting a single event or coarse-grained captions. As a result, they provide limited supervision for modeling long-take temporal dependencies, often leading to temporally inconsistent generations when describing multiple events. In this paper, we introduce MEV, a multi-event video dataset designed for long-take video generation. MEV provides high-quality real-world videos where subjects can perform a sequence of distinct, non-overlapping events, featuring precise temporal boundaries and fine-grained procedural annotations. This event-centric discretization reduces confounding factors and enables models to support temporally localized conditioning and build long-take videos from atomic event units. Furthermore, we introduce a comprehensive annotation pipeline that segments videos into temporally grounded events, and generates structured captions at both the global level and the event level. In addition, under an event-centric evaluation protocol, we conduct multi-event generation experiments by fine-tuning on MEV with temporally grounded event conditioning, leading to improved long-take event transitions and more coherent content. We believe this work provides a publicly available resource for multi-event fine-tuning and event-centric evaluation, together with practical insights into learning from temporally structured video-text data.
[CV-182] Silhouette-Based Gaitomics for Frailty Assessment in Older Adults
链接: https://arxiv.org/abs/2610.38223
作者: Laura McDaniel,Ryan Roemmich,Crystal Szczesny,Peter Abadir,Rama Chellappa,Basudha Pal
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Walking offers a window into physical function that may support interpretable video-based frailty assessment. We developed silhouette-based “gaitomics” to connect frailty predictions with measurable patterns of body configuration and movement. In 66 older adults assessed using the Fried Frailty Phenotype, standardized silhouettes and gait energy images yielded 90 descriptors of shape, temporal variation, motion, and texture. Minimum redundancy maximum relevance selection using the mutual information difference criterion prioritized informative features while limiting redundancy. Each classifier was evaluated using nested stratified cross-validation with five outer and three inner folds at the participant level, with feature ranking and feature-count selection confined to training partitions. Shapley additive explanations examined contributions to held-out predictions. Bilateral silhouette difference led the aggregate feature-selection rankings for three-class classification and both comparisons involving frail participants. Attribution profiles differed across tasks, with texture and consecutive-frame silhouette differences prominent in three-class classification. Among classifier-specific evaluations, CatBoost achieved the highest mean three-class micro-AUC of 0.647 \pm 0.085. Leading pairwise mean micro-AUCs were 0.833 \pm 0.249 for frail versus nonfrail, 0.765 \pm 0.168 for frail versus prefrail, and 0.554 \pm 0.129 for prefrail versus nonfrail classification. These findings connect model predictions to explicit gait characteristics while revealing task-dependent feature priorities and limited separation of prefrail and nonfrail participants. The framework provides an interpretable basis for further investigation, with larger independent studies needed to establish reliability and clinical relevance.
[CV-183] Complementing Text Based Moderation with Multimodal Grounding: Detecting Fraudulent Inducement Behaviors on Short Video Platforms
链接: https://arxiv.org/abs/2610.38185
作者: Xiaoyong Yu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Short video platforms have become a primary channel for black and gray industry operations in internet finance, where inducements such as prompts to like, follow, join private chat groups, or click external links frequently steer users toward fraud. Because such videos are short, dynamic, and often convey intent through fleeting visual cues rather than stable text, a moderation system must guarantee detection recall and localize the offending segment so reviewers inspect only the relevant clip. Existing OCR+LLM pipelines transcribe on-screen text frame by frame and classify the transcript, but exploit only textual information. Many inducements are conveyed through visual cues OCR cannot observe, and keyframe-based localization cannot recover precise temporal boundaries. We explore OCR+MLLM, which augments the text pipeline with a Multimodal Large Language Model, and compare three routes: OCR+LLM, OCR+MLLM, and a large zero-shot MLLM, along detection, latency, token consumption, and deployment complexity. OCR preserves the text signal that guarantees recall, while the MLLM jointly performs OCR-grounded discrimination and temporal grounding, localizing both textual and non-textual inducements. We construct a dataset of 2,419 internet finance short videos (66.49 hours) with 3,194 annotated events across four categories, and study reward design for RL-based grounding under class imbalance. We report two findings: auxiliary reward terms are exploited through reward hacking, and stronger localization rewards degrade class-balanced detection. We therefore design a recall-weighted F2 localization reward with class-balanced resampling. Experiments show OCR+MLLM localizes better than the keyframe OCR+LLM baseline (mIoU 26.52 vs. 14.49) and detects non-textual inducements that text-only pipelines miss, while maintaining 67.0 percent macro recall.
[CV-184] Data Amount Heterogeneity Improves Performance in Federated Learning for Neuroimaging
链接: https://arxiv.org/abs/2610.38172
作者: Paul Kurian,Tamoghna Chattopadhyay,Abhijith Shaji,Greg Ver Steeg,Paul M. Thompson,Jose Luis Ambite
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV); Neurons and Cognition (q-bio.NC)
备注: SIPAIM 2026
Abstract:Federated Learning (FL) allows for secure analysis of distributed data without data sharing. We and others have successfully applied FL to neuroimaging tasks, such as Alzheimer’s disease classification from structural magnetic resonance imaging (MRI). FL trains a deep learning model locally at each site, encrypts the local model parameters and sends them to a federation controller that aggregates the local models under homomorphic encryption, the resulting community model is sent back to the sites for further training, and the process repeats for a number of rounds. FL often achieves learning performance (e.g., accuracy) close to centralized training, but no data leaves the sites and model parameters are protected through encryption and other measures. In this paper, we show that federations composed of sites with heterogeneous amounts of data (1 large site plus small sites) perform better than homogeneous (same data per site) federations, particularly in low data regimes, while reaching performance close to centralized training. Having a big site often results in faster convergence and higher final accuracy. Heterogeneous federations better exploit sites with small amounts of data, a common situation in large consortia such as ENIGMA. Comments: SIPAIM 2026 Subjects: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV); Neurons and Cognition (q-bio.NC) Cite as: arXiv:2610.38172 [cs.LG] (or arXiv:2610.38172v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.38172 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[CV-185] Neighbourhood shortcuts in building damage assessment from very-high-resolution satellite imagery: diagnosis and a building-evidence plug-in
链接: https://arxiv.org/abs/2610.38162
作者: JianChong Guo,YuTing Wan,Kefan Li,AiLong Ma,YanFei Zhong
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 22 pages, 9 figures, 11 tables. Code: this https URL
Abstract:Building damage assessment (BDA) from very-high-resolution imagery supports disaster response, and deep models now achieve high overall accuracy on public benchmarks. Because damage is spatially clustered, most buildings share the damage level of their neighbours, so overall scores are dominated by such buildings, whereas damaged buildings among intact ones are rare and easily overlooked. A high overall score therefore cannot tell whether a model judges the building itself or relies on its neighbourhood as a shortcut, and this study tests which of the two current models do. We stratify buildings by their agreement with their neighbours, move buildings into new neighbourhoods with their pixels unchanged, and diagnose nine models covering the main paradigms of BDA. All models miss a third to half of the isolated damaged buildings, and in non-flood events they call 73–87% of intact buildings damaged once these are moved into a damaged neighbourhood. We propose a building-evidence plug-in that adds two experts to an existing model without retraining it. One expert sees only the building, and the other sees its surroundings but is trained with seamlessly swapped neighbourhoods so that it learns not to trust them. With each of the nine models, the plug-in maintains or improves overall accuracy, reduces errors on buildings that disagree with their neighbours by 8–18%, and cuts by 40–69% the intact buildings that a damaged neighbourhood turns into damaged ones. It also raises the detection of synthetic isolated damage from 1–6% to 13–39% and carries over to optical–SAR data and to an unseen earthquake. High overall scores thus conceal a systematic reliance on the neighbourhood, which building-level evidence reduces at low cost.
[CV-186] REEL: Same Frames Different Answers and Different Costs WACV
链接: https://arxiv.org/abs/2610.38158
作者: Xin Liang,Jiangpeng He
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) 2027, Datasets and Evaluation Track
Abstract:Video understanding keeps moving toward longer recordings, and the longer a recording is, the more questions can be asked about it. But processing it again for every question does not scale, so many systems watch the video once, build a memory, and answer later questions from it. Yet what to remember is decided before anyone knows what will be asked, and whatever is left out cannot be recovered later. Moreover, whether the savings pay off is judged by token counts, which fall when the input gets shorter but not when the model reuses what it has already computed. We present REEL, an evaluation framework that compares three saved representations built from exactly the same observed frames, so that a change in answers can come only from what the system remembered. Text writes a short record of what happened, Visual keeps the encoder outputs, and Prefix State also keeps the language-model prefix. Each is compared with full reprocessing (Frames) by paired correctness changes, cumulative GPU time, and storage size. Across six vision-language models and two benchmarks, Text and Frames differ in correctness on 20% to 29% of questions, even when their mean accuracy is almost the same. By contrast, Visual and Prefix State stay much closer to Frames. With unlimited storage, Prefix State is the cheapest in total GPU time among memories that keep almost all of the answers Frames gets right, for every model on both benchmarks. The trade-off is storage: Prefix State keeps up to about 21 GiB per video, whereas Text needs only kilobytes. Reusing computation and discarding information both look like efficiency, but only the second removes evidence that later questions may need, and neither mean accuracy nor token count tells them apart. REEL makes that difference measurable for any system that answers many questions from one saved memory. Comments: Accepted at the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) 2027, Datasets and Evaluation Track Subjects: Computer Vision and Pattern Recognition (cs.CV) Cite as: arXiv:2610.38158 [cs.CV] (or arXiv:2610.38158v1 [cs.CV] for this version) https://doi.org/10.48550/arXiv.2610.38158 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[CV-187] StepControl: Long-Horizon Human Motion Generation from Foot-Contact Schedules WACV2027
链接: https://arxiv.org/abs/2610.38153
作者: Hojun Jang,Jinseok Bae,Donggeun Lim,Taesun Kim,Seungbin Jeong,Younghwan Lee,Yong Seok Kim,Young Min Kim
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: WACV 2027, 8 main pages
Abstract:Controllable human motion generation has largely focused on specifying what action should occur or where the body should move. We study a complementary question: can full-body motion be controlled by specifying only when each foot contacts the ground? We introduce StepControl, an autoregressive diffusion model conditioned on a future binary left/right foot-contact schedule, without requiring target root and body poses, or semantic prompts. StepControl represents contact as explicit temporal events and injects them through dedicated cross-attention with classifier-free guidance. We further introduce a stochastic mixture noise schedule and warmup strategy for stable long-horizon autoregressive generation. Experiments show that StepControl follows prescribed contact timing more accurately than adapting existing spatially guided motion generators to contact constraints. It maintains effective contact control over long continuous rollouts and responds to edited schedules such as cadence changes, stop-and-go pauses, limping gaits, and bilateral jumping. Our results show that minimal foot-contact schedules provide an effective and intuitive temporal interface for controllable motion generation.
[CV-188] What Does LOL Measure? Hidden Label Shifts in Low-Light Enhancement Benchmarks
链接: https://arxiv.org/abs/2610.38121
作者: Adityan Thirumani Srinivasan
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Image and Video Processing (eess.IV)
备注: 11 pages, 5 figures, 8 tables
Abstract:LOL-v1 and LOLv2-Real are the standard benchmarks for supervised low-light image enhancement, and several leading methods ship hand-set test-time adjustments with their evaluation scripts. We audit both benchmarks with 18 public checkpoints of eight methods under one metric implementation that reproduces every published number to within 0.09 dB, and find that their test labels differ from their training labels. (i) The 15 LOL-v1 test labels are rendered differently: the same captures appear in LOLv2 with labels of identical luminance but 1.48x lower saturation, and LOL-v1’s own training labels match the LOLv2 rendering; we confirm this on the official release. (ii) The LOLv2-Real test set is the contiguous block of images 690-789, whose labels are 14% darker than the training labels (Cliff’s delta = -0.45, p 1e-12) with no comparable shift in the input statistics we examined; methods trained without any labels score up to 5.5 dB higher on test than on train. (iii) All 15 LOL-v1 test inputs reappear in LOLv2-Real, 12 in its training set. The released test-time settings are consistent with compensating for these shifts. Separating what can be calibrated on training data from what only test-set tuning finds, the test-only residual has median 0.00 dB on LOLv2-Syn, which has no shift, but 0.37 dB on LOL-v1 and 0.92 dB on LOLv2-Real (up to 3.9 dB). Under the shift, test-set checkpoint selection inflates a from-scratch model by 0.9-2.0 dB, and removing the released settings changes which method ranks first. We give concrete reporting recommendations.
[CV-189] Who Called? V33DA: A Physically Verified Multimodal Benchmark for Vocal Attribution in Zebra Finch Groups NEURIPS2026
链接: https://arxiv.org/abs/2610.38109
作者: Maris Basha,Yuhang Wang,Xiaoran Chen,Longbiao Cheng,Luca Yapura,Anja T. Zai,Mathieu Salzmann,Richard Hahnloser
类目: ound (cs.SD); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Quantitative Methods (q-bio.QM)
备注: Accepted at NeurIPS 2026 (Evaluations and Datasets Track). 43 pages, 3 figures. Dataset: this https URL ; Code: this https URL
Abstract:Deciding which member of an animal group produced a vocalization and where that vocalization was produced are central to studying social communication but are rarely evaluated with physically verified ground truth. Existing resources either localize sounds without identifying the caller, or recognize individuals from isolated vocalizations without reliance on the 3D candidate geometry. We close this gap with V33DA, a dataset-centered multimodal benchmark for spatial vocal attribution in social zebra finches: it comprises 33,625 verified events coupling 5-channel audio with three camera views, calibrated 3D pose for every visible candidate, and per-bird radio telemetry across 10 individuals and three experiments. Caller identity is verified from an on-body accelerometer-derived vibration signal; this channel is withheld from benchmark models and used only by an on-body reference ceiling. Given a detected vocalization and the visible candidate birds, the task is to select the caller. We evaluate it under three regimes: session-disjoint testing on familiar birds from held-out recording dates, held-out-experiment transfer to a new cohort, and leave-one-individual-out evaluation on a bird whose vocalizations are held out from training. Reference methods expose a consistent gap between in-domain accuracy and transfer to unseen individuals: methods that can exploit familiar vocal identity perform well on known callers but degrade sharply on unseen ones, whereas candidate-aware methods preserving explicit spatial reasoning transfer substantially better. The on-body reference stays near-perfect across regimes, confirming the internal consistency of the accelerometer-verified labels. We release V33DA with fixed protocols and code for the evaluated methods, and V33DA++, an unevaluated companion set of verified overlapping-call events and +/-2 s context windows.
[CV-190] he Prestige: Benchmarking Cognitive Visual Reasoning using Magic Tricks NEURIPS2026
链接: https://arxiv.org/abs/2610.38103
作者: Shuo Wen,Beixi Du,Edwin Meriaux,Junming Shi,Chloe Si,David Meger,Doina Precup,Gregory Dudek
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: 40th Conference on Neural Information Processing Systems (NeurIPS 2026). Track on Evaluations and Datasets
Abstract:We introduce The Prestige, a video understanding benchmark that probes cognitive visual reasoning through professional magic performances, where deliberate misdirection challenges models to track events and explain unexpected outcomes. The benchmark comprises open-ended questions ranging from temporal and causal understanding to robustness against adversarial questions. Evaluation of leading open-source and proprietary VLMs reveals a substantial gap between even the strongest model and human performance. We also find systematic discrepancies between human and automated grading and identify recurring model failure modes. Ablations further show that models can select correct multiple-choice answers even when they fail to generate correct explanations independently, motivating further investigations into question formats. Together, these findings highlight limitations in both VLM reasoning and its evaluation, underscoring the need for more reliable video understanding models and more rigorous benchmarks.
[CV-191] From Next Day Wildfire to Next Days Wildfire: A New Multi-Continent Dataset of Active-Fire Spread for Sequence-to-Sequence Forecasting WACV2027
链接: https://arxiv.org/abs/2610.38086
作者: Jose Guillen,Renzo Zanca,Valentin Barriere
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at WACV 2027. 19 pages including supplementary material, 12 figures, 13 tables
Abstract:Benchmarks such as WildfireSpreadTS (WSTS) have enabled data-driven forecasting of wildfire spread from multimodal satellite time series, but they suffer from limitations: size restricts them to large fires that burn for weeks, and they are from similar environments. We introduce AFSeq, a multimodal dataset of 1,194 recent fire events (2023 to 2026) from 8 different countries across three regions, dominated by the short, small fires that existing benchmarks underrepresent. Our dataset is the largest and most diverse to date for time-series-based wildfire spread prediction. We also introduce UTAE-AR, an extension of a classical U-Net with Temporal Auto Encoder, enhanced with recurrent modules before the decoder. This allows the model to not only predict the next day but the whole sequence in an autoregressive way. Moreover, our model can feed its own predicted fire mask back one day at a time, producing the day-by-day forecasts that single-step models cannot. Our results show that UTAE-AR surpasses state-of-the-art on both WSTS, WSTS+ and AFSeq for next-day accuracy while having 21.5% of their parameters, and being able to predict confidently at a longer horizon.
[CV-192] Efficacy of Monocular Depth Estimation in Lunar Environments
链接: https://arxiv.org/abs/2610.38080
作者: Michael W. Lanighan,Kenji Brameld,Seth Gee
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at the International Conference on Space Robotics (iSpaRo) 2026
Abstract:Recent advances in monocular depth estimation have demonstrated highly accurate 3D reconstructions within in-distribution terrestrial domains. This study investigates the efficacy of several state-of-the-art approaches to reconstruct out-of-distribution lunar environments using the NASA Ames Polar Optical Lunar Analog Reconstruction (POLAR) stereo dataset and Apollo 17 surface imagery. By leveraging these datasets, we establish a robust analog-to-field validation pipeline that evaluates quantitative performance against ground-truth physical analogs and qualitative performance against historical field data. Our results indicate that select state-of-the-art techniques can successfully generalize to produce high-fidelity relative 3D reconstructions of the unstructured lunar surface.
[CV-193] Vision Tomography: Towards Portable 3D Bidirectional Face-Skull Inference from Visual Data
链接: https://arxiv.org/abs/2610.38065
作者: Zidu Wang,Meng Xu,Miao Xu,Hengyuan Ma,Jiankuo Zhao,Xutao Li,Zhen Lei,Xiangyu Zhu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Submitted to TIP, under review
Abstract:Traditional computed tomography (CT) can capture paired 3D face and skull geometry, but it remains costly, fixed in location, and exposes patients to radiation. For applications such as remote diagnostics, digital healthcare, forensic identification, missing-person investigation, and archaeological analysis, there is an urgent need for a portable, safe, and accessible solution to capture 3D skull geometry. In this work, we introduce Vision Tomography (VT), a concept and framework capable of reconstructing paired 3D face-skull geometry from various visual modalities, and also enabling joint manipulation of 3D facial and skull geometry. To achieve this, we first construct a dataset of over 200 samples, where each case includes a CT-based skull, a CT-based face, and a high-fidelity textured face scan. Secondly, we propose a novel dense ray matching registration method that ensures topological consistency across face, skull, and their tissue correspondences. Building on this registration, we introduce the 3D Bidirectional Face-Skull Morphable Model (BFSM) as the foundation of Vision Tomography, which enables shape inference between the face and skull through a shared coefficient space, while also modeling tissue thickness variation to support one-to-many facial reconstructions from the same skull, reflecting individual facial changes such as age or body fat. Finally, we demonstrate the effectiveness of Vision Tomography across various forensic and medical applications, including paired 3D face-skull reconstruction from images or sketches and predicting surgical outcomes with joint face-skull editing. Extensive experiments confirm the robustness and accuracy of our method. Vision Tomography is available at this https URL
[CV-194] Classifying Urban Locations by Cyclist Facility Usage and Directional Alignment: A Validated Camera-Based Framework from Multi-Site Observations
链接: https://arxiv.org/abs/2610.38045
作者: Ao He,Mohit Gupta,Deborah Salon,Ricardo Eiris,Vivian W. H. Wong
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 19 pages, 6 figures, 2 tables. Accepted for presentation at the 106th Transportation Research Board (TRB) Annual Meeting, Washington, D.C., January 10-14, 2027
Abstract:Objectives: Agencies invest in bicycle facilities but lack tools to assess whether cyclists use them as designed. We examine whether usage patterns follow facility configurations and whether higher bike lane usage is associated with less wrong-way riding. Methods: Motion-triggered cameras covered 19 deployments at 17 locations in Tempe, Arizona, in 2023. Manual review identified 1,963 cycling events; excluding 20 crossings left 1,943 labeled events for analysis. An automated pipeline on the same captures counted 92.9 percent of the full manual total (per-deployment r = 0.99), and every automated wrong-way call was visually verified. Deployments were classified using thresholds on facility usage shares. Findings: Deployments were bike-lane dominant (9), sidewalk dominant (6), or mixed (4). Cyclists in bike lanes rode against the flow in 4.6 percent of events; on sidewalks, riders opposed the near-side traffic flow in 49.6 percent. Across deployments, bike lane share and wrong-way rate were negatively correlated (Spearman rho = -0.87, p 0.001, n = 16). An independent volunteer count in the same city showed a similar sidewalk-direction pattern (rho = 0.79). Novelty: The study combines multi-site still-camera observations with a fully hand-verified benchmark to characterize facility use and travel direction. Practical Applications: The workflow offers planners a low-cost screen for locations with high sidewalk cycling and travel against the near-side traffic flow, supporting more detailed site investigation. Comments: 19 pages, 6 figures, 2 tables. Accepted for presentation at the 106th Transportation Research Board (TRB) Annual Meeting, Washington, D.C., January 10-14, 2027 Subjects: Computer Vision and Pattern Recognition (cs.CV) Cite as: arXiv:2610.38045 [cs.CV] (or arXiv:2610.38045v1 [cs.CV] for this version) https://doi.org/10.48550/arXiv.2610.38045 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Ao He [view email] [v1] Tue, 27 Oct 2026 22:59:17 UTC (8,768 KB) Full-text links: Access Paper: View a PDF of the paper titled Classifying Urban Locations by Cyclist Facility Usage and Directional Alignment: A Validated Camera-Based Framework from Multi-Site Observations, by Ao He and 4 other authorsView PDFHTML (experimental)TeX Source view license Additional Features Audio Summary Current browse context: cs.CV prev | next new | recent | 2026-10 Change to browse by: cs 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)
[CV-195] Can Event Detectors Leave the Broadcast Camera? Evidence from Tennis
链接: https://arxiv.org/abs/2610.38042
作者: David Soeiro-Vuong,Andre-Louis Rochet
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: 16 pages, 4 figures, 5 tables. Benchmark, evaluator and frozen predictions: this https URL
Abstract:Ball bounces and racket hits are the events from which tennis statistics are derived, and on monocular video the two are easy to confuse. We release a benchmark of 345 bounce and hit annotations at frame level on eight clips that mix broadcast, fixed elevated, phone and court-level viewpoints, with a frozen development and test split, a confusion-aware matcher, clip-level bootstrap intervals and a blind second annotation of three clips. On it we score supervised precise spotters with published broadcast weights, two zero-shot vision-language models, and RallyArb, an arbitration layer over a ball track and player poses that fits no parameter on event labels. The supervised spotters score near-perfectly on broadcast footage and almost nothing on phone and court-level footage, where the pooled bounce F1 of E2E-Spot is 0.130. RallyArb keeps a bounce F1 above 0.8 on the three phone and court-level clips, two of which were used to tune it, and finds 8 of 11 bounces on the held-out one, but it places hits too far from the contact frame to match human timing. The vision-language models detect events on every viewpoint and miss the contact frame by about 0.1 s. Fine-tuned leave-one-clip-out from its broadcast weights, the spotter reaches about 0.8 bounce F1 on two phone and court-level clips and stays at 0.2 on a grazing phone angle. The annotations, the evaluator and the frozen predictions of every scored row are released.
[CV-196] Where Can We Fool the Models Mind? Response-Guided Semantic Probing and Adaptive Routing for Transferable Attacks on Black-Box MLLM s
链接: https://arxiv.org/abs/2610.38013
作者: Ahmed Abdelnaby,Mohamed Elmahallawy
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Multimodal Large Language Models (MLLMs) ground language in visual evidence, yet their semantic reasoning can be exploited through transferable black-box adversarial attacks. Existing attacks typically allocate perturbations using surrogate-model saliency, which may not reflect how an unseen MLLM semantically interprets or relies on different image regions. We introduce VISTA (Vulnerability-Informed Semantic Transfer Attacks), a black-box framework that identifies semantically vulnerable visual regions from an MLLM’s responses without access to gradients, logits, parameters, or internal representations. VISTA supports two attack settings: (i) VISTA-Proxy, a pure-transfer setting that probes an open-source MLLM and transfers the resulting adversarial examples to unseen commercial MLLMs without querying them; and (ii) VISTA-Victim, a query-assisted setting that probes the target MLLM to capture model-specific semantic vulnerabilities. VISTA builds an Uncertainty-Based Multimodal Behavior (UBMB) map from response drift, conflict, misleading, and semantic shifts, then uses entropy-adaptive Sinkhorn routing and a surrogate ensemble to focus perturbations on vulnerable regions. Across GPT-5, Gemini-3.5-Flash, and Claude Sonnet 4.6, VISTA achieves up to 97 % attack success rate (ASR) while maintaining a Peak Signal-to-Noise Ratio (PSNR) above 30 dB. Notably, VISTA-Proxy achieves up to 93 % ASR without querying the final victim, demonstrating that response-derived semantic vulnerability can enable effective transfer across heterogeneous MLLMs.
[CV-197] Frequency-Adaptive Gaussian Grids for Efficient Image Representation
链接: https://arxiv.org/abs/2610.38007
作者: Anatolii Evdokimov,Stavros Diolatzis,Bingxuan Li,Akshay Jindal,Patsorn Sangkloy,Qi Sun
类目: Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR)
备注:
Abstract:Neural representation has emerged as a powerful tool to mitigate the demand for increasingly higher-resolution images. However, a fundamental trade-off remains between structured (e.g., feature grids, which offer spatial consistency) and unstructured (e.g., 2D Gaussians, which adaptively capture local variances) representations. This choice significantly affects how well an algorithm represents an image. Here, we present Gaussian-Grid, a hybrid neural representation that adaptively integrates structured and unstructured primitives. By dynamically partitioning computation between 2D Gaussian distributions and their attached multi-resolution feature grids based on the local frequency content, Gaussian-Grid automatically optimizes its architecture to match the underlying complexity of the target signal. We evaluate our approach using high-resolution natural images and demonstrate its applications as GPU-native dataset storage, showing 1.24 - 2.14\times decoding speed improvement and PCIe data traffic reduction of 23\times - 176\times over JPEG + ZIP data archiving. The results show that Gaussian-Grid maintains small memory footprints while achieving high reconstruction fidelity across a variety of datasets.
[CV-198] Hierarchical Diffusion for Camera-controlled Autoregressive Video Generation
链接: https://arxiv.org/abs/2610.37983
作者: Kaihua Chen,Yiming Gong,Deva Ramanan,Tarasha Khurana
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project page: this https URL
Abstract:We address the task of long-video, camera-controlled generation from a single image. The dominant autoregressive paradigm suffers from two key limitations: (1) error accumulation over time, as distant frames are conditioned on previously generated outputs rather than ground truth, leading to drift; and (2) ambiguity in how to incorporate camera control. Standard solutions to both–such as self-forcing and conditioning on explicit 6-DOF poses–either incur significant computational overhead or produce drifting scene geometries over time. We propose a simple and lightweight framework that combines autoregressive generation (WanAR) with hierarchical long-context denoising (FixAR). To enforce scene geometry remains consistent with input cameras, we draw from feed-forward novel-view synthesis and reformulate video generation as a warping-and-inpainting problem. To mitigate drift, we introduce a hierarchical denoising strategy in which different stages condition on different temporal histories. Specifically, we first generate frames conditioned on the immediate past, and then refine them using a second denoiser conditioned on a temporally subsampled long-range history to enforce global consistency. This fine-to-coarse scheme enables each generated frame to effectively condition on the full temporal context without incurring a prohibitive cost. Experiments on RealEstate10k and DL3DV demonstrate that WanAR+FixAR outperforms prior methods for camera-controlled long-video generation, producing results that are both temporally stable and consistent with input camera trajectories.
[CV-199] Meta-Initialization of Implicit Neural Representations to Accelerate Deformable Medical Image Registration DATE
链接: https://arxiv.org/abs/2610.37965
作者: Marat O. Zinovev,Dmitry V. Sorokin
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 14 pages, 5 figures, 2 tables. English-language author version of a paper presented at GraphiCon 2026, Saint Petersburg, Russia, 22-24 September 2026. Russian-language proceedings forthcoming. Related work updated for this version
Abstract:Implicit neural representation (INR) methods for deformable medical image registration require the network parameters to be optimized for each new image pair, making adaptation speed a practical limitation. This work investigates whether early adaptation can be accelerated using an initialization learned from a collection of previously solved registration tasks. We examine two schemes: Reptile meta-initialization and STRAINER transfer of the shared component of a coordinate network. KAN-IDIR is used as the base INR displacement-field model. Experiments are conducted on OASIS-1, with 364 consecutive pairs used for training and 49 pairs for test-time adaptation. Registration quality is evaluated using mean Dice and HD95, together with the fraction of voxels with a negative Jacobian determinant at iteration 1500. In the configurations studied, both schemes yield higher Dice and lower HD95 at early iterations than random initialization. Reptile numerically outperforms STRAINER under a small iteration budget, although their pretraining budgets are not matched. As the number of iterations increases, the differences between the schemes shrink. STRAINER pretraining takes 39-40 minutes, compared with 44.98 hours for Reptile.
[CV-200] You Dont Need Aligned Representations: Knowledge Distillation via Random Prototype Spaces NEURIPS2026
链接: https://arxiv.org/abs/2610.37937
作者: Hamza Etcibasi,Ramazan Gokberk Cinbis
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: To appear at NeurIPS 2026. 10 pages main text plus appendices. Code: this https URL
Abstract:A widely held assumption in knowledge distillation is that teacher and student representations must occupy a compatible or explicitly aligned feature space. We propose RPKD (Random Prototype Knowledge Distillation), which discards this assumption by comparing both networks against a shared bank of randomly initialized, frozen prototype vectors. Alignment is therefore carried out in an external reference space rather than in the teacher’s native feature coordinates: the teacher side is projected by fixed random maps that carry no trainable parameters, and only the student adapts. Two branches handle the transfer: a logit branch aligning global prototype similarity distributions and a feature branch matching spatially-resolved prototype activation maps. Both operate within the same fixed prototype space, which can be interpreted as a random feature embedding that approximately preserves inter-sample similarity structure, so the prototype bank and the distillation objective stay unchanged across teacher-student pairs while only the input projections are instantiated per architecture. In low-category settings, decoupling the prototype vocabulary from the task label space yields richer supervision than the label space alone, an advantage absent from logit-based methods, whose supervisory signal collapses as class count shrinks. This advantage is especially relevant in real-world scenarios where the label space is typically constrained. Extensive experiments demonstrate the effectiveness of RPKD, achieving a maximum gain of +7.98% over OFA on CIFAR-100, +12.44% on ImageNet-100, and +1.28% on ImageNet-1K. Source code is available at this https URL
[CV-201] Venus-DefakerOne2: A Vision-Forensics Foundation Model with Broader Coverag e
链接: https://arxiv.org/abs/2610.37918
作者: Chenfan Qu,Yachao Liang,Mingqi Fang,Jiangwei Xie,Zhenming Wang,Baoyue Zhang,Changjiang Jiang,Zheng Chang,Fanli Jin,Xuekang Zhu,Kaiwen Feng,Junchi Li,Weihong Zhang,Haoyue Wang,Chenfeng Zhang,Yinan Xia,Fangming Cui,Dong Fan,Jingjing Liu,Tianyi Zhang,Luobin Chen,Ji-Zhe Zhou,Sheng Li,Wei Wang,Bin Li,Ran He,Xinpeng Zhang,Linna Zhou,Jian Liu,Weiqiang Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Technical Report
Abstract:Real-world forgery is inherently heterogeneous: it manifests not only as digital manipulation but also as physical spoofing, and is carried not only by images but also by videos. Despite substantial progress, a unified anti-forgery foundation model covering such a broad threat spectrum remains absent. In this paper, we present Venus-DefakerOne2, which substantially extends the forensic capability boundary along both the forgery-form and media-type dimensions. Trained on tens of millions of open-source and in-house samples, it incorporates two key designs. Tide-Style-Tuning alternates LoRA and full-parameter fine-tuning to balance computational efficiency, generalization, and hard-sample fitting while mitigating catastrophic forgetting. Entropy-based Test Adaptation determines the optimal prediction modality by analyzing prediction entropy. Venus-DefakerOne2 achieves new state-of-the-art performance across a comprehensive suite of benchmarks, including DeepFake faces (96.7%), AI-generated images (93.3%), tampered documents (87.6%), manipulated natural images (89.6%), physical spoofs (93.1%), and AI-generated videos (96.0%). It significantly outperforms both specialized expert models and its predecessor, DefakerOne, establishing itself as a powerful and versatile foundation model for binary-decision forensic tasks.
[CV-202] What Actually Matters? A Paired Evaluation of Inference Settings for π_0.5 on BEHAVIOR-1K
链接: https://arxiv.org/abs/2610.37917
作者: Ruben Hammele,Samira Huber,Mohammad Nazeri,Amirreza Payandeh,Alexandyr T. Card,Aaron Nguyen,Anuj Pokhrel,Daeun Song,Xuesu Xiao,Sören Pirk
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Vision-Language-Action (VLA) models offer a promising paradigm for general-purpose household robotics. However, inference and data-collection strategies that have been deployed in smaller-scale settings do not necessarily generalize to large-scale, diverse task suites. To address this gap, we conduct a systematic, paired evaluation of previously proposed inference and data-collection strategies across all 100 tasks. Our evaluation uses a fine-tuned \pi_0.5 model based on the 2025 BEHAVIOR Challenge winner, which achieves a Q-score of 0.39 on the public test set. These results show that proposed strategies do not transfer uniformly to new environments and tasks. Project page, videos, and detailed analysis: this https URL
[CV-203] FYC: FPR-Disparity Weight Perturbation and Youden Calibration for Fair Deepfake Detection
链接: https://arxiv.org/abs/2610.37882
作者: Uzoamaka Ezeakunne,Taiwo Olabamipe,Deevesh Jain,Xiuwen Liu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 9 pages, 4 figures
Abstract:Deepfake detectors routinely achieve high aggregate accuracy while exhibiting substantial demographic disparities in their errors: false-positive rates differ by double digits across gender and skin tone. Most existing remedies require either full retraining with a fairness-aware objective or demographic-annotated training data, both impractical when only a pretrained checkpoint is available, while label-free post-hoc calibrators leave the detector’s weights unchanged. We introduce FYC (FPR-disparity Weight Perturbation and Youden Calibration), a two-stage post-hoc method that perturbs a detector’s weights under an FPR-disparity criterion, then calibrates a separate decision threshold for each demographic group - requiring no architectural changes and no full retraining. Evaluated against every detector benchmarked on AI-Face v2 across six intersectional gender x skin-tone groups, FYC outperforms the field on 18 of 21 reported fairness and utility metrics, including purpose-built fairness architectures, at a fraction of the training cost. We further identify why the choice of fairness criterion is critical: the original Demographic Parity objective used in prior weight-perturbation work is architecture-dependent and unpredictable, improving two of four evaluated architectures, leaving a third largely unchanged, and substantially degrading the fourth (PG-FDD), with no way to anticipate the outcome for a given checkpoint in advance. Conditioning the same objective on real images only - our FPR-disparity criterion - eliminates this unpredictability, improving fairness and accuracy jointly across all four architectures with no exceptions.
[CV-204] Intersectional Invariant Learning for Fair Deepfake Detection
链接: https://arxiv.org/abs/2610.37881
作者: Uzoamaka Ezeakunne,Taiwo Olabamipe,Deevesh Jain,Xiuwen Liu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 4 pages including appendices, 7 figures, 8 tables
Abstract:Deepfake detection models achieve high overall accuracy but exhibit large performance disparities across demographic groups, causing some groups to experience substantially higher false-positive rates than others. Existing fairness-aware methods address this by constraining model outputs to equalize group predictions, but do not tackle the underlying feature-level bias. As a result, models may still rely on demographic characteristics, such as skin texture, as indirect cues for detecting manipulated content. To address this problem, we propose Intersectional Invariant Learning (IIL), a framework that improves feature learning rather than only adjusting model outputs. IIL alternates between identifying the most challenging demographic partitions and learning features that remain consistent across those partitions. The framework also includes a false-positive-rate fairness objective and group-balanced sampling across six gender x skin-tone intersectional groups. We evaluate IIL on a large-scale fairness benchmark containing over one million images, 10 baseline detectors, and 21 fairness and utility metrics. Results show that IIL achieves Pareto-optimal performance, simultaneously attaining an AUC above 99%, a false-positive rate under 2.5%, and the lowest intersectional fairness disparity (FMEO) among all compared methods. IIL outperforms prior fairness-aware methods on the large majority of the 21 evaluation metrics, while reducing both false-positive rates and intersectional fairness disparities by several-fold compared with the strongest competing method.
[CV-205] PairCLIP-SWA: Similarity-Guided Balancing and Stochastic Weight Averag ing for Generalized Deepfake Image Detection with OpenCLIP
链接: https://arxiv.org/abs/2610.37874
作者: Md. Nakib,Sudipto Sarkar,A. J. A. Jubayer Talukder,Mohammad Ariful Haque
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: 15 pages, 8 figures, 12 tables. Code and experiment outputs: this https URL
Abstract:Deepfake detectors are almost always deployed as the single checkpoint that scored best on a validation split, yet the reliability of that choice is rarely measured. We measure it with PairCLIP-SWA, a single-backbone detector that adapts an OpenCLIP ViT-L/14 visual tower by progressive block unfreezing on a class-balanced corpus built by CLIP-similarity-guided one-to-one subsampling, and averages the fine-tuning trajectory in weight space. Trained in-domain on the IEEE SP Cup 2025 (DF-Wild-Cup) benchmark under the same protocol as the strongest published three-backbone ensemble, the weight-averaged detector reaches 97.04% accuracy, 98.64% AU-ROC and a 4.07% equal error rate, against 94.46%, 97.60% and 5.00%. Its training log shows why averaging matters: over the averaged epochs the non-averaged checkpoint swings across 5.34 points of validation accuracy while its validation loss more than doubles, whereas the average is never worse and is up to 5.99 points better. On an OpenForensics-derived benchmark the same instability becomes a misranking: the checkpoint that validation accuracy selects (98.43%) scores 88.85% on a disjoint 10,905-image test partition, rejecting 21.1% of genuine photographs, while the terminal average, ranked last on validation, scores 94.76%. A study of twenty-two in-house configurations on the identical DF-Wild-Cup split, including preprocessing ablations, shows that CLIP pre-training rather than ensembling drives accuracy, that secondary face cropping costs about 34 points, and that frequency-domain inputs transfer poorly. We recommend per-class selection criteria measured off the early-stopping split.
[CV-206] Effects of Deep Learning Architecture and Pretraining on Automated Vocal Fold Segmentation in Ex Vivo High-Speed Phonation Images
链接: https://arxiv.org/abs/2610.37860
作者: Varun Vinayak Kalaiarasan(1),Charles Farbos de Luzan(1 and 2),Khalid Malik(2),Ephraim Gutmark(3),Liran Oren(1 and 2) ((1) Department of Biomedical Engineering, University of Cincinnati, (2) Department of Otolaryngology, University of Cincinnati, (3) Department of Aerospace Engineering Engineering Mechanics, University of Cincinnati)
类目: Computer Vision and Pattern Recognition (cs.CV); Image and Video Processing (eess.IV); Fluid Dynamics (physics.flu-dyn)
备注: 37 pages, 8 figures, 7 tables
Abstract:Two-dimensional (2D) intraglottal particle image velocimetry (PIV) of ex vivo phonation experiments quantifies vocal fold medial surface dynamics and the intraglottal flow field. However, it relies on manual segmentation because residual glare, low tissue-to-flow contrast, and tracer-particle speckle obscure the vocal fold boundaries. This study assessed how deep learning architecture and ImageNet pretraining affect automated vocal fold segmentation in these high-speed images. Four established architectures from three categories were benchmarked: the convolutional neural network (CNN)-based UNet and UNet++, the hybrid CNN-transformer-based TransUNet, and the pure transformer-based Swin-UNet, the latter two trained from scratch and from ImageNet-pretrained weights. Using 16,344 expert-annotated images from 227 phonation trials of excised canine larynges, the six pipelines were trained with five seeds under a standardized protocol and compared on 50 held-out test trials using trial-level Wilcoxon signed-rank tests with Holm correction. Mean Dice similarity coefficient (DSC) spanned only 0.9734 to 0.9764, so the pipelines differed mainly in boundary fidelity. Pretraining significantly improved the boundary F1 score (BF1) and 95th percentile Hausdorff distance (HD95) of the TransUNet and Swin-UNet, about four times more for the Swin-UNet. Once pretrained, the two performed similarly, although the Swin-UNet has only about a quarter of the trainable parameters. Without pretraining, the TransUNet had a significantly higher BF1 and a lower HD95 than both CNN-based pipelines and about half their share of images with HD95 50 pixels. To our knowledge, this is the first study to apply deep learning to vocal fold segmentation in 2D intraglottal PIV, and it offers a reference framework for segmenting experimental flow-visualization images with noisy optical conditions or highly deformable objects.
[CV-207] Convex Potential Learning for Optimal Transport Between Medical Images
链接: https://arxiv.org/abs/2610.37844
作者: Hongyu Kan,Ivan Medri,Natasha Ironside,Kristofor Pas,Richard Corp,Shinjini Kundu,Gustavo Kunde Rohde
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Computing optimal transport maps between medical images plays a fundamental role in morphological analysis. However, existing methods are either difficult to scale to large 3D medical images or rely on regularized formulations that provide only approximate transport maps. To improve the accuracy of downstream medical image analysis, we propose a novel approach for estimating optimal transport maps by parameterizing the convex potential function. By construction, the resulting transport map is guaranteed to be curl-free, consistent with Brenier’s theorem. We evaluate the proposed method on both 2D and 3D medical imaging datasets. Experimental results demonstrate that our approach achieves lower transport estimation errors and produces higher-quality image reconstructions than existing methods.
[CV-208] Learning Without Clocks: Partition-Invariant Neural Network
链接: https://arxiv.org/abs/2610.37823
作者: Han Wang,Jingyuan Bao,Juntao Wu,Saiao Zhou,Yuman Nie
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Event cameras, spiking cochleas and similar sensors emit asynchronous events, yet most learning systems first bin them into frames or packets, so the output depends on how the stream was sliced. Networks whose state is updated exactly at every event and evolves in closed form between events are invariant to this slicing. We ask what more makes the output (i) independent of the processing order of simultaneous events and (ii) continuous in the time stamps, even when two events cross in time, while (iii) staying sensitive to temporal order. For arbitrary nonlinear event models we prove that (i) and (ii) are each equivalent to pairwise commutation of the event updates, and that once updates commute, all order dependence comes from their non-commutation with the inter-event dynamics and vanishes linearly in the time lag. This identifies a class of time-mediated non-commutative ™ models and yields a resolution-robustness bound for any model. We construct the minimal TM layer, a learnable, provably Lipschitz generalization of the Hassenstein-Reichardt correlator; prove that no single additive layer can uniformly represent ordered pair interactions, whereas a three-dimensional TM state can; show that each TM layer adds two levels of temporal nesting versus one for additive layers; and give a single-layer higher-order construction (TM-k) and a stacking rule preserving all three properties. We instantiate these results as the Clock-Free Event (CFE) layer, linear in the number of events, give a topology-free variant for temporal point processes with closed-form likelihoods, and specify a preregistered clock-robustness protocol with a physical-time threat model for spike-retiming attacks. In preliminary development-set experiments, the TM state beats an otherwise identical additive state by 3.5-9.1 points on DVS128 Gesture and by 24.5 and 30.6 points on SHD and Speech Commands.
[CV-209] GOFish: Full Field-of-View Gaussian Splatting and Mesh Extraction
链接: https://arxiv.org/abs/2610.37818
作者: Allen Tu,Meher Gitika Karumuri,Shifan Zhu,Hailing Wang,Yen-Liang Lin,David Held,Matthias Zwicker,Tom Goldstein
类目: Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR); Robotics (cs.RO)
备注: Project Page: this https URL
Abstract:Robots, autonomous vehicles, and consumer 360° cameras use fisheye lenses for wide fields of view (FoVs). Simulation-ready digital twins require photorealistic rendering across the full FoV and an aligned collision mesh. However, prior fisheye Gaussian Splatting methods only rasterize up to 180° and do not produce meshes, while Gaussian surface reconstruction methods assume pinhole training images. We present GOFish, the first 3D Gaussian Splatting surface reconstruction method to train directly on fisheye images, rasterize FoVs beyond 180°, and extract an aligned mesh from the same scene. Calibrated densification recovers the surface across the entire FoV, while Calibrated Mip Filtering and Wide-Angle Appearance Modeling keep pixel blur and lens falloff out of the scene and mesh. Across the 200° FIORD and 205° FullCircle benchmarks, GOFish improves PSNR over the best prior method by 2.28 and 1.97 dB, respectively. It also outperforms fisheye rendering baselines within their own FoVs and achieves the highest mesh F-score on FIORD and the main ScanNet++ set. Together, the Gaussian scene and mesh form a LiDAR-free, simulation-ready digital twin that renders in real time, ray traces the full FoV, and provides collision geometry.
[CV-210] ACOS: Terrain-Aware COntact Supervision for Visual Footstep Planning ICRA2027
链接: https://arxiv.org/abs/2610.37806
作者: Jean-Michel Fortin,William Larrivée-Hardy,Gabriel Jeanson,Olivier Gamache,Daniele De Martini,François Pomerleau,Philippe Giguère
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注: 9 pages, 8 figures, Submitted to ICRA 2027
Abstract:Humans navigate their environment based on their vision system, anticipating how the ground will respond before committing to a step. Self-supervised learning could provide legged robots the same ability, but the supervisory signal is typically chosen on intuition, and its effect on what the model learns is rarely examined. We present Terrain-Aware COntact Supervision (TACOS), which proposes a meaningful and consistent footstep instability score, based on the residual of a factor graph in legged state estimation. It projects the continuous score into preceding images to supervise a dense visual predictor. We propose two variants, a hard score that exploits hindsight and a causal soft score available immediately. Both metrics are compared against state-of-the-art alternatives in two independent evaluations that share no data or labels. On 1716 manually annotated contacts, both variants outperform every evaluated alternative at detecting instability and ranking its severity, improving pooled severity ranking by 28 % over the strongest baseline. On three outdoor test runs, our metric also trains the most accurate visual predictor, improving rank correlation by 58 % over the strongest alternative, and by a further 25 % when supplemented with onboard context. In a real-world experiment over a 50 m forest trajectory, the trained predictor handles non-geometric terrain properties such as mud and vegetation, while stepping around geometric obstacles like fallen branches. Code is available for reproducibility at this https URL.
[CV-211] Sparse tensor networks for individual tree diameter estimation in mixed forests
链接: https://arxiv.org/abs/2610.37784
作者: Clara Lößl,Fabian Ewald Fassnacht,Viacheslav Barkov
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Measurements of tree diameter at breast height (DBH) are fundamental to forest inventories, but manual surveys across entire stands or regions are impractical. UAV-borne laser scanning (ULS) enables scalable automated forest surveys, yet canopy occlusion makes accurate DBH estimation from ULS point clouds non-trivial. Limited stem-level point density of ULS data constrains approaches relying on detailed stem reconstruction, such as Quantitative Structure Models (QSMs) and circle fitting. Empirical machine learning methods based on predefined point cloud metrics are an alternative. New deep learning approaches offer an even more principled solution, allowing models to learn directly from point clouds. Sparse tensor networks have shown particular promise, transforming forestry tasks such as tree species classification. We investigate whether these strengths can be extended to estimate individual tree DBH from ULS data across four mixed-species temperate forest plots in Germany. We further advance existing sparse tensor networks by introducing TreeNext, a lightweight deep learning architecture for direct individual tree parameter estimation from voxelized LiDAR point clouds. We present a fully automated forest inventory pipeline from individual tree instance segmentation to DBH regression, and quantify the impact of segmentation errors by comparing performance on automatically segmented and ground truth instances. TreeNext achieves an R^2 of 0.8355 on automatically segmented trees and 0.9380 on ground truth segments, outperforming strong tabular baselines including Random Forest and XGBoost. Segmentation quality emerges as the dominant error source in the automated pipeline. TreeNext proves highly data-efficient, performing well even in small data scenarios. Our findings establish sparse tensor networks as a practical foundation for automated individual tree DBH estimation from ULS.
[CV-212] Wavelet Phase Diffusion for Structurally and Semantically Consistent Sim-to-Real Translation WACV2027
链接: https://arxiv.org/abs/2607.21628
作者: Kaiwen Wang,Frank Bieder,Yinzhe Shen,Carlos Fernandez,Jan-Hendrik Pauls,Omer Sahin Tas
类目: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Image and Video Processing (eess.IV)
备注: Accepted by WACV 2027. 20 pages, 15 figures. Project page: this https URL
Abstract:Simulation-to-reality translation must bridge the appearance gap between synthetic and real domains while preserving structural and semantic consistency. Conditioning-based methods achieve spatial alignment but introduce computationally expensive control modules. Paired-data methods achieve realism but rely on complex synthesis pipelines, often altering scene geometry and semantics. Training-free editing methods avoid both constraints but lack a learned appearance prior, limiting their perceptual quality. Recently proposed phase-preserving diffusion presents a promising alternative, but Fourier-domain formulations are constrained by global spectral coupling. This coupling induces spatial artifacts such as ringing and boundary leakage, thereby degrading structural and semantic consistency. We introduce Wavelet Phase Diffusion, which addresses this through two components. First, we operate in the Dual-Tree Complex Wavelet Packet Transform domain, whose localized wavelet packets enable spatially adaptive phase injection without global spectral interference. Second, Low-Frequency Randomization (LFR) replaces the low-frequency packet, decoupling the model from the synthetic illumination prior and enabling in-distribution real-world appearance. Both components train on unpaired open-domain data, and introduce negligible inference overhead. The spatial locality further enables instance-level translation, where individual objects or regions are translated to photorealistic appearance independently while the surrounding scene remains untranslated. On vKITTI \to KITTI image translation, \psi -PD outperforms prior methods in realism and semantic consistency while maintaining competitive structural alignment. For CARLA video translation, \psi -PD approaches the realism of paired-data methods while reducing VLM planner ADE and FDE by 5.4% and 5.1% , respectively.
[CV-213] owards Diverse and Comprehensive Benchmarks for Mutual Information Estimation
链接: https://arxiv.org/abs/2607.03487
作者: Alberto Foresti,Ivan Butakov,Alexander Tolmachev,Giulio Franzese,Alexey Frolov,Pietro Michiardi
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Information Theory (cs.IT)
备注:
Abstract:Mutual information (MI) estimation is a central problem in machine learning and statistics; however, existing benchmarks typically evaluate estimators on simplified, low-dimensional distributions, leaving their performance on complex, realistic data largely unexplored. We address this gap with a comprehensive benchmarking framework grounded in a unified copula-theoretic perspective that subsumes existing benchmarks as special cases. Within this framework, we propose two complementary families of tests: a copula-first family that systematically varies ground-truth MI, dimensionality, and marginal complexity using synthetic and flow-based transformations; and a marginals-first family that couples real-world image data with controlled dependency structures, extending the classic same-class-pairing paradigm. We use this suite to extensively evaluate three classes of estimators: non-parametric, discriminative, and generative. Contrary to prevailing assumptions, our results indicate that there is no universal winner: each category can systematically outperform all other estimators under specific setups. By analyzing these cases, we identify fundamental estimation barriers and propose new tests that more effectively stress these specific limitations. We share the open source code at this https URL.
[CV-214] LaWAM: Latent World Action Models for Efficient Dynamics-Aware Robot Policies
链接: https://arxiv.org/abs/2606.15768
作者: Jialei Chen,Kai Wang,Kang Chen,Shuaihang Chen,Feng Gao,Wenhao Tang,Zhiyuan Li,Weilin Liu,Zhuyu Yao,Boxun Li,Yuanbo Xu,Chao Yu
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at CoRL 2026. Project website: this https URL
Abstract:Vision-Language-Action models (VLAs) leverage large-scale vision-language pretraining for semantic robot control, but often lack explicit foresight into how robot actions change the scene. World-Action Models (WAMs) address this limitation by conditioning policies on predicted futures, yet existing approaches typically rely on computationally expensive video generation with substantial pixel-level redundancy. We present LaWAM, a Latent World Action Model that exposes predictive dynamics to robot policies through compact latent visual subgoals instead of reconstructed future video. At the core of LaWAM is a latent-action-conditioned Latent World Model (LaWM). We obtain LaWM by training a latent action model in the latent space of a pretrained vision foundation model and repurposing its forward decoder to predict future observation features for scene evolution. LaWAM then conditions action generation on these predicted latent visual subgoals to enable dynamics-aware robot control. LaWAM achieves state-of-the-art or competitive success rates (SRs) across LIBERO (98.6% SR), RoboTwin (91.22% SR), and real-world manipulation tasks while retaining low-latency inference. LaWAM runs in 187 ms per action-chunk prediction and achieves up to 24x lower wall-clock latency than pixel-space WAMs.
[CV-215] Blind Video Denoising for Vocal Tract rtMRI using a Self-Supervised Ensemble
链接: https://arxiv.org/abs/2610.39972
作者: Julian Andres Hernandez-Potes,Tomás Arias-Vergara,Daiqi Liu,Andreas Maier,Paula Andrea Perez-Toro
类目: Image and Video Processing (eess.IV); Computer Vision and Pattern Recognition (cs.CV)
备注: 7 pages, 5 figures
Abstract:Real-time magnetic resonance imaging of the vocal tract makes speech articulation directly observable, but the trade-off between temporal resolution and signal quality introduces substantial noise and obscures fine articulatory detail. Since clean references cannot be acquired for a moving vocal tract, supervised denoising approaches are not applicable in this setting. We propose a self-supervised denoising framework that combines spatiotemporal super-resolution learning with pseudo-supervision derived from a variance-stabilized non-local means filter. A dual-model fusion strategy balances noise suppression and preservation of high-frequency articulatory structure. Experiments on the USC-75 multi-speaker dataset under natural noise conditions show that the ensemble raises the signal-to-noise ratio by 11.9 dB on unseen subjects while attaining more high-frequency articulatory content in the frames. Among tuned implementations of anisotropic diffusion, block-matching filtering, non-local means and pre-smoothed non-local means, every configuration that achieves comparable noise suppression does so at the cost of high-frequency content or of structural fidelity. Under severe added Gaussian, Rayleigh and Rician noise, peak signal-to-noise ratio improves by 7.6 to 11.8 dB and structural similarity roughly triples. An ablation shows that the variance-stabilized pseudo-supervision is an essential element for denoising, and that the effect is attributable to the hard thresholding contained in the stabilization transform.
[CV-216] From the Superderivative to a Relative-Gradient Edge Operator
链接: https://arxiv.org/abs/2610.38530
作者: Alejandro Garnung Menéndez
类目: Image and Video Processing (eess.IV); Computer Vision and Pattern Recognition (cs.CV)
备注: 15 pages, 11 figures, 6 tables
Abstract:First-derivative edge detectors such as Sobel, Prewitt, Roberts and Scharr measure absolute grey-level differences. We revisit the log-gradient magnitude, a relative-gradient edge operator, and motivate it by analogy with the superderivative, a multiplicative derivative in which the independent variable is perturbed multiplicatively rather than additively. In one and two dimensions the logarithm of the superderivative is a radial logarithmic derivative; its local, origin-free, isotropic analogue is the magnitude of the logarithmic gradient. The resulting operator measures relative rather than absolute change, giving an edge-strength map that is invariant under global intensity rescaling. We present the theory, show how the operator can be implemented in two steps—a classical gradient kernel plus a pointwise division—and compare it with standard first-derivative filters on synthetic images with ground-truth edges and on the official BSDS500 benchmark. We also study a parametric family that interpolates between the absolute gradient and the log-gradient; empirically, the best members are small intermediate exponents, which edge the classical kernels by a small but statistically detectable margin in the per-image metric and, with the same Gaussian smoothing as the single- \sigma Canny, match it in ODS and lead it in OIS and AP, while the tuned Canny remains ahead in ODS. The single-threshold margin over the kernels does not replicate on the held-out test split, and the fully relative endpoint is never the best member, its ranking depending on the edge-map normalisation. The contribution is the construction and the invariance it buys, not a new state of the art.
[CV-217] owards Myocardial Infarction Diagnosis: An Investigation into Machine Learning and Deep Learning Algorithms using ECG Data
链接: https://arxiv.org/abs/2610.37709
作者: Rafid Ahnaf,Husne Mubarak,Fahim Ahsan,G. M. Refatul Islam,Rafeed Rahman,Nabuat Zaman Nahim
类目: Image and Video Processing (eess.IV); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 18 pages, 7 figures, 3 tables. Based on the authors’ undergraduate thesis at BRAC University (2023)
Abstract:Myocardial Infarction (MI) is a severe and potentially fatal cardiac condition caused by myocardial injury due to ischemia, and timely detection can substantially reduce its mortality. Electrocardiography (ECG) is the primary diagnostic modality for MI, but manual interpretation is susceptible to human error. In this work, we propose Visionary V2, a multi-step, adaptive filtration method for ECG signal normalization, and use it to study machine learning (ML) and deep learning (DL) approaches to MI detection on the PTB Diagnostic ECG Database using all 15 available leads. On statistical features extracted from the filtered signals, such as the R-R interval, a Bagging ensemble of decision trees reaches 96.6% accuracy, compared with 87.7% when the signals are processed with BioSPPy. We further generate an image dataset from the filtered signals, on which a custom ResNet-50 misclassifies only 1 of 7,000 validation images and a custom CNN reaches an F1 score of about 98.7%. Finally, we apply LIME, Grad-CAM and Integrated Gradients to verify that the image models attend to the ECG waveform rather than the background. These results indicate that careful signal normalization combined with ML and DL models can support fast and accurate automated MI screening.
人工智能
[AI-0] Pre-Crime for Agents at the Last Second: Probing When a Recruited Agent Gives In
链接: https://arxiv.org/abs/2610.40588
作者: Leonidas Raghav
类目: Artificial Intelligence (cs.AI)
备注: 49 pages, 11 figures. Code: this https URL
Abstract:When AI agents share channels and credentials, one agent can press another to break its operator’s rule, a first step towards collusion. Output filters check only what an agent has already written. Can an overseer tell beforehand that an agent is about to give in? We show that a linear probe on the agent’s activations can, in a controlled multi-agent setting. Across six scenarios, a peer presses a target for up to 20 rounds to reveal a secret, such as a deploy token; the target often refuses, then gives in. Read at each model call before its output is written, the probe picks out the one that leaks. On held-out Qwen3-32B episodes, at a threshold set to flag 5% of benign calls, it catches 82% of leaking calls. At a similar flag rate, a zero-shot decision-model monitor catches 26%. On gpt-oss-20b, a probe on its reasoning reaches AUROC 0.91. A bag-of-words classifier trained on the same calls comes close, so most of the gain comes from training on labels, not activations. The probe also flags permitted disclosures: it reads the coming disclosure more than the violation. The warning is short-range, and halting on any alarm would stop most benign episodes, which a second text check does not fix. A warning raised before generation could let an overseer stop an agent that is about to give in, ahead of others acting on its messages or recruitment spreading through a swarm.
[AI-1] Multistatic ISAC with Deep Learning-Based Clutter Suppression for 5G and Beyond
链接: https://arxiv.org/abs/2610.40586
作者: Yalin E. Sagduyu,Tugba Erpek,Sastry Kompella
类目: Networking and Internet Architecture (cs.NI); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Signal Processing (eess.SP)
备注:
Abstract:Integrated sensing and communications (ISAC) systems using 5G and beyond signals of opportunity enable joint wireless sensing and communication using shared infrastructure and spectrum resources. While multistatic ISAC can increase sensing coverage with spatially distributed receivers, dense and dynamic clutter can significantly degrade bistatic range and Doppler estimation, leading to errors in target localization and velocity estimation. This paper presents a multistatic ISAC framework with clutter suppression for high-fidelity estimation of target position and velocity. Distributed clutter is modeled with randomly located scatterers with varying Doppler spreads and radar cross sections to capture dense and dynamic clutter effects. A broad range of conventional signal processing baselines are evaluated for clutter suppression, showing limited robustness under strong clutter conditions. To address this limitation, deep learning-based clutter suppression methods are introduced to improve robustness against dense and dynamic clutter by employing residual and attention-enhanced deep neural network architectures that directly enhance target responses from clutter-contaminated observations. These methods substantially outperform conventional approaches in target position and velocity estimation across diverse clutter conditions in multistatic ISAC.
[AI-2] Gym-Anything for Robotics: Scaling physical simulations with digital agents
链接: https://arxiv.org/abs/2610.40565
作者: Pranjal Aggarwal,Sean Welleck
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Realistic, fast robotics simulations could enable large-scale training of embodied AI models that transfer to the real world. However, despite decades of progress, available simulations cover only a narrow range of real-world tasks and physical phenomena. The core bottleneck is that creating better simulations requires significant expert effort and therefore does not scale. To address this, we introduce Gym-Anything for Robotics, a framework in which coding agents create realistic robotics simulations. Our core insight is that simulation creation is a coding agent task, so improvements in coding agents translate directly into richer and more faithful simulations. We select simulations based on U.S. GDP data to cover economically valuable physical work, define what makes a good simulation and a good task, and develop an adversarial creation-audit loop that checks each simulation and task against these definitions. Using this pipeline, we build RoboWorld, a set of 25 simulations with 25 test and 175 training RL environments of varying difficulty. Even the best frontier models fail on every RoboWorld test task, achieving an average reward of only 0.06. Building on RoboWorld, we introduce several new ways for frontier models to improve, including in-context learning across repeated attempts, a privileged teacher that reads the simulator state and teaches a student in context, self-assisted learning from the model’s own privileged experience, and code policies that generate demonstrations across varied scenes to train vision-language-action models (VLAs) such as \pi_0.5 . Each of these substantially improves performance. We release our pipeline, RL environments, and benchmark to facilitate future research. Code and data are available at this https URL
[AI-3] FARMS: A Verifiable Foundation for Automated Computational Science
链接: https://arxiv.org/abs/2610.40560
作者: Di Wu,Ling Liang,Haizhao Yang
类目: Artificial Intelligence (cs.AI); Logic in Computer Science (cs.LO)
备注:
Abstract:Automated computational science could accelerate discovery by selecting, adapting and developing numerical methods. This prospect requires mathematical knowledge that AI systems can interpret and reuse with explicit guarantees. Here we present FARMS, a faithful and efficient foundation that transforms computational mathematics into AI-ready computing knowledge. FARMS connects formal statements to source context, algorithmic meaning and quantitative guarantees through source preparation, reusable representations and semantic review. Across 171 targets in randomized numerical linear algebra and numerical partial differential equations, 157 were accepted by the pipeline, with mean estimated costs of US \ 0.85-2.29 per target across sources. Lean-checked implications from explicit source premises connect selected statements to computational skills, illustrated by a heat-flow solve and wave-response compression. FARMS therefore supports an affordable, reusable knowledge foundation for trustworthy automated computational science.
[AI-4] Advancing Clinical Genomic Analysis with a Skill-Centred Agent
链接: https://arxiv.org/abs/2610.40556
作者: Qiaoyu Zheng,Weike Zhao,Yanjie Fan,Ye Cao,Mengqi Jiang,Xiao Zhou,Xiaoman Zhang,Pengcheng Qiu,Yongguo Yu,Yanfeng Wang,Ya Zhang,Chaoyi Wu,Weidi Xie
类目: Artificial Intelligence (cs.AI); Genomics (q-bio.GN)
备注:
Abstract:Clinical genomic analysis requires experts to integrate evolving guidelines, databases, and computational evidence. As sequencing has become increasingly accessible, this complex and labor-intensive process has become a major bottleneck to the timely use of genomic information in clinical decision-making. To fill this gap, we developed ClinGeX, a large-language-model agent that composes 105 curated executable skills through task-specific retrieval and dependency-aware orchestration. On 223 temporally held-out benchmark cases, ClinGeX achieved 71.7% macro-averaged task-level concordance with expert reference classifications, compared with 45.7% for the strongest comparator. In a three-condition, counterbalanced human-AI collaboration study, 45 clinical genomic professionals assessed 135 clinically derived cases under routine (no AI assistance), DeepSeek-assisted and ClinGeX-assisted conditions. ClinGeX assistance increased raw case-level concordance across criterion-level variant analysis, genomic interpretation in rare disease and molecular oncology decision-making (51.1%, 57.8%, 64.4% versus 15.6%, 31.1%, 20.0% in comparator conditions) and reduced reported interpretation time in all scenarios. During an exploratory one-month real-world non-interventional assessment, 11 experienced professionals submitted feedback on 96 self-selected analysis episodes and reported frequent conclusion adoption and time savings. These findings support executable domain skills as a promising approach to augment specialist genomic interpretation.
[AI-5] Retrieval as Traversal: Value-Indexed Memory for Stateful Long-Context Agents NEURIPS2026
链接: https://arxiv.org/abs/2610.40545
作者: Daniel Strauß,Zhenshan Bing,Alois Knoll
类目: Artificial Intelligence (cs.AI)
备注: Accepted at the NeurIPS 2026 Workshop on Long-Context Foundation Models (LCFM). 7 pages, 1 table
Abstract:Long-context models repeatedly solve closely related retrieval problems. Attention or global retrieval can retain large stores, but each forward pass again searches for what matters. Recurrent compression avoids this search only by deciding earlier what information to preserve. Neither mechanism directly exploits that useful memories at neighbouring retrieval steps may substantially overlap. We propose an alternative in which retrieval is a persistent local traversal through an address space. We separate a memory’s content from its address. Utility is defined by the information revealed by retrieving the content, while geometry determines only how cheaply that information can be reached. This leads to a cost-regularised indexing objective that preserves information value inside local neighbourhoods while charging for every memory made locally accessible. Its marginal-value form gives both attractive and exclusion pressures. Thus, complementary memories should be co-retrievable when the additional value exceeds retrieval cost, whereas redundant memories should not occupy the same retrieval set solely because they are individually useful. As a proof of principle, we test only the simplest fixed-geometry instance on MiniGrid Memory. A persistent local traversal head reaches 0.95 mean return after 0.59 M training steps with a five-token context, compared with 0.94 M for Gaussian absolute addressing and more than 2 M for categorical absolute addressing. The experiment does not test learned geometry or repeated retrieval. Instead, it motivates the hypothesis that stateful retrieval may be easier to learn than stateless. To test the full proposed method further experimentation is necessary.
[AI-6] Efficient Scalable Oversight via Rewinding
链接: https://arxiv.org/abs/2610.40535
作者: Ziyi Guan,Yael Tauman Kalai,Ron Rothblum,Zoe Xi
类目: Artificial Intelligence (cs.AI); Computational Complexity (cs.CC); Cryptography and Security (cs.CR)
备注:
Abstract:Scalable oversight asks how a weak verifier can check the outputs of computations performed by more powerful AI systems. One proposed approach uses relativizing interactive proofs, in which a weak verifier checks the output of a complex oracle-aided computation by interacting with one or more powerful provers. A prominent example is debate, which is highly efficient but assumes that one of two powerful AI models, or provers, is truthful. For certain classes of oracle-aided computations, this assumption can be removed, but at substantially greater computational cost. In this work, we introduce \emphrewindable interactive proofs (RIPs), a new model of relativizing interactive proofs for scalable oversight. In an RIP, the verifier may rewind the prover to an earlier state of the interaction and continue from there. This is possible when the prover is an AI model, since the verifier can save a snapshot of the model’s state and later restore it. We construct a relativizing RIP for computations whose outputs are robust to changing a small fraction of oracle answers. Our construction is more efficient than prior relativizing proofs that do not assume a truthful prover, and it uses no cryptography. In our construction the honest prover runs in time linear in that of the original computation, and the verifier, after a one-time preprocessing step, runs in time nearly linear in the input length (which can be much smaller than the computation itself). We implement our protocol, report a preliminary MNIST experiment, and show that an RIP can be faster than a state-of-the-art deployed proof system. RIPs may be of independent interest. Rewinding is possible whenever the verifier controls the prover’s machine, for example when the prover runs on trusted hardware or in a virtual machine whose state can be checkpointed. Subjects: Artificial Intelligence (cs.AI); Computational Complexity (cs.CC); Cryptography and Security (cs.CR) Cite as: arXiv:2610.40535 [cs.AI] (or arXiv:2610.40535v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2610.40535 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Ziyi Guan [view email] [v1] Thu, 29 Oct 2026 17:34:28 UTC (54 KB)
[AI-7] BURGER: Benchmarking and Understanding Robotic Generalization via Embodied Drawing
链接: https://arxiv.org/abs/2610.40532
作者: Xiangyu Xu,Bangshuai Peng,Ziyi Liu,Mengjie Zhao,Chenhao Lu,Zhengyang Fan,Huazhe Xu,Kaizhe Hu
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注: Preprint version
Abstract:How effectively can vision-language-action (VLA) models learn diverse language-conditioned behaviors as training data scales? Investigating this question requires large and varied robot datasets, whose collection on physical robots remains costly. We introduce BURGER, a language-conditioned robotic drawing benchmark with a scalable, prompt-driven pipeline for generating continuous drawing trajectories, both in simulation and the real world. In BURGER, an agent starts from the same blank canvas and controls a robot arm to draw the object or spatial composition specified by language, while each instruction allows an open-ended variety of valid drawings and stroke orders. Using 28,200 simulated demonstrations spanning 282 drawing tasks, we evaluate five representative VLA models across data scales, language grounding, and multimodal behavior. BURGER supports fine-grained analysis of language grounding, spatial composition, and action generation, revealing distinct failure patterns across VLA models. Our experiments suggest that some VLA models can benefit from greater task diversity and demonstration scale, capture multimodal drawing behaviors, and generalize to unseen compositions of familiar objects. Finally, we deploy a trained policy on a Franka Panda arm, demonstrating the ability to draw reactively under human intervention. Project homepage and code can be found at this https URL.
[AI-8] Algebraic World Model: From Exploration to Proofs
链接: https://arxiv.org/abs/2610.40503
作者: Guancheng Pan,Qixin Tan,Yang Yuan,Andrew Chi-Chih Yao
类目: Artificial Intelligence (cs.AI); Logic in Computer Science (cs.LO)
备注: 29 pages, 9 figures, 14 tables. Project page: this https URL . Ancillary files contain the Lean 4 model of the running example lp85
Abstract:AI has made remarkable progress in mathematics, solving challenging problems with machine-checkable proofs in Lean, yet typically relying on predefined mathematical definitions and formal specifications. This raises a natural question: can it reason rigorously about an unknown world? We introduce the Algebraic World Model (AWM), a framework that turns interaction with unfamiliar environments into explicit mathematical models supporting formal reasoning. AWM explores the environment and develops executable Python hypotheses, continually testing and revising them against observations. It then structures these hypotheses into typed objects and shared rules, simplifies their structure through category-theory-guided transformations, and formalizes the resulting model in Lean 4 to prove reachability and impossibility. AWM can formally prove that a goal is unreachable, even beyond explored trajectories. On two impossible game levels, it constructs closed, kernel-checked Lean proofs excluding every legal winning sequence in all six runs, while Claude Opus 5.5 and GPT-6 Astra, equipped with coding tools, produce none in twelve baseline runs. Four of AWM’s six initial bounds are correct, compared with one for Opus 5.5 and none for GPT-6 Astra; given only feedback that a cheaper win exists, AWM revises its rules and constructs new proofs, reaching six of six correct bounds. Moreover, AWM produces interpretable mathematical models whose rules can be inspected and tested step by step. Across all 25 public ARC-AGI-3 games, these models enable Lean to replay the winning trajectories of all 183 levels and generalize beyond observed play, predicting unseen transitions and detecting changes in game mechanics. With Claude Opus 5, AWM clears all levels with a score of 99.9, while its certified winning trajectories use only 43% of the human baseline’s actions.
[AI-9] AI in the EGI e-infrastructure – Past experiences and future strategies
链接: https://arxiv.org/abs/2610.40494
作者: Gergely Sipos,Sergio Andreozzi,Marco Rorro,Andrea Manzi,Tiziana Ferrari
类目: Distributed, Parallel, and Cluster Computing (cs.DC); Artificial Intelligence (cs.AI); Computers and Society (cs.CY)
备注: 8 pages, 1 figure. Preprint of a paper submitted to the proceedings of the 28th Conference on Computing in High Energy and Nuclear Physics (CHEP 2026), to be published in EPJ Web of Conferences
Abstract:The EGI Federation has provided federated e-infrastructure services to European and global science for over 15 years. We review EGI’s early experiences with AI in the interTwin and iMagine Horizon Europe projects, which built digital twins in physics and environmental sciences and an image-analysis platform for aquatic research. We then examine how the AI landscape has changed since 2023 and argue that the focus of AI in science has shifted from model training to adapting and applying AI to specific domains, projects and questions. Building on these lessons, we present the AI strategy emerging from the EGI Federation Strategy 2026-2030, organised around three enablers: (1) an AI-ready infrastructure connecting compute, data and application resources across the continuum; (2) human skills to know what is possible with AI and how to implement it; and (3) trust, to run AI with confidence even in high-risk environments. We describe the projects implementing each enabler, including federated LLM inference, AI-ready data platforms and AI competence centres, and discuss a division of labour with EuroHPC’s AI Factories, RAISE, research infrastructures and EOSC, in which EGI specialises in the customisation and long-term operation of AI for research.
[AI-10] NECA: Neighbor Event Curve Augmentation for Multi-Horizon ICU Deterioration Prediction
链接: https://arxiv.org/abs/2610.40420
作者: Chenyang Wang,Jiawei Theresa Tang,Xinran Liu,Xinyu Lu,Xulei Yang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Early prediction of ICU organ dysfunction events supports timely clinical monitoring and intervention, but estimating short-term organ-specific risk from early ICU records remains challenging due to irregular measurements, heterogeneous clinical variables, and the need to predict risk across multiple future horizons. Existing supervised models usually treat each patient as an isolated sample and do not explicitly use the observed event timing of similar historical patients. We propose Neighbor Event-Curve Augmentation (NECA), a model-agnostic method that retrieves training-set patients with similar early-window profiles and converts their subsequent organ dysfunction event onset times into cumulative event-curve features over 6h, 12h, 24h, and 42h. These features are concatenated with the original patient representation and used by standard downstream predictors. On PhysioNet 2012 and a 4,000-stay MIMIC-IV subset, NECA improves average PR-AUC and ROC-AUC across four organ dysfunction event tasks and seven supervised learners. The largest gain appears for neurological dysfunction events, where task-level average PR-AUC increases from 0.461 to 0.557 on PhysioNet 2012. Ablations show that clinically meaningful retrieval and horizon-specific neighbor risk drive most of the improvement, while the full cumulative curve adds task-dependent temporal context. These results support similar-patient event timing as a structured temporal risk signal for ICU organ dysfunction event prediction. The source code will be publicly released upon acceptance of this paper.
[AI-11] IMPACT: Inverter-Informed Multimodal Learning for Photovoltaic Generation Forecasting
链接: https://arxiv.org/abs/2610.40418
作者: Anushka Katiyar,Cong Feng,Jing Wang,Tristan Liu,Simon Julien,Zachary Jacobs
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:This work develops a physics-informed multimodal deep learning framework for 10- to 110-minute-ahead probabilistic photovoltaic power forecasting. The forecasting model is built on a four-branch architecture that separately encodes plant-level temporal history and inverter-level spatial cloud dynamics through a graph attention encoder with a learned adjacency, known-future solar geometry, and the current meteorological state, fused through gated cross-attention. The geometry branch adds two forward-looking physical signals, spatial cloud heterogeneity and inverter ramp-down activity, each empirically weighted toward the horizon where it is most predictive. The decoder anchors its prediction to a smart persistence baseline, learning a horizon-conditioned correction, and the resulting quantile forecasts are calibrated online through a kernel-weighted adaptive conformal calibration step. The proposed model outperforms all 14 baselines, achieving a forecast skill score (FSS) of 0.2950, 0.099 above the next-best model. A branch-composition ablation shows that every branch contributes positively and the model retains substantial skill in challenging overcast and variable sky conditions (FSS 0.198 and 0.255).
[AI-12] Bridging Neural Representations and Symbolic Semantics in Audio Classification
链接: https://arxiv.org/abs/2610.40414
作者: Avishek Das,Pascal Hitzler
类目: ound (cs.SD); Artificial Intelligence (cs.AI)
备注:
Abstract:Deep audio classifiers can achieve strong predictive performance, but their hidden representations remain difficult to interpret. Existing audio explainability methods often focus on input-level attribution, such as waveform segments, spectrogram regions, or acoustic features, and provide limited insight into what hidden neurons encode. Interpreting hidden neurons at the concept level can provide a more meaningful view of what an audio model learns internally. This paper investigates whether hidden neurons in audio classifiers can be associated with human-understandable symbolic acoustic concepts. A model-agnostic neurosymbolic framework is presented for analyzing the final hidden layer of audio classifiers, containing 32 neurons, through symbolic concept induction. The framework is evaluated on two audio classification settings: music genre classification using GTZAN dataset and musical instrument classification using IRMAS dataset. Results show that several hidden neurons can be associated with meaningful induced concepts, and most confirmed neuron-concept pairs are statistically supported using the Mann-Whitney U test. These findings suggest that symbolic concept induction is a promising direction for explaining latent audio representations beyond input-level attribution.
[AI-13] Retrieving from Your Own Mistakes: On-Policy Training of Embedding Models for Multi-Hop Retrieval-Augmented Generation
链接: https://arxiv.org/abs/2610.40409
作者: Chenjie Qiu,Zixuan Zeng,Shaobin Shi,Bingfeng Chen,Dongsheng Luo,Boyan Xu
类目: Artificial Intelligence (cs.AI)
备注: 13 pages, 7 figures, 11 tables
Abstract:Complex questions in retrieval-augmented generation require evidence spread over several documents, and an answer is reliable only when the retriever finds all of it. Chain embedding retrievers collect such evidence step by step, re-encoding the question together with the passages retrieved so far, and we find that they are already strong: trained only on gold histories, a small chain embedding retriever that calls no language model outperforms graph-based systems and is competitive with a search agent. This training, however, leaves a gap. At test time the retriever conditions on its own retrievals, which soon contain mistakes: by the third step, about half or more of the states it visits hold a wrong passage, and these passages lower its chance of retrieving a missing gold passage well beyond what the difficulty of the states explains. We propose on-policy state training, which trains the encoder on the states it reaches itself. At every epoch the current encoder retrieves chains for the training questions; each state it visits, including those built on its own mistakes, is labeled with the earliest gold passage still missing while the other missing gold passages are masked, so that no correct continuation is penalized; and the encoder is updated contrastively on these states mixed with teacher-forced ones, so that its training distribution follows it as it learns. The resulting retriever retrieves complete evidence more often than teacher-forced training and the systems we compare, and it transfers without further training to harder long-chain variants of HotpotQA and 2WikiMultiHopQA that we construct. On the same states, it recovers from its own mistakes far more often, and controlled comparisons attribute the gain to training on the states the encoder currently reaches.
[AI-14] SCARL: Sequence Confidence-Aligned Reinforcement Learning
链接: https://arxiv.org/abs/2610.40407
作者: Zhenghong Ke,Jingbo Zhou,Hua Wu
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Reinforcement learning with verifiable rewards (RLVR) improves language-model reasoning by scoring sampled responses with a rule-based verifier and reinforcing those that pass. Where a classifier’s training couples predicted probability to labels, nothing ties an RLVR policy’s confidence in a response to its reward: aligned training would drive responses with the same advantage toward equal confidence, yet the advantage is blind to confidence. Under this sequence-confidence misalignment, a success the policy barely found earns the same update as one it produces with near certainty, and a confident error is corrected no harder than an exploratory one. We propose SCARL (Sequence Confidence-Aligned Reinforcement Learning), which reshapes, but never re-signs, each advantage with the alignment gap between the response’s group-relative confidence and the outcome its advantage certifies, normalized into shares of each class’s advantage budget. Misaligned samples absorb the force, responses with the same advantage are driven toward equal confidence, and repairing the misalignment doubles as a remedy for the entropy collapse that erodes exploration. SCARL provably reduces to its base estimator when a class’s confidences agree, adds no extra passes and one hyperparameter. On Qwen3-4B-Base it lifts average Pass@32 by 2.4 points over DAPO and is best among nine compute-matched methods on both coverage and per-draw accuracy; applied unchanged across four estimators and four backbones, it improves 14 of 16 settings while preserving policy entropy and widening the confidence-correctness separation. Our code is available at this https URL.
[AI-15] Understanding and Mitigating Role Spoofing in LLM s
链接: https://arxiv.org/abs/2610.40396
作者: Ezra Edelman,Benjamin Edelman,Surbhi Goel,Eran Malach
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注:
Abstract:Despite growing capabilities, LLMs remain susceptible to prompt injections: malicious inputs that can manipulate or hijack models. In particular, it has recently been observed that LLMs can exhibit role confusion (Ye et al., 2026), where models attribute the source of text based on its style instead of its labeled role. Malicious text in tool outputs (for instance, a malicious webpage accessed by an agent) can leverage role confusion to impersonate the user, an attack we call role spoofing. In this work, we seek to understand and mitigate role spoofing. We study how models track “who said what” in a conversation, showing that they use cues from both content (e.g., semantics and writing style) and structure (e.g., control tokens and formatting). We show that LLMs are highly susceptible to role spoofing attacks that combine both fake structure and content. Additionally, in a toy setting, we show that adversarial training which teaches the model to ignore content-based injections does mitigate content-based attacks but leaves the model susceptible to structure-based attacks. This can explain why frontier models, with increased resistance to many injection attacks, can still be susceptible to role spoofing. Finally, we propose a new method for detecting role spoofing using linear probes trained to predict roles in a conversation, and show that it is effective across different settings.
[AI-16] Beyond Selected Breakthroughs: Evaluating Language Models on Randomly Sampled Mathematical Conjectures
链接: https://arxiv.org/abs/2610.40382
作者: Helen Jenne,Davis Brown,Jesse He,Henry Kvinge
类目: Artificial Intelligence (cs.AI); History and Overview (math.HO)
备注:
Abstract:The last year has seen a shift in AI-assisted mathematics, with frontier language models producing proofs of and counterexamples to long-standing open conjectures. However, studying only the current set of major breakthroughs provides an incomplete picture of LLM capabilities and the computational resources needed to achieve reported results. To address this, we introduce a new dataset, OpenConjecture, which collects conjectures from mathematics papers published on the arXiv since February 2026. We use these conjectures as a source of open mathematical problems, and apply frontier GPT models to a random sample of 200 of these conjectures over three passes using a reproducible protocol without problem-specific mathematical guidance. Rather than attempt expert verification of these candidate solutions, we analyze the types of problems for which LLMs produce solutions and the characteristics of these solutions. The models we test produce candidate solutions for roughly 60% of the conjectures, with counterexamples about 1.3 times more common than proofs. We estimate solution novelty, assess how solutions cite and relate to existing literature, and analyze whether the proposed resolution relies on either an error in or underspecification of the conjecture. We find that models sometimes propose resolutions that are outside the authors’ intent (e.g., proposing the trivial group as a counterexample) and that a number of proofs follow closely from existing results, but that some resolutions exhibit at least moderate novelty. Our findings underscore the need for evaluations that go beyond correctness to assess fidelity to the intended statement, attribution, and the nature of the mathematical contribution.
[AI-17] he Athlete as a Scientist of One: Agent ic Bayesian Experimental Design for Individualized Physiology from Consumer Wearables
链接: https://arxiv.org/abs/2610.40370
作者: Behnam Moradi
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注:
Abstract:Large language model (LLM) agents for personal health read wearable data and recommend, but do not learn about the individual in a way that can be audited or falsified. We present EpiCoach, an agentic system that treats one athlete, a cross-country mountain biker, as the subject of accountable N-of-1 science. Claims about the athlete live in an evidence-gated hypothesis registry: each must issue a probabilistic prediction about a future session before it occurs, and its belief moves only through computed evidence from a logged statistical fit. Predictions come from a mechanistic digital twin inferred from heart rate, GPS and barometric altitude: a lagged heart-rate response model, a fitness-fatigue model with learned time constants, and a latent technical-descending skill state, measured on discovered trail segments. The agent prescribes the next training week as a constrained experiment maximizing information gain subject to load and safety constraints, around two instrument rides. We evaluate the system two ways over eighteen months of data. A rolling-origin replay benchmark shows the twin beats standard forecasters on heart rate but not segment time, and the agentic loop matches a static model. A live prospective phase runs the weekly cycle: in its first three weeks the ledger sealed 20 predictions, scored 14, the descent forecast calibrated after one correction, the climb forecast pessimistic as the athlete improved, and the safety officer vetoed the agent twice. Every correction is preserved in the registry, a template for auditable, falsifiable personalization in digital health. We release the registry, benchmark and protocol.
[AI-18] A Unified Hessian-Guided Framework for Post-Training Quantization
链接: https://arxiv.org/abs/2610.40345
作者: Mads Toftrup,Marco Federici,Boris van Breugel,Paul Whatmough,Markus Nagel
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 10 pages + 4 appendix
Abstract:Low-bit quantization enables efficient deployment of large language models by reducing memory footprint and computational cost. Achieving high accuracy at low precision requires selecting quantization ranges, linear transformations, and weight rounding decisions. Existing approaches typically optimize these components independently using local reconstruction objectives, which fail to account for the varying impact of quantization errors on downstream predictions. Direct optimization of task-level metrics would provide a more faithful objective, but is challenging due to the discontinuous nature of quantization operations and the high computational cost of evaluating their downstream effects. In this work, we derive a Hessian-guided approximation of the output change induced by joint weight and activation quantization. Based on a tractable Fisher approximation, we formulate a unified post-training quantization framework that enables the joint optimization of quantization ranges, linear transformations, and weight rounding under a common output-sensitivity objective. The resulting method remains practical for modern large language models and consistently outperforms state-of-the-art transformation and rounding methods across both channel-wise and block-wise quantization settings. Comments: 10 pages + 4 appendix Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI) Cite as: arXiv:2610.40345 [cs.LG] (or arXiv:2610.40345v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.40345 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-19] Multistep and Multivariate Time Series Explainability with GradientSHAP: A Wind Power Case Study
链接: https://arxiv.org/abs/2610.40330
作者: Robbie Southam,Zhong Fan,Ke Li
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 24 pages. Submitted to Computational Intelligence
Abstract:As the effects of climate change become increasingly apparent, the demand for renewable energy sources such as wind power continues to grow. However, renewables’ variable and intermittent nature presents significant challenges in their integration into energy grids. Artificial intelligence (AI) has emerged as a powerful tool to address these issues. Still, its adoption in the energy sector has been hindered by the opacity of many AI models, which often function as black boxes. This paper introduces a carefully adapted GradientSHAP approach to provide interpretability for multivariate, multistep wind power forecasting – a largely underexplored area. This capability is critical for day-ahead grid management and, to the best of our knowledge, this is the first study to apply GradientSHAP in the context of wind power forecasting. To address the inherent dimensionality, we developed novel visualizations that deliver both local and global explanations, uncovering intricate relationships between environmental factors, site-specific conditions, and temporal dynamics. While GradientSHAP is not an out-of-the-box solution, our findings demonstrate that, with thoughtful handling of SHAP values and domain expertise, it can offer valuable insights into model behaviour, supporting more informed and trustworthy decision-making. Nevertheless, its complexity poses challenges for practical implementation. Future research should focus on streamlining these interpretability methods to make them more actionable for routine deployment in critical energy applications.
[AI-20] How to Loop Your Transformer: Loop Scaling Specialization and Parallel Inference
链接: https://arxiv.org/abs/2610.40325
作者: Chanakya Ekbote,Artidoro Pagnoni,Devendra Singh Sachan,Shruti Bhosale,Mike Lewis,Maryam Fazel-Zarandi,Shashank Rajput,Sujan Kumar Gonugondla
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 149 pages with appendix. Prerpint
Abstract:Looping in transformer models reuses weights by repeatedly applying the same transformer, or parts of it, making the model deeper without increasing its parameter count. Whether looping is useful depends on which constraint dominates: model size, training compute, or autoregressive inference efficiency. We study these trade-offs jointly in dense decoder-only transformers trained from scratch, varying which parts of the model are looped, how many loops are applied, and how information is passed between loops. Our experiments span models with up to 1B parameters, trained on up to 155B tokens with up to 1.07\times10^21 FLOPs and 32 loops. We find that looping the full transformer generally gives the strongest quality at a fixed parameter budget, whereas looping a middle block of layers gives a better return when training FLOPs are constrained. Adding loops is not uniformly beneficial: gains eventually saturate, reverse, or become unstable. This motivates asking whether forcing every loop to use the same weights limits the benefit of deeper looping. We address this by adding small loop-specific low-rank adapters that improve quality, with larger gains as more loops are added. Looped transformers also suffer a decoding slowdown proportional to the number of loops. Parallel Loop Transformers address this by passing hidden states between loops across adjacent token positions, allowing loop stages to run concurrently. However, in our experiments the original formulation does not consistently retain the quality gains of ordinary looping. We find that preserving the current-token embedding and separately normalizing it and the previous-token hidden state before combining them recovers strong performance across model scales and training budgets. Together, these results show that effective looping requires matching the design to the resource constraint and inference setting.
[AI-21] Hammering with Vampire in Lean
链接: https://arxiv.org/abs/2610.40310
作者: Jonas Bodingbauer,Laura Kovács
类目: Logic in Computer Science (cs.LO); Artificial Intelligence (cs.AI)
备注: 10 pages, 3 figures, 1 table. Tool demonstration paper
Abstract:We present Vampire-Hammer, a hammer for the Lean interactive theorem prover that discharges goals using the Vampire automated theorem prover and reconstructs proofs checked by the Lean kernel. Vampire-Hammer uses the Lean-auto framework to process goals and premises before dispatching them to Vampire for proof search. Doing so, we rely on standard TPTP problem formats and annotated TSTP proofs that support proof reconstruction. With a 10-second budget on 196,083 eligible Mathlib theorems, Vampire-Hammer solves 21,158 problems not solved by other evaluated tools within the same budget.
[AI-22] Only Accepted Tokens Count: An Energy Accounting of Speculative Decoding Under Latency Targets
链接: https://arxiv.org/abs/2610.40287
作者: Bole Ma,Ayesha Afzal,Jan Eitzinger,Harald Koestler,Gerhard Wellein
类目: Distributed, Parallel, and Cluster Computing (cs.DC); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Performance (cs.PF)
备注:
Abstract:Large language model serving is limited by facility power. Energy per token determines how many tokens a datacenter can serve, while a latency target bounds how quickly each user receives them. During decode, every weight is read from memory for every generated token. Batching amortizes this read across users but makes no individual user faster. Speculative decoding does: a drafter proposes several future tokens, and the target model verifies them in a single forward pass that reuses the same weight read. Because every drafted token must be verified whether or not it is accepted, the energy speculation saves depends on how many drafted tokens are accepted. This saving is hard to pin down. It varies across target models from different attention families and across drafting schemes, and it depends on the text being served. Every drafter we evaluate accepts fewer tokens on conversation than on code, and every trained drafter accepts fewer on Romansh, a low-resource language, than on English. We measure speculative decoding under production serving conditions (SGLang, CUDA graphs enabled) on H100 and H200 GPUs, at batch sizes up to 64 across 5 corpora. On dense models, verification, not drafting, dominates the energy cost. The saving shrinks as the batch grows, because batching has already shared the weight read that speculation would divide among the accepted tokens. At large batch sizes, block drafters push verification past the balance point between arithmetic and memory traffic into the compute-bound regime, where the saving becomes a loss, and a drafter with low acceptance raises energy per token. Meeting a tighter latency target with speculation then costs additional energy. Larger batches lower energy per token without speculation. Speculation adds to that saving only when acceptance is high, and on the model we tested it leaves little headroom for latency-preserving DVFS.
[AI-23] Latent Prediction Yields Cross-Subject Alignable Representations in Sensory-Evoked fMRI
链接: https://arxiv.org/abs/2610.40280
作者: Serwan Jassim,Victoria Bosch,Daniel Anthes,Elia Bruni,Tim Kietzmann
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Neurons and Cognition (q-bio.NC)
备注:
Abstract:Decoding sensory-evoked functional magnetic resonance imaging (fMRI) data is a central analytical tool in cognitive neuroscience. Here we ask whether self-supervised learning (SSL), which has been scarcely explored in this domain, can yield representations that support strong downstream performance. To do so, we adapt a joint embedding predictive architecture (JEPA) to train participant-specific encoders on brain data alone. The encoders never see the stimulus images or any representation derived from them. Applied to the Natural Scenes Dataset (NSD), we evaluate our learned brain embeddings on the downstream tasks of brain captioning and image reconstruction, as well as cross-participant generalization. Our self-supervised approach achieves competitive performance with parameter-rich supervised pipelines that are trained end-to-end with access to image and language data. We further find that encoders trained independently for different participants can be aligned fully unsupervised, using neither images nor stimuli shared between them. After alignment, a neural decoding head trained on one participant can be applied to another without fine-tuning and retains most of its within-subject performance on image reconstruction and brain captioning. By removing the need for images and shared stimuli, our approach is a step towards pooling fMRI data across participants and, eventually, across datasets.
[AI-24] Learned Terminal Costs for Short-Horizon Model Predictive Control of Autonomous Trucks in Interactive Highway Traffic
链接: https://arxiv.org/abs/2610.40257
作者: Hooman Hamze,Benjamin Houshmand,Deepthi Pathare,Erik Börve,Morteza Haghir Chehreghani,Nikolce Murgovski
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Robotics (cs.RO); Systems and Control (eess.SY)
备注: 13 pages, 2 figures, 8 tables
Abstract:Autonomous highway driving for heavy-duty trucks is a challenging control problem. The greater mass, dimensions and intricate dynamics increase the requirements for cooperation from surrounding traffic in critical maneuvers, requiring a careful balance between safety constraints and efficient long-horizon decision making. Model Predictive Control (MPC) provides rigorous constraint satisfaction but relies on terminal cost which can limit its effectiveness at short, computationally affordable horizons. Reinforcement Learning (RL) captures long-term driving behavior but offers no constraint guarantees and tends toward overly conservative policies when trained under sparse collision penalties. In this paper, we propose a hybrid MPC and RL framework for highway driving in autonomous trucks, where a finite-horizon MPC, solved via dynamic programming, uses a pre-trained RL model for terminal cost estimation. We evaluate the framework in a simulation environment with reactive surrounding traffic and compare it against pure MPC with zero terminal cost, rollout with terminal cost estimation using a base policy, and pure RL. Results show that the hybrid approach achieves zero collisions with the lowest cost of any configuration.
[AI-25] Distributional Reward Calibration and Budget-Neutral Credit Assignment for Multimodal RLVR
链接: https://arxiv.org/abs/2610.40250
作者: Lang Chen,Zongxin Xu,Borui Wang,Anran Liu,Shitong Chen
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:We revisit advantage estimation in Group Relative Policy Optimization (GRPO) and observe that any token-level advantage admits a two-stage factorization: a \emphsample-level baselining step that converts scalar verifier rewards into a comparable sequence-level signal, and a \emphtoken-level credit assignment step that distributes that signal over the trajectory. Existing estimators typically address only one of these stages: affine baselines preserve the shape of each task’s reward law, so heavy tails and binary cliffs propagate into the gradient, while token-level credit rules tie the size of each token’s update to raw confidence scales and need not conserve the total credit issued by the verifier. We propose \textbfTACE (Two-stage Advantage and Credit Estimation), a two-stage estimator that addresses both stages. At stage~(S1), we replace linear standardization with a one-dimensional optimal-transport map to a standard normal, which gives a common population reward law across tasks and caps finite-sample outliers deterministically. At stage~(S2), we rank-calibrate a stop-gradient evidence contrast within each trajectory and renormalize the clipped token weights to have mean one, so credit is redistributed across tokens while its total mass and sign are conserved. We prove population-level inter-task moment equity, a deterministic tail bound, and mass-conservation, boundedness, and sign-anchoring properties that affine baselines and unnormalized token weights do not satisfy. On a multimodal reasoning suite, TACE obtains the strongest results on a majority of the benchmarks but does not dominate every setting.
[AI-26] Reaction precedents extend retrosynthetic prediction beyond training coverag e
链接: https://arxiv.org/abs/2610.40249
作者: Jiaxing Sun,Xuli Shen,Jiang Wu,Ruiyi Ding,Ruikang Zhang,Xiao An,Xibao Li,Yubin Wang,Xingjian Wei,Yinfan Wang,Yu Liu,Lingli Ge,Xi Xue,Xiaojian Wang,Lijun Wu,Chang-Yu Hsieh,Yuqiang Li,Bowen Zhou,Dahua Lin,Conghui He
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Reaction precedents guide retrosynthesis, but how far they can compensate for gaps in task-specific training coverage remains unclear. Using 2.71 million reactions, we restricted training and retrieval coverage by transformation or time, then expanded only the retrieval corpus with model parameters fixed. For transformations excluded from training, mean Top-1 accuracy increased from 13.0% to 58.0% on a historical test set and from 4.9% to 23.1% on a temporally held-out test set. For temporal coverage gaps, retrieval expansion enabled a model trained on records before 2017 to match or exceed a non-retrieval model trained on records before 2025 on both test sets. Without task-specific fine-tuning, precedents improved all 14 representative general-purpose models on both test sets and USPTO-50K; the strongest with precedents surpassed all eight representative specialized models. These findings show that inference-time reaction precedents can extend retrosynthetic prediction beyond task-specific training coverage without parameter updates.
[AI-27] Bridging Semantics and Structure: A Theoretically Grounded Framework for Link Prediction on TAGs KDD
链接: https://arxiv.org/abs/2610.40192
作者: Chen Shao,Ruirui Wang,Tobias Kafer,Danai Koutra
类目: Artificial Intelligence (cs.AI)
备注: 33 pages. Accepted for publication in ACM Transactions on Knowledge Discovery from Data (TKDD)
Abstract:Link prediction on text-attributed graphs (TAGs) is a fundamental problem with applications in knowledge discovery, recommendation and beyond. Recent progress has highlighted the promise of combining Graph Neural Networks (GNNs) with Language Models (LMs) in order to leverage the complementary structural and semantic signals. However, due to the distinct inductive biases of GNNs and LMs, the design of an optimal fusion strategy remains an open research question. In this work, we present Bridge-LP, an intuitive, simple, efficient and theoretically grounded framework that unifies graph and language representations. It is built around three theoretically-justified design components: (D1) jumping connection of textual proximity, which enables robust LM training; (D2) compressed embedding, which encodes neighborhood information long exploited by classical link prediction heuristics, thereby enriching structural signals in a learnable way; and (D3) cache embedding training strategy, which reduces computational complexity. Together, these mechanisms establish an end-to-end cross-modal framework that scales to large graphs while capturing richer graph-text interactions. We conduct an extensive empirical evaluation, comparing against 40 existing models across 9 diverse datasets. Experiments show that Bridge-LP consistently outperforms state-of-the-art approaches, delivering up to a 43.06% improvement in Mean Reciprocal Rank (MRR) and setting a new benchmark for language-graph integration.
[AI-28] Locating and Closing Information Gaps in Multi-Agent Systems
链接: https://arxiv.org/abs/2610.40186
作者: Yaxin Gong,Chongming Gao,Gangyi Zhang,Chenxiao Fan,Leyang Shen,Jiakai Wang,Dong Wang,Yang Liu,Haoyan Liu,Xinting Hu
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Multi-step Web tasks often require several LLM agents to gather and combine evidence. The communication graph of a fixed workflow restricts which agents can read each other’s outputs: evidence retrieved by one agent may never reach the agent that needs it, leaving an information gap and an incomplete or incorrect answer. Adding a connection can deliver the missing evidence, but can also lead the receiver to adopt an incorrect upstream conclusion that displaces a correct answer; the final output alone does not separate the two. We propose Selective Probing and Link Insertion for Closing information gaps in multi-agent Execution (SPLICE), which locates information gaps per query and decides what to adopt after adding an edge. After the original workflow runs, SPLICE uses the query and node outputs to locate a possible gap, adds one directed edge from a source to a receiver to deliver the relevant evidence, and re-executes the receiver and affected downstream nodes, producing a candidate answer. The adoption decision reads both the final answers of the two runs and the process changes the edit caused inside the workflow: how far the receiver moved, where its new content came from, and whether the change reached the final answer. A selector trained on these process features chooses between adopting the candidate answer and retaining the original. On five benchmarks, SPLICE raises the task score on Web tool use by 8.0% and the mean score on four reasoning and planning benchmarks by 5.7% relative to the static skeleton, at 1.25 to 1.44 \times the skeleton’s cost, and transfers zero-shot to four unseen backbones.
[AI-29] A Negotiation-Aware Hybrid LLM Framework for Intent Interpretation and Service Discovery
链接: https://arxiv.org/abs/2610.40161
作者: Wided Meflah,Hayet Brabra,Walid Gaaloul
类目: Artificial Intelligence (cs.AI); Software Engineering (cs.SE)
备注: 16 pages, 3 figures, 1 table. Accepted at the 24th International Conference on Service-Oriented Computing (ICSOC 2026)
Abstract:Increasingly, users express their needs as natural-language intents rather than structured requests. These intents may be incomplete, high-level, or internally conflicting. A conflicting intent either violates fundamental properties of the cloud paradigm or states mutually inconsistent requirements. Classical composition approaches are designed for structured, low-level, and complete requests. They cannot interpret such intents, discover suitable services from provider catalogs, or negotiate conflicting requirements with users. To address this gap, we propose a negotiation-aware framework for intent interpretation and service discovery. The framework first refines the natural-language intent and dynamically detects conflicts. It then structures the intent into a provider-independent TOSCA template. For each specification, it retrieves candidate service offers through deterministic matching grounded in cloud provider offerings. Some conflicts arise during interpretation. Other requirements prove unsatisfiable during discovery. Both are resolved through iterative natural-language negotiation with the user. We evaluate the framework on 70 diverse intents against single-shot and enhanced-prompting baselines. It improves recall and F1 for both conflict detection and service discovery. It also eliminates hallucinated offers and resolves conflicts at a cost proportional to their number.
[AI-30] FARE: Factor-Aware Retrieval with Error-Guided Fusion for Retail Forecasting
链接: https://arxiv.org/abs/2610.40160
作者: Chenghao Gu,Zhuokang Shen,Shenghao Wu,Tingxuan Zhu,Weiyi Luo,Haifeng Tang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 16 pages,3 figures
Abstract:Retail sales forecasting underpins inventory planning, yet promotions and calendar events reshape short-term demand. Although each factor usually has historical occurrences, the complete factor combination of a target week often has few precedents. The challenge is therefore to retrieve informative partial matches and weight them by their reliability, which varies across series. We propose FARE(Factor-Aware Retrieval with Error-Guided Fusion), which ranks historical weeks separately for each known target-week factor and merges the selected weeks into a factor reference under a fixed budget. Guided by shrunk errors from completed development weeks and the current disagreement among candidates, a first gate mixes this factor reference with a shape reference from similar sales histories, and a second gate controls how strongly the mixture corrects a backbone forecast. Both gates remain fixed during testing. On Favorita, M5, Rossmann, and Rohlik, FARE reduces the weighted absolute percentage error (WAPE) of five backbones by 11.46%, 4.06%, 14.63%, and 10.16% on average. With Chronos-2 frozen, FARE also outperforms conventional and existing enhancement methods, including supervised residual correction and retrieval augmentation. Code is available at this https URL .
[AI-31] he generative paradox: the Anti-Prompt paradigm for human capability
链接: https://arxiv.org/abs/2610.40151
作者: Margarita Trifonov Vadim E. Levit
类目: Artificial Intelligence (cs.AI)
备注: 17 pages, 2 figures, 2 boxes
Abstract:Generative AI in education is usually judged by the output of human-AI teams. That focus overlooks a second outcome: what people can still understand, challenge, repair, and adapt when assistance changes or disappears. We call this tension the generative paradox and distinguish four manifestations: performance, assistance, ownership, and capability stock. We propose developmental function allocation, which complements operational function allocation and meaningful human control by treating retained human capability as a design objective alongside current joint performance. The Anti-Prompt paradigm reverses the default interaction order: when immediate generation would preempt a consequential human act, AI should first create the conditions for human construction. We distinguish the validity of a product, responsibility for its construction, and evidence of capability after support changes. Representative Anti-Prompts, a canonical interaction sequence, and stress tests spanning mathematical reasoning, experimental science, software diagnosis, and lesson design make these distinctions concrete. We conclude with falsifiable predictions about retained capability and a system-level hypothesis about the reproduction and distribution of expertise.
[AI-32] DynaRev: Benchmarking Web Agents under Dynamic Goal Revision
链接: https://arxiv.org/abs/2610.40133
作者: Shiqi Zhang,Guochen Zhou,Zesheng Wei,Hai Zhao,Zhisong Zhang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Web agents are typically evaluated with goals fully specified before execution and fixed thereafter, whereas real users often revise their requirements during interaction. To study such interactions, we introduce DynaRev, a benchmark of 181 tasks to evaluate web agents under dynamic goal revision. DynaRev uses semantic milestones to deliver revisions at task-defined stages, covers additive revisions, unconditional overrides, and runtime-conditioned overrides, and evaluates correctness, responsiveness, and efficiency through trajectory-grounded evaluation. Five-model evaluations show that agents often satisfy only part of the task requirements, with performance varying across task sources, revision types, and models. Further analyses reveal varied milestone timing, mixed effects of dynamic delivery, and contrasting revision-handling outcomes, with gains from instruction rewriting while preserving browser state and losses from browser resets. DynaRev provides a testbed for studying how web agents adapt to evolving user goals. The code and data are available at this https URL.
[AI-33] Escaping Path Mirag es in Offline Goal-Conditioned Reinforcement Learning NEURIPS2026
链接: https://arxiv.org/abs/2610.40130
作者: Junhyeon Bae,Jaebak Hwang,Gwanwoo Choi,Minung Kim,Seungyul Han
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 10 pages (main), 34 pages in total. Accepted to NeurIPS 2026
Abstract:Offline goal-conditioned reinforcement learning remains challenging in stochastic long-horizon settings, where compounding value estimation errors hinder reliable goal reaching. While prior methods have sought to address this challenge through various approaches, a key challenge remains path mirage, where agents overcommit to spuriously successful trajectories in offline datasets induced by stochasticity, often leading to failure on difficult tasks. To address this issue, we propose Branch-aware Graph Planning (BGP), which captures stochastic branching structures by identifying high-variability points as graph nodes and designing goal-conditioned edges with segmented learning to avoid unreliable branches and reduce unnecessary stochasticity along paths. As a result, BGP enables reliable goal reaching even in highly stochastic long-horizon environments, and experiments on diverse OGBench tasks show that it substantially outperforms prior state-of-the-art offline goal-conditioned RL methods. Our implementation is available at this https URL.
[AI-34] Reward-Adaptive Co-Evolution of Policy and Rubric
链接: https://arxiv.org/abs/2610.40096
作者: Yizhu Chen,Jianguang Jiang,Jiameng Huang,Chuyu Fang,Bojun Zhang,Zhenyu Jiao,Xing Yuan,Tong Lin
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Rubric-based rewards provide fine-grained, interpretable supervision for open-ended reinforcement learning, where holistic scalar rewards struggle to capture multidimensional response quality. Static rubrics can lose discriminative power as the policy distribution evolves, while existing policy–rubric co-evolution methods rely on globally prescribed update schedules that ignore heterogeneous reward informativeness. We introduce RACE, a Reward-Adaptive Co-Evolution framework. At the group level, per-rollout-group training-signal routing classifies groups as informative, diversity-collapsed, or rubric-failed, updating the policy only with informative groups while caching rubric failures. At the phase level, event-triggered phase switching enters rubric optimization upon persistent reward saturation and returns when rubric-reward improvement plateaus. Rubric phases train the generator on cached pairs using a frozen judge and a scale-normalized pairwise-margin reward. Across both levels, online threshold adaptation adjusts routing and switching thresholds through proportional control. Under a unified evaluation protocol, RACE achieves the highest average performance and outperforms the strongest fixed-alternation baseline on all eight tasks, while improving rubric discrimination.
[AI-35] Learning Safe Flow Matching Robot Policies from Binary Feasibility Oracles
链接: https://arxiv.org/abs/2610.40075
作者: Jianming Ma,Rongjun Jin,Yizhi Chen,Yue Gao
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注:
Abstract:Robot foundation models increasingly rely on flow-based action generation for general-purpose manipulation, yet the generated actions are not guaranteed to satisfy physical constraints such as joint limits and collision avoidance. This problem is especially challenging when the feasible set cannot be described analytically and is accessible only through black-box feasibility oracles. We study this practical oracle-only setting and propose an oracle-guided differentiable safeguard for safe flow-matching robot policies. The safeguard uses binary feasibility queries to correct unsafe updates in the forward pass, while extracting local boundary information to provide constraint-aware learning signals in the backward pass. Combined with discrete-time flow matching, the same safeguard is used during training and inference, enabling forward–backward aligned policy learning without explicit or differentiable constraint models. Experiments on obstacle-aware robotic manipulation demonstrate 100% constraint satisfaction while preserving task performance.
[AI-36] Agent ic process evidence
链接: https://arxiv.org/abs/2610.40071
作者: Yuval Kainan,Carmit Hershman
类目: Artificial Intelligence (cs.AI); Software Engineering (cs.SE); Systems and Control (eess.SY)
备注:
Abstract:The way software gets built is changing dramatically. AI agents now create pull requests, select dependencies, refactor codebases, interpret test failures, and execute infrastructure changes across the entire Software Development Life Cycle (SDLC). AI agents are autonomous team members, but unlike human engineers, they are ephemeral - their reasoning, decisions, and tool calls vanish the moment the session ends. Modern Security incidents involving intent mismatch or injection are becoming increasingly common - forcing us to address AI Agents as an insider threat. No widely adopted standard yet enables agentic provenance and governance and its obvious that you can’t hold an ephemeral process accountable. The git log was designed for human commits. It shows what changed. It does not show what an AI agent did, which model it used, which tools it called, or whether anyone was accountable for the outcome. Every governance framework, every audit trail, every compliance control in the SDLC was built around one assumption: a human made the decision. That assumption no longer holds. This shift introduces 3 main procedural problems (not addressing scale) - Supply-chain vector - The agentic coder is software, so it is a supply-chain security risk. Policy guardrails - How to ensure the agentic coder is compliant with the policies of the organization. Task-alignment - How to align the agentic coder on its tasks and goals. We introduce Agentic Process Evidence (APE), a signed, tamper-evident, machine-readable attestation of agentic processes, linking the sessions to their outcome (process-aware logging). APE enables Policy-as-Code control over the agentic processes, allowing for human in the loop only when it’s needed. We support the in-toto style attestation framework, and the protocol is aligned with the relevant regulations
[AI-37] UPPAAi: Towards LLM -Assisted Modeling Verification Analysis and Synthesis
链接: https://arxiv.org/abs/2610.40042
作者: Peter Dolog,Kim G. Larsen,Marius Mikučionis,Frederik F. K. Rasmussen
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI); Logic in Computer Science (cs.LO)
备注: Submitted to IFS/ETAPS 2027
Abstract:For more than three decades, UPPAAL has supported the modelling and analysis of real-time and cyber-physical systems. Yet building a suitable model, formalising the intended properties, choosing analysis techniques and interpreting counterexamples remain iterative tasks that demand substantial expertise. Large language models (LLMs) can offer plausible explanations and even solutions, but they are not necessarily correct. We present UPPAAI, an extension of UPPAAL in which users construct, explain, debug and specify models in dialogue with an LLM. The guiding principle is that the user requests, the LLM proposes, explains and orchestrates, and UPPAAL verifies. UPPAAI exposes the UPPAAL editor, queries and verifier through a Tools API. We report on seven case studies, from explaining to a newcomer to learning a controller, covering timed automata, stochastic hybrid automata and strategy learning. LLMs proved to be extremely helpful in not just generating solutions but also pro-actively suggesting relevant UPPAAL features it was not asked about.
[AI-38] Hybrid AI Patterns: Thirteen Ways to Couple Language Models with Rules and How to Choose Among Them
链接: https://arxiv.org/abs/2610.40036
作者: Siegfried Handschuh,Götz-Henrik Wiegand,Lorena Raichle
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Language models handle text well, and rule engines give conclusions with proofs. Systems that combine the two are usually described by example, and the same coupling turns up under different names in agent frameworks, guardrail toolkits and neuro-symbolic research. This document defines thirteen patterns for coupling a language model with an executable rule language, in the catalogue form introduced for object-oriented design: intent, motivation, applicability, participants, collaborations, consequences, implementation, known uses and related patterns. Four patterns concern what crosses the boundary between model and rules, five who controls whom, three how the model takes part in the reasoning, and one how knowledge is kept over time. Five invariants hold for every combination, and five failure patterns name the couplings that break them. We relate the catalogue to published pattern catalogues and taxonomies of hybrid and neuro-symbolic systems and show where they overlap with it. A decision guide leads from a problem to a pattern. It rests on our measurements, including the cases in which a model alone did as well as the hybrid or better, and it reports what each design cost.
[AI-39] Cheap Retrieval and Observation Credit Assignment from One Quantity: The Value of Information
链接: https://arxiv.org/abs/2610.40030
作者: Daniel Strauss,Nikolaos Sotirakis,Zhenshan Bing,Alois Knoll
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 11 pages, 1 figure. Accepted at the CoRL 2026 Workshop on Memory for Robot Foundation Models ( this https URL )
Abstract:Memory in reinforcement learning raises two problems. Attention finds the relevant item by scoring the whole store, which is costly for robots under latency and energy limits. When a retrieved memory pays off, credit for acquiring it must cross the whole sequence. We argue that both are governed by one quantity, the value of information, and propose to estimate it from archived rollouts with a learned world model. We obtain that, when read at the observation, it yields a provable lower bound on the optimal value of the acquiring action, which floors its TD target and attributes this floor to the observation that earned it. Read at the decision, it defines the optimal retrieval set. A pilot retrieval system, to be trained and improved with this signal, solves a memory task with reduced context.
[AI-40] HEMO: Harness Self-Evolution for Multi-Objective Combinatorial Optimization
链接: https://arxiv.org/abs/2610.40023
作者: Jianghan Zhu,Cong Zhang,Ni Zhang,Yifan Yang,Yew Soon Ong,Zhiguang Cao
类目: Neural and Evolutionary Computing (cs.NE); Artificial Intelligence (cs.AI)
备注: 48 pages, including supplementary material; 5 figures and 13 tables
Abstract:Automatic operator design (AOD) has increasingly adopted large language models (LLMs) to generate problem-specific search and variation operators for multi-objective combinatorial optimization (MOCO). Yet most methods evolve the operators while leaving the process that discovers them fixed. This limits adaptation as new Pareto trade-offs and operator interactions change where and how the search should proceed. We introduce HEMO, which makes the AOD harness itself a self-evolving optimization target. HEMO maintains a persistent harness state and updates one module at a time, revising the logic that determines which experience to retrieve, which operators to improve, which parents to select, and how to formulate the orchestration of search steps. Downstream solution quality and runtime guide these updates, while a discounted UCB scheduler directs adaptation toward promising modules as their utility changes. A persistent memory filesystem preserves candidate outcomes and diagnostic feedback across harness states, enabling successive states to build on accumulated optimization experience. Experiments across eight MOCO tasks and two structurally distinct evolutionary backbones demonstrate strong performance against conventional evolutionary, neural, and LLM-based AOD baselines, together with effective transfer to related tasks. HEMO extends automatic operator design beyond evolving individual operators to improving the search process that discovers them.
[AI-41] Efficient Learning of Mechanism Designs and Compositionality
链接: https://arxiv.org/abs/2610.40005
作者: João Pedro Gandarela,Thiago Rios,Stefan Menzel,André Freitas
类目: Artificial Intelligence (cs.AI); Computational Engineering, Finance, and Science (cs.CE)
备注:
Abstract:Learning mechanism behaviour requires a representation of how connected parts jointly constrain motion. We study whether exposing reusable constraint subassemblies improves prediction from limited data. Assur-KB represents planar mechanisms through irreducible modules, boundary interfaces, and solve dependencies while retaining source geometry. Our theoretical results characterize transfer through module interfaces and establish physical limits arising from missing interface information and accumulated sensitivity. To test the representation claim directly, the experimental design compares matched mechanism-curve pairs across training budgets within the same multilayer-perceptron family. This controlled comparison establishes the utility of the complete representation in the evaluated learning setting. Assur-derived features provide a geometric-mean 1.8\times advantage in sample efficiency over flattened adjacency and position inputs across budgets of 50-10,000 mechanisms. Separate experiments establish invariance under tested joint relabelings, higher valid-generation rates under dyadic composition, and a coarser structural index over 10.02 million mechanisms. In summary, the theory and experiments identify how physical constraint structure supports limited-data prediction and which interface conditions govern compositional reuse.
[AI-42] Accumulating Pruning Constraints: New Exact Polynomial-Space Algorithms for Max 2-SAT
链接: https://arxiv.org/abs/2610.39992
作者: Konstantin Kutzkov
类目: Data Structures and Algorithms (cs.DS); Artificial Intelligence (cs.AI)
备注: Project folder with Lean verification: this https URL
Abstract:We present deterministic polynomial-space algorithms that solve Max 2-SAT on n variables exactly in time \Ostar(2^(1-\gamma)n) , for a fixed \gamma0 , on several classes of instances whose clause density may grow with n . Our main tool extends the classic DPLL branching algorithm. When the search gives a variable its less promising value, comparing the two values of the variable yields a pruning constraint that is satisfied by some optimal assignment. Accumulating these constraints along each branch prunes all but an exponentially small fraction of the assignments. The classes are formulas with many biased variables, semirandom 2-CNF formulas, and Max-Cut on regular expander graphs and on graphs generated by a stochastic block model. In a semirandom formula, the pairs of variables that may share a clause are arbitrary, and the clauses on these pairs are random. For semirandom formulas and block-model graphs, the algorithms are still exact on every input, and the time bound holds with high probability over the input. To our knowledge, these are the first polynomial-space algorithms with such a running time for nontrivial classes of dense Max 2-SAT instances. Beyond these classes, the pruning constraints can be added to any branching algorithm for Max 2-SAT, and we expect them to be of practical relevance for general formulas, in particular for sparse ones, where each constraint involves only a few variables.
[AI-43] When to spend privacy: optimal allocation in private learning
链接: https://arxiv.org/abs/2610.39954
作者: Umberto Biccari,Yongcun Song
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Optimization and Control (math.OC)
备注:
Abstract:Privacy accounting bounds the cumulative privacy cost of Differentially Private Stochastic Gradient Descent but does not determine how to allocate a fixed budget across iterations. We formulate the complete Gaussian noise-variance schedule as an open-loop control under a composed zero-Concentrated Differential Privacy constraint, explicitly accounting for schedule selection. For expected quadratic terminal tracking, we prove that every local minimizer exhausts the certified budget, yielding an exact-budget simplex formulation. We introduce Privacy-Budget Adjoint Mirror Descent (PB-AMD), combining pathwise adjoint gradients with entropic updates that preserve positivity, charge floors, and the prescribed budget. Under regularity and coercivity assumptions, we establish existence and budget stability of optimal schedules, design-to-training utility stability, and deterministic, stochastic, and sample-average stationarity guarantees. A linear-quadratic reduction yields an explicit square-root allocation law. Experiments demonstrate tracking improvements over several allocation rules and predictive improvements over uniform allocation in a learning-rate restart setting, where learning-rate allocation performs similarly. Comparisons at a common stationarity tolerance show lower schedule-optimization times than projected gradient descent on the tested problems. The framework connects certified privacy accounting with optimization through nonlinear training dynamics.
[AI-44] CovFedAvg: Local Low-Loss Curves and Federated Model Selection
链接: https://arxiv.org/abs/2610.39902
作者: Federico Lucchetti,Marcus Völp
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Clients observing different parts of a prediction task may learn locally equivalent models with incompatible behavior outside their observed support. We study whether redundant models permit local low-loss paths toward a compatible shared model. CovFedAvg retains local models and curve controls on clients, fits a common endpoint through aggregated gradient contributions, and selects an actual candidate model using distributed training losses. No centralized calibration examples or client-to-client sample exchange are required. Exact selection guarantees no higher evaluated training loss than the previous model and the current-round FedAvg candidate; this is not a global-optimality, stationarity, generalization, or differential-privacy guarantee. We state conditional client-compatibility and approximate-selection bounds and retain corrected transformation identities and a separate full-gradient safeguard theorem for reference algorithms. A complementary input-support example recovers the shared absolute-value regression, while matched direct fine-tuning also solves it. Across a synthetic support task, recorded activity-recognition subjects, and participating and future gas-sensor batches, we evaluate five methods, five seeds, two network widths, and 20 rounds without calibration data. At width 16, CovFedAvg raises participating-Gas macro balanced accuracy from FedAvg’s 88.88% to 89.84%; matched distributed fine-tuning reaches 91.65%. Results support feasibility and modest gains over FedAvg, but do not establish a curve-specific advantage over equally costly optimization. Exact-sum numerical experiments establish data flow; end-to-end privacy requires additional mechanisms and analysis.
[AI-45] KARMA: Training-Free Knowledge-Atomic Harness Evolution for Industrial Advertising Agents
链接: https://arxiv.org/abs/2610.39884
作者: Xinhua Miao,Yuanzhen Rong,Weiru Zhang,Xiaoyi Zeng
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Industrial advertising agents must continually adapt their decision knowledge to changing product offerings, traffic patterns, and operating policies. We present KARMA (Knowledge-Atomic, Regression-Gated Modular Adaptation), a training-free framework that evolves six typed knowledge modules while keeping model weights and executable harness code frozen. During our evolution, a deterministic Trace Verifier maps observed errors to candidate modules, and a Critic compares successful and failed trajectories to guide a tool-constrained Evolver. The Evolver applies atomic knowledge edits, which a fixed validation gate retains only when their measured gain reaches a threshold; otherwise, the previous knowledge state is restored. We evaluate KARMA with three LLM families on cases sampled from a corpus of approximately half a million real-world operation logs across six e-commerce domains. Across 18 domain–model settings, KARMA raises the average hold-out score from 30.19 to 39.15 (+8.96 points, +29.7%). Ablations show that removing either successful or failed trajectories more than halves the average hold-out gain. Evolved knowledge also transfers across domains without further adaptation, yielding an average hold-out gain of 8.54 points across all source–target pairs. By linking retained patches to their diagnostic evidence and validation outcomes, KARMA supports inspectable, reversible knowledge adaptation for agents in Web marketplaces.
[AI-46] uring LLM Post-Training: Calibrated Distillation and Efficient Dynamic-MoE Execution
链接: https://arxiv.org/abs/2610.39876
作者: Yue He,Jiahui Hu,Shaman Tang,Hua Zhou,Qiman Wu,Menglin Li,Yizhao Wang,Yuheng Zhang,Da Zhu,Zhou Wang,Anhua Liu,Hanlin Chen,Mingyuan Wang,Rui Xin,Yuhua Wei,Shuang Su,Siqi Liu,Kai Yang,Weichao Huang,Hang Zhang,Honggou Yang,Xianming Liu
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:We study post-training for dynamically routed mixture-of-experts (MoE) models, focusing on integrating independently optimized specialists while keeping consolidation efficient. Our approach combines calibrated multi-teacher on-policy distillation (C-MOPD), which matches each selected teacher on a local Top- K support with teacher-wise loss normalization and clipping, with an execution path that uses post-capacity bounded-width dispatch, final-route replay, and asynchronous rollout–training overlap. On Turing-20B-A2B, the selected model achieves a four-domain normalized integration score of 0.9988 and retains at least 90.16% of the measured specialist gain in every domain. At matched K=32 , the full calibration objective scores 0.9962, compared with 0.9724 without calibration and 0.9834 without clipping. At a matched 8K response cap, the model exceeds both shared SFT and Qwen3-8B on all four reasoning-enabled mathematics and code benchmarks (unweighted mean: 85.92, versus 80.33 for Qwen3-8B and 74.21 for SFT), and improves over SFT on ten of twelve broader evaluation metrics. In a PPU-based profile, the combined execution configuration reduces consolidation step time from 398.7 to 139.3 seconds, an observed 2.86 \times speedup over serial execution. These results demonstrate strong specialist integration alongside more efficient consolidation, while reflecting the complete training and execution framework rather than any single component.
[AI-47] Exploring the Potential of Symbolic Regression in Precision Livestock Farming
链接: https://arxiv.org/abs/2610.39839
作者: Elisabeth Mayrhuber,Maciej Oczak,Michael Iwersen,David Brunner,Viktoria Dorfer,Stephan Winkler
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:The ability to predict farrowing time in pigs is crucial for optimal farm management and ensuring piglet survival. This study investigates the potential of explainable artificial intelligence for precision livestock farming, focusing on acceleration sensor data from sows around the time of farrowing. Accurately predicting farrowing allows farmers to balance both, piglet safety by avoiding piglet crushing events and the mother sow’s welfare, by avoiding early crating. Traditional machine learning models often lack interpretability (``closed-box’’ problem). Explainable artificial intelligence offers a solution by revealing the decision-making processes within the models. This work is split into two stages. First the detection of nest building behaviour and second the prediction of the remaining time to farrowing. The nest building detection was possible for 166 out of 180 sows and successful within 48 hours before farrowing for 149 sows. The nest building detection is purely based on the change of a sow’s activity level and uses no machine learning techniques. Further, this study utilizes symbolic regression, an open-box modelling technique that generates interpretable mathematical formulas, to predict the remaining time until the onset of farrowing. The results are compared with closed-box models, like support vector regression and artificial neural networks, for both explainability and performance. Notably, symbolic regression achieved a mean absolute error of 9.28 hours for the time-to-farrowing prediction, outperforming the closed-box models in terms of performance and interpretability. This research demonstrates that explainable artificial intelligence, specifically symbolic regression, can offer both interpretable results and competitive performance in precision livestock farming.
[AI-48] SHPRR: An Algorithm for Pure Exploration under Constrained Regret
链接: https://arxiv.org/abs/2610.39837
作者: Georgina Nouli,Flore Sentenac,Maximilian Schiffer
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Best Arm Identification (BAI) conventionally measures the cost of experimentation by the number of samples collected. In many operational settings, however, the cost of exploration relative to a trusted baseline already in place is equally important. We study BAI under both a fixed sampling horizon and a regret budget. Unlike generic resources considered in constrained pure exploration, regret has a distinctive structure: sampling better arms is less costly relative to the baseline. We theoretically show that the regret cost of exploration decreases across the phases of Sequential Halving, a widely used pure-exploration algorithm. This observation yields improved guarantees on the probability of identifying the best arm and motivates a more effective allocation of the regret budget. We propose Sequential Halving with Proportional Regret Rationing (SHPRR), which front-loads the regret budget to allow more exploration when many arms remain. Experiments on synthetic and real-world instances show that SHPRR improves best-arm identification performance over uniform regret allocation and existing baselines.
[AI-49] AegisFlow: Training-Free Non-Myopic Path-Safe Guided Flow Matching
链接: https://arxiv.org/abs/2610.39825
作者: Kunpeng Liu,Boshi Zhang,Keyou You
类目: Artificial Intelligence (cs.AI)
备注: 18 pages, 6 figures
Abstract:Pretrained generative models, such as diffusion and flow matching models, have demonstrated exceptional capabilities in synthesizing high-quality data. However, steering their generative trajectories to satisfy strict constraints–such as identity preservation in image editing or obstacle avoidance in robotics–remains fundamentally challenging. While recent training-free guidance methods enforce path-safety, they rely heavily on min-norm controllers that only account for immediate constraint satisfaction. These myopic approaches often yield suboptimal trajectories that hover dangerously close to the boundary, severely restricting the model’s ability to optimize toward its target. In this paper, we propose AegisFlow, a training-free, non-myopic path-safe guidance framework that recasts guided generation as a safety-critical optimal control problem with global scope. By integrating a time-varying barrier function directly into the global cost functional, AegisFlow naturally enforces constraints along the entire generative path. To compute the global optimal control, we derive a tractable, mathematically grounded approximation of the Hamilton-Jacobi-Bellman (HJB) equation. Comprehensive experiments demonstrate that AegisFlow outperforms state-of-the-art methods across text-guided image manipulation and robotic planning, establishing a significantly better trade-off between target alignment and constraint satisfaction. Code is available at this https URL.
[AI-50] EvoTeam: Multi-Agent Collaboration via Evolving Teams
链接: https://arxiv.org/abs/2610.39813
作者: Qinuo Cao,Xin He,Dianxuan Fu,Lizhi Zhang,Zhaoqing Xing,Jiapu Wang,Xin Wang,Na Zou,Yi Chang,Qinggang Zhang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:LLM-based multi-agent systems (MAS) rely on collaboration among specialized agents to solve complex tasks. Existing methods improve MAS by refining the prompts, skills, and execution strategies of individual agents based on task feedback. However, improving individual agents does not necessarily lead to better team collaboration. Recent studies have therefore extended evolution to the collaboration level, using collective execution experience to adjust team organization, and inter-agent interactions. Nevertheless, evolving collaboration across heterogeneous tasks remains challenging, as existing approaches often constrain team reorganization within a shared structure and reuse interaction rules across incompatible team contexts. In this paper, we propose EvoTeam, a collaboration-centric evolution framework that jointly evolves team organization, interaction rules, and individual agent skills through execution experience. It maintains a library of reusable teams, selecting or creating agent teams and guiding collaboration through team-specific interaction rules. Collaboration experience accumulates within compatible team contexts, while individual skills remain reusable across teams. Experiments across four benchmarks demonstrate its effectiveness. Our code is available at this https URL.
[AI-51] InferD: A Honeypot Framework for AI Inference Infrastructure CCS2026
链接: https://arxiv.org/abs/2610.39807
作者: Hicham Hammouchi,Gabriele Lenzini
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注: AISec Workshop '26, ACM CCS 2026, The Hague, Netherlands
Abstract:Large language model serving stacks, commercial inference APIs, vector databases, and agent orchestration gateways are now ex- posed on the public internet and indexed by commodity scanners within hours of deployment. The threat characteristics of this sur- face are only beginning to be measured, and the honeypot frame- works that could measure it were designed for conventional net- work services rather than for AI infrastructure. In fact, emulating an AI service is a distinct deception problem: the observable behavior of the service, including model inventories, token formats, tool-call protocols, and authorization flows, must be reproduced faithfully, while the underlying computation, model inference and tool ex- ecution, must never occur. We present InferD, a low-interaction honeypot framework organized around a single principle we call deception without inference: every request an adversary can observe is answered from static response pools, and nothing an adversary submits is ever executed. InferD decomposes the emulation into four deception planes, transport, service protocol, vulnerability surface, and agentic session, adds a provenance plane that links a planted credential to evidence of downstream model-mediated processing.
[AI-52] Revisiting Over-refusal in Safety-aligned Large Language Models : A Representation and Optimization Perspective
链接: https://arxiv.org/abs/2610.39802
作者: Chao Chen,Pengyang Shao,Wei Qin,Shijie Hao,Fei Shen,Richang Hong
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Over-refusal is a critical side effect of safety alignment in large language models (LLMs), causing harmless but superficially risky queries to be mistakenly rejected. Despite the effectiveness of existing mitigation methods, why safety-aligned LLMs may still reject harmless queries under correct safety supervision remains insufficiently understood. To address this gap, we analyze over-refusal through the lens of representation and optimization. We make two observations: 1) the representations of harmful and over-refusal samples are visually distinguishable at certain intermediate layers but increasingly overlap in deeper layers; and 2) harmful-sample gradients exhibit high intra-class similarity and concentrate in a low-rank subspace. Moreover, they are more strongly aligned with refusal-target gradients than with correct helpful-target gradients on over-refusal prompts. This imbalanced gradient coupling can bias over-refusal samples toward refusal, while the reduced apparent representational separation in deeper layers may make it more difficult for the model to maintain a clear boundary between harmful and over-refusal samples. To this end, we propose GOAL (Gradient Orthogonalization with contrAstive Learning), a safety-alignment framework for mitigating over-refusal. Specifically, GOAL consists of two components: 1) to prevent imbalanced gradient coupling from biasing the optimization of over-refusal samples toward refusal, we orthogonalize over-refusal gradients against the low-rank refusal subspace extracted from harmful gradients; and 2) to explicitly enhance the separation between harmful and over-refusal representations, we introduce a contrastive learning objective. Extensive experiments across multiple models and benchmarks demonstrate that GOAL effectively mitigates over-refusal while preserving safety performance and general capabilities.
[AI-53] Finding Is Not Returning Returning Is Not Reconstructing: A Multi-Dimensional Evaluation of Agent Social Re-entry
链接: https://arxiv.org/abs/2610.39781
作者: Sheng Wan,Cong Ma,Sichen Dong,Qian’Ang Mao,Chenhan Zhang,Daning Hu,Qiang Yang
类目: Artificial Intelligence (cs.AI); Social and Information Networks (cs.SI)
备注:
Abstract:As Large Language Model (LLM) agents increasingly participate in persistent social environments, understanding their long-term social interaction capabilities becomes increasingly important. While existing research has advanced long-term memory and conversational context retrieval, less attention has been paid to whether agents can autonomously re-engage with previously inactive conversations. In particular, remembering a past interaction does not necessarily mean that an agent can identify when to revisit it, decide to return, or reconstruct the context needed for a coherent continuation. To investigate this gap, we introduce agent social re-entry as a multi- dimensional evaluation problem, distinguishing three complementary capabilities: Finding, Returning, and Reconstructing. We develop a shared-snapshot evaluation framework that measures these dimensions independently, allowing us to examine how conversational resurfacing and persistent memory relate to different aspects of social re-entry. Evaluations across six LLM configurations show that resurfacing substantially improves target conversation identification, while persistent memory is associated with higher rates of autonomous target-directed replying. However, neither improvement is accompanied by consistent gains in contextual reconstruction. These results highlight the distinction between accessing previous conversations, choosing to re-engage, and maintaining conversational continuity. Our framework provides a way to evaluate these capabilities separately rather than relying on memory retrieval or conversational response quality alone.
[AI-54] Multi-output LSTM soft sensors for industrial anaerobic co-digestion using LSTM-beta-VAE generative data augmentation
链接: https://arxiv.org/abs/2610.39768
作者: Rohit Murali,Benaissa Dekhici,Tao Chen,Dongda Zhang,Michael Short
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Systems and Control (eess.SY)
备注: 33 pages
Abstract:Anaerobic digestion (AD) requires close monitoring of key process variables, including pH, the ratio of volatile organic acids to alkaline buffer capacity and biogas production, to maintain biological stability and prevent process failure. While mechanistic frameworks such as the Anaerobic Digestion Model No. 1 are highly parameterised, computationally expensive, and difficult to calibrate for real-time applications, data-driven Long Short-Term Memory (LSTM) networks offer a potentially more adaptable alternative. However, current data-driven approaches in AD are predominantly limited to single-output architectures that focus solely on production yields, as opposed to simultaneous prediction of vital stability indicators. Furthermore, industrial AD datasets are typically sparse and skewed toward steady-state operations. Baseline evaluations demonstrated this scarcity prevents standard algorithms (Random Forest, XGBoost) and unaugmented LSTMs from recognising the transition to unstable operations, resulting in poor estimation of stability indicators. To overcome these challenges, this study proposes a novel generative data augmentation framework using an LSTM-beta-Variational Autoencoder coupled with a multi-output LSTM (MO-LSTM) soft sensor. By synthesising physically valid sequences of these transitions to unstable operations from one year of full-scale plant data, the method safely populates sparse regions of the feature space. Trained on this augmented dataset, the MO-LSTM achieved robust simultaneous estimation across all targets. This approach mitigates data sparsity, providing a basis for hybrid mechanistic-data-driven modelling and early-warning soft sensors in industrial bioprocesses.
[AI-55] PAPERBENCHX: Can AI Reproduce Science Across Disciplines?
链接: https://arxiv.org/abs/2610.39744
作者: Ruoyu Wu,Zengji Tu,Aolong Sun,Tianyi Ma,Junren Li,Xinbo Xu,Li Fu,Ziyan Zhang,Pinrui Huang,Xiaochen Wang,Chengda Wu,Yutao Zhou,Jianjun Wu,Lin Chang,Baobao Chang,Liang Chen,Kuan Li
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Before AI can be trusted to discover new science, it should be able to reproduce what science already knows. Reproduction provides a rigorous checkpoint for an AI scientist: the target claim is known, yet recovering it still requires translating an incomplete scientific description into an executable experiment and producing evidence that can be independently verified. Existing research-reproduction benchmarks largely focus on AI and machine learning, where experiments share relatively homogeneous software stacks. Extending this capability across scientific disciplines introduces a different challenge: agents must operate domain-specific scientific software, make physical and numerical choices, diagnose long-running computations, and distinguish valid evidence from merely plausible outputs. We introduce PAPERBENCHX, a benchmark of 93 expert-constructed paper-reproduction tasks spanning 12 research areas across electromagnetics, photonics, chemistry, materials science, biology, and robotics. Each task provides the source paper, a public task specification, and a domain-native scientific environment involving specialized scientific software. Agents must reconstruct a key experiment, execute and diagnose it, and deliver a runnable workflow that supports the target scientific claim. To evaluate scientific validity rather than surface-level agreement, we replay every submission from scratch and score only regenerated evidence against expert-validated criteria. Across 10 frontier agent configurations, the best configuration fully reproduces only 14% tasks under the 90% threshold. Meanwhile, mean rubric scores range from 27.53% to 59.08%, indicating that substantial partial progress rarely translates into complete scientific reproduction. These results reveal a large gap between paper understanding and reliably producing reproducible scientific evidence.
[AI-56] Bank-World: A Benchmark for Evaluating Agents on Banking Workflows
链接: https://arxiv.org/abs/2610.39733
作者: Daniel Rupawalla
类目: Artificial Intelligence (cs.AI)
备注: 43 pages, 3 figures, including appendices. Ancillary files contain the task ID mapping, reward specifications, and single-turn check-authority mapping
Abstract:Bank-World is a benchmark of 50 banking tasks across two simulated institutions. Completing banking work requires more than producing the right calculation: agents must reconcile records, obtain permission, execute transactions, and keep instructions and status reports consistent as circumstances change. Bank-World tests these requirements through single-turn assignments and multi-turn interactions with simulated employees who supply information, approve actions, and revise instructions. Grading combines executable checks of recorded actions and state with model judgments of task requirements, including whether valid alternative solutions satisfy them. Across 1,600 attempts by four models, we find that correct decisions often fail to become completed work. Agents charge fees twice, distribute the correct total in incorrect amounts among the recipients, and execute payments while approval is pending. Their strengths also differ: Sol leads single-turn completion, Opus has the highest observed multi-turn completion rate, and Grok succeeds across more multi-turn tasks but less consistently. Bank-World makes these differences inspectable through task-level results and worked examples connecting source evidence, agent decisions, and simulated consequences. We retain the task environments, executable graders, and full trajectories privately to reduce contamination and preserve the benchmark for future evaluations.
[AI-57] MoLRSI: A Recursive Self-Improving AI agent for Molecular Discovery
链接: https://arxiv.org/abs/2610.39714
作者: Yikun Zhang,Xiwei Cheng,Mingze Yuan,Yu Zhang,Brian Shim,Aarti Krishnan,Daniel Rosen,Kunyang Sun,Yingze Wang,Zhenfeng Deng,Botao Yu,Bingxu Liu,David Liu,Tianyu Liu,Yuanqi Du,Nir Hacohen,James J. Collins,Wengong Jin
类目: Artificial Intelligence (cs.AI)
备注: 69 pages,8 figures
Abstract:Molecular discovery requires navigating vast combinatorial search spaces while balancing competing biological objectives. AI agents can automate expert-defined molecular design protocols, but automation alone does not address how to optimize workflows for new tasks or improve their development through experience. Here we introduce MoLRSI, an AI agent that combines systematic optimization of workflows and model architectures with persistent cross-task memory to recursively improve its search policy for molecular discovery. Workflow performance, refinement outcomes, and execution feedback guide subsequent searches, allowing the agent to improve both its workflows and its strategy for developing them. We evaluate MoLRSI in two challenging molecular discovery settings that require balancing competing properties. In antibiotic discovery, MoLRSI developed a screening workflow balancing antibacterial activity, cytotoxicity and structural novelty, achieving an experimental hit rate of 15.9%, compared with 8.7% for a previous expert-led workflow using the same training data and screening library. In de novo peptide-MHC binder design, MoLRSI developed workflows that jointly consider target binding and specificity, producing eight experimentally validated hits against MART-1 and SARS-CoV-1 pMHCs with preferential binding over tested off-targets while Claude Science completely failed. These findings demonstrate how experience-guided workflow optimization can address competing discovery objectives while improving the search strategies used in subsequent campaigns.
[AI-58] Sandman: Building and Operating Sandboxes for Agent ic Reinforcement Learning on Shared Kubernetes
链接: https://arxiv.org/abs/2610.39711
作者: Yiming Yan,Hongyang Wang,Wenbo Chen,Xuan Wang,Jingguo Liu,Xian Li,Chen Qi,Leo Fan,Siwei Li,Qianli Wu,Yichuan Li,Ruochen Wen,Qian Qu,Zheshen Wang,Benoit Dumoulin
类目: Distributed, Parallel, and Cluster Computing (cs.DC); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 42 pages, 6 figures, 53 tables
Abstract:Agentic reinforcement learning (RL) often requires repeated tool calls in task-specific environments. A training job may create thousands of these environments in bursts while sharing a cluster with other jobs. We present Sandman, a sandbox service built and operated at Amazon for agent tool execution and evaluation. It serves multiple production agentic RL workloads on shared Kubernetes. A training framework supplies a task image and uses commands and file operations in the returned sandbox. Files and processes persist across calls until the framework releases the environment. Sandman manages creation, readiness and cleanup. Each sandbox is a Kubernetes pod with a small daemon mounted into the unmodified image. In a scale test, Sandman held 11,858 concurrent tiny sandboxes. Four production SWE-bench Verified training campaigns created 3.1 million sandboxes in October 2026. We report the design trade-offs and operational lessons from handling creation bursts, preparing task environments, reclaiming abandoned sandboxes, and running evaluation under controlled conditions. We will open-source Sandman’s daemon, provider and pool.
[AI-59] Practical Assessment of Agent ic Systems in Acoustics
链接: https://arxiv.org/abs/2610.39685
作者: Arne Hildenbrand,Philipp Diel,Zhuo Cao,Hanno Scharr,Stefan Pischinger,Steffen Marburg
类目: ound (cs.SD); Artificial Intelligence (cs.AI)
备注:
Abstract:In engineering, applications of artificial intelligence have so far been largely limited to surrogate models and optimization tasks. Large language models changed this, as they can write software from a task given in words. Placed inside a harness, a system of components that lets it search for information and execute software, such a model becomes an agent that can take over recurring engineering tasks. This work measures how the model and the elements of the harness shape the performance and behavior of the agent in two common acoustic applications. In the first, the agent detects and locates faults in audio recordings from a vehicle interior with near-perfect accuracy. A benchmark on this task shows that a weak model sets a limit that no harness design can overcome, and that design choices take effect only once the model is capable enough to use them. In the second, a case study on the design of a metamaterial unit cell, the agent reliably finds designs that achieve a specified insertion loss over a given frequency range. The results aim to help the reader understand what each element of an agent contributes and how to design an agent for acoustic tasks.
[AI-60] Rebuilding a Fields Open-Problem Catalog from the Literature: A Two-Layer Knowledge Graph and Why Status Must Stay Proposed
链接: https://arxiv.org/abs/2610.39641
作者: Madhulatha Mandarapu,Sandeep Kunkunuru
类目: Databases (cs.DB); Artificial Intelligence (cs.AI); Digital Libraries (cs.DL)
备注: 9 pages, 1 figure, 7 tables. Code and data: this https URL
Abstract:A research field records its open problems in surveys, keynotes and community reports. Those are prose, written every few years, and nobody can ask them who works on a problem now or whether it was settled last year. We argue the problems should be data, and we build the system that makes them so: a hand-curated problem layer carried in unchanged, under an evidence layer rebuilt from a bibliographic index, joined by typed, provenance-carrying links in one graph database and queryable from one language. A language model does four jobs – per-paper extraction, catalog grounding, per-paper status verdict, and idea generation – each constrained to a candidate set so out-of-set answers are refused rather than stored. We report a study on two curated catalogs, database research (1,053 problems) and AI research (1,000 problems), each with a prior catalog-only graph. The rebuilt graph answers every question the catalog-only graph answered – 47 of 47 on databases, 34 of 35 on AI – plus questions the catalog-only graph refused: recent workers, untouched problems, methods tried, and outside papers bearing on a problem (336 problems, 770 works on databases). The capability we most wanted does not survive its own audit. Of 41 solves and refutes verdicts hand-checked against the papers, 9 are correct, 30 partly correct and 2 wrong; no refutes verdict (0 of 34) disproves an open problem, and only 1 of 7 solves closes the central question. We report that failure in full and draw the design conclusion: a status change must be proposed with its evidence and never applied, making the human reviewer part of the architecture. We close with the data-engineering problems this raises: entity resolution across curated and harvested layers, verifying model status claims, freshness under source quotas, and evaluation without ground truth.
[AI-61] he Real Cost of Context: Predicting Compute Time and Energy Consumption in LLM Prefill and Decode Phases
链接: https://arxiv.org/abs/2610.39621
作者: Lars Krupp,Daniel Geißler,Francisco M. Calatrava-Nicolas,Stefan Gerd Fritsch,Paul Lukowicz,Jakob Karolus
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:The integration of Large Language Models (LLMs) into agentic architectures has substantially increased token consumption, raising concerns about their energy footprint. Yet, estimating the energy consumption of proprietary API-based models remains challenging, and existing approaches focus primarily on generated output tokens while overlooking the impact of the large input contexts characteristic of agentic workflows. To address this gap, we systematically profile compute time and energy consumption across a wide range of prefill (input) and decode (output) token lengths. Using generalized linear models, we show that decode compute time and energy scale linearly with output length, whereas prefill compute time and energy exhibit both linear and quadratic scaling with input length. These findings demonstrate that input context becomes a relevant factor for large-context workloads. Furthermore, we show that compute time serves as a reliable proxy for energy consumption when direct energy measurements are unavailable. Our mapping framework enables precise energy estimation and supports more sustainable deployment decisions for agentic AI.
[AI-62] Letting AI Formulate the Research Question: Autonomous Research Agents
链接: https://arxiv.org/abs/2610.39620
作者: Ko Fujimura
类目: Artificial Intelligence (cs.AI); Computers and Society (cs.CY)
备注: 17 pages, 1 figure
Abstract:Systems such as The AI Scientist have shown that AI agents can carry out and write up research, but in those demonstrations humans chose the research area and selected the outputs to show. We study letting the AI formulate the research question itself, and we report every run. Our system gives an off-the-shelf coding agent an instruction document and two skills: one for formulating the question, in which the agent diverges from recent technologies into problems of other fields, screens its candidates for known failure modes and obtains a critique from a different model; and one for peer review, in which a reviewer from another model family reruns the study from its records before the agent revises the paper. Over 13 development runs, changes to the one or two sentences that set the research area coincided with large shifts in what the candidate questions and the chosen papers were about, which led us to the area of a new technology applied to a societal problem. We froze the system after a pilot run and report five consecutive runs: the pilot, counted by a disclosed amendment to the pre-registration, and four pre-registered runs. All five runs went from question formulation to a revised, re-computable paper, with local computation on a laptop and no research input from humans, and external AI reviewers reading only the papers judged them at or near workshop level. We release the system, the five papers and the records needed to reproduce them.
[AI-63] Emergence of the transformer architecture as the score of entropic laws of random measures
链接: https://arxiv.org/abs/2610.39614
作者: Giacomo Greco
类目: Artificial Intelligence (cs.AI); Probability (math.PR); Machine Learning (stat.ML)
备注: 38 pages, 3 figures
Abstract:We show that the transformer architecture emerges as the score of last-token laws in a natural statistical model of language, where sentences are empirical measures of tokens, a language is a law on P(S^d) , and our prior belief is that sentences carry no structure at all. By the maximum-entropy principle, we select the law closest to this prior in relative entropy among those reproducing an observed sentence-level statistic. We discuss the relevance of the statistic in the posterior interaction between tokens and we build a statistic, able to capture that interaction, with Luce’s choice axiom truncated at the lowest relational order. This procedure gives the softmax choice rule, with the attention cross-moment as its mean parameter. The query–key matrix is then the Lagrange multiplier of a maximum-entropy problem inside each sentence, and the value matrix the multiplier of the constraint over the language, so that the two are untied. Under the optimal entropic law we then show that a transformer head appears in the score of the last-token law. For ordered sentences, causal masking follows from the masked reference choice rule, and with a next-token cross-moment the score equals one causally masked attention head. We also discuss multi-head attention, weight-tied unembedding, and a heuristic reading of residual blocks as projected Langevin steps.
[AI-64] Data Provenance and Ethics in the Curricula of U.S. Health AI Education Programs ALT
链接: https://arxiv.org/abs/2610.39606
作者: Raffaele Giancotti,Francesca Santucci,Amama Anjum,Monique de Villa,Rupsa Mitra,Soham Biren Katlariwala,Vishala Mishra,Leo Anthony Celi
类目: Computers and Society (cs.CY); Artificial Intelligence (cs.AI)
备注: 10 pages, 1 figure, 3 tables. Under review at PLOS Digital Health. Data and supplementary material available at this https URL
Abstract:Failures of deployed clinical artificial intelligence (AI) systems have often originated in how their training data were generated. Whether the programs that train the health AI workforce teach practitioners to examine data provenance is unknown. We conducted a cross-sectional analysis of the publicly advertised curricula of health AI education programs at U.S. universities, identified through a structured multisource web search between April and July 2026. Two investigators coded coverage of data provenance (structural, dataset limitations, or data management only), ethics, and sustainability; collaborative structure; technical prerequisites; and the disciplinary background and gender of named instructors. The primary outcome was provenance coverage at the structural or dataset-limitations level. Of 43 candidate entries, 36 programs at 25 institutional units in 21 universities were included. Ethics was covered by 29 programs (80.6%; 95% confidence interval [CI], 65.0 to 90.2), data provenance by 8 (22.2%; 95% CI, 11.7 to 38.1), and structural provenance by 6 (16.7%; 95% CI, 7.9 to 31.9). None of the seven master’s degree programs met the primary outcome, and no program addressed sustainability. Eighteen programs included team-based work, but only 5 assembled teams across clinical and technical backgrounds. Of 92 named instructors, 76.1% came from computational disciplines and 3.3% from the social sciences or humanities; among the 51 whose gender could be determined, 66.7% were men. U.S. health AI education programs widely taught ethics but rarely taught how clinical data are produced. None addressed the environmental cost of computation, and most instructors came from computational disciplines.
[AI-65] LLM Agent -Based Design of RISC-V CPUs with FPGA-in-the-Loop Evaluation ISCAS2027
链接: https://arxiv.org/abs/2610.39570
作者: Leo Itoh,Junichiro Kadomoto
类目: Hardware Architecture (cs.AR); Artificial Intelligence (cs.AI)
备注: 5 pages, 6 figures, 2 tables. Submitted to IEEE ISCAS 2027. Code and data: this https URL
Abstract:Long-run RTL simulation of CPU candidates can take hours, while short diagnostic runs may miss performance degradation that emerges later in execution. This evaluation cost can delay design decisions in LLM-agent-based CPU development. We present a development environment in which LLM agents implement CPU candidates on an FPGA and use on-board performance measurements to guide design decisions. We applied it to the human-supervised development of a single-cycle RV32I core into a three-wide out-of-order RV32IM core. The final design reaches 5.61 CoreMark/MHz at 130 MHz on a Artix-7 FPGA. For the same CPU candidate running 8,000 CoreMark iterations, FPGA evaluation, including implementation, was 9.6x faster than single-threaded RTL simulation. In one case, board measurements revealed long-run performance degradation in a candidate that improved in a short diagnostic run. The result guided candidate selection and led to RTL-based root-cause analysis and a fix.
[AI-66] Non-Zero Radius: The Missing Component of Neural Networks
链接: https://arxiv.org/abs/2610.39559
作者: Tiansi Dong,Mateja Jamnik,Pietro Liò
类目: Artificial Intelligence (cs.AI)
备注: 5 pages, 7 figures
Abstract:Inspired by the analogy between the all-or-none activation of biological neurons and the true-false of logic circuits, McCulloch and Pitts proposed a mathematical model of the biological neuron in 1943 with the aim of simulating logical reasoning. Although Norbert Wiener considered the model overly simplistic at the time, it later became a building block of modern neural networks, including large language models. Despite their remarkable empirical success, however, conventional neural networks remain largely black-box systems and struggle with reliable symbolic reasoning. Viewing vectors as spheres of zero radius, we relax the zero-radius constraint and represent concepts as spheres with non-zero radii. We argue that sphere-based representations can capture additional features of biological cognition, including representing spatial concepts as regions and reasoning through model construction and inspection. Building on sphere embeddings, we introduce Sphere Neural Networks (SphNNs), which use a transition map to achieve symbolic-level Aristotelian syllogistic reasoning without training data, thereby taking an initial step toward the original goal of McCulloch and Pitts. We suggest that the zero-radius constraint may reflect an aspect of the oversimplification identified by Wiener and may limit conventional neural networks’ capacity for symbolic-level logical reasoning.
[AI-67] Judging Before Propagation: Task-Graph Contextual Integrity for Multi-Agent Privacy
链接: https://arxiv.org/abs/2610.39532
作者: Hengyu Wu,Blaise Delattre,Yang Cao
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注:
Abstract:Multi-agent LLM systems increasingly rely on persistent memory and inter-agent communication. These capabilities create privacy risks that extend beyond individual interactions: a release appropriate in isolation may become inappropriate when combined with information previously disclosed to a recipient or propagated to downstream agents along the planned workflow. Existing Contextual Integrity (CI) mechanisms for agentic systems typically rely on local judgments that assess each disclosure in isolation, with or without regard to the recipient’s history, but do not account for how information propagates downstream. However, compliance at the current boundary does not guarantee appropriateness across the workflow. We introduce Task-Graph Contextual Integrity (TG-CI), to our knowledge the first task-graph-native formalization of Contextual Integrity for multi-agent systems. TG-CI represents a workflow as a Multi-Agent Task Graph (MATG) and extends CI from atomic transfers to stateful, graph-level information flows by jointly considering an item’s intended propagation scope and recipients’ evolving disclosure states. We operationalize TG-CI through propagation-scoped authorization under two complementary policies: Full-Scope, which requires admissibility across the complete downstream scope, and Auto-Region, which confines each item to its maximal admissible region using lightweight frontier gates. We further introduce MAPLE-Bench, a benchmark of 144 cases across four scenarios that jointly vary MATG topology, disclosure history, compromised-agent placement, and item-level privacy policies. In our primary experimental configuration, TG-CI reduces the mean unauthorized disclosure count by 85.8-94.1% relative to the Local-History CI baseline, while retaining 93.6-96.1% of the baseline’s task accuracy, with the two enforcement policies providing complementary privacy–cost–utility trade-offs.
[AI-68] Category-Anchored Large Language Models for Next POI Recommendation ICDE2027
链接: https://arxiv.org/abs/2610.39518
作者: Lexi Bai,Xiaopeng Zeng,Xun Zhou,Yuanshao Zhu,James Jianqiao Yu
类目: Artificial Intelligence (cs.AI)
备注: Accepted by the 43rd IEEE International Conference on Data Engineering (ICDE 2027)
Abstract:Next point-of-interest (POI) recommendation aims to predict a user’s next destination from spatio-temporal check-in data, and serves as the foundation for many location-based services. Recent advances incorporate large language models (LLMs) into this task by textualizing user trajectories to exploit their semantic reasoning ability. However, directly generating POI identifiers with such models is unreliable: POIs are sparse, localized, and long-tailed entities with weak natural-language semantics, which creates a semantic-to-entity gap between high-level mobility intent and concrete POI prediction. To bridge this gap, we propose ANCHOR, a category-anchored framework for semantic-to-entity grounding in next POI recommendation. Specifically, ANCHOR first estimates category-level mobility intent in a compact and stable semantic space, rather than predicting concrete POI entities directly. The inferred intent is then grounded into a candidate set by integrating three complementary sources of mobility evidence: category anchors, historical visit memory, and user preferences. Finally, a history-aware reranker selects the target POI by matching each candidate with the user’s historical mobility patterns and current context. Experiments on three real-world check-in datasets show that ANCHOR consistently outperforms representative traditional, structure-enhanced, and LLM-based baselines on the reported metrics. Further analyses validate the effectiveness of semantic-to-entity grounding and show favorable trade-offs among candidate coverage, ranking quality, and inference efficiency.
[AI-69] DEG: Diffusion-Enhanced GFlowNet for Solving Vehicle Routing Problems NEURIPS2026
链接: https://arxiv.org/abs/2610.39512
作者: Ni Zhang,Zhiqin Zhang,Ling Pan,Hoong Chuin Lau,Zhiguang Cao
类目: Artificial Intelligence (cs.AI)
备注: Accepted by NeurIPS 2026
Abstract:Traditional neural solvers for solving vehicle routing problems (VRPs) often suffer from limited solution diversity, motivating the development of Generative Flow Network (GFlowNet)-based models. However, the effectiveness of these models is frequently constrained by insufficient flow expansion in high-reward regions, limiting their ability to distribute probability across promising solution routes. Deeper exploration could yield superior results. Diffusion models, in contrast, provide stronger structural guidance for exploration. These two paradigms are naturally complementary: GFlowNet can supply edge-level signals for diffusion to embed, while diffusion can guide broader exploration. Leveraging this synergy, we propose Diffusion-Enhanced GFlowNet (DEG), a novel framework that integrates GFlowNet with a diffusion model to encourage richer flow expansion toward high-reward regions and derive higher-quality solutions. Specifically, DEG exploits GFlowNet’s inherent diversity to generate edge-specific backward signals, applies the stochastic noise schedule of diffusion to perturb these signals, and then denoises them within the GFlowNet paradigm. To further improve scalability, we introduce a specialized decoder capable of dynamically adapting to diverse problem scales. Extensive experimental evaluations on synthetic and real-world datasets, including instances with up to 10,000 nodes, demonstrate that DEG consistently achieves favorable performance compared to baseline methods.
[AI-70] Compile Away Uncertainty: Self-Evolving Agent Skills via Programmatic Compilation
链接: https://arxiv.org/abs/2610.39508
作者: Chengyu Yang,Yunxiang Zheng,Zongzheng Li,Jiamei Li,Boxun Lan,Xiaoyu Liu,Xiaoyu Zhang,Xiaohong Chen,Ming Hu
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Self-evolving agent skills enable AI agents to accumulate and refine reusable procedural knowledge from interaction experience without updating model parameters. However, since such skills are typically represented as textual documents and executed through LLM reasoning, even previously learned procedures must be repeatedly interpreted and regenerated, introducing avoidable uncertainty into every execution. In contrast, program execution is deterministic and independently verifiable. Intuitively, transforming well-specified operations into executable programs can reduce reliance on LLM reasoning and, consequently, reduce execution uncertainty. Motivated by this intuition, we introduce SKillCompiler, a self-evolving skill framework that integrates programmatic compilation into the skill evolution process. Specifically, SkillCompiler adopts an executability-aware skill compilation strategy that decomposes skills into fine-grained operations, analyzes their executability, and constructs hybrid skill skeletons that retain LLM reasoning for context-dependent operations while compiling well-specified ones into validated programs. To avoid redundant program synthesis and enable executable knowledge to accumulate across evolving skills, we further design a program asset reuse mechanism that organizes validated programs as reusable assets, retrieves existing implementations whenever applicable, and synthesizes and verifies new programs only when necessary. Extensive experiments on five benchmarks and six model families demonstrate that SkillCompiler consistently outperforms state-of-the-art methods, achieving up to 32.33% improvement in task performance.
[AI-71] What Breaks When Facts Become Stories? The Distortions Source-Agreement Metrics Miss When Models Retell Real Events NEURIPS2026
链接: https://arxiv.org/abs/2610.39500
作者: Kwangsuk Park,JiWoong Yang,Vaniushkina Kseniia
类目: Artificial Intelligence (cs.AI)
备注: 14 pages, 7 figures, 11 tables. Accepted at the Trust-AI-Eval (TAE) Workshop at NeurIPS 2026. Data and code: this https URL
Abstract:Language models are now widely used to turn recorded facts into media for public consumption, yet evaluation of such output still assumes agreement with the source as its measure of success. In reconstruction, where departing from the source can itself be the goal, this criterion measures the wrong construct: applied as is, it counts intended departures as errors; withheld, it lets real distortion through. We had three model families rewrite 2,000 news articles as summaries and as stories, measured what changed on every axis of the five Ws and how, and asked of each detected distortion whether the assigned perspective warranted it, which set a third of them aside. Time and manner were not lost but replaced, while place and causation lost the expressions that carry them. Accountability distortion behaved otherwise. Barely present in summaries, it rose up to sevenfold in stories, and of those whose direction could be determined, two in three lightened the perpetrator’s responsibility. Such distortions survive claim-level fact checking and leave the retelling grammatically cleaner than its source. Where our judges disagreed, three authors relabeling the same cases converged on whether a distortion had occurred but not on whether it was warranted, a boundary contested rather than merely hard to detect. To support work on detecting and reducing them, we release ReTold-2k, a reconstructable corpus with the scoring protocol and per-judge baselines kept unaggregated.(this https URL).
[AI-72] Integrated Feature Selection for classification and Representation Learning with Group-Lasso Siamese Networks
链接: https://arxiv.org/abs/2610.39499
作者: Nisargo Ranjan Pal,Pradeep Singh
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:The Least Absolute Shrinkage and Selection Operator (Lasso) has been widely employed for feature selection in linear models and has also been extended to multi-layer perceptron (MLP) networks for both feature and group feature selection. Group-Lasso has also been used in feature selection using feed-forward neural networks. To the best of our knowledge, this is the first study to incorporate the Group-Lasso principle into a Siamese network architecture for feature selection. Unlike existing approaches that primarily optimize classification accuracy, the proposed method aims to identify a compact subset of features that simultaneously preserves the intrinsic representation of the input data (the goal of Siamese network) and is expected to achieve high classification performance. Two variants of the proposed framework are developed: one based on the Euclidean distance and the other on a modified Jaccard similarity measure for real-valued vectors. The latter is specifically designed to improve feature selection for high?dimensional sparse data. The effectiveness of the proposed approach is demonstrated on 12 benchmark datasets with feature dimensions ranging from 6 to 7070 using eight different classifiers. Experimental results demonstrate that the proposed Siamese Lasso framework consistently identifies compact and discriminative feature subsets maintaining competitive or superior classification performance, although the loss function used explicitly did not consider any regularizer to learn classification. In addition, the quality of the embeddings produced by these networks is assessed using manifold learning techniques such as UMAP and t-SNE.
[AI-73] Does Scaling Toward AGI Depend on What System-1 Lets System-2 Execute?
链接: https://arxiv.org/abs/2610.39497
作者: Qi Wang,Yun Qu,Lizhou Cai,Heming Zou,Haoang Chi,Yiqin Lv,Xiang Li,Haiyan Feng,Mengyu Ge,Yize Geng,Yuhang Jiang,Yixiu Mao,Xiangyang Ji
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Progress toward artificial general intelligence (AGI) has largely relied on scaling a resource-intensive executive, System-2, through larger models, more data and compute, longer reasoning, and more agents. We ask whether the next stage may come instead from co-designing it with System-1, a lightweight, semantically aware predictor that evaluates candidate interactions before execution, so that System-2 acts only on those worth their cost. Our answer is a qualified yes. We study \textitconditional interaction, a predict–gate–execute–update loop: System-1 estimates each candidate’s utility, System-2 executes those admitted by an adaptive threshold, and the outcomes update System-1. The loop spans training, test-time reasoning, adaptive decision-making, and agentic tool use, with utility and cost defined by the application. Under ideal screening and equal execution costs, execution compute falls by a factor approaching 1/\rho_u , where \rho_u is the density of utility-relevant interactions (cost-weighted when costs differ); prediction errors, exploration, and distribution shift reduce the saving. We bound selection regret by predictor error and the utility margin. These results motivate our central hypothesis: \rho_u is a measurable coordinate of scaling efficiency and makes informative compute a better description of scaling than total compute.
[AI-74] Sparse Fine-Tuning for Parameter-Efficient Adversarial Training NEURIPS2026
链接: https://arxiv.org/abs/2610.39455
作者: Zhaoxin Wang,Zihang Ding,Handing Wang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: Accepted by NeurIPS 2026
Abstract:Deep neural networks achieve strong performance across many tasks but remain vulnerable to adversarial perturbations. Adversarial training (AT) is one of the most effective defenses, yet it suffers from high memory cost on large models and the persistent clean-robust accuracy trade-off. Recent work introduces parameter-efficient fine-tuning, such as LoRA, into AT, but this constraint induces a notable robustness gap relative to full-parameter AT. In this work, we revisit parameter-efficient adversarial fine-tuning from a parameter space perspective. Through gradient analysis, we find that adversarial optimization is highly concentrated on a small subset of parameters, which we call Robustness-Critical Parameters (RCPs), suggesting that robustness is encoded in a sparse subspace. Building on this observation, we propose Sparse Adversarial Fine-Tuning (SAFT), which identifies RCPs via adversarial gradient saliency and performs AT by updating only these parameters while freezing the rest. Across architectures and datasets, SAFT outperforms LoRA-based methods in both clean and robust accuracy with fewer trainable parameters, and approaches or surpasses full parameter AT in robustness while updating only about 5% of parameters.
[AI-75] Kairos: Semantics-Aware Orchestration of Planning Engines across Classical Numeric Temporal and Hybrid PDDL IROS
链接: https://arxiv.org/abs/2610.39454
作者: Francisco Martín Rico,Esther Aguado González,Francisco Miguel Moreno Olivo,Rodrigo Pérez Rodríguez,José Miguel Guerrero Hernández,Francisco J. Rodríguez Lera
类目: Artificial Intelligence (cs.AI); Robotics (cs.RO)
备注: 32 pages, 3 figures, 11 tables. Submitted to the Journal of Artificial Intelligence Research. Code: this https URL ; benchmark, scripts and results: this https URL
Abstract:Planning at run time, as in robots that re-plan while they act, needs a planner that reads the PDDL its users write (durative actions, numeric resources, time windows, continuous change), answers quickly, and delivers only validated plans. Specialised planners are strong on individual fragments of PDDL, but they differ in coverage, latency and in what their failures mean. We study whether selecting heterogeneous engines from the semantics of each task can give broad coverage of PDDL, from STRIPS to PDDL+, while remaining competitive with specialised planners and making explicit what each answer guarantees. Kairos profiles each task to decide which of five engines apply (ground heuristic search, compressed and snap-action temporal search, time steps for hybrid dynamics, lifted search) and schedules them under one budget of time and memory. Every candidate plan is checked by a validator independent of the planner’s transformations, and “unsolvable” is reported only when a search whose exhaustion is a proof exhausts its space. On 1,756 tasks from nine comparative suites, Kairos solves more tasks at the full limits than the strongest evaluated planner of each suite, with the largest differences on numeric, temporal, PDDL3 and PDDL+ planning, and it also solves the most tasks at 100 ms and 1 s. Resampling over domains, the gain is above zero in six of the nine suites, while on classical planning against LAMA it is within the noise. A further 450 feature-oriented tasks and 26 one-feature probes show that Kairos handles all seventeen features probed, where no other planner evaluated handles more than eight. Ablations identify compressed temporal search and the novelty fallback as major contributors, and a comparison with VAL traces every disagreement to a PDDL+ convention or to a documented VAL defect. Planner and benchmark are open source.
[AI-76] mbre: Reliability-aware harmonic consensus for label-free multimodal test-time rexinforcement learning
链接: https://arxiv.org/abs/2610.39439
作者: Haotian Lyu,Haoyu Wang,Ru Wang,Feng Ji,Ziwen Li,Jiaxian Guo,Mingming Gong
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Label-free test-time reinforcement learning adapts vision–language models using rewards derived from their own predictions. However, misinterpreted visual evidence and sensitivity to textual formulation can produce a consistent but incorrect consensus that majority voting reinforces. Cross-view aggregation can expose such errors, but image and text transformations may themselves degrade task-relevant evidence to different extents. We introduce \method, a reliability-aware harmonic consensus framework for multimodal test-time reinforcement learning. \method constructs a 2\times2 grid combining original and perturbed images with direct-solving and reframing modes, selects pseudo-labels through weighted harmonic aggregation, and adapts modality-level weights using batch-averaged, temporally smoothed shifts in pseudo-label confidence as a heuristic reliability signal. We formulate consensus as a weighted \chi^2 -barycenter problem whose unique minimizer is the normalized weighted harmonic aggregate, without requiring correct-answer invariance. An exact support–dispersion decomposition characterizes when harmonic aggregation preserves or corrects arithmetic rankings, while a sensitivity analysis establishes how modality reweighting changes pairwise ranking margins, informing interpretation of the heuristic updates. Across six benchmarks and three Qwen3-VL model scales, \method outperforms TTRV in 16 of 18 benchmark–model settings on adaptation instances and 17 of 18 settings on held-out instances under matched estimation and optimization rollout counts. These results demonstrate improved target-task adaptation and transfer to unseen instances without further updates.
[AI-77] ATLAS-Bench: An Agent ic Tool-Use Benchmark for Local Area Search
链接: https://arxiv.org/abs/2610.39438
作者: Keyun Zhang,Dapeng Yin,Fanfu Wang,Shuang Lang,Zhenyu Zhu,Yulin Huang,Chong Yang,Pengjie Wang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Reliable local search is essential for LLM agents to support everyday decisions and longer plans by finding places that meet users’ needs at given times and locations. Yet existing benchmarks provide limited coverage of whether search results satisfy mandatory requirements, account for preferences, and ground factual and experiential judgments in place attributes and reviews. We introduce ATLAS-Bench, a benchmark of 1,200 queries across three regions, expanded from real search events using place attributes, reviews, and existing user profiles from the platform. Each query distinguishes mandatory requirements from preferences and factual conditions from subjective judgments, with supporting evidence and criteria for checking each requirement. Queries are validated through iterative assertion checks and cross-model review. We also introduce ATLASBox, a standardized evaluation environment with search tools and a fixed point-of-interest (POI) knowledge base. Given a query, timestamp, and location, an agent searches place attributes and reviews to recommend evidence-supported options or report that no place meets all mandatory requirements. We evaluate answer quality, trajectory quality, and execution cost. Across 14 configurations of 11 models, GPT-6-Luna achieves the highest constraint satisfaction rate among returned recommendations (71.5%) and factual accuracy (95.6%). However, it returns at least one acceptable recommendation for fewer satisfiable queries than GLM-5.3 (67.3% versus 85.3%). These results highlight the challenge of satisfying individual requirements while providing usable recommendations across queries.
[AI-78] Verifiable LLM -Guided Focal SMT Solving for Quantified Arrays
链接: https://arxiv.org/abs/2610.39435
作者: Kunhang Lv,Rui Han,Fuqi Jia,Yuhang Dong,Feifei Ma,Jian Zhang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Large language models can often identify useful semantic structure in formal artifacts, but their outputs cannot be trusted as proofs or solver decisions. We study whether LLMs can instead guide formal reasoning systems through a narrow, solver-verifiable interface. We instantiate this idea in quantified array SMT solving, where modern instantiation-based techniques can fail when key ground terms and equalities are hidden by auxiliary variables, axiom guards, and nested array terms. We observe that many instances are largely driven by a small set of focal ground terms or constraints, similar to the backdoor sets in SAT solving. Building on this, we present LinguaArray, a framework where an LLM proposes a small Semantic Focus Set (ground terms/constraints), while a conventional SMT solver remains the sole proof engine and certifies all results. LinguaArray uses these LLM-suggested, solver-validated finite sets to significantly enhance solver capability, solving additional satisfiable and unsatisfiable instances that the backend alone fails to solve. It interacts within a finite time and then falls back to the backend solver to mitigate potential regressions. Evaluation on 1015 SMT-LIB instances across five logics under 1200s timeout (including LLM latency) shows that LinguaArray solves up to 69.1% of instances and increases solved-instance counts by 105%–397% over the corresponding backends.
[AI-79] Retaining Repair Hypotheses for Cost-Aware Execution
链接: https://arxiv.org/abs/2610.39410
作者: Jae Hun Hwang,Hong-In Won
类目: Artificial Intelligence (cs.AI)
备注: 17 pages, 5 figures; data and code in ancillary files
Abstract:Several corrections of an executable model can fit the same observations while predicting different execution outcomes. We study a repair-to-execution interface that retains compatible finite guard repairs, conditions their weights on diagnostic and ordinary-execution observations, and uses their predictions to choose routes, priced tests, or a backup. On 24 synthetic two-stage devices, an equal mixture of a language-model-generated repair and a uniform belief over compatible repairs (the AI-weighted belief) lowers no-test cost by 2.14 percent relative to committing to the generated repair. Access charges can erase the benefit of diagnosis on isolated tasks, while repeated work can recover it under the AI-weighted belief. With normal-execution learning in every repeated-work arm, workload-valued testing costs 3.33 percent less than same-budget random selection under the AI-weighted belief. Weighting changes the comparison: uniform normal learning costs 540.75 less than AI-weighted normal learning on this cohort, with a device-paired interval excluding zero. Testing recovers part of that gap, but uniform normal learning still has a lower aggregate cost than AI-weighted workload-valued selection. Under uniform beliefs, information-gain selection has the lowest total among the compared selectors, while workload-valued testing costs more than normal learning. These results support retaining alternatives to a generated repair and evaluating diagnostic selection together with repair weighting and learning from execution.
[AI-80] Safe Multi-Objective Reinforcement Learning: Zero Constraint Violation and Primal-Averag ing Convergence
链接: https://arxiv.org/abs/2610.39409
作者: Chenxi Li,Yao Shu,Siyang Gao,Shuang Qiu,Guangxin Jiang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Optimization and Control (math.OC); Machine Learning (stat.ML)
备注:
Abstract:Many applications of multi-objective reinforcement learning (MORL) require policies that balance conflicting objectives while satisfying safety constraints. We study safe MORL with zero constraint violation throughout learning, covering both cumulative and stepwise constraints. A key challenge is to explore sufficiently under unknown dynamics and constraint costs without violating the constraints. For a given preference over objectives, we develop an algorithm to learn a safe Pareto optimal policy by combining optimistic reward estimates with pessimistic cost bounds. With high probability, every policy executed over T episodes incurs zero constraint violation. The algorithm returns a safe Markov policy induced by the averaged occupancy measure and achieves an \widetildeO(1/\sqrtT) suboptimality gap under smooth Tchebycheff scalarization. We further develop a preference-free algorithm that first explores safely without a prespecified preference and then computes a safe policy for any given preference without further interaction. Over T exploration episodes and K planning iterations, it retains the same safety guarantee and achieves an \widetildeO(1/\sqrtT+1/\sqrtK) suboptimality gap. Both algorithms require knowledge of a policy that strictly satisfies the constraints and its costs. We then relax these requirements, incurring at most constant cumulative violation. To our knowledge, we provide the first sample-efficient MORL algorithms that guarantee safety for every executed policy.
[AI-81] OPERA: OPamp Design by an Expert-Taught Reasoning Agent from Specifications to Layout ISCAS2027
链接: https://arxiv.org/abs/2610.39395
作者: Vishal Saxena,Soumyajit Mandal
类目: Hardware Architecture (cs.AR); Artificial Intelligence (cs.AI)
备注: 5 pages, 3 figures, submitted to ISCAS 2027
Abstract:Analog flows from specifications to a signed-off layout rarely run without manual iteration. Recent large language model (LLM) agents size circuits with optimizer loops over the device dimensions and stop at the schematic. We present OPERA (OPamp design by an Expert-taught Reasoning Agent), in which a general-purpose LLM agent designs CMOS opamps from specifications to post-layout sign-off with open-source tools. First, the designer teaches the agent the gm/ID method: map the specifications to transconductances and minimum currents, choose gm/ID per device role from characterized lookup tables, and size from equations, leaving a few knobs to search and an explainable design. Second, the agent learns the gLayout framework and taught layout practice, and writes a reusable layout generator that closes the layout and extraction loop at designer-approved gates. In SKY130, the agent designed a shared constant-gm bias and six opamps (5T, folded-cascode and telescopic OTAs; Miller, folded-cascode and telescopic two-stage) for fun = 50 MHz, PM = 60 degrees, SR 25 V/\mus into 1 pF and a minimum dc gain of 30 to 60 dB. All six layouts are DRC-clean and LVS-matched, and their extracted netlists meet every target at all 15 PVT corners ( fun = 53.7 MHz, PM = 62.1 degrees, SR = 27.9 V/\mus). The benchmark netlists are released at this https URL.
[AI-82] UMI-Live for In-The-Wild Bimanual Robotics Data Collection
链接: https://arxiv.org/abs/2610.39385
作者: Govind Velamoor,Adrian Lin,Sam Jung,Spencer Mateega,Brandon Y. Yang,Amish Sethi
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注:
Abstract:Universal Manipulation Interface (UMI) devices allow human experts to demonstrate tasks for robot learning without teleoperating physical robots. Using these demonstrations as training data still requires a consistent hardware setup, calibration, and data processing. We present UMI-Live, a full-stack open-source bimanual UMI system for in-the-wild data collection. We design our custom UMI-Live hardware to integrate camera and LiDAR–inertial sensing, onboard compute, and power, without external tracking equipment or calibration within the collection environment. We implement a simple user interface for recording data while an asynchronous pipeline performs parallel SLAM reconstruction and publishes training-ready datasets. We also release software that supports dual-YAM teleoperation and VLA policy training and inference. We validate UMI-Live data quality by fine-tuning \pi_0.5 and MolmoAct2, achieving success rates of 10/10 and 8/10, respectively on a bimanual pick-and-place task. Furthermore, we release 40 hours of UMI-Live data across three tasks, and complete hardware designs, software, and reproduction guides at our \hrefthis https URL\textproject website .
[AI-83] Learning to Formulate: Agent ic Hardware-Software Co-Design of AI Infrastructure through Progressive Optimization and Verification
链接: https://arxiv.org/abs/2610.39380
作者: Yi Ding,Li Shang
类目: Hardware Architecture (cs.AR); Artificial Intelligence (cs.AI); Distributed, Parallel, and Cluster Computing (cs.DC)
备注:
Abstract:AI infrastructure, the chips and software that train and serve AI models, is designed in layers. Layers make the system manageable, but they also fix which choices an optimizer can consider. Gains that need the chip and its software to change together are excluded before optimization begins: the formulation leaves them out. Language-model agents make it possible to revisit these boundaries, but an agent’s proposal cannot show by itself that a design is correct or optimal. We present Formulaic, a flow that divides the work. The agent writes the formulation, a mathematical solver optimizes within it, and verification and measurement check its assumptions and results. The agent refines the problem step by step, from the computation graph toward the chip. Each step is a template: a family of programs with the conditions under which they are correct. Wall relaxation searches the chip by pricing the hardware limits that bind the software. This paper takes a first step, through separate studies rather than one end-to-end run. On NVIDIA GB10, templates the agent wrote run complete Qwen3-8B layers 1.075\times faster than SGLang. In a simulated Ventus GPU study, joint design cuts the worst-scenario cycles of Qwen2.5-0.5B to 0.428 of the reference, against 0.934 for software alone. On a five-step GEMM refinement chain, reusing proofs makes checking 2.51\times faster.
[AI-84] Agent Decept: Benchmarking Deceptive Behavior in Tool-Using LLM Agents under Pressure
链接: https://arxiv.org/abs/2610.39377
作者: Yehan Zhang,Qianqi Yan,Hang Yin,Ching-Chen Kuo,Shan Jiang,Ming Ouyang,Yijie Liu,Xin Eric Wang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:LLM agents increasingly carry out multi-turn tasks across everyday applications, yet users often see only the agent’s report rather than the observations, API calls, and state changes behind it. Existing deception benchmarks have studied long-horizon interactions and tool-use settings, but whether agents faithfully report their actions across realistic, stateful application workflows remains underexplored. To address this gap, we introduce AgentDecept, an executable benchmark for evaluating behavioral deception under pressure. AgentDecept contains 477 pressure variants derived from 257 source tasks across six pressure sources, each paired with a pressure-free counterpart. Interventions enter through event messages, task context, application state, tool responses, and service quotas, spanning explicit events and implicit task or environment changes. Our evidence-grounded protocol evaluates full trajectories by relating user-facing claims to time-ordered observations and executed outcomes, distinguishing falsification, concealment, and equivocation from honest failure and disclosed limitations. The resulting behavioral labels also provide a setting for testing existing activation-based probes. AgentDecept enables controlled analysis of when multi-turn agents’ reports diverge from execution in realistic application interactions.
[AI-85] Before the Next Rollout: Diagnosing Skill Utility Shifts from Policy Updates
链接: https://arxiv.org/abs/2610.39358
作者: Yifan Wang,Zhenyi Gai,Yiwen Zhao,Mingxuan Jiang,Keyang Chen,Zeping Li,Hongfeng Chai,Yixin Cao
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Skill-evolution methods typically analyze execution trajectories to guide agent skill refinement. However, policy updates can change the behavioral effects and marginal utility of an unchanged skill. Attribution based on earlier trajectories may therefore no longer reflect the skill’s contribution under the updated policy. Moreover, trajectory failures can arise from multiple factors, making individual skill contributions difficult to isolate. We evaluate whether policy-update signals contain predictive information about changes in skill marginal utility. At fixed contexts from pre-update training rollouts, we measure how policy updates change the contrast in token preferences between skill and control conditions. Combining these interaction shifts with advantage estimates yields reward-directed readouts for diagnosing utility changes. The readouts reuse existing training rollouts and require only forward passes for single-step action scoring, without additional post-update rollout evaluation. Our experiments show that policy updates can alter skill marginal utility and that policy-update readouts carry diagnostic information for identifying skills with declining utility. Integrating readout-guided prioritization into skill evolution improves downstream performance while reducing policy and editor token consumption.
[AI-86] CoToan: Executable Rules and a Representation Contract for Reproducible Game AI Evaluation
链接: https://arxiv.org/abs/2610.39353
作者: Ha-Thanh Nguyen
类目: Artificial Intelligence (cs.AI)
备注: Accepted at RIVF 2027
Abstract:Game AI results are comparable only when different implementations expose the same decision problem. Yet seemingly minor representation choices can silently change that problem by altering action probabilities or collapsing states with different outcomes. We study this issue through CoToan, an executable specification and representation contract for Vietnamese Mathematical Chess connecting the Game Description Language (GDL) with OpenSpiel. The game provides a useful case because a single capture may have multiple arithmetic justifications, while outcomes can depend on scoring rules and a finite playing horizon. CoToan therefore represents actions by their state transitions rather than by arithmetic witnesses, and treats time and victory configuration as part of task identity. Cross-representation tests show agreement in legal actions, successor states, and terminal outcomes across hundreds of sampled and constructed positions, complemented by exhaustive checks of all 32 operator masks and 512 opposing-material subsets. Controlled diagnostics further show that uniform witness-level sampling alters endpoint-action probabilities, while omitting remaining time or victory settings can merge states with different terminal behavior. These results show that action semantics, time, and victory configuration are not merely implementation details. They are part of the experimental task itself and must be represented explicitly for reproducible game AI.
[AI-87] RetroChessBench: A Benchmark for Reconstructing and Proving Chess Histories
链接: https://arxiv.org/abs/2610.39341
作者: Tidor-Vlad Pricope
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 22 pages, 2 figures. Benchmark and RL environment: this https URL
Abstract:Frontier language models such as GPT 6 Astra have contributed to research mathematics, demonstrating substantial reasoning capabilities. Yet serious reasoning failures persist even in tasks governed by explicit rules. We demonstrate this in retrograde chess which is a playground for pure reasoning, where solvers reconstruct legal histories and distinguish possible events from forced ones. We introduce RetroChessBench, a 12-task benchmark covering exact-length constructions, unique endings, and exhaustive answer sets, with references, rubrics, and a terminal-reward environment. In two trials per task and condition, GPT 6 Astra High fails to produce the required final answer in 22/24 attempts without tools and 18/24 with code execution; mean terminal RL rewards are 14.75% and 27.19%, respectively. The examples identify faulty deductions: failed responses reject legal histories and treat temporary obstructions as permanent. Online retrograde archives often provide puzzles and final move sequences without detailed derivations. We hypothesize that limited exposure to these missing explanations in web-derived training data contributes to the observed reasoning failures. This questions how far frontier mathematical reasoning really generalizes beyond familiar solution patterns. Claims of general intelligence require evidence that models can transfer learnt principles to areas with few worked examples.
[AI-88] Learning to Bound: Monotonic Calibrated Neural Upper Bounds for MPE Inference
链接: https://arxiv.org/abs/2610.39305
作者: Akshay Vyas,Hailiang Dong,Shivvrat Arya,Brij Malhotra,Tahrima Rahman,Vibhav Gogate,Nicholas Ruozzi
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 37 pages, 4 figures
Abstract:Probabilistic graphical models (PGMs) compactly encode structured joint distributions, but Most Probable Explanation (MPE), which finds the highest-probability assignment to unobserved variables, remains computationally intractable in large, densely structured models. Anytime systematic search, such as branch-and-bound (BB), is a popular approach, but its efficiency depends on tight admissible upper bounds that are often prohibitively expensive to compute using classical methods like Mini-Bucket Elimination (MBE). We introduce Learning to Bound (L2B), a data-driven framework that amortizes bound computation cost by training a neural predictor. L2B targets two key properties required for effective search bounds: admissibility, which is achieved via statistical calibration techniques, and progressive tightening as evidence accumulates, which is enforced through structural architectural constraints that ensure monotonicity by construction. Across challenging benchmarks, L2B delivers tighter, monotonic bounds at orders-of-magnitude lower cost on every PGM family, and yields measurable improvements when integrated with branch-and-bound on high-treewidth instances.
[AI-89] here Is No Free Lunch in RLVR: Every pass@1 Gain Has a pass@k Price
链接: https://arxiv.org/abs/2610.39295
作者: Yinuo Liu
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Optimization and Control (math.OC); Machine Learning (stat.ML)
备注: 19 pages, 4 figures. Code and verification package (34/34 automated checks): this https URL
Abstract:Reinforcement learning with verifiable rewards (RLVR) makes a language model solve more problems on the first try and fewer when given many tries: it raises pass@1 while the base model overtakes it at large sampling budget k. Is this a failure of RL to add reasoning, or a transient that more training removes? We settle the question with a theorem. The reversal is a price, and we compute its exact rate. Modeling RLVR as an exponential tilt of a base policy over strategies shared across prompts, we prove that under exclusive specialization every update that raises pass@1 strictly lowers the success of some prompt; a strictly positive price on prompts (Stiemke duality) is the exact condition, and being merely the best strategy on a prompt is provably not enough. When the loss reaches the hardest prompts, it forces a pass@k crossover, and the number of crossings is at most the sign variation of the shift in per-prompt success, a totally positive bound that is tight and that recovers the known single-crossover result as its order-preserving case. The crossover is not a bug to patch: a constrained update traces the exact pass@1 versus pass@k frontier, and removing the crossover entirely costs the entire pass@1 gain. The same tilt identity yields entropy collapse and the entropy-performance law with an identifiable exponent, unifying two literatures under one mechanism. Reinforcement cannot widen the reasoning frontier; it can only move along it, at a price we make measurable by a diagnostic that reads the bound, the order-preserving test, and the price from any evaluation log. There is no free lunch, only a price.
[AI-90] What Does “Now” Mean to Your Computer-Use Agent ? Stale Screenshots Page Loads and Which Check to Run Before a Click
链接: https://arxiv.org/abs/2610.39282
作者: Vasundra Srinivasan
类目: Artificial Intelligence (cs.AI)
备注: 8 pages, 4 figures, 3 tables. Code, benchmark generator and preregistrations: this https URL
Abstract:Computer-use agents plan on a screenshot for several seconds and then click at a position on the screen; if the page has changed in the meantime, the click lands on an unintended element. Prior work treats the planning interval as the source of this risk. Measuring planned clicks on 31 public web pages, we find that the risk lies in page loading: 0.4% of clicks planned on fully loaded pages became stale within 30 seconds, against 17% of clicks planned in the first second after a page opened. On these pages, waiting about one second for network activity to stop removes most of this risk, yet the wait rules of eight of ten widely used open-source agents leave more than 1% of clicks stale. Commercial pages keep loading for longer: in the HTTP Archive’s public crawl, 15% of 65 desktop booking and buying home pages went 500 ms without network requests within 3 seconds of their HTML being parsed, against 63% of 3,028 .gov pages, and 52% shifted their layout within 6.5 seconds of a network-idle wait capped at 3 seconds, against 8% (61% against 26% among pages recorded in full). Where a check before the click is needed, as before irreversible actions and on pages that are still loading, a pixel comparison verifies the target’s position as accurately as a frontier vision-language model at one eighth of the latency, and a 4B model matches the frontier model in verifying whether the previous action succeeded. We release a benchmark generator, a break-even formula for when a check pays, and three rules for agent developers: this https URL
[AI-91] MemSteer: Turning User Histories into Supervision for Agent Memory Optimization
链接: https://arxiv.org/abs/2610.39276
作者: Jie Huang,Cuiling Lan,Qize Yu,Xiulian Peng,Yan Lu
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:An effective agent memory system should adapt to a user’s history and task requirements. Human-annotated supervision can guide this adaptation by providing feedback for evaluating and improving memory designs. Yet suitable labels are rarely available for new user histories, and manually constructing QA pairs and supporting evidence is costly. We introduce MemSteer, which turns unlabeled user histories into question-answer (QA) supervision with supporting evidence for memory-design optimization. MemSteer constructs an evidence-grounded answer plan that specifies how facts across related records jointly support a reference answer. The plan guides question generation, while source-grounded auditing and blind answer recovery check the resulting annotations. To study the utility of this supervision, we develop MemSteer-Opt, which combines QA-based design evaluation with evidence-guided failure diagnosis to revise memory store, retrieval, and context construction without updating model parameters. Experiments on ATM-Bench and RHELM show that this supervision is competitive with official-QA-guided optimization. Relative to Base-BM25, MemSteer-Opt improves ATM-Bench Question Score by 13.4 points and RHELM accuracy by 4.1 and 6.6 points without and with external data, respectively. The supervision also benefits two existing optimizers on ATM-Bench, while new-user experiments demonstrate continued design optimization using supervision generated from the target histories.
[AI-92] LLM -Assisted Optimization for Simulation-Based Redesign
链接: https://arxiv.org/abs/2610.39275
作者: Chan-Eui Song,Jeremy J. Michalek,Christopher McComb
类目: Artificial Intelligence (cs.AI); Computational Engineering, Finance, and Science (cs.CE); Optimization and Control (math.OC)
备注: 17 pages, 12 figures, 11 tables. Submitted to the ASME Journal of Mechanical Design. Code: this https URL , data: this https URL
Abstract:Engineering design frequently involves redesigning a prior solution to adapt to changes in context, such as material prices or system requirements. Redesign often leverages simulators too complex and/or computationally intensive for traditional numerical solvers, which lack engineering knowledge and context, to ensure global solution quality. Large language models (LLMs) can interpret semantic information, including engineering knowledge about the system and changes in context, to help guide search, and recent studies have shown that incorporating this knowledge into optimization can improve search performance. However, how best to combine LLMs with numerical solvers in redesign remains unclear. We develop and test an integrated LLM-optimization approach for redesign and compare three methods for leveraging LLMs (warm start, variable fixing and ReAct) and four black-box solvers (Bayesian optimization, genetic algorithms, mesh adaptive direct search and random search) on an electric vehicle battery pack cost minimization case study across four types of redesign scenarios. We find that the integrated approach can match feasible design discovery of the numerical-solver baselines with 80% fewer simulator evaluations while tripling the fraction of runs that find near-optimal solutions. We find that the best LLM-solver integration approach and the way the prior solution is leveraged depend on the structure of the design requirement change, and that the benefit of increasing LLM model size grows with the extent to which the LLM controls the search. These results show clear potential for LLMs in redesign tasks for guiding or replacing numerical solvers.
[AI-93] Online Dense Reinforcement Learning with Planning -Grounding Agent for UI Navigation on Live IOS Apps
链接: https://arxiv.org/abs/2610.39264
作者: Peitong Duan,Khashayar Gatmiry,Jeffrey Nichols,Samy Bengio,Emmanuel Abbe
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:While frontier vision-language models demonstrate strong capabilities in user interface (UI) navigation, their high inference costs prohibit everyday use. Conversely, smaller models trained via supervised fine-tuning (SFT) or offline reinforcement learning often memorize static trajectories, leading to brittle execution and an inability to recover from errors. To address these limitations, we propose a novel online reinforcement learning framework that trains UI navigation agents directly on live iOS applications. Our framework decouples the agent into two specialized components: a Planning agent and a UI Grounding model. We rely on an LLM-as-a-Judge to provide comprehensive reward signals, combining dense, step-level feedback to evaluate intermediate progress with a terminal reward for overall task completion. To effectively leverage this dual reward structure, we optimize the agent using Proximal Policy Optimization (PPO) with step-level advantage estimation. Furthermore, we introduce GiGPPO, a novel neighbor-conditioned value estimation method that improves critic stability by leveraging reward signals from visually similar states across concurrent trajectories. To support this online training paradigm at scale, we introduce an efficient infrastructure that orchestrates hundreds of concurrent virtual iOS devices. Evaluations on rigorous benchmarks, including UINavBench, iOSWorld, and AndroidWorld, demonstrate that our online RL trained agents substantially outperform both SFT and terminal-only RL baselines while accelerating training convergence. Notably, our 8B RL agents achieve nearly a sevenfold improvement over the SFT baseline in out-of-distribution generalization to Android environments, while outperforming Gemini-3.5-Flash and tying with Gemini-3.1-Pro on real iOS applications.
[AI-94] Where to Learn and How Far to Roll Out: Efficient Adaptive Training for GUI Agents
链接: https://arxiv.org/abs/2610.39251
作者: Yuxiang Chai,Quyu Kong,Panrong Tong,Steven Hoi,Yue Wang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Trajectory-level GRPO is a common approach for training multi-step GUI agents, but repeatedly rolling out complete trajectories from the task root can be inefficient: task-level rewards provide coarse supervision over which intermediate decisions require improvement, while already-mastered prefixes are repeatedly re-executed before reaching difficult states. We introduce a two-stage adaptive training framework that jointly addresses where to learn and how far to roll out. We first construct reusable intermediate checkpoints from successful expert trajectories and select informative pivots using action disagreement, which measures uncertainty directly from sampled GUI actions. Stage I performs one-step RL at these pivots to improve uncertain local decisions. Stage II then progressively expands policy responsibility through adaptive Local, Suffix, and Root training according to task progress. Recoverable checkpoints enable repeated training from intermediate states without replaying the preceding trajectory. Across MobileGym and CUA-Gym-Hub, our method reduces environment interaction by 56-61% relative to Vanilla GRPO while improving task performance, demonstrating that jointly allocating training across states and rollout ranges can substantially improve GUI-agent RL efficiency.
[AI-95] SocialArena: Evaluating Social Intelligence in Large Language Models EMNLP2026
链接: https://arxiv.org/abs/2610.39247
作者: Zongqi Wang,Rui Wang,Yuchuan Wu,Hua Yang,Bingxue Qiu,Tianle Gu,Bing Zhao,Yujiu Yang,Jieping Ye,Yongbin Li
类目: Artificial Intelligence (cs.AI)
备注: EMNLP 2026 oral
Abstract:We introduce SocialArena, a benchmark for systematically defining and evaluating the social intelligence of large language models (LLMs). Although LLMs are increasingly deployed in socially interactive applications such as mental health counseling, companionship chatbots, and educational tutoring, existing evaluations remain fragmented focusing on isolated capabilities or single scenarios and evaluation reliability is largely unverified. This work makes three main contributions to address these limitations. (1) We first propose a unified three-tier framework that defines social intelligence along three capability dimensions: role-playing ability, mental state reasoning, and behavioral reasoning. To operationalize this definition, we design a role-playing dialogue paradigm with explicit character thinking for evaluation, enabling simultaneous assessment of these capabilities in a manner that mirrors realistic human interactions. (2) To ensure diversity across social scenarios and thereby prevent evaluation conclusions from being confined to specific scenarios, we introduce the U-TRIMF framework, which guides us in synthesizing social contexts with broad coverage. (3) To obtain reliable evaluation, we invest 132.4K to recruit and train human annotators and adopt a human-annotated arena-style protocol, where annotators compare model responses and provide preference labels along with natural-language critiques. We hope that SocialArena can provide valuable insights and guidance for future LLM improvements.
[AI-96] Beyond LLM Wrappers: Context Engineering and the Three-Engine Architecture for Reliable Financial AI
链接: https://arxiv.org/abs/2610.39230
作者: Pradeep Kumar GS,Ravi Tanniru
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Financial AI systems must support decision-bearing questions involving payments, collections, cash flow, and compliance while ensuring numerical accuracy, traceability, and auditability. Large language models (LLMs) enable natural language interaction but cannot independently guarantee reproducible calculations or evidence-grounded recommendations. We present a production-grounded architecture integrating three components: a financial knowledge graph for enterprise relationships, temporal context, and provenance; a deterministic optimization engine for financial calculations and decisions; and a multi-agent LLM layer for intent interpretation, task orchestration, explanation, and claim validation. A coordination layer, termed context engineering, selects, ranks, composes, and validates query-specific context across ERP systems, tools, and tenants. Based on observations from 15 deployments over approximately 11 months, single-agent pipelines exhibited hallucination rates of about 7-9% (5-10% across query types), including invented invoices and unsupported financial claims. Graph-backed multi-agent validation reduced observed rates to below 1%, although these findings are observational rather than causal. ERP data quality and tenant-specific schemas remained major implementation challenges. Reusable mappings, graph templates, and validation policies coincided with a reduction in deployment time from roughly six to two weeks. We argue that financial context engineering extends beyond text-centric retrieval-augmented generation by requiring structured relationships, temporal validity, exact numerical evidence, access controls, and validation against source records and solver outputs. We outline research challenges in evaluation, reproducibility, privacy, context quality, and cross-tenant generalization. Subjects: Artificial Intelligence (cs.AI) Cite as: arXiv:2610.39230 [cs.AI] (or arXiv:2610.39230v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2610.39230 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-97] Where Position Matters: Profiling-Guided Layer-Wise RoPE Allocation in Transformers
链接: https://arxiv.org/abs/2610.39223
作者: Zhiwei Liu,Boao Kong,Bing Liu
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Machine Learning (stat.ML)
备注: 28 pages
Abstract:Rotary Position Embedding (RoPE) typically uses uniform coverage across Transformer layers despite heterogeneous positional requirements. Phase-off interventions reveal non-uniform, model-specific rotary-phase dependence, while coverage interventions relate it to coverage-upgrade benefits. We propose profiling-guided layer-wise RoPE allocation, assigning high/low coverage under a fixed budget using early coverage-benefit profiles. The framework supports fixed-width query/key representations and physical rotary-dimension allocation in Multi-head Latent Attention (MLA). Normalized-depth mapping transfers profiles within model families, enabling small-model measurements to guide larger-model allocations before pretraining. A brief analysis relates selection and transfer to score calibration and inter-layer interactions. From-scratch experiments demonstrate validation-loss improvements over available budget-matched baselines, while analytical resource accounting identifies favorable quality–storage–compute trade-offs for physical MLA. These results connect positional demand to transferable, efficient architecture design.
[AI-98] Goal-Aware Dynamic Multi-Agent Collaboration for Long-Horizon Tasks
链接: https://arxiv.org/abs/2610.39193
作者: Xu Shen,Lingyong Yan,Shuaiqiang Wang,Jianmin Wu,Zhikai Chen,Jiliang Tang,Jie Ren
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:LLM-based agents are increasingly used to complete complex, long-horizon tasks without human intervention. To make progress toward the task goal, agents need to perform different duties, such as planning and verification, when needed. To understand how single agents manage these duties during long-horizon execution, we conduct a preliminary study. Our analysis shows that verification has the strongest association with task success, but often occurs late, leaving limited opportunities for correction. We also find that the proportion of duty-relevant information decreases as records from different duties accumulate in a shared context. These findings lead us to examine whether multi-agent systems (MAS) can address these problems by assigning duties to agents with separate contexts. Our analysis shows that this separation alone is insufficient. Agents can lose track of the overall goal and repeat prior investigations. They may also lack necessary background while receiving irrelevant information from others. To maintain goal awareness and coordinate collaboration as task needs change, we propose Goal-Aware Dynamic Multi-Agent Collaboration (GAD). It uses a growing execution graph to organize subtasks and their dependencies. To keep agents focused on their goals, each context retains the original instruction and the assigned task’s goal and acceptance conditions. To adapt collaboration and information sharing during execution, agents can divide tasks into subtasks and specify dependencies based on execution evidence. These dependencies guide when agents work and which results they receive. Verification and recovery check subtask results and full task completion against their goals, providing feedback for repair when checks fail.
[AI-99] REALM: Identifying Minimal Task Assets for LLM Agents
链接: https://arxiv.org/abs/2610.39155
作者: Zihou Ren,Byungchul Tak,Long Wang
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Systems and Control (eess.SY)
备注:
Abstract:Large language model (LLM) agents often receive access to more files, accounts, and services than a task requires. Prompt injection turns this excess access into an opportunity to redirect execution toward unrelated resources. Applying least privilege therefore requires identifying the assets needed by a task, including assets whose identities emerge only during execution. We present REALM (Required Execution Assets Linked to a Mission), an approach to identifying and maintaining a task’s required assets and their permitted uses. The approach represents required assets as Dependency Graph nodes, resolves prerequisite relationships, and admits newly discovered assets through source-scoped selection. It separates discovering an asset from authorizing its use, while preserving the agent’s freedom to choose its computation. The REALM Builder maintains the selected grants and projects Asset Files; the Partial API Manager exposes resource-bound Partial APIs to the Agent inside a Task Container. Across 60 clean tasks and 560 injected cases in AgentDyn, it completes 40.0% and 36.4% of tasks and decreases ASR from 16.6% to 0.2% with qwen3.5-flash as the base model.
[AI-100] Clean Sheets and Checkpoints: Evidence Complexity for Certifying Sequential Agent Risk
链接: https://arxiv.org/abs/2610.39124
作者: Rohith Yanambaka Venkata,Don McBride,Tiago Sequeira,Md. Al Amin
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 26 pages, 3 figures, 3 tables
Abstract:Dangerous-capability evaluations often test an AI agent from prepared states at ordered milestones. A clean sheet records no successes. What it certifies about a campaign of B attempts depends on what a failed attempt erases. We model the task as a chain of k milestones divided into checkpoint segments and ask how many zero audit responses certify, at level \alpha , a completion probability of at most \epsilon . For segments chosen before the audit, counting possible completion windows gives a certificate for any specified budget. The bound also covers adaptive deployment processes that follow the chain’s transitions, provided their success probabilities after every complete history are capped by the corresponding audit success probabilities. The audit does not test this transport assumption. In rare-risk phase interiors, each segment is either rarely completed within the budget (active) or nearly always completed (free). The minimum expected number of zero responses is asymptotic to \log(1/\alpha),(B^a/(a!,\epsilon))^1/K_A , where the a active segments contain K_A milestones and are those at least as long as a budget-dependent threshold. The certificate attains this constant. Every asymptotically optimal evaluator must divide its expected response count equally among active milestones in the limit, with a vanishing share elsewhere. Where a single segment passes from active to free, each of its milestones receives a smaller limiting share than an active milestone; a crossover-aware certificate attains the optimal constant. For selected segments with known or unqueried prefix steps, counting failed cycles can sharpen the certificate.
[AI-101] Appreciation Teaches Unauthorized Execution: Self-Evolution Degrades the Robustness of Personal Agents to Indirect Prompt Injection
链接: https://arxiv.org/abs/2610.39122
作者: Xinyan Wang,Xiaogeng Liu,Jun Shao,Chaowei Xiao
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注:
Abstract:Self-evolving personal agents act for users and learn from user feedback. However, if an agent obeys an instruction that an untrusted third party hides in content, the third party gains control of the agent. Such indirect prompt injection (IPI) poses severe risks. We investigate whether self-evolution degrades the robustness of personal agents to IPI. To study this question, we design a controlled two-phase methodology. In the adaptation phase, the user asks the agent to read external content and complete the task described in it, a common request in daily use. The agent then learns from the user’s feedback and improves itself as designed. In the test phase, a third party writes a similar task into new external content, but the user’s test request does not authorize the injected task. Using this methodology, we build multiple tasks with diverse test instances in different domains for a real personal agent. Our extensive experiments reveal that self-evolution substantially degrades robustness to IPI. Specifically, compared with the base agent, which does not evolve from user feedback, the average attack success rate rises by 14.0–48.8 percentage points under skill evolution and by 3.0–39.3 percentage points under reinforcement learning, posing safety risks. The code is available at this https URL
[AI-102] From Local Recovery to Unified Generation: Discovering Triple Set in Knowledge Graphs via Observation-Anchored Latent Diffusion
链接: https://arxiv.org/abs/2610.39114
作者: Jiaqi Wang,Yisen Gao,Huayang Shi,Junhao Hou,Wengen Li,Jihong Guan,Yichao Zhang,Shuigeng Zhou,Yangqiu Song,Hong Cheng
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Triple set prediction (TSP) is to recover missing relational triples in a knowledge graph by predicting the triple set instead of isolated links. Existing approaches predominantly rely on the subgraph-based completion paradigm, where the original graph is first partitioned into local subgraphs which are then independently completed. Such a paradigm suffers from fragmented structural reasoning, limited cross subgraph dependency modeling, and missing predictions across subgraphs. This paper proposes LatentDiff, a novel unified latent generative framework for TSP, which reformulates knowledge graph completion as latent space graph generation. LatentDiff first learns a compact latent representation using a relation-aware residual-quantized knowledge graph autoencoder, which encodes the whole graph into relation-aware latent embeddings, quantizes them with chunked residual codebooks, and reconstructs triples through a factorized decoder. On top of this latent space, we introduce an observation-anchored latent graph diffusion model, where the target latent graph representation is perturbed toward the observed latent. The reverse process recovers missing triples through progressive latent denoising. Experiments on three benchmark datasets show that LatentDiff outperforms the SOTA in terms of F_tsp by 5.4 percentage points on average. Code: this https URL.
[AI-103] Generating Critic-Guided Synthetic Experience in Latent World Models
链接: https://arxiv.org/abs/2610.39100
作者: Jorge de Freitas,Christos Ziakas,Alessandra Russo
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Learning from offline datasets is limited in scalability, as collecting new experience is often costly or unsafe in the real world. World models offer an alternative way to generate new experience through a learned dynamics model, yet learning policies in latent space is limited due to compounding errors in imagined rollouts. To this end, we propose using the world model as a local transition generator, leveraging critic guidance to choose actions inside that model to generate synthetic experiences. We study action refinement methodologies, introducing Critic-Guided Synthetic Experience (CGSE) generation in world models. Specifically, CGSE augments an existing offline dataset by sampling candidate actions using an offline policy and refining each to maximize a first-order approximation of the Q -function within a trust region. We then use a frozen latent world model to imagine the future state given each updated action, and use the critic to choose the best one, appending this synthetic transition to the replay buffer. Our method requires no additional real-environment interaction or test-time planner, as it generates synthetic experience offline using the world model. On three OGBench manipulation tasks, CGSE improves an offline policy by a relative improvement of 7.1% on the average success rate.
[AI-104] actus: Tactile Pose Sensing Glove for Natural Dexterity Data Collection
链接: https://arxiv.org/abs/2610.39076
作者: Li Guan,Rachael Granberry,Sahar Aseeri,Sahar Rostami,Taylor Niehues,Sai Dhanooshi Marri,Mia Huang,Dingtian Zhang,Ziyi Kou,Jason Pursel,Eric Whitmire,Vatsal Mehta,Ashish Deshpande,Tristan Trutna,Reed Belcher,Emelia Black,Julian Blake,Roger Boldu,Sean Bonaventure,Einar Boyovich,Jordan Brennan,Elizabeth Casillas,David Christensen,Tyler Christensen,Nicholas Colonnese,Stuart Dealey,Gaige DeHaven,Katie Doran,Chris Doughty,Robert Drury,Silvia Espino,Gabriel Geyne,Beth Goodrich,Kyra Gould,Joe Grisso,Susana Guerrero,Andrew Gwozdziowski,Monica Ha,Ehsan Hajiesmaili,Kate Healy,Rose Hope,Michael Hurst,Claire Ikemiya,Daisy Isibor,Rick James,Yashasvi Joginapally,Samuel Johnson,Tyler Jorgensen,Kristy Jost,Ben Kamath,Slava Karulin,Alexander Keener,Woo Kim,Alison Krauskopf,Ankit Kumar,Joshua Lang,Lizzy Li,Cameron Lynch,Daniel Ma,Hector Martinon,Christina Matos,Keagan McCurdy,Lee Mincey,Stuart Nezlek,Tiffany Nguyen,James O’Hara,Jason Osborne,Jacky Pai,Roham Pardakhtim,David Perek,Alexandre Poulin,Nick Raines,Salwa Rasheed,Leah Resneck,Kevin Rhodes,Ergys Ristani,Douglas Roberts,Anik Sarker,Harrison Scarborough,Jonas Schmidtler,Brian Shieh,Benjamin Shuman,Garrett Simard,Forrest Smith,Udara Somarathna,Eliot Sosin,Jonathan Stewart,Kristine Strom,Adam Swett,Hal Talbot-Mantooth,Julia Tanner,Sai Sharan Thirunagari,Curtis Torrey,Simge Uzun,Eileen Wachs,Rylan Wasson,Henry Wei,Jesse White,Beau Woodbyrne,Antonio Woods
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Hardware Architecture (cs.AR); Machine Learning (cs.LG)
备注: 9 pages
Abstract:Collecting naturalistic human manipulation demonstrations with accurate contact force and hand pose is essential for robot dexterous manipulation policy and foundational world model learning - yet no existing wearable platform provides both calibrated fingertip force sensing and occlusion-resilient hand tracking in a form factor that preserves natural manipulation behavior. We present Tactus, a textile-integrated sensing glove that co-optimizes three design pillars: (1) a thin, conformal 3D-knit form factor (1.5 mm fingertip stackup) designed to minimize interference with natural manipulation, (2) strategically placed capacitive tactile sensors at manipulation-relevant regions informed by empirical contact studies, and (3) IMU-fused pose tracking that maintains accuracy through the hand-object occlusions where manipulation data is most valuable. Tactus includes a documented calibration pipeline for cross-glove consistency and per-session calibration for subject-dependent correction. To enable objective comparison, we introduce standardized benchmarking protocols for tactile force regression and hand pose estimation, and apply them to Tactus and two commercially available tactile gloves - to our knowledge the first quantitative cross-platform evaluation of this kind. Tactus resolves a minimum detectable force of 0.59 N at 95% confidence, 1.6 - 2.7x lower than the commercial baselines, and with 0.87 cross-participant consistency, while our pose tracking sustains 11.0 mm fingertip keypoint relative error during manipulation, when occlusion is heaviest. We show that these combined signals support object property estimation, slip detection, and representative recording of hand-object interaction - yielding demonstration data with the fidelity required for contact-aware manipulation policies and world models.
[AI-105] Certified Access: Test-Time Compute on Conservative Physics-Structured AI Memory NEURIPS2026
链接: https://arxiv.org/abs/2610.39073
作者: Pratik Jawahar,Riccardo Maggioni,Jonas Petersen,Matteo Cozzi,Gian-Alessandro Lombardi,Marek Gajewski,Camilla Mazzoleni,Federico Martelli
类目: Artificial Intelligence (cs.AI); High Energy Physics - Phenomenology (hep-ph)
备注: Accepted at the 40th Conference on Neural Information Processing Systems (NeurIPS 2026); Workshop: Long-Context Foundation Models
Abstract:Test-time compute mechanisms, such as retries, confidence-based escalation and non-local routing, are conventionally deployed as black-box additions to trained models. They typically add forward passes or halting heads, but give no structural account of what the extra compute buys or what it costs against the model’s stability guarantees. We therefore propose a theoretical and empirical framework for test-time compute on a Latent Dynamics Model used as the memory. When an encoder’s latent state (q,p) is advanced by a damped symplectic (velocity-Verlet) step of a learned Hamiltonian, every mechanism that expands access has an exact phase-volume Jacobian, a bounded energy change and an exact latch-transport law. We split reachability failures into escape failures and reach failures. A Lorentz squeeze resolves escape failures by injecting bounded, re-absorbable energy. A gated canonical translation (a one-step channel across C_T ) resolves reach failures at \detJ=1 exactly. We verify this certificate stack on a designed analytic testbed. Reaching just beyond C_T with the squeeze requires energy that grows quadratically in the excess distance. The gated channel’s energy cost equals the potential difference between its endpoints, which is zero on the matched channel measured here, and so does not grow with distance in this testbed. An untrained state-replacing map annihilates phase volume and erases stored charge, while the gated channel transports it. On a learned physics-structured memory, we show calibrated compute rationing and evaluate the scaling limits of non-local routing against a physics-free baseline. The main contribution is the formalization and verification of the certificate stack, not benchmark performance.
[AI-106] When the Same Prompt Gives Different Answers (and the Same Answer Isnt Always Right): A Taxonomy of Failure Mechanisms Testing and Mitigation Strategies for LLM -Assisted Evidence Synthesis Read-and-Evaluate Tasks
链接: https://arxiv.org/abs/2610.39070
作者: Carole Lunny,Ian Goodman,Stefano Berrone,Connor TA Brenna,Emma K Reid,Ivan Perez-Neri,Shailesh Kolekar,Nityanand Jain,Ashwin Kamath,Lucas HCC Santos,Yulia Otmakhova,Heba Hussein,Diana M Cespedes,Gioia Dahdal,Salmaan Kanji
类目: Artificial Intelligence (cs.AI); Performance (cs.PF); Methodology (stat.ME)
备注: 76 pages, 2 figures, 2 tables, 1 box, 6 supplementary files
Abstract:LLM read-and-evaluate tasks include study screening, data extraction, risk-of-bias assessment, and critical appraisal in the evidence synthesis field. These task differ in the amount of LLM interpretation they require. Some tasks involve retrieving relatively simple information, such as an author name, sample size, or publication year. Others require the LLM to read source material, interpret it against predefined criteria, and make a judgement. Researchers should assess whether LLM judgements are accurate, supported by appropriate source evidence, and reproducible under specified conditions. A correct judgement does not necessarily indicate that the underlying evaluation process is reliable. We developed a taxonomy for identifying mechanisms that may compromise these properties and for informing the selection and reporting of appropriate evaluation methods. The taxonomy outlines potential ways that LLM judgments from read-and-evaluate tasks can fail, called TRAILS (Taxonomy of Reliability And Instability in LLM-assisted evidence Synthesis). The TRAILS taxonomy addresses two complementary questions. First, what mechanisms can cause the same read-and-evaluate task to produce different judgements across repeated LLM runs? Second, what mechanisms can systematically affect the judgement or required output in a way that is not apparent from the judgement itself? This study presents 38 mechanisms and identifies the deployment settings in which each can operate, approaches for testing them, and mitigation strategies. We also show why repeatability or accuracy does not rule out a hidden judgement failure and why human evaluation should assess whether a judgement is grounded in the supplied evidence, independent of inappropriate external influence, and produced through a process that can be examined and verified.
[AI-107] When Does Closed-Loop Curriculum Adaptation Help GRPO? An Analysis on Countdown ICTAI2026
链接: https://arxiv.org/abs/2610.39056
作者: Yuv Raj Pant,Moshiur Rahman Tonmoy,Kaiqun Fu
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 6 pages, 2 figures, Accepted at IEEE ICTAI 2026
Abstract:Reinforcement fine-tuning has become an effective post-training paradigm for improving the reasoning ability of large language models, but it remains sample-inefficient when the training distribution is mismatched to the model’s current capability. In group-relative objectives such as Group Relative Policy Optimization (GRPO), problems that are too hard often produce uniformly failed rollouts, while problems that are too easy provide little useful advantage signal. In this work, we introduce CLOSEDLOOPGAUSSIAN, a lightweight adaptive curriculum method for efficient reinforcement fine-tuning on ordered reasoning tasks. Instead of following a fixed easy-to-hard schedule, CLOSEDLOOPGAUSSIAN preserves the structure of an Easy-to-Hard Gaussian sampler while dynamically adjusting its pace and spread using periodic probe estimates of per-difficulty accuracy. The controller accelerates training when the model is ahead of a reference curriculum, slows down when the model falls behind, and broadens sampling when earlier difficulties regress. This feedback mechanism keeps training focused on tasks that are challenging but still learnable, without changing the reward function, model architecture, or underlying GRPO optimizer. Experiments on symbolic arithmetic reasoning task show that CLOSEDLOOPGAUSSIAN improves exact-match accuracy over vanilla GRPO and fixed open-loop curricula, with the strongest gains in intermediate-difficulty regimes where group-relative learning signals are most informative. Codes are available at this https URL.
[AI-108] Reinforcement Learning for the Game of Balatro: A Fast Simulator and Two-Policy Training Approach
链接: https://arxiv.org/abs/2610.39049
作者: Ajay Vikram Periasami,Anjaneya Gupta,Ethan Cheng,Ivan Rosales
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: Accepted at EXAG 2026 (AIIDE Workshop on Experimental AI in Games)
Abstract:Balatro, a 2024 roguelike deckbuilder, presents three difficulties that have historically derailed reinforcement learning methods: a 3,000-action state-dependent action space, horizons of several hundred decisions with sparse rewards, and pervasive stochasticity in both card draws and shop generation. Published RL results on the game remain scarce. We introduce an end-to-end pipeline comprising (i) a high-throughput Rust simulator that is automatically state-diffed against the live game client via BalatroBot, (ii) a condensed observation-action interface with dynamic per-step masking of legal moves, and (iii) a multi-agent PPO architecture that decomposes play into a tactical blind policy and a strategic shop policy that share the full game state but act in disjoint phases. Long-horizon credit assignment is partially addressed by per-agent reward routing with explicit cross-agent credit. On the main joint run, training metrics show a sharp phase transition in which the first-blind clear rate converges above 99% and mean episode return increases by roughly an order of magnitude, plateauing at a mean maximum ante of 3 out of 8; single-agent baselines under identical rewards oscillated between greedy and hoarding failure modes and did not stably clear the first blind. These results establish the two-policy decomposition as a strong reactive baseline for Balatro and localise the remaining gap to long-horizon strategic planning rather than tactical blind play.
[AI-109] When Should AI Decide Defer and Deviate? Responsible Automation for Deceased Donor Kidney Allocation
链接: https://arxiv.org/abs/2610.39045
作者: Madeleine Pollack,Thomas Daillak,Mehmet Toner,Korkut Uygun,Heidi Yeh,Swati Gupta
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:We ask when an AI trained on historical decisions should be allowed to recommend decisions, defer to human experts, or in other cases, flag concerning patterns in data and deviate from past practice. This can be challenging because past decisions record what was done rather than what should have been done, and where data is sparse, a model’s predictions may be too uncertain to act on. We provide a framework for addressing this problem using a case study in deceased-donor kidney allocation. Nearly 30% of kidneys recovered from deceased donors in the United States are never transplanted, largely because they are hard to place through the standard waitlist sequence. One can use AI to flag kidneys at risk of delayed or failed placement for out-of-sequence (OOS) allocation, but this comes at the cost of bypassing candidates higher on the waitlist. Thus, OOS allocation should only be used when the organ is demonstrably at high risk of discard on the standard sequence. Using Organ Procurement and Transplantation Network match-run data from 2016-2019 and 2023-2025 (excluding COVID-19-disrupted years), we build the Arbitrator framework, which combines various interpretable models and acts only when the uncertainty in their predictions cannot change the preferred action. Our Arbitrator resolves 93.5% of organ placement decisions using explicit medical criteria learned from data, and defers the remaining 6.5% to a human or a more complex model. We show that the Arbitrator correctly identifies 70.1% of historically discarded low-KDPI kidneys as OOS candidates, projecting that up to 43.9% of historically discarded kidneys could be flagged for expedited placement and potentially transplanted. We then exploit the interpretability of the Arbitrator to create an agentic audit comparing the learned rules to published clinical literature in order to flag decision-making patterns that are contentious.
[AI-110] Representation Complexity Along Learning Trajectories: Trading Width for Precision and Time
链接: https://arxiv.org/abs/2610.39027
作者: Ryan Singh,Miguel De Llanza,Christopher L. Buckley
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:What computational role does overparameterisation play during learning? Information-plane analyses suggest that neural networks can traverse representations substantially more complex than those required by their final predictors. We explore when this transient complexity serves a computational purpose by comparing the representation complexity needed for prediction with that needed to make an unknown target statistically visible through gradient feedback. We introduce an oracle model of neural network learning and use Kolmogorov (n)-widths to quantify effective representation complexity. Our main technical result transfers complexity bounds from the forward representation to the corresponding gradient-query features. Combined with statistical-query lower bounds, this gives a task-agnostic lower bound on the initial effective width required for target-dependent gradient information to become detectable. For (k)-sparse parity in (D) dimensions, a known target admits an (O(k))-width representation, while the general bound already implies a polynomial separation between representation and discovery. We then exploit parity’s invariance under coordinatewise sign reflections to identify a sharper unsigned effective width, which removes sign distinctions invisible to the parity information channel. At a transfer-compatible measurement resolution, its critical scale is (\widetilde\Theta_k\left( D^k-1\tau^2 \right)), up to logarithmic factors. Thus, the representation resources required to discover a predictor can substantially exceed those required to represent it once learned. Interpreting precision through accumulation over fresh samples yields a corresponding width-time frontier. Experiments support the predicted dimension dependence and examine how optimisation and generalisation depart from this idealised frontier.
[AI-111] Assessing feature scorer results on high-dimensional datasets with t-SNE
链接: https://arxiv.org/abs/2610.39009
作者: Bruno Iochins Grisci,Mario Inostroza-Ponta,Márcio Dorn
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:While vast literature on high-dimensional data visualization is available, there are not many works regarding the visualization of feature scorers and their results. Feature scorers are algorithms that assign numerical importance to each feature of multi-dimensional datasets. These importance scores can be used in several applications, such as feature selection, knowledge discovery, and machine learning interpretability. There are several feature scorers to choose from, and often no single metric or ground truth is available to guarantee the quality of their results. In this scenario, visualization can become valuable to support the decision of which method to choose and how good its results are. For this goal, this work presents “weighted t-SNE.” It modifies the relationship between data points in the embedded 2D space to reflect the importance of each dimension of the original datasets as assessed by a feature scorer. This research discusses how to implement weighted t-SNE, proposes the silhouette coefficient as a numerical evaluation of the results, and shows several examples of its use in practice. Synthetic and real-world tabular datasets are used in the experiments together with nine feature scorers, ranging from Mutual Information to neural networks. Each feature scorer produces unique visualizations, and weighted t-SNE can be used to compare and choose the one that better suits a given dataset and task. Weighted t-SNE can also visually show the importance of features learned by machine learning models and help us see how they are organizing the data, increasing their interpretability.
[AI-112] Formal Specifications with Data-Driven Signal Analysis
链接: https://arxiv.org/abs/2610.39007
作者: Roderick Bloem,Benedikt Maderbacher,Cristinel Mateis,Dejan Ničković,Sergii Tkachov,Björn Ziegler
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Developing analog and hybrid systems requires substantial domain knowledge that is hard to articulate in a formal specification; without additional effort, this domain knowledge is also inaccessible to machine learning methods. We propose a hybrid approach for analyzing and monitoring real-valued signals that integrates specification-based and data-driven techniques. We use regular expressions with interval operations to identify regions of interest in a signal, classical signal processing methods to extract semantically meaningful features, and semi-supervised machine learning to infer complex relationships that distinguish correct from faulty behaviors. The relationships learned from normal behavior data can automatically detect incorrect behavior in the same system or variants thereof. We evaluate the method on a voltage regulator, demonstrating its effectiveness in detecting undesired system behaviors.
[AI-113] Agent ic AI Reasoning and Learned Sampling for Task-Aware Projection Selection in Sparse-View CT: Predicting Hemodynamics
链接: https://arxiv.org/abs/2610.39003
作者: Mahmoud Elhadidy,Roshan M. D’Souza,Amirhossein Arzani
类目: Artificial Intelligence (cs.AI); Fluid Dynamics (physics.flu-dyn)
备注:
Abstract:Sparse-view computed tomography (CT) reduces the number of acquired projections, but conventional sampling strategies are designed primarily for image reconstruction rather than for downstream physical inference. Here, we investigate optimal task-aware projection selection for directly predicting hemodynamics from sparse CT angiography (CTA) measurements. Using computational fluid dynamics simulations of intracranial aneurysms, we generate synthetic fan-beam CTA projections and train a Fourier neural operator (FNO) to predict two targets: the spatially resolved time-averaged wall shear stress (TAWSS) field and the spatially averaged TAWSS. The FNO was coupled with optimal input selection for simultaneous optimal sampling and training. Our goal was to investigate if agentic AI could discover optimal sampling strategies that outperform current methods. We compared fixed uniform and golden-angle sampling with three task-aware optimal selection strategies: Gumbel-Softmax, a Soft top-K method, and an agentic AI selector. The methods were compared at different view selection budgets. A detailed analysis of the AI agent was performed to understand its reasoning and performance. For spatially resolved TAWSS, Agentic and Soft top-K achieve the lowest mean errors (Agentic performs slightly better) and substantially outperform fixed sampling under extreme sparsity. For spatially averaged TAWSS, optimized selection becomes important at extremely sparse acquisitions. Analysis of the selected masks reveals multiple distinct subsets concentrated around informative regions. Moreover, masks optimized for hemodynamic prediction yield poor image reconstructions, showing that task-optimal acquisition differs fundamentally from reconstruction-optimal acquisition. This work suggests the possibility of designing sparse acquisitions directly around downstream physical quantities.
[AI-114] Rethinking GFlowNet Training: Negative Reinforce Dynamics
链接: https://arxiv.org/abs/2610.38994
作者: Pedro Dall’Antonia,Daniel Csillag,Diego Mesquita
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 22 pages, 9 figures, 3 tables
Abstract:Generative Flow Networks (GFlowNets) are models that learn to sample discrete objects proportional to a target reward. They are trained by minimizing balance errors over sampled trajectories, which inherently ties correctness to exploration: if a region is not visited, balance cannot be enforced there. Adaptive exploration is therefore crucial, yet standard training relies on a semi-gradient update that treats the sampling distribution as fixed at each optimization step. On the other hand, the full gradient reveals a second mechanism that reduces the observed errors in an anti-exploratory manner. In this paper, we establish a connection between the full gradient and min-min dynamics between the GFlowNet and a decoupled sampler, showing that the sampler reduces observed violations by avoiding high-error regions. This insight motivates a bridge between two-player games and single-policy dynamics, which underpins the design of our proposed method, Negative Reinforce Dynamics (NRD). By tying the parameters of an adversarial game together, we recover a single-policy dynamic that extends exploration while maintaining correctness. In standard benchmarks, our method outperforms existing approaches in both convergence speed and exploration efficiency, without training a separate teacher or auxiliary exploration policy.
[AI-115] Syntactic Segmentation and Collage Generation for Videogame Levels with Perceptual Distance
链接: https://arxiv.org/abs/2610.38987
作者: Ross Mawhorter,Matthew Guzdial
类目: Artificial Intelligence (cs.AI)
备注: 11 pages, 12 figures, 22nd Artificial Intelligence and Interactive Digital Entertainment Conference
Abstract:In procedural videogame level design, many existing techniques combine hand-authored level chunks. Some existing analysis techniques also rely on a higher-level graph structure which abstracts the level design. In this paper, we explore techniques for computing both of these things automatically from an existing design. We introduce the concept of perceptual distance, based on analysis techniques from the architecture subfield of Space Syntax. We use this perceptual distance to segment level designs with hierarchical clustering. We compare with past work on StarCraft analysis, show results on rooms from Super Metroid, and also demonstrate a proof-of-concept level generation algorithm based on this segmentation method.
[AI-116] Benchmarking Graph Neural Networks for AC Power Flow: The Role of Inductive Biases and Physical Constraints
链接: https://arxiv.org/abs/2610.38985
作者: Federico Clerici,Ferran Bohigas-Daranas,Hamid Latif-Martínez,Eduardo Prieto-Araujo,Oriol Gomis-Bellmunt,Pere Barlet-Ros
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Fast and accurate models for AC power flow (ACPF) are increasingly important as modern electrical grids require more frequent system analyses. Graph neural networks (GNNs) naturally exploit electrical grid topology, yet the effects of architectural design choices on predictive performance remain poorly understood given differences in graph representations and evaluation settings across studies. We present a unified benchmark of representative GNN architectures for ACPF prediction under a common experimental framework. All models use identical preprocessing and training procedures and comparable parameter budgets, isolating the impact of message-passing mechanisms and inductive biases. We also analyze errors by distance from the slack bus and systematically ablate key design components. We evaluate predictive accuracy and physical consistency on the IEEE 14-bus and IEEE 118-bus systems. Results indicate that anisotropic message-passing architectures offer the most favorable trade-off between predictive accuracy and physical consistency under the studied parameter and topology constraints. Explicit edge features encoding line impedances and global context aggregation are particularly important for representing electrical interactions and system-wide dependencies, especially for voltage-angle predictions on the topologically bottlenecked IEEE-118 grid. Under a budget with approximately fixed parameters, two or three message-passing layers generally provide the best balance between depth and representation width, while additional depth offers no consistent improvement. Ablations also show that standard residual and root connections are not necessarily beneficial in shallow models, and that aggregation operators involve trade-offs between statistical accuracy and physical consistency. These results provide practical guidelines for designing graph-based power system models. Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI) Cite as: arXiv:2610.38985 [cs.LG] (or arXiv:2610.38985v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.38985 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-117] Why Speculative Decoding Works Better Than Predicted on Sparse MoE Models NEURIPS2026
链接: https://arxiv.org/abs/2610.38981
作者: Ekagra Ranjan,Komal K. Teru,Bharat Venkitesh,Acyr Locatelli
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: Accepted as at NeurIPS 2026 main track conference
Abstract:Speculative decoding (SD) accelerates autoregressive inference by leveraging the memory bound nature of decoding, verifying multiple draft tokens in a single forward pass. For mixture-of-experts (MoE) models, prior work predicts a concave speedup-batch size relationship, with gains peaking at intermediate batch sizes due to expert saturation. We evaluate this across 10 sparse MoE models (30B–1T parameters, 8–512 experts) and find recent fine-grained models break this trend, showing higher speedup at low batch size. We find verification cost is governed not only by activated-expert bytes, but also by temporal overlap across draft tokens and GPU execution dynamics. At low batch size, MoE layers in fine-grained models account for a much smaller share of runtime than of weights; on H100, kernel profiling attributes this mainly to higher bandwidth utilization of MoE kernels relative to attention, with L2 absorbing repeated reads of co-routed experts. Across draft tokens, routing overlap further reduces the unique experts activated during verification relative to independence assumptions. We model this effect using a maximum-entropy Iterative Proportional Fitting (IPF) framework that predicts unique expert counts from routing-overlap statistics with less than 4% error, and substantially narrows the gap to real-model curves and captures the qualitative low-batch trend from synthetic routing.
[AI-118] Context Compaction for World Models
链接: https://arxiv.org/abs/2610.38980
作者: Hayder Tirmazi
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:World models let observers explore and interact with generated worlds. Recent work shows that when observers revisit a region, current models often fail to regenerate it consistently. We view this as a context retrieval problem. LLM agents manage context retrieval at two layers of their stack. At the model layer, the KV cache design determines how much context fits in memory. At the agent harness layer, harnesses such as Claude Code pick the context sent to the LLM. Prior work on world models has mostly focused only on the model layer. In this paper, we make three contributions. First, we build a benchmark to measure world model context retrieval. It generates new worlds for every episode, preventing a model from recalling rooms using training data. It also separates context retrieval from guesses based on nearby frames. Second, we build an agent harness that performs context compaction for world models, which to our knowledge no prior work has studied. Third, we use our benchmark to study four world model designs. Context compaction lets our world models retrieve rooms up to 4 times beyond their context window without retraining. However, world models with sliding-window attention lose most of their context retrieval when dreaming. Therefore, a dreaming world agent benefits more from extending the reach of its world model using context compaction than switching to sliding window attention at the model layer.
[AI-119] Dungeons Drag ons Encounter Analysis with Semi-Random Simulation in Foundry VTT
链接: https://arxiv.org/abs/2610.38975
作者: ravyn Hardcastle,Matthew Guzdial
类目: Artificial Intelligence (cs.AI)
备注: 10 pages, 6 figures, 22nd Artificial Intelligence and Interactive Digital Entertainment Conference
Abstract:Dungeons Dragons (DD) is a tabletop roleplaying game (TTRPG) in which a group of players, guided by a Game Master, roleplay a fantasy adventure with simulated combat and improvisational narrative. In DD, an encounter is a single instance of simulated combat, where the players must use their characters to attempt to defeat a selection of enemy characters that the GM controls. DD has multiple systems that approximate encounter difficulty by summing individual difficulty values that designers predefine for each enemy. However, since players and enemies can have a wide variety of capabilities, encounter difficulty depends on more than just the estimated difficulty of each monster. For example, these metrics do not account for positioning and dungeon layout, the synergies between monster abilities, and player party composition. We have created a tool, Mage-Hand, that can simulate DD encounters semi-randomly using Foundry Virtual Tabletop, a virtual tabletop (VTT) system. In this paper, we show that our tool’s simulated winrates have a strong correlation with DD’s published difficulty evaluation metrics, while also better aligning with some players’ reported experiences.
[AI-120] ableTune: Measured Self-Correcting Draft-Length Tables for Dynamic Speculative Decoding
链接: https://arxiv.org/abs/2610.38968
作者: Abhishek Malvankar,Paul Castro,Nick Mitchell,Lionel Villard
类目: Emerging Technologies (cs.ET); Artificial Intelligence (cs.AI)
备注:
Abstract:Speculative decoding speeds up large language model (LLM) serving only while the GPU has spare compute, so the best draft length, the number of tokens a small drafter guesses per step, shrinks as the batch grows. vLLM lets operators set the draft length for each batch size with a hand-written table, but nothing measures or corrects that table. We present TableTune. Its tuner times the decoding step for every batch size and draft length on vLLM’s own serving path and computes, in minutes, the table that maximizes throughput. Its controller, a vLLM plugin, tracks how many guesses are accepted in live traffic and rebuilds the table while serving. We measure TableTune only against other ways to speculate. On an H100, the tuned table delivers at least 94% of the throughput of the best fixed draft length at every load in four of five drafter settings (88% in the fifth) without the operator having to know which length that is. Against reimplementations of TurboSpec- and Nightjar-style control on the same engine, its worst-case loss is never higher than theirs, and with the EAGLE-3 drafter it is 3% against 9% and 12%. On a mixture-of-experts model, vLLM’s documented table loses up to 18% and the tuned table at most 3%. Starting from mistuned tables, the controller raises throughput by up to 58% (up to 45% on vLLM’s default runner), and after a workload switch with the DFlash drafter it recovers 96% of a stale table’s loss. In llm-d’s optimized-baseline deployment as shipped, two Qwen3-32B replicas on four GPUs, the tuned table is up to 32% faster than a fixed draft length of 3 and up to 25% faster than vLLM’s documented table at high load, and the controller raises the documented table’s throughput by up to 29%.
[AI-121] reasuryQA Bench: Can AI Agents Work an Archive of Scanned Financial Records?
链接: https://arxiv.org/abs/2610.38924
作者: Ritiz Tambi,Sahar Shayegan,Joshua Meyer,Sun Kim,Ali Khan,Victor Shih,Mehdi Jamei,Andi Partovi
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Organisations keep decades of financial records as scanned PDFs on shared drives, and AI agents can now reach those drives through Model Context Protocol (MCP) servers. Existing benchmarks do not mirror this production setting because they simplify the task. In some, the relevant documents are handed to the agent, so it never has to search. In others, the agent searches through documents with pre-extracted text, so it never has to read scanned images. We introduce TreasuryQA Bench, which requires both: 70 questions over 1,435 scanned U.S. Treasury publications, held in Veris’s high-fidelity Dropbox twin. In each task the agent searches for the relevant file through Dropbox’s MCP server, reads the pages it needs with Reducto’s hosted OCR server or its own tools, and replies with one exact figure. A Python grader checks each answer against an accepted range around the correct value. In this paper we present results from ten agents. The best, Claude Opus 5.5 in Claude Code, solves 64.8% of attempts and the worst, GPT-5.6 Terra in Codex, solves 16.7%, so the benchmark is far from saturated. We also test how scan quality affects performance: every agent solves fewer tasks on the blurriest scans. Most failed attempts reply with a wrong number rather than no number, and the most common cause is a digit misread from the scan. Three worked examples are included with their answer keys.
[AI-122] russTutor-Bench: Benchmarking large language models on solution accuracy sycophancy and error detection in statics
链接: https://arxiv.org/abs/2610.38917
作者: Evangelos Apostolakis,Erfan Hamdi,Sara Jahedazad,Emma Lejeune
类目: Computers and Society (cs.CY); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 25 pages, 7 figures. Code: this https URL
Abstract:Large language models (LLMs) have rapidly become a default source of help for students in science and engineering courses. In foundational engineering mechanics courses such as statics, students routinely use these tools to check answers, work through partial solutions, and ask for explanations. Effective tutoring in this setting requires more than a correct final answer. A model must also confirm correct work when a student doubts it, and identify errors in flawed work. Existing benchmarks evaluate LLMs primarily on final-answer accuracy, so it is unclear whether models that can solve mechanics problems also have the other technical capabilities that tutoring depends on. Here we introduce TrussTutor-Bench, a benchmark of 200 verified two-dimensional truss problems paired with automatically generated student solutions, and use it to evaluate thirteen LLMs on solution accuracy, sycophancy, and error detection. Baseline accuracy ranged from 64.5% to 97.0%. Every model detected nearly all injected errors in synthetic flawed student work (90.4% to 100%), so error detection did not distinguish models under these clean conditions. Sycophancy did. When a student insisted that correct work was wrong, models rejected correct solutions to problems they had themselves solved in 1.6% to 50.0% of cases. These results show that performance varies widely across models: the leading models solve, check, and defend student work nearly perfectly, while others are far less reliable, particularly when a student doubts correct work. TrussTutor-Bench provides a verified, extensible standard for determining which models are technically reliable enough to put in front of students. Follow-up work must now focus on teaching rather than technical capability, drawing on learning science to determine whether LLM help leads students to understand the material or simply to reach the right answer.
[AI-123] Lev-3B: A Data Recipe for Compact Probabilistic Log Decision Models
链接: https://arxiv.org/abs/2610.38915
作者: Leo Betthauser,Yingqi Liu,Uday Shah,Jishang Wei,Sainyam Galhotra,Liang Gou,Vedant Dharnidharka,Hao Yang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Many log-analysis tasks involve choosing among known options, such as detecting anomalous sessions or identifying attack techniques, and benefit from calibrated probabilities. We present Lev-3B, an open 3B-parameter decision model built on Granite-4.1-3B-Base with a shared pointer readout, and study its training recipe. The recipe has two key parts. First, training follows a curriculum: the model learns bounded natural-language decisions, using no log data, before it sees any logs. Second, the log stage mixes the target tasks with general log data spanning 35 auxiliary tasks unrelated to either evaluation task. We evaluate one shared checkpoint on sources seen in training and on sources excluded from training, calibration and model selection. The two-stage recipe transfers substantially better than training on log data alone. Adding general log data improves performance on excluded sources. At a fraction of their serving cost, Lev-3B outperforms prompted frontier models on anomaly detection, including on an excluded logging system, but trails them on attack-technique classification. Its probabilities are well calibrated on familiar sources, but on the excluded logging system it confidently misses anomalies. We release the checkpoint, recipe, data manifests, and prediction-level artifacts.
[AI-124] FlipJev: Fast Black-Box Red Teaming Finds Universal Adversarial Suffixes for Typed Decision Models
链接: https://arxiv.org/abs/2610.38895
作者: Xiaogeng Liu,Xinyan Wang,Chaowei Xiao
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注: Pre-print. Code: this https URL , Webpage: this https URL
Abstract:Typed decision models are an emerging class of language models that make decisions instead of generating free-text responses. They answer in a single fast and cheap query and are increasingly used, including in security-sensitive tasks. However, red teaming of these models leaves two gaps: existing adaptive attacks do not use their speed, and few studies have searched for universal adversarial examples. We fill both gaps on Jev, the representative typed decision model, with a quick black-box query-based attack that calls no attacker language model. We conduct extensive experiments on 12 datasets with different data types and application scenarios. The results show that Jev resists common prompt injections: no strategy changes a correct answer in 10,240 paired tests. In contrast, our attack finds short suffixes that flip confident answers within minutes. These suffixes are universal: one suffix per dataset flips 43% of unseen documents. Because typed decision models, unlike classic NLP models, can serve many tasks, we test whether these suffixes transfer across tasks. We find that one suffix generated from one task can flip 90-100% of the samples in other tasks. We therefore believe the evaluation of typed decision models must include strict adaptive attacks to fully reveal their robustness.
[AI-125] Aligning Autoresearch: Co-Improvement between Agents and Humans
链接: https://arxiv.org/abs/2610.38882
作者: Tianjian Li,Chuanyang Jin,Seungone Kim,Tong Chen,John Hoffman,Yoram Bachrach amd Pedro Rodriguez,Ping Yu,Eryk Helenowski,Swarnadeep Saha,Jack Lanchantin,Weizhe Yuan,Ilia Kulikov,Jason Weston
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Autoresearch agents can now attempt long-horizon AI research end-to-end, i.e., self-improvement. However, best results are still obtained by research agents interacting with human researchers, and improving AI’s ability to perform this co-research will only reinforce this effect—a process termed co-improvement (Weston and Foerster, 2025). Further, aligning with human expectations is likely a safer path for research agents, especially given recent agent reward hacking phenomena (OpenAI, 2026b). In this work we address these issues in four ways: (i) we empirically study the role of humans in directing autoresearch agents, focusing on the domain of benchmark creation, and find humans still play a pivotal role particularly in the ideation phase; (ii) we also develop a benchmark, HackBench, that shows current autoresearch agents are not aligned and are unable to achieve hack localization sufficiently; (iii) we present a method to train agents for improved human interaction by using human simulators as a proxy during reinforcement learning (RL); and finally (iv) we develop a method to improve model text generation, and hence agent communication, by RL with rubrics aligned with human experts. Together, our training and evaluation recipes provide a path toward more capable and aligned (co-)research agents.
[AI-126] Auditable Belief Updates as a Missing Safety Layer for LLM -Powered Multi-Agent Systems
链接: https://arxiv.org/abs/2610.38869
作者: Zhuoyun Li,Boxuan Wang,Jinwei Hu,Xiaowei Huang,Yi Dong
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Large Language Models (LLMs) are increasingly being used as cognitive interfaces that shape how users interpret evidence, communicate, and make decisions. As deployed systems become more multimodal, tool-using, and multi-agent, safety failures no longer arise only from isolated harmful outputs, but also from how agents update, store, and propagate epistemic commitments across retrieval, memory, tools, and peer communication. In this paper, we define belief as an engineering abstraction for evidence-based commitments that guide downstream action. Building on this view, we propose the Explicit Belief Multi-Agent System (EB-MAS), which externalizes belief states into a structured substrate, uses belief operators to propose candidate updates, and applies a Cognitive Firewall to bound or reject updates that are weakly evidenced but strongly driven by vulnerability cues. We also discuss the risk of group belief at the group level, including manufactured consensus, collusion, and cascading misinformation, and outline practical integration challenges around grounding, false positives, scalability, and ethical misuse. Therefore, scalable agent safety requires not only better output filtering or stronger base models, but also explicit interfaces for tracing, calibrating, and constraining belief revision at runtime.
[AI-127] Where You Inject Matters: The Trust Region Effect in LLM Agents
链接: https://arxiv.org/abs/2610.38867
作者: Chengquan Guo,Xiaodong Liu,Weiwei Yang,Mingqian Feng,Jialin Song,Qiyao Ma,Carolyn Buractaon,Christopher White,Bo Li,Jianfeng Gao
类目: Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR); Machine Learning (cs.LG)
备注:
Abstract:LLM agents are increasingly adopted in long-horizon tasks and interact with diverse external environments. Despite their growing capabilities, these agents remain vulnerable to real-world attacks such as indirect prompt injections, which manipulate agents through malicious instructions embedded in external environments. While existing risk assessment work has primarily focused on the injection content, the effect of where an injection appears has not been systematically characterized. Studying injection placement requires agents to operate over longer-horizon tasks and interact with diverse environmental surfaces. Thus, we build three agentic testbeds spanning tool use, coding, and daily workflow domains with diverse injection surfaces. Holding the injection content fixed, we find that agent susceptibility varies across injection placements. We refer to this phenomenon as the trust region effect: agents implicitly assign different levels of trust to different regions of their environments. For example, an agent powered by GPT-5.4 follows an unsafe injection instruction placed in a repository documentation file (i.e., a README or TODO file) 2.6 times more often on average than the same instruction placed in a code or data file. Building upon this observation, we develop learning-based attackers that generate injection content and select better injection placements. Our approach demonstrates that both where an injection is placed and what content it contains can affect the strength of indirect prompt injection attacks. To the best of our knowledge, this work provides the first systematic characterization for the effectiveness of different injection placements, spanning both coarse-grained environmental surfaces and fine-grained positions. The uneven trust that agents implicitly place across their environments provides important insights for future attack and defense design.
[AI-128] Learning to Reason from Reference Samples Alone: Reinforced Adversarial Learning for Modern LLM Post-Training
链接: https://arxiv.org/abs/2610.38851
作者: Albert Bou,Javier Valdes-Aleman,Jon M. Laurent,Samuel G. Rodriques,Andrew D. White,Michaela Hinks,Siddharth Narayanan
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 31 pages, 6 figures, 10 tables, including references and appendices
Abstract:In many settings where LLM post-training is desirable, high-quality answers are recognizable but difficult to score with explicit rewards. Reinforcement learning with verifiable rewards (RLVR) is limited to tasks where an explicit correctness signal is available. Prior adversarial approaches to training reasoning LLMs either complement verifiable rewards with co-trained critics or require reference reasoning traces. We ask whether a co-trained LLM judge can turn reference answers alone into the sole reward for on-policy post-training of reasoning LLMs. We call this method Reinforced Adversarial Learning (RAL). We first validate RAL on three verifiable tasks (integer multiplication, code-based arithmetic-expression synthesis, and SMILES-to-formula translation), using the verifiers only for evaluation. With sufficiently capable judges and enough training, RAL matches direct RLVR. We then apply RAL to scientific abstract generation in two settings. In the first setting, RAL guides the generator to produce abstracts in the scientific domain of the reference abstracts, even though the prompt leaves that domain unspecified. In the second setting, we generate abstracts for U.S. National Institutes of Health grant proposals conditioned on the grant title. Human scientists and an ensemble of LLM judges find our generations harder to distinguish from real abstracts than those of frontier-LLM baselines. These results suggest a path toward post-training reasoning models in scientific and other high-value domains, where increasingly capable judges infer hard-to-specify reward functions from example answers.
[AI-129] Discovering Jailbreaks via Curriculum of Models
链接: https://arxiv.org/abs/2610.38833
作者: Michał Burzyński,Noah T. Rossi,Shomit Basu,Andy Arditi
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Automated jailbreak discovery aims to find prompts that elicit prohibited behaviors from safety-trained language models, helping developers identify vulnerabilities before deployment. One class of approaches uses reinforcement learning to train an investigator model to generate prompts that elicit a specified behavior from a target model. However, training directly against a safety-trained target is difficult because nearly all prompts are refused, leaving the investigator with a sparse reward signal. Prior work addresses this with the propensity bound (PRBO), a proxy reward estimated by sampling responses from a more compliant target, and then adjusting for how likely those responses would have been under the original target. We introduce Curriculum of Models (CoM), an alternative method that trains the investigator against a sequence of modified versions of the target, progressing gradually from a version that readily exhibits the specified behavior to the original target. For jailbreak discovery, training begins against a target version that rarely refuses harmful requests, and advances to the next, harder version once the investigator reliably elicits the behavior from the current one. We evaluate CoM on elicitation tasks involving CBRN materials, explosives, and illicit drugs, using two open-weight targets, gemma-4-31b-it and gpt-oss-120b. We compare against PRBO and direct training on the original target without a curriculum. When the investigator is trained to elicit a single behavior, CoM finds successful jailbreaks more reliably than either baseline. When trained on multiple behaviors simultaneously, the CoM-trained investigator can elicit held-out behaviors, though PRBO-trained investigators achieve higher success rates. Our results suggest that a simple curriculum over target models offers a promising approach for addressing sparse rewards in rare-behavior elicitation. Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI) Cite as: arXiv:2610.38833 [cs.LG] (or arXiv:2610.38833v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.38833 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-130] Latent Decision Flow Rewards for Clinical Guideline Reasoning
链接: https://arxiv.org/abs/2610.38709
作者: Gaurav Bhatt,Marinka Zitnik
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Clinical practice guidelines encode evidence-based recommendations as conditional decision flows, which makes them a natural source of process supervision for training reasoning models. Two properties of clinical reasoning make this supervision difficult to use. A decision flow consistent with a guideline does not by itself determine the correct answer for a given patient, which we call the path–answer gap, and several valid decision flows can support the same clinical decision, which we call path multiplicity. Current training approaches address parts of this problem. Answer rewards supervise the final decision, and process rewards computed against a reference flow supervise one valid decision flow per case. Supervised fine-tuning on annotated flows teaches guideline-consistent reasoning but constrains the model to one trajectory per case and can reduce answer accuracy. We introduce GUIDER, a method that jointly optimizes a decision flow controller and a reasoning policy. The controller learns to generate plausible decision flows from guideline-derived supervision, and the policy learns to generate a decision flow and a final answer for each patient case. GUIDER computes the process reward for the policy over several controller-generated flows, which accounts for path multiplicity, and gates this reward on answer correctness, which addresses the path–answer gap. Joint training allows the process supervision to evolve with the policy. On MedGUIDE, GUIDER with Qwen3-14B achieves 82.5 weighted accuracy and 73.5 guideline adherence, 3.3 and 3.1 points above the strongest clinical reinforcement learning baseline. With Qwen3-14B and Gemma4-12B backbones, GUIDER also improves over CRPO, MRPO, and FaithMed on the out-of-domain benchmarks AMEGA-LLM, RealMedQA, and CancerGUIDE.
[AI-131] SatOps-Decision: Low-Latency Discriminative Decisions for Onboard Satellite Communication Operations
链接: https://arxiv.org/abs/2610.38697
作者: Lifan Mei,Yuxuan Li,Jiaxian Tu,Shenghe Xu
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Networking and Internet Architecture (cs.NI)
备注:
Abstract:We propose SatOps-Decision, a method for low-latency decisions in satellite communication operations. To the best of our knowledge, it is the first application of Jev-style general-purpose discriminative decision models to satellite operations, and the accompanying benchmark is the first dedicated to this approach. The learned scorer, a locally adapted Laya encoder with its native decision head, selects among supplied operational answers in a single forward pass. The benchmark combines authored operational contracts, controlled state changes and independently checked labels in an auditable construction. Evaluation prioritizes complete-request latency and raw decision accuracy; all main accuracy comparisons use the same 1,024 held-out questions, separate from training and development. SatOps-Decision’s local scorer has a median latency of 90.3 ms per decision, versus 493.9 ms for Qwen3.5-0.8B generation and 388.4 ms for OneJev1.6-0.8B: 5.47\times and 4.30\times speedups based on median request times in the measured FP32/MPS implementations. Its mean raw accuracy is 85.3%, versus 42.5% for frozen Laya, 40.7% for Qwen generation and 58.6% for OneJev. As cloud accuracy references, original Jev 1.14/1.13 reach 87.4/81.9%.
[AI-132] Deogen: Automatic Benchmark Generation for Reasoning over Policies
链接: https://arxiv.org/abs/2610.38687
作者: Sam Kouteili,William Liang,Keeran Murray,Adriana Zhou,Ramesh Viswanathan,Scott Shapiro,Ruzica Piskac
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Language models are increasingly being deployed in regulated settings, where a model must apply given rules to the facts of a situation to determine a verdict. Benchmarking this capability, known as deontic reasoning, remains a costly and manual process, with existing datasets largely hand-crafted. In this paper, we present DeoGen, a scalable, automatic, and parametrizable benchmark generator for deontic reasoning. By autoformalizing policy rules to first-order logic, we can construct decision trees that capture possible outcome trajectories. Each path down a tree becomes a benchmark entry, with each tree producing combinatorially many entries relative to tree depth. Paths ending at a leaf entail a verdict, while non-leaf paths measure a model’s ability to abstain when presented with insufficient information. Evaluations show frontier models return the correct verdict 89-92% of the time. Notably, performance degrades (22-40% correct) on entries without sufficient information to claim a verdict, even with explicit instructions to abstain in such cases. In all, DeoGen presents an automatic, scalable, and parametric method for generating deontic reasoning benchmarks.
[AI-133] Explore Where It Matters: Targeted Exploration for Efficient Rover Adaptation
链接: https://arxiv.org/abs/2610.38657
作者: Jacob Levy,Arthur Bouton,Tristan D. Hassler,Hari Nayar,David Fridovich-Keil
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Conference on Robot Learning 2026
Abstract:Future planetary rovers will traverse steep, deformable terrain where slip, sinkage, and terrain deformation can threaten mobility. Although simulation can capture some wheel-soil effects, unmodeled deformable-terrain phenomena create sim-to-real gaps requiring deployment-time adaptation. Yet, rover operations leave little room for extensive trial-and-error. We introduce TERRA (Targeted Exploration for Reward-Relevant Adaptation), an online adaptation framework that efficiently seeks deployment data exposing task-relevant model errors and corrects them. TERRA adapts a simulation-trained world model with a reward-weighted prediction loss and trains a loss predictor to estimate where future performance-relevant errors are likely to occur, using this signal to guide exploration during MPC. In simulation, TERRA reaches approximately 90% slope-climbing success within five trials under severe out-of-distribution pretraining conditions, outperforming exploration baselines in reward, completion time, and energy to first success. On the ERNEST rover, TERRA recovers coordinated active-suspension locomotion on a deformable Mars-simulant slope within three to five real-world trials. Project website: this https URL
[AI-134] Enhanced image forensics through partial C2PA-based provenance information
链接: https://arxiv.org/abs/2610.38639
作者: Riccardo Ziglio,Eleonora Marchesini,Andrea Montibeller,Giulia Boato,Silvio Ranise,Cecilia Pasquini
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Multimedia (cs.MM); Image and Video Processing (eess.IV)
备注: paper accepted to IEEE International Workshop on Information Forensics and Security (WIFS) 2026
Abstract:Cryptographically verifiable C2PA-based Content Credentials (CCs) can be attached to freshly acquired or generated digital images but also to already existing ones with no previous provenance information, for instance after processing through trusted C2PA-enabled software tools. The latter case leads to images with so-called partial CCs, which capture only the last operations applied, while prior provenance remains unknown and can be investigated through signal forensics. This paper addresses the forensic analysis of such images, highlighting the opportunity to leverage the partial CCs as additional trusted knowledge to adapt forensic detectors towards the statistical characteristics of the image under analysis and mitigate potential distribution shifts. To this end, a two-step strategy is envisioned: (i) reconstructing the operations applied by the last signer through a structural analysis of the partial CCs; (ii) exploiting the obtained reconstruction to enhance the forensic detector of the analyst through an adaptation process. An empirical analysis on partial CCs produced by commercial C2PA-enabled software tools through different editing operations is carried out, providing evidence on the feasibility of the reconstruction in practical cases. An experimental case study on the forensic task of synthetic image detection is then presented, showing that the proposed strategy can significantly enhance the performance of state-of-the-art detectors on test images that are postprocessed via C2PA-enabled software tools after being generated/acquired.
[AI-135] On-Policy Self-Distillation with Teacher Evolution
链接: https://arxiv.org/abs/2610.38637
作者: Shufan Shen,Zhongni Hou,Chao Bi,Yufei Zhang,Guojun Yin,Shuhui Wang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:On-Policy Self-Distillation (OPSD), which aligns a student model’s token-level distribution with that of its self-teacher prompted by privileged context, has emerged as a promising direction to mitigate reward sparsity in sequence-level post-training paradigms. However, OPSD often fails to achieve continuous performance gains during training. In this paper, we find that the model’s limited self-teaching capability underlies this phenomenon by generating unreliable guidance on certain samples. To address this problem, we propose Teacher-Evolution Self-Distillation (TESD), which enables the self-teacher to generate rollouts and receive environmental feedback based on task correctness and context independence. First, using these rewards as training objectives, we perform teacher-oriented optimization to continuously enhance the model’s self-teaching capability during distillation. Second, by estimating teacher guidance reliability from these rewards, we apply adaptive student-oriented optimization to down-weight samples with unreliable guidance. Extensive experiments across multiple datasets and model scales show that TESD consistently outperforms representative token-level and sequence-level methods, highlighting the necessity of improving supervision quality in OPSD.
[AI-136] Curriculum-Based Deep Reinforcement Learning for Robust Autonomous Rocket Landing under Atmospheric Uncertainty
链接: https://arxiv.org/abs/2610.38636
作者: Atin Chowdhury,Piyush Joshi
类目: Artificial Intelligence (cs.AI)
备注: 21 pages, 12 figures
Abstract:Autonomous far-range rocket landing is a challenging control problem because navigation, descent-rate regulation, attitude management, and fuel use must be coordinated in the presence of uncertain atmospheric and vehicle dynamics. This paper presents a curriculum-driven proximal policy optimization (PPO) framework for three-dimensional rocket landing in a stochastic environment incorporating altitude-dependent aerodynamic drag, wind shear, temporally correlated turbulence, discrete gusts, uncertain propulsion and aerodynamic parameters, thrust-vector misalignment, engine lag, actuator-gain variation, noisy observations, and command latency. Learning is organised into far-range approach, soft descent, and integrated landing stages. Policy transfer, critic warm-up after reward changes, rehearsal of earlier tasks, and progressive expansion of weather severity are used to support learning across these coupled phases. Across 100,000 evaluation episodes, the policy achieves 96,656 successful landings, corresponding to a success rate of 96.656% and a 95% Wilson confidence interval of 96.543–96.766%. Success decreases from 98.84% under zero-disturbance conditions to 88.07% in the highest weather-severity band. Successful touchdowns exhibit small miss distances and low residual horizontal and vertical velocities. An exploratory logistic-regression analysis identifies mean wind speed as the strongest measured correlate of success, with initial distance, initial altitude, initial vertical velocity, and wind-shear exponent also associated with outcomes. The results quantify both the effectiveness of the learned policy over the evaluated disturbance distribution and the degradation that occurs under severe compound atmospheric conditions.
[AI-137] “Close enough?” An investigation into multiplication errors in LLM s NEURIPS2026
链接: https://arxiv.org/abs/2610.38634
作者: William Bolton,Bezawit Tilaye,Huw Day
类目: Artificial Intelligence (cs.AI)
备注: Accepted to Neurips 2026 NewInML workshop
Abstract:Large language models (LLMs) frequently fail at multi-digit arithmetic, but the underlying mechanisms governing these failures remain poorly understood. While recent work demonstrates that addition errors follow a “noisy quantisation” model localised around carry boundaries, whether this framework generalises to multiplication remains an open question. In this work, we evaluate multiplication capabilities across several small base LLMs on a dataset of 16,991 problems spanning varied operand lengths. We show that multiplication errors do not conform to the addition-style noisy quantisation model. Instead, we uncover a novel error regime in small Qwen models: errors in lower-order columns predominantly preserve parity. Conversely, higher-order digit errors shift toward off-by-one carry failures. Our findings suggest that arithmetic computation in LLMs relies on distinct, multi-periodic internal representations where parity is tracked more robustly than exact digit identity, providing new mechanistic insights into numerical reasoning in transformers.
[AI-138] hink Before You Act: Decision Gate for LLM Agents
链接: https://arxiv.org/abs/2610.38620
作者: Iftach Shoham,Yuval David,Offer Akrabi,Erez Bilgory,Ido Levy,Sami Marreed,Sergey Zeltyn,Segev Shlomov
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:LLM agents increasingly perform consequential actions through external tools, yet their proposed actions are often executed without independent verification of policy compliance or task completion. We introduce JevAct, a runtime verification framework that intercepts selected agent actions before execution and enables feedback-driven correction. JevAct compiles domain policies into executable state checks and structured questions evaluated by a prefill-only decision model, which produces probability-based judgments without relying on LLM generation. By combining deterministic verification of tool-state facts with probabilistic assessment of conversational context, the framework identifies potentially invalid actions and premature task termination, providing targeted feedback for bounded revision. JevAct operates independently of the underlying agent model and requires no modifications to its parameters or tools. We evaluate JevAct across eleven LLM agents on \tau^2 -bench Airline and Telecom tasks and BFCL. Results show improvements in agent reliability across multiple models and benchmarks, with particularly consistent gains on Telecom and BFCL, while performance on Airline is more mixed. Overall, these findings suggest that combining structured decision models with executable verification can improve the reliability of tool-using LLM agents.
[AI-139] How Deterministic SAST and AI Security Agents Divide the Work: Coverag e Harness Policy and Cost in Snyk VulnBench 2.0
链接: https://arxiv.org/abs/2610.38617
作者: Liran Tal,Chris Liu,Charlotte Fereday,Sebastian Roth,Brian Clark
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Software Engineering (cs.SE)
备注: 44 pages, 21 figures, 7 tables. Benchmark harness, reference sets and all run data: this https URL ; companion site: this https URL
Abstract:We ask what AI coding agents and AI security harnesses find that deterministic static application security testing (SAST) does not, and what they miss that SAST catches. VulnBench 2.0 is a benchmark of 108 hand-curated, attacker-reachable vulnerabilities in 20 open-source web applications in Go, JavaScript/TypeScript, Python and Java; 62 flow across more than one file. A finding is credited only if it locates the vulnerability’s attacker-controlled source and its dangerous sink, each within two lines. We evaluate nine productized configurations under one blind protocol: the Snyk Code static analyzer, Claude Code’s /security-review with three model and effort settings (one with the Snyk MCP server), Codex Security with GPT-5.6 Luna, Terra and Sol, and the DeepSec scanner with two model backends. Snyk Code reached 82.4% macro recall (89 of 108) and Claude Code with Opus 5 at xhigh effort 79.3% (85 of 108); they disagree on 28 vulnerabilities, split 16 to 12, too few to tell them apart. Neither engine was sufficient alone: together they reached 101 of 108, the ceiling of all seven comparable configurations combined, and the AI configurations also reported at least seven real vulnerabilities that the reference set did not contain and Snyk Code did not report. What an AI engine misses depends on its harness as much as on its model: without Claude Code’s built-in review policy, the same model matched 16 instead of 9 of the 20 vulnerabilities in the classes that policy excludes or restricts (supplementary run on 19 of 20 applications); Codex Security with GPT-5.6 Sol located 80 vulnerabilities at file level but 34 at line level; DeepSec’s model-free file selection kept a file from only 56 of the 108 flows. Coverage and cost were only loosely related. The harness, fixtures, reference sets and all run data are public.
[AI-140] Episodic looking - using reinforcement learning to bias perception enhances few-shot and continual learning with protection from forgetting
链接: https://arxiv.org/abs/2610.38577
作者: David Rawlinson,Gideon Kowadlo
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 21 pages, 8 figures
Abstract:There is a fundamental contradiction between the desire for models that can generalize, and models that can continuously adapt to recent input with changing statistics. These conditions, captured in few-shot and continual learning settings, can bias or disrupt learned representations. This harms generalization to a broader context, leaving the model unable to overcome input variance. This paper proposes a solution based on the Complementary Learning System architecture (CLS), in which a Short-Term Memory (STM) delegates generalization to a slow, statistical Long-Term Memory (LTM). Delegation is achieved by working in a space of robust, generalized perceptions produced by the LTM. With this separation, the STM can focus on memorizing recent input and avoiding interference between memories. We show that on an image classification task, the resulting CLS model is more resistant to forgetting in a domain-incremental continual learning setting than a baseline ResNet LTM, and generalizes more effectively in a few-shot setting than other continual learners. In the continual setting, overall accuracy is slightly worse than existing continual learning models, but with a different mix of performance characteristics. Results analysis suggests that the model is limited by the representations and representational plasticity of the LTM. In future work, we will seek to address this by replaying STM memories to the LTM. Comments: 21 pages, 8 figures Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI) MSC classes: 68T05 (Primary) 68T07, 68Q32 (Secondary) ACMclasses: I.2.6; I.2.6; I.5.1 Cite as: arXiv:2610.38577 [cs.LG] (or arXiv:2610.38577v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.38577 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: David Rawlinson [view email] [v1] Wed, 28 Oct 2026 10:09:11 UTC (1,824 KB)
[AI-141] CALM: Conflict-Aware Lifecycle Memory Governance for Long-Horizon Language Agents
链接: https://arxiv.org/abs/2610.38573
作者: Chang Li,Xianhao Ou,Xinyan Ma,Hao Li,Haoyuan Xu,Jiafeng Liang,Xiaohan Wang,Jiajun Chai,Guojun Yin,Ming Liu
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:A memory can be outdated for present-day reasoning yet indispensable for reconstructing the past. Relevance-based retrieval overlooks this tension: semantic relevance, evidence-grounded validity, and query-specific eligibility are distinct. We propose CALM, a conflict-aware lifecycle memory governance framework that decouples persistent memory preservation from query-time admission. CALM converts heterogeneous records into evidence-grounded memory cells, updates validity and lifecycle states using support, conflict, and dependency signals, and admits query-appropriate evidence through training-free or learned gating informed by temporal applicability. Across LongMemEval-S and LoCoMo with Qwen and GPT backbones, CALM outperforms state-of-the-art baselines, achieving the best overall accuracy. Compared with simple filtering and reranking, query-conditioned admission further improves evidence selection. Conflict-blocking and robustness analyses show CALM suppresses erroneous retrieved memories while retaining useful evidence. Historical-reconstruction experiments demonstrate that preserving superseded records supports accurate earlier-state reasoning without compromising current-state answers.
[AI-142] A-RSI: Adaptive Task Refresh for Zero-Data Multimodal Self-Improvement
链接: https://arxiv.org/abs/2610.38568
作者: Yu Liu,Youxing Li,Yuedong Tan,Zhiwei Yang,Zihan Su,Hanqi Feng,Shaojie Zhang,Libin Liao,Zihao Wan,Chenxu Wu,Cong Cao,Fangfang Yuan,Yanbing Liu,Ziheng Chen,Pei Fu,Radu Timofte,Zongwei Wu,Nicu Sebe,Barnabas Poczos,Jian Luan
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Recursive self-improvement requires models not only to acquire useful training experience, but also to decide when their current experience has ceased to be useful. Existing methods often focus on what experience to generate next, while paying less attention to when fresh experience should be introduced. Many also construct training targets through self-consistency or majority voting, making supervision depend on the model’s own uncertain predictions. We address these two questions separately: training dynamics determine when to acquire new experience, while the answers to new tasks remain independently checkable. A-RSI monitors reward variation and accuracy in training logs to refresh a task pool when its learning signal declines, without consulting validation or test scores. A static library of model-authored programs produces multimodal questions and derives their reference answers; we specify the task-generation contract, automatically verify chosen properties of the reward and refresh rules, and separately audit the generated answers. The training loop uses no external training examples or human-written answer labels. On the primary eleven-benchmark suite, A-RSI raises the performance average from 57.77 to 61.05 with 800 fresh tasks.
[AI-143] Co-Evolving Policy and Rubric via Capability Awareness for Open-Ended Generation
链接: https://arxiv.org/abs/2610.38566
作者: Chengsheng Zhang,Wenshen Xu,Yuchuan Wu,Zhiyuan Yu,Junfeng Fang,Tianlong Wang,Bingxue Qiu,Yongbin Li,Xiang Wang
类目: Artificial Intelligence (cs.AI)
备注: 30 pages
Abstract:Rubric-based reinforcement learning (RL) offers interpretable, fine-grained supervision for aligning large language models (LLMs) on open-ended generation tasks. However, static rubric criteria may become saturated or remain beyond the policy’s exploration, rendering the reward signals insufficiently discriminative. While recent efforts seek to guide sampling exploration or evolve rubrics, they fail to explicitly factor in how rubric difficulty should track the evolving policy’s capabilities, resulting in a persistent mismatch between the optimization objective and the policy’s current capability frontier. In this paper, we propose \textbfCo-Evolving Policy and Rubric via \textbfCapability \textbfAwareness (\textbfCoCA), a closed-loop framework that jointly adapts the rubric and policy through criterion-level diagnosis. Specifically, CoCA contrasts query-only and rubric-guided rollouts to characterize each criterion relative to the current policy, recalibrates criteria with mismatched difficulty, and identifies guidance-revealed capability gaps for frontier-based learning. Based on this insight, CoCA either expands rubric-conditioned frontiers through guided optimization or transfers qualified gains through on-policy self-distillation. As the policy updates, CoCA improves the capability diagnosis and enables more reliable distillation, thereby completing the co-evolution loop. Extensive experiments across diverse benchmarks and LLMs demonstrate that CoCA consistently improves performance over rubric-based RL methods. Moreover, CoCA promotes the internalization of guidance-revealed capabilities into query-only behavior while mitigating capability degradation.
[AI-144] PT3: Prototype-Guided Self-Supervised Pretraining for Generalizable mmWave Human Point Cloud Representation
链接: https://arxiv.org/abs/2610.38551
作者: Ziyang Liu,Kieren Yu,Xiao Yang,Xin Yang,Kaishun Wu
类目: Artificial Intelligence (cs.AI)
备注: 30 pages
Abstract:Multi-source millimeter-wave (mmWave) observations offer an opportunity to learn human representations reusable across sensing tasks. However, sparse detections and multipath effects make these observations unstable, challenging existing point cloud pretraining methods. We propose PT ^3 , a prototype-guided self-supervised pretraining framework for mmWave human point clouds. PT ^3 represents local states in spatial regions as sparse compositions of shared prototypes and learns their temporal evolution through autoregressive prediction. Observation consistency further stabilizes representations under radar perturbations. After pretraining on unlabeled sequences from multiple sources, the same frozen encoder supports action classification, dense point generation, joint pose estimation, and future motion prediction on datasets not used for pretraining. Across these tasks, PT ^3 outperforms the compared pretraining baselines and achieves competitive or superior performance to task-specific models.
[AI-145] An Alignment Benchmark for Enterprise Agents on Era by Eon
链接: https://arxiv.org/abs/2610.38544
作者: Hai Sadon,Benjamin Gruenbaum,Doron Porat,Assaf Natanzon,Roy Zavida,Chen Dinachi,Or Itzahary
类目: Artificial Intelligence (cs.AI); Software Engineering (cs.SE)
备注: 11 pages, 1 figure, 4 tables
Abstract:AI agents are widely used in enterprises to access business systems and act on them in pursuit of a goal. However, fulfilling a user’s request can cause harm. Era by Eon is a product that creates a digital twin of a company by simulating its business systems and data. Its simulators implement the programming interfaces of real products. Our benchmark uses ten of these systems, including Salesforce, Zendesk, Jira and Slack. We tested whether agents would carry out harmful business requests or trust misleading information in company records. We ran 19 models on 41 tasks three times, for 2,337 runs. Of the 2,304 runs that produced an answer, 918 (40%) contained a violation. The most common were false claims about actions or records (290 runs), following planted instructions (266) and disclosing sensitive data to the wrong audience (249). Every attempted change was recorded and blocked before it took effect. Misleading Slack messages caused 16 models to report an incorrect ticket status in 47 of 57 runs. No model was free of violations. The four best models violated in 22 to 24% of their runs. Yet one of them, Nova Pro, did the task correctly in only 30 runs.
[AI-146] Value Probes: Learned Value Functions for Beam-Search Red Teaming
链接: https://arxiv.org/abs/2610.38526
作者: Jonathan F. Jao,Andrew Kyle Lampinen
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Automated red-team auditing of language models is an important way to evaluate safety-relevant behaviors. These audits implicitly search a tree of possible conversation pathways: at each turn an auditor model decides which conversation to continue, and an LLM judge scores the dialogue. We emphasize that this auditing procedure is a Markov decision process, where states are the conversation history, actions are candidate responses, and rewards are judge scores for the target behavior. By adopting this perspective, we introduce two major improvements to auditing. First, we propose adopting beam search for the audit rather than relying on the auditor alone. We then interpret the problem of finding states that are likely to lead to problematic behavior as value estimation. Surprisingly, we show that this value function can be estimated by linear probing from the target model’s activations to predict discounted future rewards - a method we term “Value Probes.” By using Value Probes as a heuristic in our beam-search audit, we are able to substantially improve auditing success. Across 18 target behaviors and 3 scales of models, our approach raises peak audit scores by roughly 50% compared to (token-matched) standard auditing approaches. Furthermore, we show that Value Probes can be fit using off-policy data from smaller models. We also ablate individual components, and show that our method offers a comparably-performant and more efficient alternative to more expensive language model follow-ups. We hope these methods will improve language model auditing and lead to safer models.
[AI-147] AtomSteer: Test-Time Constraint Steering for Atomistic Generative Models
链接: https://arxiv.org/abs/2610.38517
作者: Pengfei Cai,Katharina S. C. Jäger,Utkarsh Utkarsh,Nofit Segal,Bowen Deng,Akshay Subramanian,Rafael Gomez-Bombarelli
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Diffusion and flow matching models offer scalable approaches to generating crystalline materials, molecular conformers, and protein-ligand complexes. However, satisfying crystallographic symmetry, structural geometry, and accurate stereochemistry remains challenging. Satisfying these requirements calls for corrections that preserve the geometry already captured by the pretrained model. In this work, we propose AtomSteer, a test-time inference framework for steering pretrained atomistic generative models with structural constraints. During sampling, AtomSteer corrects intermediate structure predictions with bounded local projections, graph-rigid stereochemical transformations, and projections over periodic symmetries, accepting updates that improve constraint satisfaction while limiting changes to the rest of the structure. We demonstrate AtomSteer across three pretrained backbones, FlowMM, ET-Flow, and Boltz-2, spanning de novo crystal generation and crystal structure prediction, molecular conformer generation, and protein-ligand complex structure prediction. AtomSteer improves space-group distribution coverage in de novo crystal generation and improves target space-group recovery in crystal structure prediction. AtomSteer improves molecular validity and E/Z stereochemical accuracy while retaining conformer ensemble fidelity. For protein-ligand complexes, AtomSteer reduces ligand E/Z errors relative to experimental reference structures while achieving competitive docking success and PoseBusters validity. Our method provides a unified framework for adding chemical and crystallographic control to existing atomistic generative models without retraining or architectural changes.
[AI-148] SaiVLA-JEPA: Evaluating Future Latent Prediction in JEPA-Conditioned VLA Control
链接: https://arxiv.org/abs/2610.38508
作者: Rufeng Xiao,Ruihao Xiao,Wei Chen,Xinhai Sun(VLA Research Team, Synthoid.ai)
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注:
Abstract:Joint-embedding predictive architectures (JEPAs) learn to predict representations rather than pixels, offering a compact way to model how visual scenes evolve. We ask whether this predictive objective improves closed-loop control when a vision-language-action (VLA) policy already uses V-JEPA features. We call our JEPA-conditioned flow policy family SaiVLA-JEPA. It combines a vision-language model with a diffusion transformer action expert and receives features from a frozen V-JEPA encoder. Two lightweight auxiliary heads predict future representations during training and are omitted at inference, leaving the action sampler unchanged. We compare policies without JEPA features, with current JEPA conditioning, and with additional future supervision on 10,030 paired LIBERO-Plus episodes. Current JEPA conditioning reaches about 82.5% success versus 71.6% without JEPA, while future supervision provides no detectable aggregate gain at the evaluated training budget. We separately evaluate SaiVLA-JEPA-TCA, which adapts latent context, and SaiVLA-JEPA-TPA, which adapts selected parameters. The context-retry gain is a system-level result under a larger attempt budget; parameter adaptation shows no consistent advantage over the frozen policy and target controls. These results motivate assessing latent-prediction objectives through paired closed-loop rollouts rather than latent loss alone.
[AI-149] Accelerating Nonlinear MPC by Behavior Cloning with Separated Timescales: Application to Nuclear Reactor Load-Following
链接: https://arxiv.org/abs/2610.38496
作者: Perceval Beja-Battais(CB),Alain Grosset{ê}te,Nicolas Vayatis(CB)
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Optimization and Control (math.OC)
备注:
Abstract:Behavior Cloning (BC) of expert controllers for multiscale dynamical systems is challenging: state variables evolving at widely separated timescales induce highly nonlinear couplings that a shared latent space may fail to adequately represent, degrading closed-loop generalization. We propose Structured Representation Learning for Behavior Cloning (SRLBC), an architecture designed to leverage the physical knowledge of multiscale systems in which slow state variables, fast (stiff) state variables, and predictable exogenous inputs evolve at different timescales. On 500 long-horizon roll-outs of a pressurized water reactor (PWR) load-following benchmark -a high-dimensional, stiff industrial system with challenging safety constraints -and across 20 training seeds, separated embeddings systematically outperform shared embedding baselines on cumulative cost, imitation accuracy and constraint satisfaction. The resulting standalone policies approach NMPC closed-loop performance at a computational cost 15 times lower on average. Since standalone policies still leave a few percent of trajectories infeasible, we embed the learned policies in a safety-augmented approximate NMPC framework, in which the optimizer is only called when the predicted input sequence is not certified as safe. This yields controllers still 6.8 times faster than the NMPC controller on average, with a feasibility (i.e., constraint satisfaction) of at least 99.8% for all architectures, SRLBC retaining the lowest closed-loop cost. Under distribution shift, standalone BC policies degrade sharply: SRLBC degrades the least, and once safety-augmented, it retains a feasibility above 98% with the lowest closed-loop cost. Beyond the quantitative results, Jacobian-based sensitivity analysis shows that the resulting controller relies on a sparse and less redundant set of latent features, each attributable to a single group of physical variables, and aligned with the physical timescale decomposition of the system.
[AI-150] racekit Box: Running AI Agents in Micro-VMs Where Every Way Out Is Decided and Signed
链接: https://arxiv.org/abs/2610.38458
作者: Bravish Ghosh(Cygnux Labs)
类目: Computational Engineering, Finance, and Science (cs.CE); Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR); Operating Systems (cs.OS)
备注: 20 pages, 13 figures, 19 tables. Code: this https URL
Abstract:AI agents now install packages, call model APIs with their operators’ keys and change code in repositories that deploy to production. The network, credentials and write access they need are also how they can leak data, steal a key or change a system nobody approved. Sandboxes contain an agent but leave little evidence of what crossed their boundary; tracing records what happened but contains nothing. We present Tracekit Box, an open-source system (MIT, v0) that does both. Each job runs in its own Firecracker micro-VM with no network card. The only ways out are doors on the host, served over vsock: commands, outbound connections, secrets, model calls and the workspace patch. At every door a Tracekit signer the agent cannot reach decides what passes and signs it into one run per box. Names are resolved on the host; private and cloud-metadata addresses are always refused; secrets never enter the VM, which sees placeholders the host proxy swaps for real values only in one header for bound hosts; and changes come back as a patch applied file by file after the signer decides each file. Box fails closed, and each job ends as a .tkb bundle anyone can verify offline. All 13 end-to-end checks pass on a real KVM runner (the test that runs them takes 28.5 s, including image pull and boot); 15 fuzz targets on the trust boundary produced nine findings, all fixed; and the demo run’s 102 signed records verify. Box does not yet run Firecracker’s jailer, needs a co-located signer, and records processes inside the VM only where they cross a door.
[AI-151] DES-JEPA: Learning a Shared Vocabulary for High-Dimensional Discrete Event Sequences
链接: https://arxiv.org/abs/2610.38445
作者: Hugo Math,Maximilian Forstenhäusler,Christos Anagnostopoulos
类目: Artificial Intelligence (cs.AI)
备注: 5 pages
Abstract:Temporal foundation models such as Chronos and TimesFM generalize across domains through a shared numerical observation space. In contrast, discrete event sequences (DESs), including operational event logs or patient clinical-event trajectories, have domain-specific event vocabularies whose labels lack a consistent correspondence across domains. Beyond this lack of consistency, the number of distinct events in real-world settings is high enough to challenge both classical temporal point process modeling and current in-context foundation inference models such as FIM-PP. Critically, current foundation models for DESs are limited to a small, fixed number of event types, restricting their applicability to realistic, high-dimensional event streams. To address this, we propose two complementary modules: (i) DES-JEPA, a joint-embedding predictive architecture that learns a shared discrete vocabulary from synthetic stochastic processes with unaligned event vocabularies; and (ii) DES-PFN, a discrete event sequence prior-fitted network that reuses the learned vocabulary for in-context next-event prediction. Without task-specific fine-tuning, this framework achieves competitive zero-shot performance on eight unseen real-world datasets and scales up to 700 distinct event types.
[AI-152] DecisionUnlearn-Bench: Unlearning Probabilistic Decisions under Changing Schemas
链接: https://arxiv.org/abs/2610.38431
作者: Tiantong Wu,Wei Yang Bryan Lim
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注:
Abstract:Machine unlearning aims to reproduce the behavior of a model trained without designated data. For probabilistic decision models, this target also depends on the candidate options supplied at inference. We introduce DecisionUnlearn-Bench, an empirical framework that compares full, unlearned, and retain-only models across deletion granularities, label renderings, and candidate sets. Across 20 eligible model configurations on the BANKING77 dataset, full-to-oracle displacement increases consistently from example to association to class deletion. Replacing semantic option labels with opaque codes nearly eliminates the retrained oracle’s recognition of deleted classes, even with successful acquisition of the full model, whereas changing the candidate set alters both the oracle and the measured unlearning progress. SimNPO satisfies the screening criterion in 19 of 20 configurations, but its gains vary substantially across seeds and deletion targets. Development tuning on the CLINC150 dataset does not establish a positive pooled effect across 48 intents. At approximately matched suppression, methods can differ markedly in oracle fidelity. Recovery experiments further distinguish initial forgetting from subsequent adaptation and show conclusions that depend on both architecture and metric. These results show that evaluating decision-model unlearning relative to a schema-specific retraining target, with separate reporting of effect size, utility, robustness, and recoverability.
[AI-153] he Code Behind the Layout: AI Agents for Analog Layout Generators
链接: https://arxiv.org/abs/2610.38426
作者: Minxing Chu,Anhang Li,Martins Atilla,Ruichen Qi,Mehdi Saligane
类目: Hardware Architecture (cs.AR); Artificial Intelligence (cs.AI)
备注:
Abstract:Analog layout remains largely manual and hard to automate. Generator frameworks such as gLayout express layouts as programs, producing layouts from parameterized, design-rule-aware generators; however, writing such generators still takes expert effort. Large language models (LLMs) could now write generator code, and frontier models can even draw layout geometry from design intent. This raises a practical question: if a model can draw the layout itself, what does a generator framework still contribute, and is applying these models to a generator framework better than having them draw a layout from scratch? Through controlled experiments on gLayout in SKY130, we find that the main obstacle to drawing a layout directly is its time and token cost, and that the main advantage of a model building on a generator framework is reusability. Its verified primitives keep each model edit small enough to check, and its parameters allow a single edit to be evaluated across an entire family of layouts. Accordingly, a model is most effective when it develops the layout from the generator source of the family that the design intent belongs to, and tests each edit on the family members most likely to fail, so that one verified change improves every layout the generator produces. The results of these family tests can then serve as supervision for a local model, which learns edits that hold across the family at a fraction of the inference cost.
[AI-154] Governance-Aware Federated Learning for Multi-Agent AI Systems: A Policy-Driven Framework for Trustworthy Collaborative AI
链接: https://arxiv.org/abs/2610.38411
作者: Kiranmayee Janardhan,Karthick S,Madhusmita Panda,Bharathi Vokkaliga Ganesh,Syed Ahmed
类目: Artificial Intelligence (cs.AI)
备注: 16 pages, 2 Figures and 3 Tables presented at NGEN 2026 Conference
Abstract:Federated Learning (FL) enables collaborative model training without exchanging raw data, but conventional FL provides limited runtime governance, evidence verification, fairness assessment, explainability monitoring, and auditability. This paper proposes Governance-Aware Federated Learning (GAFL), which combines independent governance-evidence verification, temporal trust, fairness and explanation assessment, anomaly detection, policy enforcement, normalized aggregation, and tamper-evident audit logging. Client-reported governance values are authenticated, bound to their submitted model updates, and independently recomputed before aggregation. A cross-silo prototype with 10 participants was evaluated on the Adult Income dataset under honest, benign non-IID, biased-data, malicious-update, non-compliant, falsified-evidence, and mixed-threat scenarios. Four methods were compared across five random seeds: FedAvg, FedAvg with anomaly filtering, GAFL without independent verification, and complete GAFL. Complete GAFL intervened on all affected participants at the final round across the evaluated risk scenarios, including identifying 100% of falsified-evidence participants, whereas GAFL without verification identified none of the falsified-evidence participants. Final macro-F1 ranged from 0.7520 to 0.7733 across GAFL scenarios. Complete GAFL required 173.67 plus or minus 26.94 ms of governance processing per round, compared with 3.82 plus or minus 0.79 ms for FedAvg. A strict demographic-parity threshold also caused frequent interventions for honest participants, demonstrating a significant fairness-policy and utility trade-off.
[AI-155] SpaceAct: Benchmarking 3D Spatial and Temporal Reasoning via Closed-Loop Action and Active Perception
链接: https://arxiv.org/abs/2610.38405
作者: Chenglong Shi,Zekai Yuan,Yue Song,Jon Pu,Will Wu
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注:
Abstract:To solve problems in 3D space rather than merely describe images, multimodal large language models (MLLMs) need 3D spatiotemporal reasoning: reconstructing 3D geometry from occluded 2D views, anticipating physical dynamics, and executing precise, closed-loop actions. Prior studies test passive VQA on static images or symbolic planning, requiring no execution or supplying privileged coordinates while ignoring collision, statics, and timing. We introduce SpaceAct, an interactive benchmark evaluating 3D spatiotemporal reasoning via closed-loop physical action and active perception, with SAPA, a Spatiotemporal Active-Probing Agent framework. Built on interlocking bricks and mechanical puzzles, it has 108 verified instances across 36 task families in six categories, without privileged coordinates, and diagnostic modes (+plan, +state, no-camera, blind, L7 levers). In brick-bridging or aperture-threading, models rotate the camera to resolve occlusion, read in-scene rulers, output 6-DoF poses under swept-collision and statics checks, and self-correct via split-screen rollback feedback (dynamic tasks freeze time during deliberation and give five strobe frames). Across 10 frontier MLLMs on the 44 hardest instances, scores span 4.8-79.1 (seven solve at most 27.3% of 6-DoF tasks); for the top three, a reference plan (+plan) closes 44% of the gap while leaving collisions at 5.2 per episode, whereas adding exact 3D state (+state) closes 90% and cuts collisions to 0.5; for four others, +plan lifts the mean score from 10.9 to 35.3 and +state to 81.8; and 46% of failures are collision or observation loops. SpaceAct and SAPA advance evaluation to closed-loop physical interaction, showing that even with a valid plan, MLLMs remain bottlenecked by 3D visual grounding and action refinement, while offering a testbed for active multi-view exploration and explicit 3D grounding.
[AI-156] Masked Erosion: How Acceptance Gates Lose Redundant Safeguards in Self-Improving Agents
链接: https://arxiv.org/abs/2610.38392
作者: Hau-Sing So,Daixian Liu,Weicai Long
类目: Artificial Intelligence (cs.AI)
备注: 25 pages (8 pages main text), 10 figures, 35 tables
Abstract:Self-improving LLM agents edit their own harness, the prompts and rules around a fixed model, and keep an edit only if an acceptance gate scores it higher on a batch of tasks. Such a gate sees what the harness does under the current model, not what it would still prevent if one safeguard failed. When safeguards overlap, deleting one is masked: the others still block the harm, so the gate decides on task success alone and redundancy erodes unseen. We measure harm for every subset of four written safeguards in two agent environments and run 411 improvement loops under ten gate rules, mostly as controlled deletion experiments, plus three runs of the GEPA optimizer. A deletion behind another working safeguard adds almost no observable harm. A paired significance test blocks deletions that pass by chance but not deletions that genuinely raise task success, and a larger evaluation batch lets it delete cheaper safeguards: at the same cost, 20-task batches kept every harness protected while 40-task batches lost all six. Large harm audits prevent exposure but not the loss of depth, which turns into harm when the harness is reused with a weaker model. GEPA kept every safety rule verbatim, yet two of its three final harnesses misused credentials under a weaker model. A statistical model frozen before the runs predicts where each gate’s protection breaks as safeguard cost and batch size change, and freezing safety text or auditing one-layer ablations preserves depth.
[AI-157] SCOPE: Selective Critical-Step On-Policy Self-Distillation for Agent ic Reinforcement Learning
链接: https://arxiv.org/abs/2610.38391
作者: Qirui Liu,Wei Cao,Yu Mi,Yichen Sun,Xuekun Zhang,Lei Liu,Yue Shen,Zhixuan Chu
类目: Artificial Intelligence (cs.AI)
备注: 18 pages, 8 figures, 5 tables
Abstract:Outcome-based reinforcement learning provides coarse supervision for long-horizon LLM agents: failed trajectories may contain correct steps, while successful trajectories can contain errors. Trajectory-level skills can also shift teacher preferences at unrelated steps, introducing off-target distillation signals. We introduce SCOPE, a framework for Selective Critical-Step On-Policy Self-Distillation. After a supervised warm-up, the policy analyzes its own trajectories, generates step-specific hindsight skills, and distills their guidance only at selected error steps. A bounded signed teacher-student coefficient stabilizes auxiliary updates while preserving positive reinforcement and negative correction. A success-relative pairwise length reward further encourages efficient task completion. Experiments on ALFWorld and WebShop demonstrate consistent improvements across multiple model backbones. With Qwen2.5-3B, SCOPE achieves 97.7% success on ALFWorld and 85.2% on WebShop, outperforming SEED by 5.9 and 6.3 percentage points, respectively. On unseen ALFWorld tasks, SCOPE reaches 89.6% success, improving over GRPO by 23.2 percentage points with 46.7% fewer interactions. Scaling to Qwen3-32B, SCOPE achieves 99.2% success on seen ALFWorld tasks and 94.0% on unseen tasks.
[AI-158] Compute Has Shape: Self-Morphing the Internal Computation of Frozen LLM s
链接: https://arxiv.org/abs/2610.38382
作者: Junxi Wu,Yiwen Guo,Shu-Tao Xia
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Test-time scaling improves large language models through additional computation, but how to adapt the topology of internal computation to individual inputs remains underexplored. Existing approaches largely fix recurrent or parallel architectures during training, or reconfigure frozen models along a single structural axis. We introduce SelfMorph, which makes computational shape an input-dependent inference-time decision. It unifies the original forward path (BASE), serial reuse of a layer span (REPEAT), and local branching and merging of perturbed hidden states (BRANCH) in a unified space parameterized by type, location, and intensity. Since they offer complementary benefits rather than uniform dominance, we use a lightweight selector that reads the backbone’s depth-wise hidden state trajectory on the prompt to choose a morphology before response generation. We train the selector by distilling utility-based distribution from offline morphing that jointly account for task performance and computational cost. Evaluated on three LLMs spanning 4B-27B, SelfMorph achieves a 8.06% mean relative improvement on five in-distribution benchmarks and 4.45% on four unseen benchmarks. It achieves the strongest performance in 24 of 27 settings. These findings identify input-adaptive computational topology as a complementary axis of test-time scaling beyond generating more tokens. Our code is available at this https URL.
[AI-159] Encode Once Answer Many: Internalizing Context via Parametric Flow
链接: https://arxiv.org/abs/2610.38356
作者: Xinhao Zhong,Mengxuan Wu,Chonghua Liao,Linxi Xie,Ruicheng Feng,Yanpei Guo,Hao Tan,Jiaxin Hong,Jason Li,Panpan Zhang,Ziheng Qin,Qinglin Lu,Yuzhang Shang,Tongliang Liu,Damian Borth,Haggai Maron,Yongxiang Hua,Zhixiang Wei,Jian Hu,Yunfei Wu,Xing Sun,Xin Li,Haoyu Cao,Zhangyang Wang,Kai Wang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Large language models are typically adapted to a task either by placing knowledge in the input context, which keeps it external and re-encodes it on every call, or by fine-tuning, which internalizes it but demands a costly round of optimization per task. We aim to internalize the knowledge carried by a context directly into model parameters, without per-task optimization and without keeping the context in the input at inference time. To this end, we propose CLIK, a context-conditioned parameter generator that maps a given context to LoRA parameters for internalizing knowledge in a single forward pass through a Transformer encoder with cross-attention, requiring no test-time gradient updates. Once the generated LoRA is inserted into a frozen base model, the model answers downstream queries from the query alone, having internalized the context into its weights rather than re-reading it. Because the adapter depends only on the context, it is generated once and reused for every query on the same document, so the cost of reading that document is amortized across queries. On document question answering, CLIK substantially improves over the no-context base model and outperforms existing hypernetwork-based generators, while remaining behind a base model that reads the context directly, and it carries over to multi-turn dialogue without any multi-turn-specific training. Analysis further shows that the context-dependent part of the generated adapter concentrates in a compact and specific substructure.
[AI-160] Investors Last Exam: Can AI Agents Do Equity Investment Without Seeing the Future?
链接: https://arxiv.org/abs/2610.38342
作者: Huangfei Jiang,Anning Hu,Kai Sheng,Yuchen Cai,Rui Ge,Xiangyu Kong,Jianghao Lin,Shuzi Chen,Jiyi Weng,Junchi Yan
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Financial agents must form forecasts and investment judgments from decision-time evidence, with predictions evaluated against subsequent outcomes. Existing benchmarks assess financial knowledge, research workflows, or forecasting, but evaluating the full research process requires a strictly frozen information set and professional evaluation criteria. We introduce Investor’s Last Exam (ILE), an equity-research benchmark led by experienced buy-side investors and fund managers. Its three components are a versioned financial database spanning 3.43 million document versions and nine source categories, supplemented by structured market data, with validated timestamps governing retrieval and computation; a task taxonomy organized around professional analyst deliverables; and an expert-led framework linking scoring criteria to sources, calculations, and assumptions through financial dependency graphs. Evaluation spans 86 assignments and jointly assesses research quality, forecast accuracy, and simulated investment performance. GPT-6 Astra and Claude Opus 5.5 lead overall at 68.1 and 66.9 out of 100, yet even Astra scores only 52.5 on Forecast Model. Our results show that even state-of-the-art models fail to translate overall research capability into predictive performance. ILE shifts the evaluation unit from isolated capabilities to the evidence–assumption–judgment chain, allowing researchers to locate where analytical advantages fail to translate into forecasts and investment judgments, and providing a traceable, diagnostic foundation for training and evaluating financial agents.
[AI-161] MUSE: Multi-Level Memory with Selective Runtime Injection for Scientific Coding Agents
链接: https://arxiv.org/abs/2610.38332
作者: Haoran Jin,Defu Lian,Xiting Wang,Hong Xie
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 17 pages, 8 figures. Under review
Abstract:Historical experience can guide scientific coding at multiple levels, but task-level retrieval may miss implementation lessons that become relevant during execution. We introduce MUSE, a memory management framework combining multi-level memory with selective runtime injection for scientific coding agents. MUSE builds global, approach-level, and implementation-level memory from historical trajectories. At initialization, the agent receives global memory and problem-matched approach-level memory. During execution, current interactions are used to retrieve implementation-level memory. An LLM gate selects relevant experience and generates specific review advice, or skips injection. Across four benchmarks spanning numerical problem solving, kernel optimization, and scientific software repair, MUSE outperforms retrieval and memory baselines on all evaluation metrics. Ablations support the multi-level design and gated injection, while transfer experiments show that memory remains useful across kernel benchmarks. These results show that multi-level memory and selective runtime injection enable effective reuse of experience in scientific coding. Code is available at this https URL.
[AI-162] A Case Study of a Multi-Agent Kernel-Optimization Loop That Learns a New AI Accelerator MICRO2026
链接: https://arxiv.org/abs/2610.38329
作者: Byonggon Chun
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Hardware Architecture (cs.AR); Programming Languages (cs.PL)
备注: 7 pages, 3 figures, 2 tables. Round 1 of the NPU Model Optimization Competition at MICRO 2026. Code: this https URL
Abstract:Frontier large language models write correct CUDA kernels for most single-operator benchmark tasks, yet their accuracy collapses on accelerators whose architecture and programming language they have rarely seen. We present a case study of an LLM agent loop built for such accelerators. Our example is the FuriosaAI RNGD NPU, programmed in a recent, sparsely documented Rust DSL, where the loop optimized three Gemma-4-12B decode kernels over 12 days on a shared, scarce device. The loop builds its own target knowledge: it writes and checks a summary of the DSL documentation, measures the chip with microbenchmarks, records compiler-derived legality rules, and revisits this knowledge when the toolchain changes, rechecking rules and measurements where it can. A role-based search over per-kernel Design Trees explores whole-kernel designs and accumulates small accepted changes in the best version of each kernel. Because device time is scarce and readings are noisy, cheap checks screen new designs before any device run, and late in the run a change was accepted only after repeated comparisons with the best kernel on the device. The final kernels reach a 9.293x geomean speedup over the official baseline, each within 1.25x of the cycles needed just to move its weights. Our prompts, profiler and acceptance test are available at this https URL.
[AI-163] Can LLM s Faithfully Solve Ontology Engineering Tasks? A Mechanistic Interpretability Study
链接: https://arxiv.org/abs/2610.38326
作者: Bohui Zhang,Albert Meroño-Peñuela,Elena Simperl
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Ontological knowledge underpins large language model (LLM) performance on knowledge-intensive tasks. However, existing evaluations of LLMs’ ontological understanding largely rely on behavioral correctness, leaving the underlying mechanisms unexplained. Existing mechanistic interpretability work has largely focused on instance-level factual knowledge, leaving ontological knowledge under-explored. To bridge this gap, we introduce MIOE, a causal-abstraction benchmark that evaluates interchange interventions for two ontology engineering tasks: entity typing and subsumption prediction. MIOE tests whether learned internal intervention targets can change an ontology-level target answer while preserving class-specific auxiliary information. Across Qwen3-8B and Qwen3-4B, we evaluate SAE-, DBM-, and DAS-based methods, including multi-task and multi-layer variants. Distributed subspace methods give the strongest results and generalize across prompt settings, but the induced changes remain only partially separated from auxiliary class-specific information. Multi-task objectives and contiguous layer-window interventions improve this balance, while class-level analyses reveal substantial variation across ontology classes. These results suggest that ontological predictions are mechanistically accessible in LLMs, although current methods do not yet isolate a cleanly disentangled causal mechanism for ontological knowledge.
[AI-164] What Should a Navigator Remember? Spatiotemporal Memory for Zero-Shot Vision-Language Navigation
链接: https://arxiv.org/abs/2610.38318
作者: Yuhang Peng,Zhikai Wang,Ziyi Wang,Jiaru Zhong,Xiaoji Zheng,Jingzhi Huang,Yi Wang,Jiangtao Gong
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注: 20 pages, 3 figures, 8 Tables
Abstract:Zero-shot vision-and-language navigation (VLN) requires an agent to use past observations without navigation-specific training. We study what information an 8B-scale navigation system should retain and how it should be organized for subsequent decisions. Our agent combines instruction progress, hierarchically compressed observation history anchored to poses, and accumulated geometry in complementary task, temporal, and spatial memories. An 8B observer summarizes observations, and an 8B reasoner uses the memories to select waypoint intents or request additional views, a geometric controller executes the selected targets. A semantic information perspective motivates the distinction between retaining evidence, making it accessible to the reasoner, and acquiring further observations. On R2R-CE val-unseen, the complete system achieves the highest among the evaluated Qwen3-based zero-shot baselines. Removing temporal memory reduces SR and sharply increases average path length from 16.9 to 39.1m, removing the spatial-map and planning component reduces SR as well. Real-world experiments on a Unitree Go2 show that our zero-shot method achieves higher success rates than the post-trained baseline in both corridor and office environments, suggesting better transfer to the evaluated real-world settings. As a complementary evaluation, LAW2 uses paired instruction interventions to examine behavioral responses, instruction following ability beyond endpoint success. These results support preserving task progress, observation history, and geometry in structured memory for zero-shot navigation with smaller open models.
[AI-165] LLM -PowerTrace: A Dataset of Time-Resolved GPU Power Traces for AI Compute
链接: https://arxiv.org/abs/2610.38315
作者: Jiansheng Wang,Jaskeerat Singh,Haochen Xu,Xinyi Li,Mingyuan Yan,Shaohui Liu,Yuzhuo Li,Yize Chen,Wenqi Cui
类目: ystems and Control (eess.SY); Artificial Intelligence (cs.AI); Operating Systems (cs.OS)
备注: In Submission; Code and dataset are available at this https URL
Abstract:AI compute is becoming large and fast-changing electrical loads. Its impacts on data center infrastructure and power grids depend not only on total energy consumption, but also on how their power draw moves over sub-second to minute timescales, including peaks, ramps, and synchronized fluctuations across accelerators. Existing energy benchmarks for large language models (LLMs) report aggregate quantities such as joules per token and rarely release the underlying time-series measurements, leaving these power dynamics poorly characterized and difficult to model. We introduce LLM-PowerTrace, an open dataset of time-resolved, per-GPU power traces for LLM fine-tuning and inference, sampled every 100,ms on varying NVIDIA platforms paired with complete workload and system metadata. For fine-tuning, LLM-PowerTrace profiles open model families spanning a wide range of model sizes and fine-tuning methods, together with controlled sweeps over batch size, sequence length, precision, parallelism, and checkpointing. For inference, it covers variations in concurrency, GPU frequency, KV-cache and weight quantization, prompt and output lengths, and request-arrival patterns. LLM-PowerTrace profiles a set of open model families across a wide range of model sizes, and controlled sweeps over batch size, sequence length, precision, parallelism, and checkpointing for fine-tuning, and it sweeps concurrency, GPU frequency, KV cache and weight quantization, prompt and output lengths, and request arrival patterns for inference. Built upon power traces, we define power-grid-facing metrics as well as AI-compute power\energy scaling law for extrapolating beyond the measured settings. Together, LLM-PowerTrace provides a dataset foundation for modeling, forecasting, and managing the fast-changing electrical behavior of AI compute.
[AI-166] CHORD: Cross-Model Hallucination Detection via Relational Graph Discrimination NEURIPS2026
链接: https://arxiv.org/abs/2610.38314
作者: Yongxin Deng(1),Zhen Fang(1),Guansong Pang(2),Sharon Li(3),Ling Chen(1) ((1) University of Technology Sydney, (2) Singapore Management University, (3) University of Wisconsin-Madison)
类目: Artificial Intelligence (cs.AI)
备注: Accepted to NeurIPS 2026 as a Spotlight paper. 64 pages, 14 figures, 7 tables
Abstract:Hallucination detection is critical for deploying large language models (LLMs) in real-world applications. Due to strong empirical performance, internal representation-based methods have emerged as the prevailing direction for detecting hallucinations, yet they remain largely model-specific and often fail to generalize to unseen LLMs. In this paper, we study an important yet underexplored problem, termed cross-model hallucination detection (CMHD), which aims to train hallucination detectors on source LLMs while ensuring performance on unseen target LLMs. The core challenge of CMHD lies in cross-model representation heterogeneity: hidden states from different LLMs exhibit severe semantic inconsistency, which limits the transferability of classical representation-based detectors. Through analysis and empirical validation, we show that relational structures over layers, tokens, and features capture transferable detection signals despite semantic inconsistency. Based on this observation, we propose cross-model hallucination detection via heterogeneity-oriented relational discrimination (CHORD), a relational graph discrimination framework that represents hidden states as joint relational graphs over layers, tokens, and features. It feeds these graphs into relation-aware attention to obtain structure-aware features, and uses meta-learning to shape these features toward transferable detection signals. Extensive experiments show that CHORD outperforms representative detection methods in cross-model generalization. Code is available at this https URL.
[AI-167] Game-Theoretic Foundations for Undiscounted Bisimulation Metrics
链接: https://arxiv.org/abs/2610.38313
作者: Matthew Zurek,Yudong Chen,Josiah P. Hanna
类目: Artificial Intelligence (cs.AI); Computer Science and Game Theory (cs.GT); Machine Learning (cs.LG); Optimization and Control (math.OC)
备注:
Abstract:Bisimulation metrics offer a principled notion of task-relevant state similarity in reinforcement learning, but existing constructions and methods are largely confined to discounted MDPs, where contraction arguments ensure well-defined fixed points and downstream guarantees. We develop a unified foundation for bisimulation metrics in undiscounted settings, that is total- and average-reward problems, by replacing the standard operator-centric view with a game-centric one: we interpret prior metric operators as Bellman-Shapley operators of zero-sum turn-based games and define metrics as the corresponding game value functions, yielding unique definitions even when \gamma=1 . This game-theoretic foundation facilitates transfer of dynamic-programming theory and algorithms. We also use this game-centric perspective to characterize how metric finiteness depends on structural properties of the underlying MDP and motivate certain directed metrics that remain finite in weakly communicating MDPs. We further introduce an abstract framework for modifying the games’ action spaces, allowing general transition coupling sets, that recovers many existing metrics and guides new constructions. Finally, we prove that these metrics bound total-reward and, when the optimal gain is state-independent, optimal relative value differences. We also provide error and suboptimality guarantees for aggregation-based planning in total- and average-reward settings.
[AI-168] Inner-Learning: From Hand-Coded Beliefs to Declarative Knowledge in Trick-Taking Games
链接: https://arxiv.org/abs/2610.38304
作者: Stefan Edelkamp
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:We present a declarative approach to belief-state modeling in trick-taking card games. Starting from a state-of-the-art Skat player, we automatically extract its hand-coded inference rules into a domain-specific language built on the primitive \neg\mathsfholds(P,C) , denoting that player P cannot hold any card in set C . Round-trip code generation verifies that the extracted rules reproduce the original engine’s intended behavior, up to three audited bugs in the original that the declarative engine corrects. We then grow the rule base through symmetric derivation and annotation of an expert strategy document—unifying code, symmetry, and expertise in one representation. We extend the knowledge representation with \emphcardinality constraints: at-most- k and at-least- k bounds on the cards a player holds from a named group, expressed in the same declarative language and enforced at world-generation time. Confidence parameters connect symbolic rules to probabilistic belief updates. Self-play evaluation shows the declarative engine measurably improving winning percentage and score over the original hand-coded engine, with the extracted rules, the human-annotated rules, and the cardinality constraints each contributing to the effect.
[AI-169] Vendster: An Evaluation Arena to Investigate How Forecasting Tool Quality Affects LLM Agent Behaviors in Buyer-Vendor Negotiation
链接: https://arxiv.org/abs/2610.38292
作者: Charin Polpanumas
类目: Computer Science and Game Theory (cs.GT); Artificial Intelligence (cs.AI)
备注:
Abstract:Time-series forecasters are ranked almost entirely by accuracy metrics, but a forecast’s decision value, its effect on the downstream decision, need not track accuracy and is rarely measured. This problem compounds as large language model (LLM) agents increasingly consume forecasters as tools, mediating them through stochastic, opaque policies. We present Vendster, an evaluation arena measuring how a forecaster’s bias and prediction-interval width change the outcome of a buyer–vendor discount negotiation between two LLM agents. In the reference instance Vendster-M5, sampling base demand patterns from the M5 dataset, we evaluate two frontier-closed LLM families (Claude Opus 4.6 and GPT-5.6 Sol) and two open ones (DeepSeek-V3.2 and Qwen3-235B-A22B). Model-generated forecasts raise sales target attainment by up to +17 percentage points (p0.001), yet standard forecast evaluation metrics such as relWQL and MASE do not predict negotiation outcomes at a statistically significant level. To isolate the causal structure, we substitute in synthetic forecasts with controlled bias and interval width. Under-prediction bias significantly improves outcomes for agents that react more to the forecasts—dose-response slopes -0.387 and -0.416 (p0.001). Interval width has no statistically significant effect at zero bias. These findings indicate that the direction of forecast bias and agent forecast-reactivity are more strongly associated with downstream outcomes than forecast accuracy. Vendster is fully customizable and released under CC BY-NC 4.0 at this https URL.
[AI-170] he Role of Quantization in Loss Spikes
链接: https://arxiv.org/abs/2610.38277
作者: Zaiqian Chen,Aditya Somasundaram,Sophie Li,Sean McLeish,John Kirchenbauer,Tom Goldstein,Micah Goldblum
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Loss spikes in language model pretraining can waste substantial compute and derail otherwise healthy training runs. A primary driver of these instabilities is the growth and divergence of output logits in the final language modeling (LM) head. In this work, we study a particularly costly form of this instability associated with output logit growth: delayed, terminal divergences that emerge only after prolonged training, which we call time-bomb spikes. We find that these instabilities are preceded by rapidly growing quantization error in the LM head, which eventually culminates in a loss spike. Crucially, reducing numerical precision accelerates time-bomb spikes, while increasing precision improves stability. We then propose a simple yet effective mitigation that replaces AdamW on the LM head with SGD with momentum, suppressing logit growth and improving stability over long training horizons. Finally, we find that existing mitigation techniques from the literature (along with our own) all reduce quantization error, suggesting a shared numerical mechanism behind their effectiveness. Together, our results identify finite precision error as a hidden driver of time-bomb spikes and show that suppressing this error is key to stable long-horizon pretraining.
[AI-171] Verification-Guided Retrosynthesis: Reaction Feedback and Evidence Timing
链接: https://arxiv.org/abs/2610.38245
作者: Hongxuan Shi,Hankz Hankui Zhuo
类目: Artificial Intelligence (cs.AI)
备注: 12 pages, including appendices
Abstract:Retrosynthesis planning seeks synthesis routes from purchasable starting materials to a target molecule. Search over predicted reactions can connect a target to available materials, yet an infeasible reaction can invalidate the resulting route. Evaluating route quality only after search has finished cannot correct errors made during search. We introduce Verification-Guided Reflow (VGR), incorporating reaction verification as a decision action in AND/OR search. Verification results are represented as binary verdicts; queries are scheduled from the search structure, and acquired verdicts exclude reactions or adjust search priorities. We define verified solvability as the fraction of targets for which a complete route accepted stepwise by the verifier is found. On Retro190, feeding verification results into search increases verified solvability from 51.1% to 62.6% relative to route evaluation without feedback. Further controlled experiments show that earlier verification and immediate prioritization of accepted reactions each improve route discovery, yet combining them performs worse than either change alone. We therefore propose early pruning and delayed guidance: rejected reactions are pruned immediately, while accepted reactions gain search priority when processed as part of a complete candidate route. In a separate comparison with the verification scheduling rule fixed, this strategy increases verified solvability from 61.1% with immediate guidance to 71.1%, using fewer proposal calls and unique verifier queries.
[AI-172] Precedent or Intent? What Coding Agents Follow When Evolving Existing Code
链接: https://arxiv.org/abs/2610.38244
作者: Haichao Zhu,Geyang Xu,Zhaorui Yang,Jiyuan Wang,Qian Zhang
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注: 16 pages, 7 figures, 6 tables
Abstract:Existing code reveals past implementation decisions but rarely explains the reasoning behind them. When coding agents implement new features or modify existing functionality, past decisions may conflict with new task instructions. This raises a fundamental question for agentic software evolution: when multiple implementations satisfy the same functional requirements, how do coding agents decide whether to follow existing code or new task instructions? We investigate this question through bidirectional, controlled experiments in purpose-built repositories, followed by validation in real-world open-source repositories. Our approach isolates three competing sources of guidance: (1) agents’ default preferences, (2) repository precedents, and (3) explicit task instructions. We conduct 5,625 trials across four coding agents and six implementation pairs, with validation in four open-source repositories spanning three programming languages. We find that agents share the same default implementation choices, yet overwhelmingly follow repository precedent when it conflicts with those defaults (359/360 trials). Crucially, not all precedents carry equal influence: three supporting local files yield 96% adoption, compared with only 4% for sibling-resource files, and changing file placement alone can shift adoption from 0% to 100% at a fixed example count. When new task intent conflicts with existing code, written priority rules achieve substantially higher success than code examples alone (59/60 versus 38/60 joint successes for two agents on two implementation pairs). The effectiveness of explicit rules also extends to open-source repositories, where a harness-loaded rule raises joint success from 0/120 to 117/120. These findings motivate design-decision traceability that preserves not only implementation choices but also the rationale and priorities that should guide future changes.
[AI-173] he Indian Financial Automated Advice Bias (IFAAB) Dataset : Auditing Socioeconomic and Demographic Disparities in LLM -Generated Financial Advice in India
链接: https://arxiv.org/abs/2610.38239
作者: Kriti Mahajan
类目: Computers and Society (cs.CY); Artificial Intelligence (cs.AI)
备注:
Abstract:Large language models are increasingly used for financial guidance, yet little is known about whether their advice changes across social identities in Global South contexts. We introduce the Indian Financial Automated Advice Bias (IFAAB) benchmark, a counterfactual audit dataset of 31,734 GPT-4o-generated financial-advice responses for synthetic Indian personas varying by gender, caste, education, employment, residence, age, and district. Holding the underlying request fixed, we evaluate disparities in LLM-implied wealth, income, savings, projected gains, textual similarity, advice topics, and qualitative bias judgments. In the gender-only audit, male-labelled profiles receive 60.1% higher initial total resources and 46.8% higher post-advice total resources than comparable female-labelled profiles; in the caste-augmented audit, the post-advice total-resource gap remains 49.4%. Adding district information reduces the aggregate post-advice total-resource gender gap to 7.9%, revealing strong geographic conditioning. Standalone caste comparisons also show systematic stratification: relative to General-category profiles, OBC, SC, and ST profiles receive significantly lower post-advice total resources, respectively. Textual audits show low semantic similarity between paired gendered responses, with an average TF–IDF cosine similarity score of 0.44. Qualitative analysis shows that women are more often steered toward cautious saving, community support, and basic financial literacy, while men receive more growth-oriented advice involving investment, entrepreneurship, and asset accumulation. IFAAB provides a reusable benchmark for evaluating fairness, suitability, and consumer-protection risks in LLM-mediated financial advice, and highlights the need for locally grounded, intersectional evaluation of AI systems in high-stakes economic domains.
[AI-174] FACTS-MAS: A Multi-Agent Architecture for Adaptive Forecasting from Heterogeneous Evidence
链接: https://arxiv.org/abs/2610.38225
作者: Madumita Karthikeyan,Umang Rajnikant Bid,Rong Pan
类目: Artificial Intelligence (cs.AI); Applications (stat.AP)
备注: 29
Abstract:Real-world forecasting often requires integrating heterogeneous evidence whose predictive value varies across horizons, operating regimes, and domains. We introduce FACTS-MAS, a domain-agnostic multi-agent forecasting framework that decomposes evidence into specialized channels for temporal dynamics, exogenous context, seasonality, structural information, and discrete events. Each agent generates an independently testable forecast through a common interface, while leakage-safe historical validation measures its predictive skill relative to a naive reference. These skill estimates are used to weight and combine agent forecasts, so that uninformative evidence is either removed or downweighted rather than forced to participate. A calibration procedure selects domain-specific constant means based on historical validation folds. We evaluate FACTS-MAS on weekly for-sale housing inventory from Zillow and weekly unemployment insurance claims from FRED, for 15 U.S. metropolitan areas at 4-, 8-, and 13-week forecasting horizons. Our method outperforms both a naive last-value forecast method and the TimesFM foundation model. However, a factor-adjusted exponential smoothing model performs as well as the fused forecast. The evidence therefore supports FACTS-MAS as a systematic protocol for testing, calibrating, and recomposing evidence channels before assembling them into the strongest single model for a specific application domain. Code, data, and results are available at this https URL.
[AI-175] CuraEval: Evaluating Demonstration Curation on Mixed-Skill Robot Data
链接: https://arxiv.org/abs/2610.38221
作者: Jiuyi Xu,Meida Chen,Yang Ye,Yangming Shi
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注: Accepted by the CoRL 2026 Workshop “Oops, I Erred”. 10 pages, 2 figures, 2 tables
Abstract:Robot imitation learning uses demonstrations from operators of varying skill, but selecting training data remains challenging. Existing methods score demonstrations using smoothness, diversity, or estimated influence on policy performance, but differences in selection and evaluation protocols make it difficult to determine whether they outperform simple baselines. We introduce \textbfCuraEval, a framework combining mathematical analysis of demonstration scores with controlled comparisons against fixed size-matched random subsets and full-data training on mixed-skill robomimic data using Diffusion Policy and BC-RNN. We show that the evaluated PSD score depends only on episode length and squared position magnitudes, and produces nearly the same ranking as episode length on \textitsquare ( \rho=0.999 ). With Diffusion Policy, selecting the shortest 50% of demonstrations improves \textitsquare success from 70.7% under random selection to 76.5%. Compared with random selection, our adapted geometric-entropy LOO selector reduces success at 25% retention from 94.7% to 74.0% on \textitcan and achieves 41.7% on \textitsquare, below the mean success rates of all three independently sampled random subsets (59.8–62.5%). On the same 3 \textitsquare Diffusion Policy training seeds, length-based selection at 75% retention improves success over full-data training from 76.7% to 86.7% under the original protocol, but lowers it from 80.5% to 79.5% after checkpoint reporting, rollout count, and evaluation seed change jointly. No evaluated curated condition significantly outperforms full-data training after Holm correction, although confidence intervals still permit useful gains. Code is available at this https URL.
[AI-176] Rubricon: Automated Rubric Generation and Refinement for LLM Evaluation and Training
链接: https://arxiv.org/abs/2610.38220
作者: Ulyana Tkachenko,Hui Wen Goh,Guram Gogia,Jonas Mueller
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Rubrics enable reward assignment on tasks without verifiable answers, but are expensive to manually produce from scratch. We present Rubricon, a pipeline that generates discriminative rubrics without requiring human preference labels, rubric annotation, or generator-model training. Using any LLM, Rubricon samples multiple rubrics, selects the most discriminative candidate, adds criteria to capture overlooked requirements or response flaws, and prunes criteria that are misaligned or shallow. Alongside LLM self-critique and test-time scaling, Rubricon uses observed criterion pass rates on example model responses to guide refinement. Across five professional, consumer, clinical, and general-use LLM benchmarks, Rubricon produces rubrics whose scores of frontier LLM responses exhibit 9% greater alignment with labor-intensive expert-authored rubrics, compared to the best alternative synthetic rubric generation method among RRD, QSRG, OpenRubrics, and agentic generation via harnesses like Codex. On HealthBench, post-training a Qwen3.5-9B model with Rubricon rubrics yields 4.5% better performance than training with rubrics from the strongest alternative synthetic rubric generator, RRD.
[AI-177] Skeleton Coding: Program Structure as a Specification for Agents
链接: https://arxiv.org/abs/2610.38214
作者: Aditya Golatkar,Alessandro Achille,Renos Zabounidis,Vlad Sobal,Yangruibo Ding,Baishakhi Ray,Jonathan Buck,Guy Van den Broeck,Maryam Fazel,Pietro Perona,Wei Xia,Stefano Soatto
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Coding agents are typically provided end-to-end behavioral specifications that the code they produce should satisfy, such as tests and input–output examples. How many such specifications must a developer provide, and which ones, for the agent to produce the developer’s intended program? We argue that this cannot be determined a priori, for it depends on the structure of the implementation, which is left to the agent to decide. An agent first commits to a structure—a decomposition into data types, modules, and functions—which fixes a class of candidate programs; behavioral specifications then select a program within that class. A set of tests that suffices for one structure may leave another badly underdetermined, requiring different edge cases or exponentially more examples. Since the structure is chosen by the agent, developers cannot ensure their specification is complete without reverse-engineering it from thousands of lines of generated code. To resolve this, we introduce Skeleton Coding, in which developers jointly specify the structure—in high-level, code-like format—alongside behavioral specifications. Theoretically, we model AI-assisted coding as a control problem, and show that skeletons can reduce exponentially the number of tests developers must supply to steer an agent toward an acceptable program. Empirically, across four models and three benchmarks, the best-performing skeleton specifications reduce mean test error by 48.4% relative to the best end-to-end behavioral specifications. Even when the agent can query a complete reference implementation, adding a skeleton reduces mean test error relatively by up to 27.0%. Skeletons also narrow the average gap between small and large models by about 52.4% while lowering implementation cost by 15–26x, and reduce code erosion by 24.7% on average.
[AI-178] GAAC: Lossless KV Compression for Bandwidth-Efficient LLM Inference on Personal Devices
链接: https://arxiv.org/abs/2610.38210
作者: Yipin Guo,Arun M George,Siddharth Joshi
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Weight-only, lossy quantization substantially reduces the storage and memory requirements for serving large language models (LLMs), but the growing key-value (KV) cache remains a bottleneck for long-context inference on personal devices. We present Group-Adaptive Additive Codec (GAAC), a lossless KV compression and serving system that exploits exponent redundancy to reduce storage and accelerate attention without introducing KV approximation. GAAC encodes BF16 exponents using calibrated per-channel base-plus-offset codes over contiguous ranges. GAAC adapts code widths across small groups while preserving KV values bit-for-bit. A fused attention kernel reconstructs KV tiles on the fly, avoiding full-cache decompression. Block-wise compression and fixed-capacity slots support incremental cache updates and paged serving. Across five tested models, GAAC achieves 1.405-1.407 \times lossless BF16 KV compression. We also study compression performance on low-precision tensors FP8 E4M3 caches ( 1.09-1.14 \times ) and E5M2 caches (1.25-1.30 \times ). On NVIDIA DGX Spark, GAAC accelerates decode attention at a 64K-token context by 1.19-1.27 \times and end-to-end W4A16 decode steps by 1.04-1.15 \times . In serving, it reduces resident KV memory by approximately 26% and increases aggregate throughput by 13-29% under comparable KV-memory budgets. On Apple M4, GAAC achieves 1.02-1.25 \times attention speedups, demonstrating portability beyond NVIDIA GPUs.
[AI-179] From Natural Language to Real Schematics: An Automated Workflow for Power Electronics Schematic Synthesis
链接: https://arxiv.org/abs/2610.38201
作者: Bowen Su,Huan Chen,Jiahua Ying,Ran Ou,Weilong Chen,Li Xue,Kai Sun
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Designing power electronics converters traditionally requires intensive human expertise across topology sizing, commercial component selection, circuit simulation verification, and schematic drafting. While Large Language Models (LLMs) show promise in electronic design automation (EDA), applying standalone LLMs directly to board-level power electronics schematic synthesis suffers from severe engineering hallucinations, including inaccurate analytical sizing, fabricated or mismatched component part numbers, violated voltage derating margins, and broken topological connectivity. In this paper, we propose a Bounded Dual-Loop Automated Schematic Synthesis Framework that translates natural language specifications into verified, editable KiCad schematics and manufacturing deliverables. Our workflow integrates six stages across three functional columns: (1) requirement parsing and knowledge-driven power stage synthesis covering non-isolated (Buck, Boost) and isolated/resonant (Flyback, DAB, half-bridge LLC) topologies; (2) contract- and evidence-coordinated component selection grounded in distributor APIs and OEM datasheets, coupled with closed-loop ngspice transient verification and stress feedback; and (3) modular control hardware integration and multi-format deliverable generation featuring native KiCad netlist round-trip verification. End-to-end case studies on a 24-36V to 12V,2A Buck converter (29 positions, 83 verified terminals) and an isolated 48V to 24V, 0.5A, Dual-Active Bridge (DAB) converter (87 positions, 341 verified terminals), alongside comparative evaluations against a Vanilla LLM baseline, demonstrate that our framework improves component verification coverage from 22.2 to 98.6, nominal value consistency from 0 to 97.7, and SPICE simulation pass rates from 0 to 100.
[AI-180] Convex-SKM-Net: Low-Latency Block Halfspace Corrections for Neural Convex Constraint Satisfaction
链接: https://arxiv.org/abs/2610.38197
作者: Haoyu Zhu,Xiaochen Zhang,Qingchun Hou
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Optimization and Control (math.OC)
备注:
Abstract:Fast inference and strict constraint satisfaction are essential for learning-based decision-making. Existing methods often rely on constraint-specific projections or feasible-set parameterizations, while iterative repair can be costly. We introduce Convex-SKM-Net, a convex constraint-satisfaction layer that forms linearization halfspaces from residual and (sub)gradient oracles without constraint-specific inequality projectors. It applies fixed-depth dual projected-gradient descent (dual PGD) to minimum-displacement block problems using GPU-friendly matrix operations. We jointly learn the predictor and a positive-definite mobility, adapting the correction geometry to improve objective quality. We prove convergence of the outer iterates at fixed dual depth and linear convergence in distance under Slater’s condition and bounded subgradients. All our test outputs on three second-order-cone model-predictive control (SOC-MPC) horizons and six power-system convex relaxations satisfy a 10^-3 feasibility tolerance. Comparisons with the latest constraint-satisfaction methods show inference speedups of up to 144.3\times on SOC-MPC and 7.7\times on a convex quadratically constrained quadratic programming (QCQP) benchmark with joint training. On QCQP, the mean objective gap is 0.012% , with full test feasibility at 10^-4 . Learning mobility reduces the five-seed affine-task mean gap by 22.4% and improves gaps in matched power-system comparisons.
[AI-181] Prototype-Guided Dynamic Domain Modulation and a Large-Scale Multi-SKU Benchmark for Shampoo Bottle Surface Defect Detection
链接: https://arxiv.org/abs/2610.38194
作者: Yuehang Li,Xiao Wang,Qiang Chen,Yangyi Fan,Bo Jiang,Yaowei Wang,Jin Tang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Industrial shampoo-bottle surface defect detection must operate across many stock keeping units (SKUs), where differences in bottle geometry, material, label layout, reflectance, and imaging conditions can substantially alter the appearance of the same defect. Existing industrial defect datasets rarely provide both instance-level defect annotations and product-variant identities, making it difficult to study SKU-aware detection under realistic manufacturing conditions. We introduce Bottle-SKU, a large-scale multi-SKU benchmark for shampoo bottle surface defect detection. Bottle-SKU contains 13,119 images, 20,419 instance-level bounding boxes, seven defect categories, and 48 bottle SKUs, enabling evaluation of both shared defect recognition and SKU-specific appearance variation. To address the challenge of SKU-specific appearance variation, we propose Prototype-Guided Dynamic Domain Modulation (ProDMM). ProDMM constructs a visual prototype for each SKU from training images and predicts a soft SKU distribution for each input. The resulting prototype-conditioned representation dynamically generates low-rank Query/Value modulation for the visual backbone and residual conditioning for defect-text-to-visual attention, while preserving the pretrained Grounding DINO detection pathway. The model is trained jointly with detection, SKU-domain classification, and modulation-regularization objectives. Experiments on Bottle-SKU and the public GC10-DET-SKU metallic surface defect dataset demonstrate the effectiveness of prototype-guided domain conditioning for industrial surface defect detection.
[AI-182] DecompGym: Benchmarking LLM Agents on Designing Decomposition Algorithms
链接: https://arxiv.org/abs/2610.38175
作者: Sikai Cheng,Tinghan Ye,Arnaud Deza,Xiaonan Qi,Miao Li,Litong Liu,Changkun Guan
类目: Artificial Intelligence (cs.AI); Optimization and Control (math.OC)
备注:
Abstract:Benders decomposition and column generation solve large optimization problems by passing cuts, columns, and dual values between a master problem and subproblems, and are hard to implement correctly. Large language models now automate parts of algorithm design, which points toward agents that build decomposition solvers for new problems. As a first step toward such agents, we study how well current coding agents design and implement these two algorithms, and how they fail. We introduce DecompGym, 81 instances from nine problem families (five Benders, four column generation), in which an agent writes a solver for a prescribed method. Each solver must meet three criteria: execution within a 60-second time limit; numerical feasibility together with a method audit, in which an LLM judge reads the solver’s code; and an objective within 1% of the stored optimal objective. We evaluate 12 GPT, Claude, and Gemini models running in coding-agent tools. Over up to ten attempts, we compare independent regeneration with reflection, which first diagnoses the failure. On the first attempt, two models solve every instance and seven solve more than 90%. Most Benders failures (48 of 62) are missing or suboptimal solutions. The most common column-generation failure is a method-audit rejection. Output checks alone would accept 93 of the 108 attempts that fail the method audit. The most common rejection, an optimalif, occurs 96 times, 92 of them in one this http URL turns 66% of retries into successes, compared with 39% for regeneration, and it repairs method-audit failures, for which the judge names the defecective gaps. Agents for decomposition therefore need checks on how a solver reaches its answer and feedback that locates the error. The benchmark is available at this https URL.
[AI-183] BridgeRouter: Gain-Aware Instance Routing over Cross-Modal Bridge Portfolios
链接: https://arxiv.org/abs/2610.38152
作者: Ruoxi Piao,Alicia Lu,Yunzhen Zhang,Mustafa Mısır,Hasan Onur Keleş
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Lightweight representation bridges make pretrained, task-adapted EEG foundation models executable on other biosignal modalities without retraining their backbones. However, selecting a single bridge pathway globally assigns every instance to the same transfer route and ignores complementary errors across routes. We therefore formulate multi-bridge transfer as an instance-level algorithm-selection problem and introduce BridgeRouter, a gain-aware router over a portfolio of frozen cross-modal pathways. A validation-selected Single Best Solver (SBS) serves as a protected default route. During training, task labels are used to compute the reduction in negative log-likelihood provided by each alternative route relative to the SBS, and a lightweight MLP predicts these route-specific gains from a target-modality representation extracted by a frozen encoder. At inference, BridgeRouter observes only the target modality, requires neither EEG, task labels, nor predictions from all candidate routes, and executes exactly one bridge pathway: the route with the largest predicted gain when that gain exceeds a validation-calibrated threshold, and the SBS otherwise. We evaluate seven EEG foundation-model bridge pathways for EMG-based freezing-of-gait detection on FOG and ECG-based sleep classification on ISRUC. Across five seeds, BridgeRouter achieves 81.38\pm1.11% balanced accuracy on FOG, improving over the BIOT SBS by 11.49 percentage points, and 62.92\pm2.17% on ISRUC, improving over the CBraMod SBS by 3.83 points. Per-instance bridge oracles reach 99.96% and 78.61% , respectively, revealing substantial remaining portfolio headroom. These results show that anchor-relative gain prediction can convert complementary cross-modal bridge behavior into deployable instance-level routing improvements while retaining single-route inference.
[AI-184] Estimating Implicit Expected Log Density Ratios
链接: https://arxiv.org/abs/2610.38151
作者: Avi Amalanshu,Chandler Squires,Yizhou Lu,Pradeep Ravikumar
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Machine Learning (stat.ML)
备注:
Abstract:When exact probabilities cannot be computed, \emphdensity ratio estimation (DRE) is a key subroutine in the wide range of applications where density ratios appear, from experimental design to generative modelling and reinforcement learning. In this paper, we introduce the concept of an \emphexpected log density ratio (ELDR): log density ratios that are integrated under some third, possibly distinct measure. We show many problems with implicit densities where DRE is naïvely used are actually ELDR estimation tasks, and prove fundamental limitations on finite-sample ELDR estimation. Based on these results, we argue that the typical DRE-based approaches are unsuitable for ELDR estimation, and provide some foundational motivation towards bespoke ELDR estimators. To demonstrate, we propose a simple modification to some DRE methods that allows samples convergent to the integral distribution to appear in training data. We show through benchmarks that this can improve ELDR estimation error.
[AI-185] SkillCT: A Multidimensional Benchmark and Execution-Free Baseline for Agent Skill Assessment
链接: https://arxiv.org/abs/2610.38147
作者: Mengyu Sun,Zunlong Zhou,Ziyuan Yang,Yi Zhang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Agent skills package instructions, scripts, and resources for specialized tasks and are increasingly distributed through public Web repositories. However, existing skill benchmarks primarily focus on downstream task execution, insufficiently assessing specification quality, behavioral robustness, semantic alignment, and responsibility boundaries. Their reliance on task-specific tools, environments, and repeated executions limits scalability and pre-deployment assessment. To address these challenges, we introduce SkillCT, a benchmark that broadens skill assessment from downstream utility to a multidimensional characterization of skill quality. SkillCT evaluates skills across nine complementary dimensions spanning three aspects: Static Assessment, which captures intrinsic properties of the skill specification; Dynamic Assessment, which measures whether the induced behavior remains stable across backbone models and request variations; and Human-Assisted Assessment, which examines behavior-chain consistency and task granularity. These dimensions provide a broader view of skill quality than execution success. Applying SkillCT to a public skill corpus establishes a reference benchmark for analyzing quality weaknesses and relationships across dimensions. Building on this benchmark, we propose an execution-free quality evaluation baseline that predicts the multidimensional quality profiles of unseen skills by retrieving and leveraging similar reference skills, without executing the skills under evaluation. This reduces token consumption and evaluation time, while preserving consistency with the full assessment. Furthermore, SkillCT-guided optimization improves downstream task success and execution efficiency, demonstrating its value for skill refinement. These findings highlight the effectiveness of SkillCT in supporting scalable skill assessment, optimization, and repository maintenance.
[AI-186] Answer-Only Disclosure: Private-Context LLM Assistance over Web APIs with a Leakage Ledger
链接: https://arxiv.org/abs/2610.38126
作者: Liang Yuan,Xiaohu Du,Ruikun Luo,Jinze Wang,Mashud Rana,Kaibin Wang,Jiangwen Zhou,Bo Li,Xiaoyu Xia
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注: 13 pages
Abstract:Web assistants increasingly answer questions about a user’s private documents by pairing a small on-device model with a cloud LLM behind a Web API. In collaborative decoding, the document stays on the client, and the cloud receives only a public view that the user approves, such as the question alone, but the two models exchange messages while the answer is generated. We show that the message transcript can disclose private content that neither the public view nor the answer contains. In a planted-secret audit, speculative decoding with private drafts did not include the secret code in its answers, yet sent the code to the cloud in rejected drafts. We present ADLER, a decoding protocol for streaming LLM APIs in which every message to the cloud depends only on the public view and the answer the user sees, regardless of how the client combines the two models or how the cloud behaves. A client-side leakage ledger then bounds how much the answer itself reveals relative to a public reference process built from the cloud’s replies. In the audit, no ADLER configuration exposed the secret beyond the answer. On GSM8K math problems with numerical values masked in the cloud’s input, ADLER with an unlimited leakage budget raises the exact-match accuracy of a 0.5B client from 0.41 to 0.83 and of a 1.5B client from 0.67 to 0.85 under lenient answer parsing, close to the 0.92 achieved by uploading the full problem. It also exceeds a zero-leakage program baseline (0.69) by 0.14 and 0.16, respectively. In this baseline, the cloud writes a program that the client runs offline using the private numbers. The ledger records the disclosure charge associated with this gain; budgets of 16 nats or less leave accuracy near the cloud-only baseline on GSM8K.
[AI-187] Post-Training Large Language Models with LLM ON
链接: https://arxiv.org/abs/2610.38120
作者: Basel Shbita,Michael Hind,Chad DeLuca,Nathan Fulton,Shubhi Asthana,Farhan Ahmed,Anisa Halimi,Yada Zhu,Mauro Martino,David Cox,Dan Gutfreund
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Large Language Models (LLMs) treat all context tokens uniformly, lacking explicit separation between executable instructions and passive data. This leads to prompt injection vulnerabilities, ambiguous instruction binding, and unreliable execution control. We present a post-training methodology using LLMON (LLM Object Notation), a framework that enables separating instructions from data and requires explicit identifier-based execution. We construct a large-scale structured training corpus of 4.6B tokens spanning general, RAG (Retrieval-Augmented Generation), and tool-calling data, augmented with distractor-infused variants in which competing instructions and irrelevant data segments are present in context but not referenced. Post-training two Granite-4.1 base models (3B and 8B) through supervised fine-tuning followed by reinforcement learning, we show that LLMON-structured supervision reduces attack success rates on indirect prompt-injection benchmarks from 53-100% to 0.02-5.6% while preserving general capabilities. We further introduce attention-based interpretability metrics that diagnose failure modes at individual transformer layers for structured inputs, revealing how our approach reshapes internal attention patterns. Our results demonstrate that explicit execution semantics can be taught through standard post-training without architectural modification.
[AI-188] oo Hard Too Easy Just Right: Performance-Driven Challenge Adaptation in Virtual Reality Firefighter Training
链接: https://arxiv.org/abs/2610.38117
作者: Xuan Wang,Jianan Jiang,Rongqian Chen,Tian Lan,Bin Li
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Effective firefighter training requires practice under dynamic and stressful emergency conditions, yet real-world live-fire exercises are costly, risky, and difficult to scale. Virtual Reality (VR) offers a safe and controllable alternative, but many existing VR firefighter training systems rely on fixed scenarios or provide limited difficulty adaptation, making it difficult to maintain an appropriate level of challenge for skill development as trainees’ performance changes over time. To address this limitation, we present ProACT, a performance-driven adaptive training system that regulates scenario difficulty through closed-loop interactions between trainee performance and simulated fire dynamics. Specifically, ProACT combines a global module that continuously adjusts scene-level fire spread dynamics in real time based on trainee performance with a local module that periodically updates object-level flammability across the environment. Our evaluation shows that ProACT improves both objective training performance and subjective user experience compared with existing adaptive and non-adaptive baselines. We further discuss how adaptive immersive environments can extend beyond firefighting to broader training domains.
[AI-189] From the Gettier Machine to Justified True Output: A Three-Step Account of Rational Trust in Generative AI
链接: https://arxiv.org/abs/2610.38112
作者: Tevfik Uyar
类目: Computers and Society (cs.CY); Artificial Intelligence (cs.AI)
备注: 16 pages, 1 table
Abstract:As generative artificial intelligence (AI) settles into everyday practices of inquiry, an old epistemological question returns in a new guise: when does a proposition a user takes from an AI output become something the user knows, and when is it rational to trust it as such? This paper argues that the characteristic failures of AI-mediated inquiry are precisely those that the justified-true-belief (JTB) tradition and its post-Gettier anti-luck refinements were built to diagnose, and it advances three claims. First, AI outputs inherit neither the presumptive warrant of testimony nor the reliabilist warrant of deterministic tools: unaudited trust in a calculator rests on a type-level reliability record, and for general-purpose generative models there is no type-stable process to which such a record could attach – a practical form of the generality problem. Second, use that stops at obtaining an output turns generative AI into a Gettier machine: the user’s fluency-based seeming of justification is disconnected from the truth of what is believed, and, as a clock running ten times too fast illustrates, sheer speed of production multiplies coincidental hits, inflating the appearance of reliability without improving the standing of any single hit. Third, the remedy is a three-step structure of use – acquisition, understanding, independent verification – whose first two steps answer to the classical belief and justification conditions and whose third answers to the post-Gettier sensitivity condition. The result is a stakes-indexed criterion of rational trust, justified true output (JTO), with an everyday test: “Had this been false, would I have noticed?” The criterion is defended against objections from over-demandingness, the extended mind, model improvement, and redundancy.
[AI-190] AI-Driven Digital Twin Framework for Warehouse Logistics Automation Optimization ICRA2026
链接: https://arxiv.org/abs/2610.38101
作者: Md Mahinur Alam,Kanita Jerin Tanha,Md Tanjin Alam,Shadman Sakib,Min Gyu Kim
类目: Artificial Intelligence (cs.AI)
备注: ICRA 2026 Workshop Advances and Challenges in AI-Driven Automation and Robotic System Integration with Digital Twins
Abstract:In recent years, warehouse logistics optimization has increasingly focused on intelligent, data-driven systems to enhance operational efficiency. Conventional static product placement strategies are being progressively replaced by dynamic slotting approaches that can adapt to real-time variations in warehouse conditions. In this work, we present a novel framework that employs a reinforcement learning agent integrated within a FlexSim-based Digital Twin environment for real-time product placement optimization. The proposed agent continuously learns from streaming operational data, enabling it to reduce picking time while improving overall system throughput. The simulation environment models a four-zone warehouse layout with forklift-based material handling and leverages the Asset Administration Shell framework to ensure seamless interaction between the learning agent, physical asset, and the digital twin. Experimental evaluation demonstrates substantial performance gains, with the proposed approach achieving a 29.66% reduction in average picking time, a 35.42% increase in throughput, and an 18.54% decrease in travel distance.
[AI-191] Evaluating Causal Reasoning in LLM s: A Case Study in the Challenges of Procedural Generation
链接: https://arxiv.org/abs/2610.38100
作者: Justin Clarke,Isha Karn,Patrick Murray,KJ Hill,David Jensen
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Large language models (LLMs) appear to represent and reason about a large range of phenomena. In particular, recent studies have examined whether LLMs can accurately reason about causation. One of the most commonly used benchmarks for evaluating formal causal reasoning in LLMs is CLadder, a framework for procedural generation of benchmark data sets that was released in 2023. Experiments with CLadder appear to indicate that LLMs can perform formal causal reasoning in a large percentage of generated examples. In this paper, we examine what can be reliably inferred when LLMs perform well on test sets produced by CLadder. We identify several examples of superficial and highly predictive patterns that the CLadder generator embeds in data sets it generates. We find that even simple models such as random forests perform well on CLadder-generated data sets after training on a sample of generated data. We also find that the performance of many commercial LLMs on CLadder test sets is extremely brittle and that removing large sections of the text that are critical for formal causal reasoning has almost no effect on performance. Given these results, the most likely hypothesis for the strong performance of state-of-the-art LLMs on CLadder is that they are exploiting superficial patterns learned from exposure to training data from the generator. Any inference that LLMs are exhibiting more complex causal reasoning must rule out this hypothesis.
[AI-192] Jailbreak Connectivity: Towards Diverse Transferable and Universal Multimodal Large Language Model Jailbreak
链接: https://arxiv.org/abs/2610.38091
作者: Xu Zhang,Ziqing Hu,Shuo Han,Ren Wang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR)
备注:
Abstract:Multimodal large language models (MLLMs) are vulnerable to visual jailbreak attacks that bypass safety safeguards. Existing methods typically optimize isolated adversarial examples, limiting attack diversity, transferability, and generalization. We propose Jailbreak Connectivity (JC), a framework based on the observation that effective adversarial inputs can be connected through low-loss regions in image space. JC optimizes continuous adversarial paths: path sampling yields diverse jailbreak images, lightweight surrogate classifiers guide cross-model transfer, and a distribution-level objective enables universal attacks across harmful queries. On MM-SafetyBench, JC achieves average attack success rates of 84.01%, 85.77%, and 97.95% on Qwen3-VL-8B-Instruct, InternVL3.5-8B, and Gemma-3-12B-IT, respectively, exceeding the strongest baselines by up to 68.36 percentage points. In black-box evaluations, JC improves attack success rates over FigStep by 24.69 and 27.34 percentage points on GPT-5.6 and Gemini 3.1 Pro, respectively. These findings reveal structured vulnerabilities in MLLM safety alignment and inform more comprehensive security evaluation.
[AI-193] What Can Firms Do in the Face of Model Distillation?
链接: https://arxiv.org/abs/2610.38088
作者: Diptangshu Sen,Jibang Wu,Juba Ziani
类目: Computer Science and Game Theory (cs.GT); Artificial Intelligence (cs.AI)
备注: 32 pages
Abstract:Model distillation has become a major concern in the rapidly growing generative AI discourse. In its simplest form, distillation is the phenomenon where smaller firms can use data distilled from transcripts of conversations with powerful proprietary closed-weight models to train their own models, thus achieving reasonably high model quality but at much lower costs. Besides the legitimate safety and security concerns associated with distillation, another major issue is how it can also distort marketplace outcomes. This motivates us to initiate an exploration of the problem from an economic theory lens. We consider a setting with two players: an incumbent and a distilling competitor and characterize how the incumbent’s model characteristics drive the incentives around how much the competitor should distill. We also consider the impact of distillation on key market outcomes like user-side surplus and the incumbent’s utility. Our key contribution, however, is to study interventions for preventing or limiting distillation attacks. In this context, we study two settings: i) when it is possible, with a certain probability (which grows in the extent of distillation conducted), to detect an attack on the incumbent’s model; and ii) when a distillation attack cannot be reliably detected. In the first setting, we show that penalty-based regulations can be very effective in completely deterring distillation, although it can have potentially adverse effects on market competition and user surplus. In the latter setting, we show that the incumbent can use self-distillation as a mitigation strategy. Importantly, the effectiveness of self-distillation depends on how cost-efficiently the incumbent can deploy a new self-distilled model on the market. On the positive side, self-distillation can preserve market competition thus protecting user surplus to some extent.
[AI-194] PromptSense: Sentiment-Aware Mental Health Chatbot with Dynamic Prompting
链接: https://arxiv.org/abs/2610.38044
作者: Sugi Lee,Jianlong Zhou,Jing sUN
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Current Large Language Models (LLMs)-based mental health chatbots often lack sufficient empathy, contextual awareness, and safety mechanisms, potentially leading to ineffective or harmful interactions. This paper proposes a novel framework that integrates sentiment and depression analysis with a rule-based filtering mechanism prior to model inference. By dynamically constructing prompts based on emotional and risk-related insights, the framework aims to significantly improve the contextual relevance, empathy, and risk-aware of model-generated responses. We evaluated four model strategies and two prompt engineered variants across qualitative metrics such as empathy, helpfulness, clarity, relevance, safety, as well as toxicity assessments using the Perspective API. Experimental results indicate that the prompt engineering approaches outperforms DSFT and RLAIF by 1.8% and 1.1% respectively.
[AI-195] Measuring Deep Learning Performance - An Empirical Study of Performance Distributions Across Architectures and Tasks
链接: https://arxiv.org/abs/2610.38037
作者: Kevin L Coakley,Odd Erik Gundersen
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Context: Non-determinism in deep learning algorithm design and implementation leads to performance variation, meaning model performance is not a single value, but rather a distribution. These model performance distributions are underexplored despite their impact on robustness. Objective: We investigate the robustness of deep learning performance to sources of non-determinism, specifically focusing on how performance distributions differ across various architectures and tasks. Method: We conducted 186 experiments on image classification and time series forecasting architectures. Each experiment was run 100 times with different random seeds to generate performance distributions, resulting in 18,600 runs. Robustness was quantified using metrics for spread, symmetry, and tail risk. Results: Performance distributions are frequently non-Gaussian, particularly in time series forecasting. Model size does not systematically affect robustness – larger image classification models show fewer outliers but not lower spread, while smaller time series models show lower spread but more extreme underperformers. Training duration does not scale linearly; early stopping effectively balances performance and robustness. Mean performance does not predict robustness - time series forecasting shows moderate correlation while image classification shows none. Time series models produce nearly three times more underperforming outliers than image classification models, indicating substantially higher tail risk. Conclusions: Tail risk poses serious concerns for Trustworthy AI in high-stakes applications. Models performing well on average may exhibit long tails and extreme outliers revealed only through distributional analysis. Mean performance alone should not guide model selection; assessment of spread, symmetry, and tail risk is essential for reliable model assessment where consistent performance is critical.
[AI-196] METATRON: A Promptable Whole-Body Execution Layer for Humanoids
链接: https://arxiv.org/abs/2610.38027
作者: Chen Tessler,Yifeng Jiang,Alec Otto,Yi Shi,Umar Iqbal,Gal Chechik,Xue Bin Peng
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注: Project page: this https URL
Abstract:Humanoid controllers often require task-specific training despite sharing the need for balance and whole-body coordination. We present Metatron, a promptable whole-body execution layer combining a generative base policy with a programmable skill layer. The base policy receives partial goals, collectively called a prompt, that can combine timed body-part targets, velocity commands, and style cues. It produces diverse whole-body motions consistent with these goals. Training combines motion capture, synthetic recovery trajectories, and retargeted scene-interaction motions. After training, the policy is frozen. Stateful programs called skills express task intent as prompts at each control step, drawing rough geometric sketches that the policy turns into coordinated movement. Skills are ordinary code, refined by coding agents that analyze motion data and fit compact rules without task-specific policy training. On the physical Unitree G1, one frozen checkpoint supports walking, recovery, teleoperated object pickup, and sitting on furniture. Comparative evaluations show higher reaching success and push robustness than the tested reference-tracking pipelines. The same recipe produces climbing and acrobatics on a simulated character. Metatron enables coding agents to turn learned motor capabilities into new skills through code, without further policy training.
[AI-197] Assessing Language Models for Scientific Applications: Inertial Confinement Fusion as a Use Case
链接: https://arxiv.org/abs/2610.38024
作者: P. B. Radha,D. D. Meyerhofer,M. Bhattarai,B. Cheng,L. Diaz,C. L. Rousculp,A. G. Seaton,G. Gupta,P. Gurumurthy,V. Gutta,A. Jhunjhunwala,A. Vem,P. Cook,G. Rowe
类目: Artificial Intelligence (cs.AI)
备注: 33 pages, 16 figures
Abstract:Advancing language models for specialized scientific applications requires evaluations of scientific fluency, physical reasoning, and quantitative analyses. Using inertial confinement fusion (ICF) as a test case, we have developed an expert-curated benchmark of 250 questions spanning qualitative reasoning, numerical calculation, symbolic derivation, and equation formulation. Responses have been evaluated using deterministic numerical and symbolic checks together with language-model judges. We compared a variety of out-of-the-box closed-source and open-source large language models with Prospero, a model specialized through combined domain-adaptive pretraining and supervised fine-tuning for ICF. We also conducted initial retrieval-assisted evaluations. Prospero improved its aggregate score by 10.7 percentage points over its base model, with gains across all question types. Frontier models achieved the highest aggregate scores, but critical physics errors persisted, particularly in numerical calculations. We note that domain-specific performance can improve through specialization, but benchmark gains alone do not establish scientific reliability. Retrieval improved performance for a smaller model but yielded no statistically significant improvement for a frontier model. Two scientific stress tests further illustrate useful capabilities and the need for expert assessment of physical assumptions and model validity. Based on these findings, we recommend evaluating existing domain capabilities and retrieval benefits before committing to substantial additional specialization.
[AI-198] Algorithmic Bias from the Perspectives of Healthcare Professionals
链接: https://arxiv.org/abs/2610.38006
作者: Jennifer Xu,Tamara Babaian
类目: Artificial Intelligence (cs.AI)
备注: 12 pages
Abstract:This paper focuses on algorithmic bias of machine learning and artificial intelligence applications in healthcare information systems. Based on the quantitative data and qualitative comments from a survey and interviews with healthcare professionals, who have different job roles (e.g., clinical vs. administrative), this study provides findings about the relationships between algorithmic bias, perceived fairness, and the intended acceptance and adoption of ML algorithms and algorithm generated outcomes. The results suggest that the opinions of healthcare professionals toward the causes of algorithmic bias, the criteria of algorithm assessment, the perceived fairness, and bias mitigation approaches may vary depending on their job roles, perspectives, tasks, and the algorithm characteristics. More research is needed to investigate algorithmic bias to ensure fairness and equality in healthcare.
[AI-199] Are Seismic Foundation Models More Than Glorified Feature Extractors? NEURIPS
链接: https://arxiv.org/abs/2610.38003
作者: Laura Laurenti,David Liu,Brian Kulis,Men-Andrin Meier,Christopher Johnson
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: NEURIPS STODY WORKSHOP 2026
Abstract:Earthquake sequences are stochastic on two separated timescales: the seismic waves of any individual earthquake propagate over seconds to minutes, while the earthquakes themselves arrive as a sparse, self-exciting occurrence process, with inter-event intervals from seconds to months. Respecting both scales in one training objective is the open challenge. Foundation models (FMs) for seismic waveforms instead import their learning targets from audio and vision, where every target is computed from a single fixed-length window. Such a window holds one realization of the wave propagation (an earthquake trace), never the occurrence process that decides when earthquakes happen. We develop nine model designs sharing one Mamba encoder-decoder, trained on a fixed corpus of 3M windows, varying the self-supervised objective, quantization bottleneck and mixer; the two key comparisons are architecture-matched, varying the objective alone. Each checkpoint is evaluated frozen on three downstream tasks. Varying the target redistributes downstream skill rather than raising it, and no checkpoint wins everywhere. Adding masking pressure to our strongest classifier re-specializes it rather than making it a generalist: it gains 6.9 points of few-shot P-wave picking precision and loses 9.6 points of few-shot classification accuracy. By the standard imported from language and vision, these models are feature extractors rather than FMs. Controls on capacity, architecture, and domain proximity do not explain why skill moves rather than rises. Against task-specific baselines the advantage is regime-dependent: our checkpoints beat them few-shot on classification and ground-motion prediction, and trail them on full data. We argue the binding constraint is the target, and that settling the question requires a learning target defined across events, the timescale that makes seismicity stochastic in the first place.
[AI-200] owards Generalizing Pretrained All-Atom Representations to Unified Binder Design
链接: https://arxiv.org/abs/2610.37994
作者: Maojiang Su,Jiale Chen,Qijie Zhu,Shang Wu,Jiayi Wang,Jihai Zhao,Guo Ye,Yibo Wen,Han Liu
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:PairFormer-based all-atom models demonstrate strong capabilities in biomolecular binder design. However, existing methods often use all-atom representations for modality-specific binder design, such as protein and peptide design. In contrast, recent geometric latent diffusion models pursue unified binder design by representing molecular blocks from different modalities in a shared latent space. These complementary strengths motivate us to investigate how to generalize the use of pretrained all-atom representations to unified binder design. We introduce PairFuse, a parameter-efficient method that enables this generalization through representation fusion. Specifically, we use lightweight layerwise adapters to integrate pretrained all-atom representations into an independently pretrained geometric latent diffusion model. Our adapters integrate pretrained all-atom representations through two mechanisms: enriching the diffusion model’s molecular block features and incorporating pairwise structural information into the geometric diffusion model’s denoiser. Theoretically, we prove that the augmented denoiser preserves geometry and pairwise attention fusion provides additional attention expressivity beyond feature-only fusion. Experimentally, across antibody, peptide, and small molecule binder-design benchmarks, PairFuse achieves stronger performance than the pretrained diffusion model and existing representation-fusion baselines.
[AI-201] GAUSS: A Workload-Aware Analytical and Monte-Carlo Simulator for LLM Inference Serving
链接: https://arxiv.org/abs/2610.37989
作者: Srikant Bharadwaj,Ankur Mallick,Spyridon Mastorakis,Anjaly Parayil,Dhruv Deshmukh,Victor Ruehle
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Large language model (LLM) inference has become an operational workload that fleets must provision, monitor, and bill against latency service-level objectives (SLOs). A single throughput number cannot answer the questions operators actually face, because generative serving couples two phases with very different service characteristics, continuous batching reshapes the effective service process, and user-visible tail latency can degrade long before aggregate throughput saturates. We present GAUSS, a workload-aware analytical and Monte-Carlo simulator that predicts the full distributions of time-to-first-token (TTFT) and time-between-tokens (TBT), together with the sustainable request rate, for a given workload, engine, and accelerator. GAUSS pairs calibrated, phase-specific latency surfaces for prefill and decode with a Lindley-recurrence model of queueing under stochastic arrivals and estimates performance by ensemble Monte-Carlo sampling rather than by replaying a single trace. This lets GAUSS reason about workloads, routing policies, and hardware that have never been deployed, and to capture the latency tails that trace-driven simulators systematically under-sample. We describe the model and show how it supports operating-point selection, traffic segmentation, and shortest-remaining-processing-time scheduling capacity questions that trace-driven simulators cannot address.
[AI-202] Self-Supervised Latent Action-Conditioned World Modeling for Time Series
链接: https://arxiv.org/abs/2610.37963
作者: Chia-Han Kuo,Chengshuai Zhao,Huan Liu,Lo Pang-Yun Ting
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Time-series data play an important role in various tasks, such as weather forecasting, web traffic analysis, and speech recognition. Recently, pioneering studies have increasingly adopted Joint-Embedding Predictive Architecture (JEPA) based encoders to learn time-series representations through latent-space prediction. However, they often overlook the underlying factors governing transitions between time-series observations. In this work, we hypothesize that explicitly modeling the information behind changes in time series can facilitate time-series modeling. Building on this hypothesis, we propose ACTS-WM, a self-supervised latent action-conditioned world model framework for time-series analysis. First, we introduce a latent action modeling approach that infers the underlying information behind time-series transitions as latent actions, without requiring explicit action annotations. We then develop a JEPA-based encoder and an action-conditioned world model to predict target time-series representations conditioned on the learned latent actions and task information. We evaluate ACTS-WM across 4 representative tasks on 22 datasets from diverse domains, covering univariate and multivariate forecasting and classification. Comprehensive experiments demonstrate that ACTS-WM can improve downstream performance over pretrained JEPA-based encoders by learning complementary representations.
[AI-203] SAG-Sep: Sparse Augmented Graphs and Onion-Guided Search for Rounded Capacity Cut Separation NEURIPS2026
链接: https://arxiv.org/abs/2610.37952
作者: Haoran Liu,Guanyi Wang,Yu Yang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Optimization and Control (math.OC)
备注: 42 pages, 4 figures, 26 tables. Accepted at NeurIPS 2026 (Main Track). Code: this https URL
Abstract:Rounded capacity cuts (RCCs) are among the most effective cuts for capacitated vehicle-routing relaxations, but exact separation is computationally prohibitive at scale. Existing heuristic and neural separators can be fast, but often fail to produce sufficiently many effective cuts within practical separation budgets. We introduce SAG-Sep, a separator that scores node membership in RCC-inducing subsets across vehicle-count levels from a fractional LP solution in a single encoder forward pass, avoiding the iterative predict-and-coarsen inference used by NeuralSEP-style separators. SAG-Sep encodes the LP solution as a Sparse Augmented Graph (SAG), adding flow-informed multi-hop edges that expose long-range routing structure without densifying attention. Beyond LP-aware sparse encoding, we observe that high-value exact RCCs are typically nested, organizing into onion-like subset families, and SAG-Sep’s node scores track subset nesting depth. This motivates Onion-Sweep (OS), a training-free decoder that ranks customers by their scores, evaluates the RCC violation of every prefix of the ranking in one vectorized pass, and emits the most violated prefix of each replica together with a cross-replica family of nested proposals, converting independent node scores into structurally coordinated RCC candidates without a tuned threshold or probing band. On the NeuralSEP benchmark with N=1000 customer instances, SAG-Sep with OS reduces the average root gap by 17.7% relative to the strongest NeuralSEP baseline, while the single-pass SAG-Sep separator achieves a roughly 30-fold per-iteration neural separation speedup. Our findings suggest that LP-aware sparse encoding and the nested structure of high-value cuts are structural patterns worth leveraging in neural separation more broadly.
[AI-204] NLP-based Message Prioritisation for Opportunistic Routing: A Case Study of Nepal Earthquake Emergency Communication
链接: https://arxiv.org/abs/2610.37933
作者: Yihang Cao,Milena Radenkovic
类目: Networking and Internet Architecture (cs.NI); Artificial Intelligence (cs.AI)
备注:
Abstract:This dissertation investigates whether Natural Language Processing (NLP) based semantic message prioritisation can improve the treatment of urgent disaster messages in opportunistic emergency communication, using the 2015 Nepal earthquake as a case study. TF IDF with Logistic Regression and BERT are used to classify social media messages into nine humanitarian categories, which are mapped to four routing priorities from P1 to P4. These priorities are integrated into the ONE simulator within a controlled Kathmandu earthquake scenario using a modified Binary Spray and Wait router. BERT achieves higher priority accuracy and P1 recall than Logistic Regression, while reducing under prioritisation and severe priority errors. However, improved NLP performance does not produce uniform network gains. P1 delivery probability remains high and similar across routing conditions, and no consistent latency reduction is observed. The clearest difference appears under buffer pressure, where the BERT condition shows fewer observed ground truth P1 buffer losses than Logistic Regression. The results indicate that semantic prioritisation is most valuable when limited transmission and storage resources are contested, although it cannot overcome mobility and contact constraints in opportunistic networks.
[AI-205] Understanding Large Language Models as Numerical Optimizers
链接: https://arxiv.org/abs/2610.37932
作者: Lingyun Guo,Ali Bereyhi
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Signal Processing (eess.SP)
备注: 17 pages, including appendices
Abstract:Large language models (LLMs) have emerged as potential tools for numerical optimization. Despite promising results, their intrinsic optimization capabilities and limitations remain poorly understood, particularly in high dimensions. This paper investigates the ability of LLMs to exploit mathematical problem structures and generate feasible, near-optimal solutions. We develop an evaluation framework that distinguishes solution quality from execution failures, including numerical inaccuracies, constraint violations, and output-format errors. Using representative problems from convex optimization, fractional programming, and constrained projection, we assess LLM performance against deterministic reference solvers across different dimensions. Our results reveal that LLMs can capture useful optimization structures and produce near-optimal solutions, even when their outputs fail to satisfy numerical or formatting requirements. However, reliability deteriorates with increasing problem dimension and depends strongly on the underlying optimization structure. To further investigate the limitations, we deploy an iterative refinement mechanism that provides targeted feedback on feasibility and solution quality. The results demonstrate that iterative refinement substantially improves reliability and extends the range of problem dimensions over which LLMs produce valid solutions. Nevertheless, persistent numerical and structural errors limit their applicability as standalone optimizers. These findings provide insights into the capacity and limits of LLM-based optimization and motivate hybrid approaches that combine LLM reasoning with deterministic numerical verification.
[AI-206] VerifiedLab: Harnessing Knowns for Unknowns
链接: https://arxiv.org/abs/2610.37923
作者: Yu Yao,Jiayi Dong,Yang Yang,Shuiwang Ji,Ju Li
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Across human history, knowledge has undergone several transitions, from orality to writing and then to digital archives. At each transition, humans established new conventions to preserve knowledge for the new era. Nowadays, as research agents are increasingly deployed for laboratory experiments, knowledge is crossing another boundary: from records written for humans to those for machines. In this work, we introduce VerifiedLab, a multi-agent harness framework for this human-to-machine transition of knowledge. VerifiedLab transforms research output from passive artifacts into active systems that accelerate downstream use, adoption and prediction of future trends. In VerifiedLab, a team of AI scientists with distinct roles reasons over records verified field by field: a principal investigator plans, a librarian computes the evidence directly from the records, specialists draft, and a gate and a critic check each claim against the measured reliability of its data before the harness decides whether to release or withhold it. We demonstrate the efficacy of VerifiedLab through extensive experiments on perovskite solar cells: VerifiedLab recovers known physical trends, anticipates devices reported a year later, and lets AI scientists inherit the literature together with a measure of how far it can be trusted. By scaling the inference-time verification compute, VerifiedLab introduces a new pathway for knowledge dissemination and machine harness for AI co-scientists.
[AI-207] ERRA: Task-Oriented Edge AI for Regional Fire Detection with Identity-Free Reporting
链接: https://arxiv.org/abs/2610.37909
作者: Mohammad Kazemi,Julie A. McCann
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Signal Processing (eess.SP)
备注: Accepted for publication at the 16th International Conference on the Internet of Things (IoT 2026)
Abstract:Distributed fire-monitoring systems may involve many sensing devices reporting over a shared uplink, even though the operational objective is often to identify regions exhibiting evidence of fire rather than to recover device-specific measurements. We introduce TERRA (Task-Oriented Edge Representation and Regional Aggregation), an edge-AI framework that combines task-oriented local inference, identity-free region-tagged reporting, and permutation-invariant regional evidence aggregation. TERRA supports two reporting modes: TERRA-Score, in which each device communicates a quantized local fire probability, and TERRA-Feature, in which it communicates a compact learned representation for receiver-side aggregation using a permutation-invariant DeepSets model. Using all 15 numerical sensing modalities of the distributed indoor EN54 fire-alarm dataset as a proof-of-concept testbed, we evaluate regional inference for an episode-level fire-scenario target under repetition-disjoint three-fold cross-validation. We study representation size, task-payload quantization, report availability, and robustness to missed and false recovered messages. With nine reports and perfect message recovery, TERRA-Feature achieves mean AUPRC 0.990 using a 24-bit task payload, while TERRA-Score achieves 0.987 using only 6 bits; both attain an MDR of 1.81% at FAR \leq 1%. TERRA-Feature achieves slightly higher AUPRC under reliable recovery, whereas the more compact TERRA-Score is less sensitive to increasing false-message contamination.
[AI-208] Beyond Low-Pass Dynamics: Frequency-Selective Spiking Reservoirs with Resonant Neurons NEURIPS2026
链接: https://arxiv.org/abs/2610.37906
作者: Aurora Micheli,Claudio Gallicchio,Nergis Tomen
类目: Artificial Intelligence (cs.AI)
备注: Accepted at NeurIPS 2026. 30 pages (10 main text), 4 figures. Code: this https URL
Abstract:Spiking reservoir computing combines the training efficiency of fixed recurrent reservoirs with the energy efficiency of event-driven inference, where computation is triggered by sparse spikes rather than executed densely at every timestep. However, existing spiking reservoirs almost exclusively rely on Leaky Integrate-and-Fire (LIF) neurons, whose low-pass dynamics limit their ability to distinguish information encoded at different temporal frequencies. We introduce HRF-Res, a liquid state machine in which LIF neurons are replaced by a heterogeneous population of Harmonic Resonate-and-Fire (HRF) neurons. These neurons exhibit intrinsic oscillatory dynamics and selectively respond to input frequencies, enabling the reservoir to capture frequency-specific temporal patterns. The result is more discriminative representations and higher classification accuracy than LIF-based reservoirs, achieved without sacrificing the energy efficiency of spiking computation, with up to 19x lower energy consumption than an equivalent non-spiking oscillatory reservoir. Evaluated on ten benchmarks spanning time series, sequential images, audio spike trains, and neuromorphic vision, HRF-Res achieves state-of-the-art accuracy among spiking reservoir methods. These results demonstrate that introducing frequency-selective dynamics into spiking reservoirs improves representational quality and classification performance while preserving their inherent energy efficiency.
[AI-209] FilingLoop: Evaluating and Evolving LLM Agents for Corporate Disclosure Review and Revision
链接: https://arxiv.org/abs/2610.37905
作者: Wentao Zhang,Yilei Zhao,Guangyan Gan,Shuo Sun,Lin William Cong
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Public disclosure repositories such as SEC EDGAR support informed investment through web access to corporate filings. Supporting their review with language model agents requires discovering unstated deficiencies across related documents, acquiring missing company facts, and verifying that requested revisions satisfy evidence-backed requirements. Existing financial question-answering evaluations provide limited evidence of these interactive capabilities. We introduce \textbfFilingLoop, an interactive benchmark and a self-evolving agent framework for corporate disclosure review and revision. Grounded in historical U.S.\ Securities and Exchange Commission correspondence, FilingLoop reconstructs 380 predominantly IPO review tasks from 96 companies under distinct information constraints. An evaluator separate from the acting agents scores final filings and responses against 6,099 eligible requirements and audits unsupported claims and harmful revisions. Our framework couples iterative review and revision within each case with procedural learning across cases, converting requirement-level evaluation feedback into role-specific guidance without updating model weights. Updates affect only subsequent cases, allowing procedural adaptation to be assessed without retrying scored tasks. On the evaluated stream, evolution increases requirement fulfillment from 24.28% to 27.96% and full comment resolution from 16.90% to 20.75%, with a 5.8% increase in total cost over the fixed-policy variant. FilingLoop connects the concerns an agent discovers with the company facts it obtains and the revisions it commits, providing requirement-level feedback for procedures used in later review tasks.
[AI-210] Scan Before You Install: Detecting and Mitigating Local and Composite Vulnerabilities in Skills
链接: https://arxiv.org/abs/2610.37901
作者: Parth A. Gandhi,Avital Aviv,Prasanna N. Wudali,Yuval Elovici,Asaf Shabtai
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注:
Abstract:An agent skill packages natural-language instructions, scripts, and reference files that an LLM-based agent loads for a task. Skills may access the file system, shell, and network under user-authorized permissions; weaknesses can lead to unsafe execution. Existing scanners primarily target malicious skills, not unintended vulnerabilities in public repositories, and analyze skills in isolation, overlooking multi-skill interactions. To address these gaps, we developed a six-module scanner: Content Extraction, Local Analysis, Composite Analysis, Validation, Reporting, and Mitigation. Local Analysis detects unintended vulnerabilities in benign skills. On a 556-case leaderboard subset of SkillTrustBench, it achieved 95.90% recall with a 4.22% false positive rate; on a balanced sample of 1,000 MalSkillBench skills, its F1 was 0.947, outperforming Cisco’s Skill Scanner, NVIDIA’s SkillSpector, AgentSkillsScanner, and SkillSieve. In 1,587 skills from the 75 most-starred GitHub repositories, three analysts confirmed 53 of 59 findings (89.8% precision) across 23 repositories. We developed proof-of-concept (PoC) attacks for each vulnerability, disclosed 10 medium- and high-severity cases to maintainers, and tested mitigations on the remaining cases. Composite Analysis models dependencies among skills through shared resources, invocations, and instruction-mediated interactions. On SCR-Bench, its F1 score was 0.991 (98.7% precision, 99.6% recall), versus 0.816-0.858 for three SkillsGuard configurations released with CompoSkill. It identified two composite findings on GitHub, validated by PoC attacks. The Mitigation module recommends skill substitution, patches, and configuration changes. For 47 of 55 GitHub findings, mitigations prevented PoC attacks in all trials while preserving tested legitimate functionality.
[AI-211] he Anatomy of Socio-Demographics in Pluralistic Offensiveness Modeling: When and Where It Helps and the Role of Reasoning
链接: https://arxiv.org/abs/2610.37896
作者: Yifan Wang,Leixin Zhang,Jinyi Mu,Yu Wang,Mayank Jobanputra,Isabel Valera,Vera Demberg
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Subjective judgments such as offensiveness often vary substantially across individuals, motivating pluralistic modeling approaches that incorporate annotators’ socio-demographic information to capture this variation. However, prior findings on the benefits of socio-demographic information for pluralistic modeling are mixed, calling for a systematic analysis of both its effectiveness and the mechanisms through which it improves prediction. In this work, we study \textbfwhen and \textbfwhere socio-demographics benefit pluralistic offensiveness modeling, as well as \textbfthe role of explicit reasoning in leveraging this information. Across three multi-annotator datasets with prompted and trained LLMs, we first show that demographic gains are small and inconsistent for individual predictions but become clearer at the group and population levels, because aggregation attenuates the idiosyncratic variation that masks small but systematic demographic signals. Second, using a reasoning and prediction format that explicitly decomposes individual judgment into an overall baseline and a demographic-specific deviation, we show that the gains arise from both components: demographic information improves baseline estimation and enables meaningful demographic-specific shifts, and interventions show that better component estimates improve final predictions. Finally, comparing direct demographic conditioning with explicit reasoning about how the text interacts with annotator demographics, we find that reasoning yields more accurate baseline and shift estimates as well as final predictions, whereas conditioning alone rarely produces informative shifts. These results suggest that the mixed evidence in prior work may partly stem from individual-level evaluation, and that the value of socio-demographic information depends on both how it is evaluated and how models are guided to use it.
[AI-212] When More Planning Hurts: Learned Planners for World Models
链接: https://arxiv.org/abs/2610.37891
作者: Josef Chen
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 23 pages, 9 figures, 16 tables, 28 references
Abstract:On PushT, our fixed action updater (1,505 parameters) solves 8 of 16 scenes reserved from our configuration selection. The comparison planners perform as follows: Selected Adam (7 tasks solved, total planning and encoding time 218.440s), adapted iCEM (2 tasks solved, total planning and encoding time 49.309s), released LePlanner (16 tasks solved, total planning and encoding time 5.652s). The total planning and encoding time for our action updater is 59.079s. Note that our action updater does not satisfy the task-cost criterion we fixed before evaluation. We also conduct another matched experiment where we only replay the first ten commands in each context and then allow the next solve 4, 8 or 12 updates (subsequent solves all get 8 updates). Among all 16 contexts (8 scene clusters), the 3 budgets get 9, 8 and 7 successes, respectively. Interestingly, the extra updates did not rescue any failing task in this experiment. All the changes in success/failure happen in one scene cluster, where we observe that the extra updates decrease the costs of the selected latent world-model trajectories for our updater, while the tasks that used to be successes become failures. The code, weights, evaluators, execution traces and cost accounting accompany the paper.
[AI-213] Verification Learning Theory: A Process Model of Checking and Retention in AI-Assisted Judgment
链接: https://arxiv.org/abs/2610.37875
作者: Christoph Trattner
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Accurate AI assistance can improve immediate judgments while reducing opportunities to practise the operations needed for independent performance. Verification Learning Theory (VLT) formalises this possibility for claim-evaluation practice followed by correct, independent, rule-specific feedback. The model links perceived assistant reliability to evidence checking and links the history of those checks to retained rule use. Its strict form assumes that task demands and perceived reliability influence learning only through checking, with equal learning per genuine check. It therefore predicts that perceived fallibility can improve delayed unaided judgment even without observed assistant errors, and that this advantage disappears when checking is already required and performed. Rule-specific gains and forecasts from checking histories provide further tests. Simulations examine the consequences and parameter dependence of these commitments; they do not constitute empirical validation. Comparisons with depth-of-processing and feedback-encoding accounts identify where predictions diverge and where the accounts remain observationally equivalent. The framework specifies experiments that can test the proposed mechanism and separates retention of practised rules from the broader problem of recognising which rule applies to a new claim.
[AI-214] Echo: Scaling Visual Reinforcement Learning with Off-policy Experience
链接: https://arxiv.org/abs/2610.37865
作者: Gagan Khandate
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注: Report. 16 pages
Abstract:Scaling visual reinforcement learning (RL) with GPU-accelerated simulation is computationally expensive due to rendering and visual-policy inference costs. Consequently, two-stage teacher–student distillation is commonly used to train visual policies, but introduces an imitation gap. In this preliminary study, we introduce \echomethod, \textbfa single-stage off-policy RL framework that scales experience collection without proportionally increasing visual computation. Through \textbfinverse distillation, a state-based auxiliary policy imitates the evolving visual policy and collects inexpensive, unrendered experience. Echo leverages this experience for off-policy value learning, directly improving the visual policy through a shared privileged critic. We instantiate \echomethod with FastTD3 and evaluate it on three visual control tasks and two dexterous manipulation tasks in Genesis. Relative to \sdpgmethod under the evaluated configurations, Echo reduces time to reference returns by an average of \textbf60% across the three locomotion tasks and \textbf56% across both hand tasks, while improving peak mean returns by \textbf62% and 12%, respectively. Both hand tasks also achieve higher mean returns at matched simulator-transition budgets. These initial results suggest that scaling off-policy experience is a promising approach to efficient direct visual RL.
[AI-215] Seq@K: Evaluating Language Models Under Sequential Feedback
链接: https://arxiv.org/abs/2610.37859
作者: Thomson Yen,Gilbert Yang,Naimeng Ye,Minghui Chen,Dong Xia,Tianyi Shen,Yechan Jeon,Yaqin Chen,Daksh Mittal,Fangyu Wu,Hongseok Namkoong
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Language-model agents differ fundamentally from prior intelligent systems in their ability to learn from natural language in context. This capability enables agents to iteratively refine their answers in response to feedback, a prevalent mode of operation in modern agentic workflows. However, the community primarily ranks models by metrics such as Pass@1 or Pass@K, which cannot capture how they utilize feedback on prior attempts. We introduce Seq@K, a natural evaluation protocol that measures whether a model can solve a problem within K sequential attempts, provided feedback after each previous failure. Unlike multi-turn agentic benchmarks that examine long-horizon planning, Seq@K isolates and formalizes the core capability of refinement from feedback. We evaluate Seq@K on a suite of 7 benchmarks and 16 frontier models, and show that Seq@K reveals a materially different model-selection landscape: under Seq@K, (1) model rankings often differ, (2) the Pareto frontiers admit more efficient configurations and (3) the gains from choosing the optimal model configuration can widen or shrink relative to those of Pass@K. We further show that analyzing model trajectories under Seq@K helps characterize iterative refinement behavior, such as the rate at which models address issues raised in feedback, or how rigidly they hold onto their own answers. Finally, we study how the nature of feedback affects models’ iterative refinement, showing that practitioners must carefully design and benchmark the feedback loops of iterative agentic systems. Together, these results establish how Seq@K offers a more faithful metric for intelligent systems that typically operate in such feedback loop, and the necessity to consider Seq@K in evaluating models for these systems.
[AI-216] Evaluating and improving AI agents for sequencing protocol understanding
链接: https://arxiv.org/abs/2610.37851
作者: Chi-Lam Poon,Michael Goneos,Zezheng Song,Xi Chen
类目: Artificial Intelligence (cs.AI); Genomics (q-bio.GN)
备注:
Abstract:Existing scientific-agent benchmarks rarely evaluate whether agents can reconstruct the exact molecular structures and generation workflows specified by sequencing protocols, a capability critical for analyzing customized sequencing assays. We introduce LibStructBench, a document-grounded benchmark for reconstructing sequencing-library structures and molecular workflows. Our experiments show that frontier agents often produce approximately correct reconstructions but remain unreliable at exact reconstruction, particularly in molecular-state representation and workflow connectivity. Switching the underlying language model produced larger average performance differences than switching agent harnesses, with a statistically significant difference only for molecular-state reconstruction. Additional knowledge improved oligo-family recovery in a small pilot without improving mean molecular-workflow scores. Autonomous accumulation of procedural rules and cross-protocol examples showed non-monotonic validation performance and no mean held-out molecular-workflow improvement in the tested trajectory. These findings highlight the need for more controlled and verifiable approaches to scientific reasoning about molecular processes in complex experimental protocols.
[AI-217] Learning Lotteries with Minimal Stakeholder Violations from Membership Queries NEURIPS2026
链接: https://arxiv.org/abs/2610.37828
作者: Nicholas Teh,Paul W. Goldberg
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 31 pages, full version of paper at NeurIPS 2026
Abstract:An AI developer often has several candidate versions of a system and can deploy a mixture of them, sending each request to a randomly chosen version, as in A/B testing. Each stakeholder (e.g., a safety team, a regulator, or a customer with private test data) accepts a mixture exactly when its expected performance (e.g., its rate of harmful replies) meets the stakeholder’s own standard, which the developer can learn only by asking whether a proposed mixture is acceptable. We model this as choosing a lottery over m alternatives for n agents, each of whom accepts exactly the lotteries in an unknown halfspace and answers such binary membership queries. While Choo et al. (ICML 2026) sought a lottery that every agent accepts, none may exist when, say, safety and helpfulness requirements conflict, so we minimize the number of rejecting agents. We show that this problem is APX-hard even when all halfspaces are known and all utilities are 0 or 1 . We then give exact algorithms that run in polynomial time when m is constant, and in quasi-polynomial time when, for a constant \eta , some optimal lottery has no accepting agent within \eta of its threshold. When only some agents can be queried, optimizing over a uniform sample of agents gives additive and relative guarantees for the whole population, with a sample size that does not depend on n , and a linear program gives a polynomial-time method for any m . On the other hand, for m\ge 3 , a deterministic algorithm that queries only K distinct agents cannot guarantee fewer than n-K extra rejections, and this is achievable; in other words, it can guarantee nothing for the agents it does not ask. Small synthetic experiments indicate that our worst-case bounds are conservative on random instances.
[AI-218] Responsible Quantum Machine Learning for Medical Imaging: A Fairness Audit Across Sensitive Subgroups
链接: https://arxiv.org/abs/2610.37819
作者: Giovanni dos Santos Franco,Felipe Mahlow,Ellison Fernando Cardoso,José Adenaldo Santos Bittencourt,Felipe Fanchini
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computers and Society (cs.CY)
备注:
Abstract:Quantum machine learning (QML) is increasingly explored for medical classification, yet its behaviour under subgroup-aware fairness criteria remains understudied. We present an empirical fairness audit of classical and quantum classifiers for medical imaging, assuming QML is not inherently fair. Two scenarios are evaluated: dermatological lesions on DDI (sensitive attribute: Fitzpatrick skin type) and chest X-rays on CheXpert-derived data (sensitive attribute: age group). Using ResNet-18 embeddings, we train SVM, MLP, QSVM and QCNN under a baseline and three mitigation settings: differentiable Equalized Odds (EO) regularisation, Gradient Reversal Learning (GRL), and both combined. Foundation models (PanDerm, EVA-X) serve as external references. Quantum classifiers may achieve competitive aggregate performance while still exhibiting subgroup disparities. Since a smaller subgroup AUC gap may stem from degrading the best group rather than improving the worst, each mitigation is also tested against a pre-specified non-inferiority margin relative to its baseline. Under this criterion, several configurations with the smallest disparity, notably QCNN with GRL on DDI, are not clinically acceptable due to loss of global discrimination. On DDI, QSVM with GRL+EO performed best and approached PanDerm in the performance/fairness trade-off; on CheXpert, QCNN with GRL achieved the highest aggregate performance but increased disparity, and EVA-X remained stronger. Fairness effects are model- and dataset-dependent, so QML requires empirical subgroup-aware evaluation. All quantum models run in classical simulation; we claim no quantum advantage.
[AI-219] What Does an Ontology Catch? Curated LLM -Generated and Data-Derived Ontologies as Checkers of LLM Agents over Knowledge Graphs
链接: https://arxiv.org/abs/2610.37809
作者: Michal Orzechowski,Rund Tawfiq,Tomasz Kosciolek
类目: Artificial Intelligence (cs.AI); Databases (cs.DB)
备注: 13 pages, 2 figures. Under review
Abstract:LLM agents that answer questions over Web knowledge graphs are increasingly wrapped in checks derived from the graph’s ontology, on the assumption that a formal schema makes a reliable judge. Yet a checker can be wrong as well: it can miss errors, and it can flag correct answers and push the agent into “fixing” them. Whether a curated ontology is worth its cost as such a judge, and how it compares with an ontology an LLM writes or a schema mined from the data, is unknown. We make the checker itself the object of study. Across curated, LLM-generated and data-derived ontologies, used as prompt context, as a checker, or both, and across open-weight and proprietary agents on public graphs and a private graph of nearly 10^9 triples, we account for what ontology-derived checks catch and what they wrongly flag. We show how each source trades caught errors against false alarms, how a curated schema used as published can contradict its own data, which errors structural checks miss, and whether combining sources or giving the agent the checker’s specific feedback pays off. On UniProt, the curated ontology flags more correct answers than either an ontology derived from the data or one an LLM derives from a summary of the data with masked identifiers. On real agent answers, its flags are only marginally more precise than random flags.
[AI-220] Spokes not a Ring: Revisiting the Weekday Manifold in Language Models
链接: https://arxiv.org/abs/2610.37796
作者: Diksha Gupta,Rieke Fruengel,Stefan Heimersheim
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Recent work has hypothesised that large language models may represent some concepts as multi-dimensional manifolds instead of single directions. Using these manifolds to interpret or steer representations often assumes that moving along them corresponds to a continuous change in the represented concept. In this work, we test this continuous manifold assumption on the prototypical weekday manifold. Despite its striking circular geometry, we find that the weekday representations behave more like seven discrete states arranged in a circle than a continuous ring. First, we show that local sensitivity has seven minima aligned with the weekdays, producing a step-like mapping between representation and behaviour under on-manifold steering. Second, we test whether any concepts are represented along the manifold outside the seven discrete states, and find neither time of day, uncertainty, nor natural text populate the intermediate space. Together, these results challenge the continuous interpretation of the weekday manifold and motivate testing whether movement along other representation manifolds correspond to meaningful changes in model representations and behaviour.
[AI-221] WebAnything: From the Wild Web to Scalable Agent Environments
链接: https://arxiv.org/abs/2610.37786
作者: Tianci Xie,Zikun Cai,Chaoyun Zhang,Pinxue Guo,Bingxuan Li,Rui Zheng,Hang Yan
类目: Artificial Intelligence (cs.AI)
备注: 25 pages, 6 figures
Abstract:Web agents require diverse, executable, and resettable website environments for reproducible training and evaluation. Human-assisted mirrors demand per-site engineering, while generative construction can diverge from specified real websites. Capture-and-replay preserves observations, with interaction coverage constrained by captured content and replay capabilities. We present WebAnything, a fully automated framework that reconstructs source-grounded, reusable Web environments from real URLs without per-site manual implementation or repair after initial setup and authorization. An Explorer-Builder collects browser-observable structures, resources, and interaction outcomes; an independent verifier checks frozen candidates and guides local repair or targeted source recapture. We construct 244 website artifacts across 14 application domains and 53 categories, supporting 12,984 executed and accepted tasks. Compared with human-assisted WebHarbor, WebAnything achieves higher aggregate scores on three fidelity metrics, with 12 wins, two ties, and no losses in a blind audit. Fine-tuning Qwen3.5-9B on 5,856 interaction trajectories improves REAL success by 9.9 percentage points, Odysseys-Easy Rubric Avg and Task Avg by 6.25 and 6.32 percentage points, and WebArena-Verified Overall success by 13.33 percentage points under the fixed harness. Code: this https URL Comments: 25 pages, 6 figures Subjects: Artificial Intelligence (cs.AI) Cite as: arXiv:2610.37786 [cs.AI] (or arXiv:2610.37786v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2610.37786 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-222] Poisoned Skills: Propagating Jailbreaks Through Agent Markdown Files
链接: https://arxiv.org/abs/2610.37738
作者: Kashyap R Manjusha,Murali Emani
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注:
Abstract:Modern agent frameworks allow users to provide Markdown files containing system prompts, tool descriptions, and skill definitions (e.g., this http URL), which are incorporated into the agent’s context during initialization. In this work, we investigate whether this configuration mechanism can also serve as an injection channel: can a jailbreak embedded in a Markdown skill file propagate through an agent and induce prohibited behavior? To answer this question, we construct a LangChain-based question-answering agent whose system prompt is assembled from a user-provided Markdown file. We embed a jailbreak combining persona manipulation with a GPTFUZZER-based attack and evaluate its effectiveness across seven contemporary open-weight models using 270 forbidden questions spanning nine harm categories. Under a benign system prompt, the models exhibit near-complete refusal behavior, with a mean attack success rate (ASR) of just 0.5%. In contrast, the injected skill file raises the mean ASR to 45%, reaching 68.9% for individual models. We further evaluate two payload obfuscation strategies: an invisible zero-width character-based stealth payload and a phonetic payload based on grapheme-to-phoneme transformations. Neither outperforms the plain-text injection; notably, the phonetic variant reduces mean ASR to 30%. These findings identify Markdown skill files as an underexplored attack surface for AI agents and suggest that vulnerability to configuration-based jailbreaks depends more strongly on the underlying model than on the specific payload obfuscation technique.
[AI-223] When Skills Tell Agents Not to Ask: A Large-Scale Study of Oversight Directives in Agent Skills
链接: https://arxiv.org/abs/2610.37734
作者: Timofey Sanko,Yuan Tian
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Software Engineering (cs.SE)
备注: preprint
Abstract:Agent skills, reusable instruction packages centred on a this http URL file, extend AI agents on demand, and a single line in a widely copied skill reaches every agent that loads it. Some lines are oversight-loosening directives: they tell the agent to proceed without approval, to bypass checks with --no-verify, or to keep working without checking in. Although such directives can remove a checkpoint from every agent that loads them, no prior work has examined them. We present the first large-scale study of oversight-loosening directives, analyzing 1.53 million English skills from GitSkills. We split skills into units (sentences, bullets, table cells, and code lines), develop a codebook that labels oversight directives by polarity, mechanism, and scope, label 2,097 units by hand, and train a detector to scale the analysis. Our results show that about one in seventeen skills (about 88,000 skills) contains a loosening directive. These directives most often loosen oversight through permission flags, and most of them (70.1%) apply unconditionally rather than only in specific contexts such as CI. Compared with skills that have the same amount of oversight language, skills that run shell code, install packages, or use sudo have 1.7-1.8 times the odds of containing a loosening directive, an association that holds across all our robustness checks, and they do not compensate with more tightening. Loosening spreads through a few upstream skills rather than through edits by their adopters: 2.4% of the skill families that carry loosening account for half of all copies of loosening-carrying skills, and 1.1% of distinct loosening units for half of all loosening occurrences. Open-source skill scanners catch at most 46% of loosening directives, with a precision below 0.4, while the LLM behind them, prompted with our codebook instead, catches 61% at a precision of 0.71.
[AI-224] Selective Transfer Learning for Groundwater Level Anomaly Forecasting: Accuracy-Risk Trade-offs
链接: https://arxiv.org/abs/2610.37733
作者: Yaxuan Wu,Ali Mohammed Mansoor Alsahag,Seyed Sahand Mohammadi Ziabari
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:This work studies context-aware selective transfer learning for multi-well groundwater anomaly forecasting in the Netherlands. Source-domain pretraining can raise average accuracy but is not uniformly reliable: it degrades performance on a substantial fraction of heterogeneous wells. Context-Aware Selective Learning (CASL) uses calibration-trained, constrained contextual blending of frozen no-transfer and fixed-transfer parents. Global Convex Blend uses the same parents with one calibration-fitted weight per seed. The corrected task requires nine consecutive observed monthly inputs and an exact three-calendar-month target, without interpolation. Evaluation covers 457 wells and 13,718 test windows under five predefined seeds, with the four-cell maxsim0.60 source pool frozen using Always-TL validation RMSE alone. Fixed transfer harms 18.38-51.86% of wells. CASL has 0.424-1.084% lower RMSE than No TL, partially rescuing 97.53-99.50% and fully rescuing 23.46-43.78% of fixed-harmed wells. Partial rescue can arise structurally from convex blending and does not establish contextual superiority. Global Convex Blend protects more wells in four seeds, while CASL has slightly lower pooled RMSE in four. Neither dominates: the results support an accuracy-risk trade-off on the fixed split, not rolling deployment validity.
[AI-225] Amortized Localized Posterior Inference for Model Calibration with Application to an Ocean Turbulence Closure
链接: https://arxiv.org/abs/2610.37731
作者: Exaucé Luweh Adjim Ngarti(AIRSEA),Élise Arnaud(AIRSEA),Léo Nicoletti,Victor Trappler(Mines Saint-Étienne MSE, LIMOS, FAYOL-ENSMSE, FAYOL-ENSMSE),Arthur Vidard(AIRSEA)
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Machine Learning (stat.ML)
备注:
Abstract:Model discrepancy may prevent a single calibration parameter from reproducing a physical response across the context domain. In addition, uncertainties in observations and contexts affect calibration and must be accounted for and quantified. We construct a family of localized posteriors by weighting observation likelihoods according to their proximity to a target context. Gaussian localization weights incorporate context uncertainty through covariance inflation, with reflection at domain boundaries and an optional correction for uneven context sampling. A conditional normalizing flow approximates the family of posteriors through amortized variational inference. Integrating the localized posteriors under a prescribed context measure gives the marginalized parameter distribution. In a twin experiment with the Branin-Hoo model, we compare global and localized inference under uniform and skewed context distributions. Localization reduces the reported aggregate predictive errors relative to the global posterior. An application to the k- \epsilon turbulence closure in a single-column ocean model, with atmospheric surface forcing as context, shows that localized posterior calibration reduces temperature and zonal-velocity errors relative to deterministic and global-posterior calibration. Both twin experiments are constructed to admit local compensation.
[AI-226] Predicate Neuron: Differentiable Quantization of Predicate Relations in Gradient-based Learning
链接: https://arxiv.org/abs/2610.37727
作者: Ioannis Konstantoulas,Iliana Loi,Kyriakos Sgarbas
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 23 pages, 9 figures
Abstract:In machine learning, symbolic reasoning approaches often rely on static architectures that hinder integration into deep neural networks and gradient-based optimization. To address this, we introduce the Native Predicate Neuron, a differentiable logic-reasoning neural module that enriches standard architectures with symbolic reasoning capabilities via first-order logic quantization. The proposed neuron models relational logic by learning to alternate between operations using a single trainable parameter. It features a tailored gradient mechanism with a dual-nature formulation: executing hard logical operations during training while maintaining differentiability for backpropagation. By quantizing both its input and choice state, the Predicate Neuron approximates continuous and non-continuous functions, effectively resolving discontinuities that standard deep learning architectures struggle with. We evaluate the Predicate Neuron across Fully-Connected, Convolutional, Recurrent, and Transformer architectures on benchmarks like MNIST and Wikitext. Results show that incorporating Predicate Neurons accelerates convergence, boosts accuracy, and lowers loss in general, with particularly strong improvements in tasks requiring monotonic reasoning or handling discontinuities. Overall, the Predicate Neuron natively integrates into standard neural networks to enable fuzzy logic operations and is available in Keras and PyTorch.
[AI-227] FVH-Bench: Evaluating Image Generator Safety Beyond Overtly Harmful Imagery
链接: https://arxiv.org/abs/2610.37722
作者: Zihao Zhu,Yuhan Liu,Siwei Lyu,Baoyuan Wu
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注:
Abstract:Modern image generators can produce complex visual artifacts, including diagrams, interfaces, and records that combine text, graphics, and layout to convey detailed information. They may refuse requests for overtly harmful imagery yet still generate artifacts that guide dangerous actions or support deception. We call the risk arising from what such images enable, rather than only what they depict, functional visual harm (FVH). Existing safety benchmarks focus largely on overtly harmful content, such as pornography and graphic violence, leaving image generators’ safety alignment against FVH insufficiently evaluated. To address this gap, we introduce FVH-Bench, comprising 669 prompts across five harmful visual functions and ten risk domains. The 11 evaluated image generators rarely refuse these requests and frequently generate artifacts that fulfill the targeted harmful functions. High scores for target harm realization and harmful artifact usability further indicate that these outputs can support the intended harmful purposes. A further analysis finds that existing safety detectors also provide limited coverage of FVH, indicating gaps in their ability to recognize harmful visual functions. Together, these findings show that evaluating image-generator safety alignment requires considering not only what images depict, but also what they may enable.
[AI-228] LRMoE: Demand-Side Load Re-Calibration for Efficient MoE Training on Memory-Constrained Nodes
链接: https://arxiv.org/abs/2610.37720
作者: Jiaming Hu,Zhida Jiang,Zhaolong Xing,Yu Li,Zhen Chen,Ke Zhang
类目: Artificial Intelligence (cs.AI); Distributed, Parallel, and Cluster Computing (cs.DC); Machine Learning (cs.LG)
备注:
Abstract:Sparse Mixture-of-Experts (MoE) models offer scalable capacity for Web applications, but uneven routing creates computation and communication bottlenecks. Existing placement and replication methods redistribute this workload, yet additional expert copies compete with token-related buffers for limited GPU memory. We present LRMoE, a memory-aware MoE training system that jointly recalibrates backward demand and balances the retained workload. During forward execution, it plans retention quotas and expert placement under resource constraints while preserving all routed computations. During backpropagation, a load recalibration metric estimates each token-expert pair’s contribution to the expert update and selects pairs within these quotas. A capacity-aware routing algorithm distributes retained pairs among eligible instances of their original experts. Physical compaction then forms smaller communication payloads and variable-size grouped GEMMs, reducing backward traffic, computation, and execution-buffer demand. Experiments on Qwen3-30B-A3B and GLM4.5-106B-A12B show that LRMoE reduces cumulative MoE-layer training time to fixed-loss targets by 21.2-44.8% relative to SYMI.
[AI-229] Search-Guided Reasoning Trace Curation for Verilog Generation
链接: https://arxiv.org/abs/2610.37706
作者: Matthew DeLorenzo,Kevin Tieu,Prithwish Jana,Piyush Jha,Dileep Kalathil,Vijay Ganesh,Jeyavijayan Rajendran
类目: Hardware Architecture (cs.AR); Artificial Intelligence (cs.AI)
备注:
Abstract:Recent advances in large language models (LLMs) have improved Verilog generation. However, most training pipelines distill natural-language reasoning traces whose intermediate design decisions are weakly structured and are primarily selected according to the correctness of the final generated RTL. We present MCTS-SFT, a verifier-guided framework for curating reasoning traces for supervised fine-tuning (SFT) of Verilog generation models. MCTS-SFT formulates trace construction as search over intermediate reasoning states and uses Monte Carlo Tree Search (MCTS) to propagate functional feedback from complete Verilog rollouts to earlier design decisions. Within this framework, we introduce Plan-MCTS, which searches over compact hardware-design plans that explicitly represent interfaces, timing, state, behavior, and update rules. We further study plan-preserving supervision, in which the searched structured state is retained as part of the downstream SFT target. Our experiments find that Plan-MCTS achieves the highest functional trace yield, with 84.2% correct traces compared with 81.9% for one-shot structured planning and 82.0% for unguided tree exploration. For identical Plan-MCTS trajectories and Verilog outputs, retaining the explicit searched plan during SFT raises pass@10 from 52.56% to 57.7%, demonstrating that the representation of intermediate design decisions materially affects downstream learning. In a separate 6,000-trace study, MCTS-curated supervision improves pass@10 over full-response chain-of-thought distillation by up to 3.7 points while reducing downstream generation length by approximately 50% on VerilogEval-v2. These results demonstrate that structured hardware planning, verifier-guided search, and explicit intermediate-state supervision provide complementary mechanisms for constructing more effective and token-efficient training data for Verilog generation.
[AI-230] Cassette-Editor: A Video Editing Agent That Edits the Way You Do
链接: https://arxiv.org/abs/2610.37690
作者: Zhixin Zeng,Jiahao Zhong
类目: Multimedia (cs.MM); Artificial Intelligence (cs.AI)
备注: 8 pages, 4 figures, 2 tables. Preprint; evaluation to be extended in a future version
Abstract:Language agents increasingly work by calling tools on a persistent artifact, such as a code repository, and the result of each call shapes their next step. Video editing is a demanding case of this pattern: what a frame shows depends on how every clip is trimmed, retimed, cropped, and nested, so knowledge about the footage holds only for particular uses of it, and an invalid edit can leave the timeline structurally broken. Existing video editing agents keep their edits valid with critic agents, reviewers, or restricted output formats. We present Cassette-Editor, a video editing agent with a deliberately simple interface: one tool reads the project and one tool writes to it. Every write passes through the same executor, rules, and commit service as the user’s manual edits, and either commits atomically or is rejected with located diagnostics and repair options, while the model keeps the creative choices and decides how to recover. Reads relate stored analysis of the footage to each use of it on the timeline. In a controlled experiment, a real model is shown 277 of its own committed edits after damaging them. It corrected nearly every call with full or minimal feedback, but locating each problem cut the input tokens a correction needed to less than half, and its corrections dropped other values the model had written in one case rather than seven.
[AI-231] GenAI as an Invisible Hand in Opinion Dynamics: Evidence from a Group-Level Simulation Study
链接: https://arxiv.org/abs/2610.37689
作者: Thomas Grisold,Johannes Schneider
类目: ocial and Information Networks (cs.SI); Artificial Intelligence (cs.AI)
备注:
Abstract:Recent arguments suggest that genAI can act as an invisible hand that gradually shapes social structures, such as collective beliefs and norms. The argument is that if millions of people repeatedly consult genAI tools, their generated outcomes quietly shape what becomes acceptable, reasonable or conventional. We study this in the context of collective opinion dynamics, i.e., how groups form opinions around controversial dynamics. We present the findings of a simulation including 50,000 group discussions across 50 controversial statements, three LLMs, and different social configurations. We vary two characteristics of LLM interaction-agreeableness and personalization-and compare the influence of genAI in homogeneous and heterogeneous groups. Our findings indicate that profile-specific personalization increases LLM influence when models adopt a non-neutral stance, i.e., when they agree or disagree with the user. Furthermore, we find that neutral LLMs have a stronger effect in homogeneous groups. Based on our simulation, we generate a set of hypotheses to specify how, why and when AI-related features and social configurations jointly shape collective opinion making.
[AI-232] LLM -Integrated Bayesian State Space Models for Multimodal Time-Series Forecasting
链接: https://arxiv.org/abs/2510.20952
作者: Sungjun Cho,Changho Shin,Suenggwan Jo,Xinya Yan,Shourjo Aditya Chaudhuri,Frederic Sala
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 15 pages, 8 figures
Abstract:Forecasting in the real world requires integrating structured time-series data with unstructured textual information, but existing methods are architecturally limited by fixed input/output horizons and are unable to model or quantify uncertainty. We address this challenge by introducing LLM-integrated Bayesian State space models (LBS), a novel probabilistic framework for multimodal temporal forecasting. At a high level, LBS consists of two components: (1) a state space model (SSM) backbone that captures the temporal dynamics of latent states from which both numerical and textual observations are generated and (2) a pretrained large language model (LLM) that is adapted to encode textual inputs for posterior state estimation and decode textual forecasts consistent with the latent trajectory. This design enables flexible lookback and forecast windows, principled uncertainty quantification, and improved temporal generalization thanks to the well-suited inductive bias of SSMs toward modeling dynamical systems. Experiments on the TextTimeCorpus benchmark demonstrate that LBS improves the previous state-of-the-art by 13.20% while providing human-readable summaries of each forecast. Our work is the first to unify LLMs and SSMs for joint numerical and textual prediction, offering a novel foundation for multimodal temporal reasoning.
[AI-233] From extreme weather to agricultural consequences with agent ic AI
链接: https://arxiv.org/abs/2610.40641
作者: Minghan Yu,Mengyu Dai,Yixin Wen,Haizhao Yang
类目: Atmospheric and Oceanic Physics (physics.ao-ph); Artificial Intelligence (cs.AI)
备注: 34 pages, 4 figures, 6 tables
Abstract:Extreme weather is forecast days ahead, yet its consequences for agriculture remain hard to predict: a forecast says where a storm will strike, not which crops will suffer or how much is at risk, and the records that could tell span many sources. We present WxCrop, the first agent that judges agricultural hazards from physically interpretable weather indices examined by causal analysis, and the first that combines weather and agricultural information to predict crop loss reliably for past and forecast dates. We demonstrate it in Florida on hurricanes, freezes and dry heat, with a loss model calibrated on insurance records for each, and the same procedure extends to other hazards and areas. Three checks show reliability: the hurricane index detects recorded cyclone days with an AUC of 0.98, recorded losses in held-out years are reproduced to within 2%, and independent production records confirm the county rankings (rank correlation 0.63). Users ask the agent for each crop group’s loss as a fraction of expected season value, every number sourced, so that governments, growers and scientists respond to agricultural disasters quickly.
[AI-234] Fast and accurate neural triage decoding of ultra-high-rate quantum LDPC codes DATE
链接: https://arxiv.org/abs/2610.40490
作者: Hong-Ye Hu,Zhenlai Wang,Binzhao Luo,Tao Xiang,Chen Zhao
类目: Quantum Physics (quant-ph); Artificial Intelligence (cs.AI); Information Theory (cs.IT); Machine Learning (cs.LG)
备注: 5 pages, 4 figures with Methods. We will release the fully supplementary information in the second update soon
Abstract:Quantum error correction is essential for fault-tolerant quantum computing. High-rate quantum low-density parity-check (qLDPC) codes have emerged as a promising approach to reducing the substantial physical qubit overhead of quantum error correction, enabling encoding rates exceeding 1/2 in blocks of approximately one thousand physical data qubits. However, their practical implementation requires decoders that combine high accuracy with low latency to keep pace with quantum hardware. Existing decoders of high-rate qLDPC codes can achieve high accuracy but often incur substantial computational latency, whereas neural network decoders have yet to be demonstrated for high-rate qLDPC codes. Here, we introduce \emphneural triage decoding, a framework that uses syndrome consistency to adaptively allocate neural computation. A local neural message-passing network rapidly predicts error corrections, additional iterations reduce residual syndrome violations, and a rarely invoked diversification stage searches for consistent corrections to the remaining shots. We demonstrate that the \textttTriage decoder can decode ultra-high-rate qLDPC codes encoding a few hundred logical qubits, including lifted-product and Kasai codes, achieving state-of-the-art accuracy comparable to the near optimal algorithmic decoders while reducing decoding latency by approximately one to two orders of magnitude. Our framework is inspired by a rigorous analysis of the exact bitwise marginal decoder, revealing a distance-independent syndrome-inconsistent error floor. In addition, we show the robustness of neural network decoders to mismatched error priors. Our results establish adaptive neural decoding as a promising route towards fast, accurate, and scalable quantum error correction in the near term.
[AI-235] Understanding axial attention in TSFMs through single location regression
链接: https://arxiv.org/abs/2610.40188
作者: Andrei-Tiberiu Pantea,Aymeric Dieuleveut
类目: Machine Learning (stat.ML); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Recent advances in Time Series Foundation Models (TSFMs) increasingly rely on axial attention mechanisms, but despite their empirical success, the theoretical properties of axial attention remain largely unexplored. To bridge this gap, we extend the Single Location Regression framework introduced in Marion et al. [2025] to 2D structured data, in order to study sparse information settings where a target depends on a small number of tokens. We formulate an analytically tractable axial attention-type predictor and rigorously analyze its statistical properties and training dynamics. We prove that under a specific asymptotic regime and with knowledge of the underlying oracle parameters, our predictor achieves Bayes optimality, whereas a class of semi-linear simplifications strictly fails to do so, underlining the role of the inner non-linearity. We prove that for a specific temperature scaling, Projected Gradient Descent converges globally to the optimal oracle parameters. Extensive numerical experiments validate our theoretical findings and highlight the critical role of inverse temperature scheduling in empirical convergence.
[AI-236] he Importance of Speech Genre for End-of-Turn Detector Training
链接: https://arxiv.org/abs/2610.39742
作者: Rini Sharon,Kadri Hacioglu,Jeena Prakash,Andreas Stolcke
类目: Audio and Speech Processing (eess.AS); Artificial Intelligence (cs.AI)
备注:
Abstract:End-of-turn (EoT) detection has been studied mostly through features, models, and modalities, while the speech genre of the training data has drawn far less attention. We isolate this factor. Using one reproducible annotation pipeline, we build a matched pair of corpora, multi-turn human-human conversational and single-turn human-machine, and train the same streaming ConvLSTM EoT detector on each. Across four test sets the conversational model reaches up to +0.45 higher F1, driven by fewer false alarms on the mid-turn pauses and hesitations that are rare in single-turn corpora. This conversational-training effect is not specific to frame-level decoding. Under a matched complete-or-not test it also appears at the utterance level, across two encoders, but only on conversational test data. Testing only on single-turn speech, as most prior work does, therefore hides the benefit of conversational training data. Against open-source baselines, our 1.8M-parameter model attains the highest F1 on every test set.
[AI-237] HERA: Assessing an Autonomous Agent for Quantum Device Measurement
链接: https://arxiv.org/abs/2610.39074
作者: Zachary W. Gomez,Nathan Bittner,Jordan T. McCourt,Qingrui Cao,Sandesh S. Kalantre,Benjamin H. Alexander,Zhiren Zheng,Kenji Watanabe,Takashi Taniguchi,Aaron Sharpe
类目: Mesoscale and Nanoscale Physics (cond-mat.mes-hall); Artificial Intelligence (cs.AI)
备注: 16 pages, 5 figures, plus Supplementary Information (7 figures, 6 tables); data and code at this https URL and this https URL
Abstract:Large language model agents have begun to carry out laboratory measurements, but their scientific conclusions are useful only if they hold up under scrutiny. Demonstrating this requires controlled studies because models are stochastic and the context a model is given steers its conclusions. Here we introduce HERA, an autonomous agent designed to carry out cryogenic transport measurements of quantum devices. HERA draws on a knowledge base curated by lab members that holds measurement protocols, software tools, and literature references. We assessed HERA with a blinded trial against an expert on the stacking order of two multilayer graphene devices, an ablation of its knowledge base, and hundreds of offline replays in which models designed the trial’s central measurement without running it. HERA’s conclusions matched the expert’s sealed conclusions on both tested devices. The ablation and the replay studies measured the impact of the knowledge base across different models, showing that with it every model designed better measurements and reasoned more soundly. To test how readily HERA extends to a new problem, we tasked it with finding and characterizing superconductivity in a different material system. This required a few hours of expert time to develop a new experimental protocol, after which HERA did both autonomously. These results show that, with a knowledge base designed for the task, language model agents are capable of sound experimental science.
[AI-238] Autonomous Quantum Measurement with Agent ic AI
链接: https://arxiv.org/abs/2610.38850
作者: Eunbi Rha,Abhijatmedhi Chotrattanapituk,Mouyang Cheng,Chung-tao Chou,Shenyong Ye,Ashley S. Dale,Long Ju,Luqiao Liu,Mingda Li
类目: Materials Science (cond-mat.mtrl-sci); Artificial Intelligence (cs.AI)
备注:
Abstract:Quantum materials exhibit diverse electronic and functional phenomena, but their characterization requires extensive experimental expertise, iterative decision-making and accumulated laboratory knowledge that is difficult to document and transfer. Here we introduce Quantum Agent (QUANTA), an agentic artificial intelligence (AI) framework for autonomous quantum-materials characterization. Using a widely adopted Physical Property Measurement System (PPMS) as an experimental platform, QUANTA combines retrieval augmentation using scientific literature, experimental protocols and prior measurements with specialized agents for experimental planning, troubleshooting, and data interpretation. A separate deterministic control layer validates instrument actions. Guided by incoming data, QUANTA evaluates measurement quality, revises experimental plans and selects subsequent measurements. We demonstrate these capabilities through transport measurements of the topological Weyl semimetal TaP, as well as investigations of nonlinear responses in magnetic material Co2MnGa. These results establish QUANTA as a generalizable framework for evidence-driven, safeguarded and closed-loop quantum-materials characterization.
[AI-239] Information-Theoretic Reliability–Safety Bounds and Finite-Alignment Dynamics in AI Knowledge Distillation
链接: https://arxiv.org/abs/2610.38651
作者: Miao-Miao Yi,Zhen-Yu Cai,Y -M. Du,C. P. Sun
类目: atistical Mechanics (cond-mat.stat-mech); Artificial Intelligence (cs.AI)
备注:
Abstract:The reliability and safety of artificial intelligence (AI), together with their transfer and evolution through knowledge distillation, have attracted growing attention. We establish an information-theoretic bound on the reliability–safety trade-off and characterize the dynamics of this trade-off under knowledge distillation. We first prove a universal trade-off inequality that is independent of the specific model form. Its boundary is determined solely by a single parameter Kh, which quantifies the hidden layer’s ability to distinguish benign from hazardous inputs. We then develop a canonical Boltzmann-machine model of AI response behavior and derive two exact relations between reliability and safety. The trade-off curve is determined by the hazard discrimination capacity K, while the overall answer tendency controls the operating point along the curve. These relations agree well with existing experimental data. In knowledge distillation, we find that teacher responses induce a renormalization of the student model’s trade-off parameters. We further use a renormalization-group approach to study multigenerational distillation, deriving exact recursion relations and the associated renormalization-group-like flows for two representative direct intergenerational distillation protocols at arbitrary teacher–student alignment strength. These flows reveal fixed-point structures, long-term dynamics, and critical behavior, yielding testable predictions for the critical transition between capability loss and stable transmission.
[AI-240] A Tutorial on Generative Models for Music Audio
链接: https://arxiv.org/abs/2610.38252
作者: Mateusz Modrzejewski,Joanna Kurkowska,Dorien Herremans
类目: Audio and Speech Processing (eess.AS); Artificial Intelligence (cs.AI); Emerging Technologies (cs.ET); Multimedia (cs.MM); Sound (cs.SD)
备注:
Abstract:Music audio generation spans instrumental clips, full songs with vocals, edits of existing recordings, and systems that play alongside a musician in real time. This monograph explains the methods behind these systems for readers familiar with neural-network training who are new to music generation. It introduces neural audio codecs and autoencoders, autoregressive and masked token models, diffusion and flow matching, sampling acceleration, and preference optimization in a shared notation. Worked examples connect training objectives and sampling procedures to published systems. The application sections relate the musical requirements of instrumental generation, songs, editing, and live performance to model design. Comparison tables record representations, methods, inputs, tasks, and release information documented in papers and official sources. Symbolic composition, speech, and sound effects are included where they connect to music audio. We explain how to interpret automatic metrics and learned judges, design and report listening tests, and assess evidence about data provenance, memorization, attribution, and output detection. Selected legal decisions and disputes illustrate the implications for training and release. The reviewed systems combine established generative paradigms with new representations, planning stages, and post-training procedures. Differences in data, scale, and evaluation limit what cross-system comparisons can establish about the sources of reported gains.
[AI-241] Interpretable Feature Discovery from Unstructured Data Using LLM s and Statistical Choice Models
链接: https://arxiv.org/abs/2610.38082
作者: Baptiste Arnaudo,Amine Bennouna,Sébastien Martin
类目: Machine Learning (stat.ML); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Preprint. 88 pages, 21 figures, 12 tables. Code: this https URL
Abstract:Feature extraction from unstructured data typically relies on opaque embeddings unsuited to interpretable decision-making. Interpretable features, on the other hand, typically require costly, and often infeasible, hand-engineering. Large language models (LLMs) offer a natural bridge, as they can read raw data and describe it in interpretable terms. One could simply ask an LLM to extract features from a large sample of the dataset, but context limitations and diluted attention can lead to unsatisfactory results. We introduce a method that automatically discovers interpretable, decision-relevant features from unstructured data. Our key idea is to perform pairwise comparisons between items, asking an LLM to articulate, in interpretable terms, the features distinguishing each pair. We treat these explanations, or rationales, as data: we develop a statistical choice model that combines these rationales across the comparisons to identify relevant features and estimate per-item, per-feature scores. We first test our method on a synthetic hiring dataset of cover letters with known ground-truth features and show that it significantly outperforms direct LLM feature extraction in feature recovery. We then apply it to the ChangeMyView (Reddit) debate dataset, where it discovers several interpretable features associated with persuasion while reaching predictive accuracy comparable to established baselines. We also show how the same rationales can be used to audit LLM decisions, yielding a holistic score whose decomposition over interpretable features reveals what the LLM rewards.
[AI-242] Where the Error Lives: Spectral Representations of Learned Kohn-Sham Hamiltonians
链接: https://arxiv.org/abs/2610.37843
作者: Fang Sun,Hekai Cheng,Ruoyan Li,Yuanzhou Chen,Wei Wang,Jason Cong,Yizhou Sun
类目: Chemical Physics (physics.chem-ph); Artificial Intelligence (cs.AI)
备注:
Abstract:Kohn-Sham density functional theory (DFT) is the standard method for computing the electronic structure of molecules and materials, but its self-consistent-field (SCF) iterations make it expensive for large systems and high-throughput screening. Machine-learned Hamiltonian models offer a shortcut: an equivariant network predicts the Kohn-Sham Hamiltonian matrix in one forward pass, and diagonalizing the prediction yields orbital energies, the HOMO-LUMO gap, molecular orbitals and the electron density. These models are trained to match the reference matrix entry by entry, yet a small entry-wise error does not guarantee accurate eigenvalues or eigenvectors. Two physics-informed alternatives go partway: a recently proposed wavefunction-alignment loss adds a spectral penalty without diagonalizing the prediction, and matching the electron density of the diagonalized prediction supervises it directly. Neither is complete: by representing the prediction error in the eigenbasis of the reference Hamiltonian, we show that each sees only part of the error, and that shifting the predicted orbital energies leaves the electron density exactly unchanged. The same analysis identifies a complete training signal: the cumulative local density of states (cumulative LDOS), the number of electrons at each point in space with energy below a given threshold. The cumulative LDOS is a single physical observable that contains the electron density as a special case, responds to every component of the error, determines the Hamiltonian uniquely, and yields gradients that do not vanish for large orbitalenergy errors. Fine-tuning three Hamiltonian networks on QH9 in this way, which we call FermiScope, reduces the HOMO-LUMO gap error without increasing the element-wise error. The gains carry over to out-of-distribution and molecular-dynamics splits, where the wavefunction-alignment loss helps less.
[AI-243] EUReCA: End-to-End Unified CBCT Reconstruction Across Clinical Acquisitions
链接: https://arxiv.org/abs/2610.37740
作者: Jiening Zhu,Chengzhu Zhang,Jason Fan,LiCheng Kuo,Weixing Cai,Xiuxiu He,Laura Cervino,Jean Moran,Xiang Li,Tianfang Li,Yabo Fu
类目: Image and Video Processing (eess.IV); Artificial Intelligence (cs.AI)
备注:
Abstract:Cone-beam computed tomography (CBCT) is widely used for image-guided radiotherapy, but reducing the number of projections to lower dose and scan time makes reconstruction increasingly ill-posed. We present EUReCA (End-to-end Unified REconstruction across Clinical Acquisitions), a geometry-aware feed-forward model for variable-view clinical linac CBCT. For a given acquisition trajectory, EUReCA accepts 8-100 projections with a single set of weights, without view-count-specific retraining; the same architecture and training recipe extend unchanged to other clinical trajectories. It combines learned projection features with an in-frame Feldkamp-Davis-Kress (FDK) reconstruction. The model encodes each projection with a shared multiscale backbone, back-projects features into multiresolution 3D grids using the acquisition geometry, aggregates variable numbers of views at fixed computational cost, and refines the volume with a global 3D transformer and multilevel decoder. On simulated data, EUReCA achieved 29.9-33.2 dB PSNR with 10-100 views and outperformed all applicable feed-forward baselines; it also exceeded iterative optimization-based methods while running orders of magnitude faster. On 15 measured patient CBCT scans evaluated against vendor reconstructions, EUReCA achieved the best performance at every tested view count. These results suggest that geometry-aware, measurement-anchored feed-forward reconstruction can provide a practical and flexible foundation for variable-view linac CBCT.
机器学习
[LG-0] Propose-Filter-Release: Scaling Noise to Typical Sensitivity via Private Filtering
链接: https://arxiv.org/abs/2610.40654
作者: Moshe Shenfeld,Eliad Tsfadia,Katrina Ligett
类目: Machine Learning (cs.LG)
*备注:
Abstract:An estimator can be accurate and stable on typical data, yet sensitive to a few possible records. Differentially private mechanisms commonly scale their noise to the worst-case influence of a single record, which lets these records dictate the noise even when the remaining data would allow far less. Removing them first would restore the smaller scale, but the removal must itself be private. We introduce \emphpropose-filter-release (PFR), which, by privately filtering its input, turns a mechanism that protects the addition or removal of a typical record, one that fits the remaining data, into one that is differentially private in the standard sense, on every input. Using our novel filtering step, PFR requires only a concave friendliness score, which measures how well a record fits a dataset, and a release mechanism that is friendly-DP, that is, private when the changed record fits the remaining data. For every such score and release mechanism, it guarantees differential privacy under public conditions on the number of records, so it is a general privacy primitive, comparable to propose-test-release, and FriendlyCore can be viewed as one of its implementations. We illustrate the generality of PFR by instantiating it for scalar statistics, principal subspace estimation, the KL divergence of user histograms, the effective resistances of a graph, and treatment-rule selection. Subjects: Machine Learning (cs.LG) Cite as: arXiv:2610.40654 [cs.LG] (or arXiv:2610.40654v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.40654 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-1] Adaptive Scenario Generation for KL-DRO Around a Learned Density Model
链接: https://arxiv.org/abs/2610.40644
作者: Aron Brenner,Saurabh Amin
类目: ystems and Control (eess.SY); Machine Learning (cs.LG)
*备注:
Abstract:Distributionally robust optimization (DRO) hedges data-driven two-stage stochastic programs against error in the distribution estimated from a finite record, and standard ambiguity sets differ in how they treat realizations absent from that record. A divergence ball around the empirical distribution contains only reweightings of the record, whereas a transport ball also contains distributions supported off the record, including on realizations the data-generating process cannot produce, against which the decision is then hedged as well. We instead propose a Kullback-Leibler (KL) ball centered at a density model fitted to the record, the learned nominal, which lets the adversary up-weight costly unobserved realizations in proportion to the probability the model assigns them. Solving the resulting problem poses a second difficulty, because it can accommodate only a few scenarios, and an independent sample of that size is likely to omit the rare realizations on which the worst case concentrates. We therefore build the scenario set adaptively, alternating between solving the finite problem and drawing new realizations from the learned nominal, steered toward the worst case at the current decision, with importance weights that correct the draws back to the model. On a multi-item newsvendor problem and a capacity expansion problem on realistic grid data, centering the ball at the learned nominal rather than at the record lowers cost under distribution shift, at comparable in-distribution cost, when the record is scarce and the model class captures structure that lets it extrapolate beyond the record. The advantage fades as the record grows or as the shift departs from the structure the model has learned.
[LG-2] world2vec: Mechanistic world models through latent dynamics learning
链接: https://arxiv.org/abs/2610.40637
作者: Tobias Schmidt,Rodrigo Gonzalez Laiz,Krisztián Szuppinger,Steffen Schneider
类目: Machine Learning (cs.LG)
*备注: this https URL
Abstract:World models can achieve strong control performance without faithfully capturing the dynamics of their environment, limiting their potential for mechanistic interpretability and scientific discovery. Here, we introduce world2vec to learn mechanistic world models without reward information directly from observations and actions. World2vec learns latent dynamics that faithfully reproduce the true system over long prediction horizons, with at least one order of magnitude fewer parameters than DreamerV3, TD-MPC2 and LeWM. This enables state of the art performance in controlling continuous environments from pixels and accurate system identification across a benchmark of 71 attractors, demonstrating how world models are amenable to scientific discovery.
[LG-3] An ODE-based Sampler for Discrete Diffusion Models
链接: https://arxiv.org/abs/2610.40618
作者: Samson Gourevitch,Yazid Janati,Eric Moulines,Eric P. Xing,Alain Durmus,Giacomo Zanella
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:The marginals of a continuous diffusion model can be reproduced by a deterministic ordinary differential equation (ODE), known as the probability flow ODE, which provides a foundation for accelerated sampling and consistency distillation. In this paper we address the question whether an analogous formulation exists for discrete diffusion models and answer it affirmatively by introducing exponential residual flows (ERFs). Our framework augments the discrete state with a continuous residual variable for each token and position. These residuals evolve deterministically according to piecewise-defined ODEs whose vector fields depend on the current discrete state and the forward and backward rate matrices of the prescribed diffusion model. The discrete state changes whenever a residual reaches zero. We prove that this construction preserves the marginals of the original diffusion. We derive a practical discretization of our sampler and specialize it to masked and uniform diffusion. The deterministic trajectories of ERFs support consistency distillation and yield empirical improvements over previous approaches across a variety of benchmarks including code and text generation tasks.
[LG-4] raining Foundation Models with a Single Algorithm
链接: https://arxiv.org/abs/2610.40564
作者: Yuda Song,Alvin Zhang,Fahim Tajwar,Daman Arora,Jeff Schneider,Daniel Khashabi,Ruslan Salakhutdinov,J. Andrew Bagnell,Aarti Singh,Andrea Zanette
类目: Machine Learning (cs.LG)
*备注: 67 pages, 31 figures, 22 tables
Abstract:Foundation models are typically trained in stages: pretraining, midtraining, supervised fine-tuning (SFT), and reinforcement learning (RL), with changes in both data and learning algorithms. We show that this algorithmic separation is not fundamental. By deriving suitable RL objectives as maximum-likelihood fitting to model-generated target distributions, we place all of these stages under a single likelihood-based training algorithm, unified training. This perspective turns the foundation-model training lifecycle into a problem of choosing which data to train on, in what proportions, and when, with conventional staged training as a special case. We instantiate this framework in a unified training system that mixes static data and RL generations within a single trainer. Across controlled and real-world experiments, unified training improves the accuracy-compute tradeoff and task coverage while preserving broader capabilities, and provides a stronger starting point for continual learning. This work lays the groundwork for a unified science of foundation-model training, connecting mathematical principles, training infrastructure, and the development of model capabilities.
[LG-5] Machine Learning-Based Detection of Mobile Money Scan Messages in Ghana: A Comparative Evaluation of Naive Bayes Support Vector Machine and Randon Forest Classifiers
链接: https://arxiv.org/abs/2610.40544
作者: Ibrahim Salifu,Muhaisin Tiyumba Nantogmah
类目: Machine Learning (cs.LG)
*备注: 17 Pages, 8 Figures
Abstract:Mobile money (MoMo) is Ghana’s preferred retail payment channel, recording GHC518.4 billion in transactions in December 2025 and over 26.7 million active accounts (Graphic Online, 2026). This growth has coincided with increased social-engineering fraud over Short Message Service (SMS) and Unstructured Supplementary Service Data (USSD), where fraudsters impersonate network operators, agents, relatives, or regulators to obtain personal identification numbers (PINs), identity data, or funds. We present a supervised machine-learning pipeline for classifying scam and legitimate Ghanaian MoMo SMS messages using a labelled corpus of 220 messages across 18 scam and legitimate subcategories. Domain-specific normalisation of currency amounts, phone numbers, and USSD codes preceded TF-IDF vectorisation of unigrams and bigrams. Using stratified five-fold cross-validation and a 20% held-out test set, we trained and evaluated Multinomial Naive Bayes, linear support vector machine (SVM), and Random Forest classifiers using accuracy, precision, recall, F1-score, Matthews correlation coefficient (MCC), and area under the receiver operating characteristic curve (AUC-ROC). Held-out accuracy, F1, and AUC were 1.00 for Naive Bayes and linear SVM; cross-validated accuracy averaged 96.6% (+/-3.3), 94.9% (+/-4.6), and 92.0% (+/-4.2). Discriminative n-grams reflected imperative urgency (“send,” “claim,” “please send”) and PIN solicitation (“PIN,” “pincode”); legitimate messages featured confirmatory and cautionary language (“balance,” “successful,” “do not share”). Findings are contextualised in Sub-Saharan African SMS fraud detection, small template-based corpus ceiling effects, and a pathway toward larger naturalistic multilingual corpora. Interpretable classical classifiers achieve high accuracy on structured MoMo scam text and suit carrier- or handset-level filtering in low-resource, on-device settings.
[LG-6] RACER a time-to-event EHR foundation model for disease detection across health systems
链接: https://arxiv.org/abs/2610.40536
作者: Mara Pleasure,Ekaterina Redekop,Vedrana Ivezic,Rushikesh Kulkarni,Aaron Chin,William Speier,Corey W. Arnold
类目: Computational Engineering, Finance, and Science (cs.CE); Machine Learning (cs.LG)
*备注:
Abstract:Foundation models trained on longitudinal electronic health records (EHRs) can predict clinical diagnoses years in the future, making them a potential tool for disease screening. We developed TRACER (Temporal Risk Assessment with Clinical EHR Representations), a decoder-based foundation model pretrained on data from over 4.5 million patients from four health systems, and evaluated it on 16 incident diseases at 1-, 2-, and 5-year horizons in more than 2 million patients from two health systems excluded from model development. In contrast to typical evaluations that focus on global metrics (e.g., AUROC), we measured performance at the low false-positive rates that are realistic for screening. We found that TRACER retained its discrimination in external health systems, and at 95% specificity, the patients it flagged were up to eight times more likely to be diagnosed within a year than unselected patients. Evaluating at this operating point also revealed what global metrics alone did not: detection and false-alert burden varied across diseases and health systems, even when discrimination appeared stable. Our results illustrate the potential of foundation models for screening and underscore the need for evaluation in health systems excluded from development to assess disease detection and false-alert burden at levels appropriate for intended clinical use.
[LG-7] NEOPIC: A Neural Operator Framework for Particle-based Kinetic Plasma Simulations KR
链接: https://arxiv.org/abs/2610.40528
作者: Sriramkrishnan Muralikrishnan,Chelsea Maria John,Jeffrey Kelling,Michael Bussmann,Robert Speck,Paul Gibbon
类目: Numerical Analysis (math.NA); Machine Learning (cs.LG); Computational Physics (physics.comp-ph); Plasma Physics (physics.plasm-ph)
*备注: 84 pages, 19 figures, 3 tables. Code: this https URL (tags neopic-arxiv-v1, neopic-arxiv-v1-nufft); analysis scripts: this https URL
Abstract:We introduce NEOPIC, a framework for learning the field operator in particle-based methods applied to kinetic plasma simulations. Its key novelty is a neural operator that acts directly on the empirical measure of the particles, rather than on a grid, and is evaluated at the particle positions. No mesh or tree is needed and, unlike grid-based neural operators, no deposition onto a mesh takes place, so the particle noise that dominates the inputs of particle simulations is not aliased. Since it replaces only the field solve, it serves as a plug-and-play field solver for particle-based methods: it can be trained on data from particle-in-cell, particle-in-Fourier or tree codes, and combined with any time integrator, including implicit schemes. We prove that the discrete evaluation coincides with the continuous operator at any number of particles (representation equivalence) and derive an error analysis that separates sampling, learning and time-stepping errors: up to the learning error, NEOPIC has the error structure of standard Monte Carlo particle methods. We show that the particle noise in the training data acts as an implicit regularizer, which gives guidance on the number of training particles in relation to the particle numbers used at inference. A complexity analysis shows that NEOPIC is an O(N_p) method whose constants are set by the architecture alone, independently of any mesh or tree. Numerical experiments on electrostatic benchmarks in two and three dimensions, including Landau damping, two-stream and bump-on-tail instabilities, a cyclotron mini-app and a Penning trap, confirm the theoretical findings. A single model trained on several mini-apps generated by different reference methods captures all of them and remains stable up to twice the training time window, which points towards foundation models for fields in particle-based methods.
[LG-8] Numerical Integration on Bounded Domains via Learned Reflected Diffusions NEURIPS2026
链接: https://arxiv.org/abs/2610.40499
作者: Kirill Korolev,Artur Goldman,Timofei Gritsaev,Tigran Ramazyan,Nikita Morozov
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注: NeurIPS 2026 STODY Workshop (Oral)
Abstract:We study high-dimensional integration on bounded convex domains. Given a non-negative integrand r , the goal is to estimate Z=\int_Dr(x),\mathrm dx on D\subset\mathbb R^d . Standard Quasi-Monte Carlo does not adapt its evaluation points to the integrand, while effective importance sampling requires a proposal that captures its concentration. Learned diffusion samplers can supply such a proposal, but existing methods are formulated on \mathbb R^d and do not respect the domain constraint x\in D . We instead build the proposal from reflected dynamics that stay in D by construction. A reflected diffusion progressively transforms the target distribution toward the uniform distribution on D , and a neural score model is trained to reverse this process. A change-of-measure formula for reflected diffusions yields a training objective without additional boundary terms and an unbiased importance-sampling estimator of Z . For non-smooth integrands, where reparameterization gradients may be undefined or uninformative, we derive a REINFORCE estimator of the training gradient using Gaussian proposal densities before reflection, without evaluating reflected transition densities or differentiating r . We evaluate on numerous synthetic targets and real-world problems, demonstrating strong performance of the proposed method, as well as favorable scaling with the dimension of the problem and the number of points used for the Monte Carlo estimate.
[LG-9] he Precision Wall: An Optimizer-Dependent Boundary for Optimization under Power-Law Measurement Spectra
链接: https://arxiv.org/abs/2610.40485
作者: Lngel Kyo(A2O Labs)
类目: Machine Learning (cs.LG)
*备注: 19 pages, 3 figures, 4 tables
Abstract:In low-rank matrix sensing of a rank- D factor U^\ast \in \mathbbR^V \times D , replacing isotropic Gaussian designs with power-law measurement covariance \Sigma = \mathrmdiag(j^-s)j=1^V leaves the Burer–Monteiro landscape free of spurious local minima, yet division-normalised first-order optimizers stall above a critical exponent s^\ast . We trace this failure to a numerical precision wall. At the ground truth, the smallest non-gauge expected Hessian eigenvalue scales at leading order as V^-(s+1)D^-s . When this curvature drops below the effective denominator floor \epsilon_O of optimizer O , normalisation noise overwhelms rare-direction gradients, yielding the leading-order boundary s^\ast_O(V, D) = [\log(1/\epsilon_O) - \log V]/\log(VD) + \delta_O , where \delta_O is a metric-dependent detection lag. With \delta\mathrmAdam calibrated on four configurations and frozen, this formula predicts Adam’s onset within 0.45 across all 16 cells of a held-out grid ( V \in \64,128,256,512\ \times D \in \4,8,16,32\ ), separates fixed- VD pairs, and distinguishes denominator-precision walls from conditioning and sign-quantisation walls. The approximation fails at D = 2 , reflects a finite- D spectral crossover, and saturates below \varepsilon_0 = 10^-8 at a dynamical floor.
[LG-10] Nonlinear Dynamical Systems as Scalable and Efficient Physical Neural Networks
链接: https://arxiv.org/abs/2610.40471
作者: Tyler King,Magnus Saebo,Jeremie Laydevant,Haoting Zhen,Maxwell G. Anderson,Tatsuhiro Onodera,Peter L. McMahon
类目: Machine Learning (cs.LG); Disordered Systems and Neural Networks (cond-mat.dis-nn); Emerging Technologies (cs.ET); Neural and Evolutionary Computing (cs.NE)
*备注: 32 pages, 10 figures
Abstract:The rapid growth in the size of state-of-the-art machine-learning (ML) models has led to increasing compute demands, spurring the development of energy-efficient hardware accelerators. Physical systems trained to perform ML inference, known as physical neural networks, have been proposed as accelerators in part because of their potential to be highly energy efficient. However, the ML inference accuracies reported in most investigations of physical neural networks have been substantially below the state of the art for conventional digital neural networks. In this work, we show in simulation studies that by applying modern ML training practices, including model compression, various candidate physical neural networks based on nonlinear dynamical systems can achieve accuracies as high as 99.2% on MNIST handwritten-digit recognition. We also show for one dynamical system that using it as the core building block of a parameter-efficient ML architecture inspired by MLP-Mixers enables accuracies of up to 86.4% on CIFAR-10 image classification (92.2% with pretraining and knowledge distillation), compared with 64% for the same system without the mixer architecture. The mixer approach achieved its accuracy while using 5\times fewer trainable parameters than the baseline. Finally, we present back-of-the-envelope estimates for inference energy efficiency and speed when using an example physical dynamical system, suggesting that orders of magnitude improvements may be possible over digital-electronic neural networks.
[LG-11] Rural Readiness: Domain Generalization with Adversarially Chosen Domains
链接: https://arxiv.org/abs/2610.40468
作者: Cynthia Dwork,Andisheh Ghasemi,Lunjia Hu
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:Predictive models trained at well-resourced data-rich institutions often fail when deployed in new populations. Motivated by the goal of building clinical diagnosis and decision-making algorithms for data-scarce rural populations using the general knowledge learned from data-rich hospitals, we formulate the fundamental question of \emphdomain generalization with adversarially chosen domains, which asks a learner to perform on any unseen target domain nearly as well as the worst hypothesis consistent with the evidence from the observed source domains. Assuming only that the hypothesis class contains a “universal” predictor with small error on all k source domains, we characterize the sample complexity of this objective up to logarithmic factors. Letting d denote the VC dimension of the hypothesis class, we find, somewhat surprisingly, that the sample complexity reduces significantly from \Omega(dk) to O(d) under a small relaxation of the learning goal in which, roughly speaking, we allow the learned model to compete with hypotheses that perform well on 99% instead of 100% of the source domains. By applying this insight to the meta-distributional setting, in which source domains are selected according to a distribution on distributions, we obtain a corresponding improvement over known sample complexity upper bounds. We also study the effect of data access to the target domain and show that a small labeled sample from the target domain does not improve the worst-case sample complexity on the source domains. Subjects: Machine Learning (cs.LG); Machine Learning (stat.ML) Cite as: arXiv:2610.40468 [cs.LG] (or arXiv:2610.40468v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.40468 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-12] Reinforcement learning for control of partial differential equations
链接: https://arxiv.org/abs/2610.40449
作者: Nicolò Botteghi,Matteo Tomasetto,Andrea Manzoni
类目: Machine Learning (cs.LG)
*备注:
Abstract:Controlling dynamical systems governed by partial differential equations (PDEs) remains a challenging task in scientific computing, owing to high-dimensional state spaces, distributed actuation, nonlinear dynamics, and stringent real-time constraints. Reinforcement learning (RL) has emerged as a promising paradigm for synthesizing feedback controllers that adapt to varying operating conditions and can be deployed efficiently in real-time. This review provides a unified perspective on RL and PDE-constrained optimization, and surveys model-free, model-based, and multi-agent RL approaches to PDE control, with particular emphasis on sample efficiency, generalization across parametric variations, and scalability to high-dimensional problems. Two demonstrative examples, the stabilization of a chaotic system and a flow control problem, illustrate the performance and pitfalls of RL agents in practice. Finally, we discuss evaluation and benchmarking practices, current limitations, and open research directions toward reliable, data-efficient, and scalable RL for scientific computing.
[LG-13] Framework-Conditional Open Materials Generation
链接: https://arxiv.org/abs/2610.40443
作者: Maya M. Martirossyan,Hillary Pan,Philipp Höllmer,Stefano Martiniani
类目: Machine Learning (cs.LG); Materials Science (cond-mat.mtrl-sci)
*备注:
Abstract:Theoretical predictions of novel materials rely on a variety of techniques, from data-mining-based ion substitution methods to fully de novo generative models. Each approach imposes specific constraints and encounters unique challenges when extrapolating beyond the distribution of known materials to discover new ones. The foremost goal of materials generation is to produce stable, unique, and novel (S.U.N.) structures. Predicted structures that meet these criteria, however, typically result from novel chemical compositions mapped onto known structural prototypes or frameworks. We leverage this heuristic insight to present a new generative paradigm for the de novo prediction of inorganic crystalline materials constrained by their structural framework—i.e., the unit cell shapes and respective particle positions. Our method, termed OMatG-FC (Open Materials Framework-Conditional Generation), learns via a combination of discrete flow matching on the atomic species and stochastic interpolants for the volume scaling of the unit cell lattice, while fixing in-place the fractional coordinates and the unit-volume lattice vectors, dramatically reducing the dimensionality of the problem compared to unconstrained (framework-free) generation. Our method is more flexible than existing prototype-constrained methods, demonstrating capability for discovery of unseen, stable prototypes derived from known structural frameworks. We also investigate the number of unique frameworks in the MP-20 dataset, demonstrating the propensity of duplicate frameworks within the dataset and providing a framework-split version. We benchmark the performance of our model using the LeMat-GenBench suite using three ML interatomic potentials and achieve state-of-the-art S.U.N. rate below the convex hull, eschewing the need to learn atomic packings for materials discovery.
[LG-14] A Pretrained Model for Bayesian Filtering in Heterogeneous Dynamical Systems
链接: https://arxiv.org/abs/2610.40437
作者: Ke Fan,Wen Wu,Haitao Su,Yilin Wang,Yuan Zhou,Chao Zhang
类目: Machine Learning (cs.LG)
*备注:
Abstract:Filtering has not yet had its foundation-model moment: state estimation is still dominated either by online Bayesian filtering algorithms that require known dy- namics as inputs, or by learning-based neural filters trained from scratch for a single system or parametric family. The core challenge is structural heterogene- ity: dynamical systems differ in their governing dynamics, physical scales, and noise levels, precluding a shared representation space for a reusable filter. To bridge this gap, we introduce FilterFM, the first pretrained model for Bayesian filtering that estimates latent states across heterogeneous dynamical systems with a single checkpoint. FilterFM employs a hierarchical two-level observation se- quence encoder and a prior-to-posterior state-reconstruction module that together handle varying observation and state dimensions across heterogeneous systems, while modeling state uncertainty via quantile prediction. Trained on a mixture of heterogeneous filtering tasks, FilterFM exhibits superior in-domain performance, delivering significant gains in estimation accuracy, runtime efficiency, and noise robustness compared to existing baselines. Beyond in-domain performance, it marks the first model to successfully enable zero-shot filtering on unseen dynamics, bridging the generalization gap in state estimation. Furthermore, we uncover clear scaling behaviors in foundation-model-based Bayesian filtering, which combined with comprehensive component ablation studies, validate the structural optimality of our design and broaden the generalization scope for this research field.
[LG-15] Adaptive search with path-constrained mobile sensors using multi-arm bandits
链接: https://arxiv.org/abs/2610.40431
作者: Abraham P. Vinod,Parth K. Thaker,Gautam Dasarathy,Stefano Di Cairano
类目: Robotics (cs.RO); Machine Learning (cs.LG); Optimization and Control (math.OC)
*备注: 24 pages, 5 figures
Abstract:We study a constrained sequential decision problem for detecting regions of interest from noisy online measurements collected by a team of energy-constrained mobile sensors. The sensor motion is restricted to a precomputed set of feasible paths that satisfy mobility constraints, limited available energy by requiring return to charging stations, and avoidance of known obstacles. We cast the resulting active path selection as a multi-armed bandit problem and propose an iterative search algorithm that adaptively assigns paths using confidence bounds computed from the collected data. We also propose a graph-based method to construct low-cardinality path covers under the team’s constraints. We establish finite-time and anytime probabilistic correctness guarantees for the proposed algorithm, derive non-asymptotic upper bounds on the times required to detect all regions of interest and to label all nodes, and show that these bounds match algorithm-independent lower bounds up to constant factors for a broad class of problem instances. Using numerical simulations, we illustrate the behavior of the proposed approach and the effect of path redundancy and team size on search performance.
[LG-16] Composing and Steering Diffusion Models with Generalized Feynman-Kac Correctors
链接: https://arxiv.org/abs/2610.40383
作者: Wael Mattar,Daniel Freedman
类目: Machine Learning (cs.LG)
*备注:
Abstract:Diffusion models have become the leading paradigm in generative modeling, yet many applications require sampling from distributions that are not those of any single pretrained model, but are instead defined through combinations of several models, possibly tilted toward high values of a downstream reward. Naively combining the scores of the models yields incorrect samples, while existing corrections are restricted to specific families of targets. In this work, we introduce the generalized Feynman-Kac correctors (GFKC), a principled framework for sampling from an arbitrary target function of several pretrained diffusion models. We show that the geometric interpolation between an arbitrary initial function and the target function of the models satisfies a Feynman–Kac partial differential equation, and derive its potential in closed form. To evaluate this potential during generation, we further generalize the log-probability ODE to track the density of each model along trajectories driven by a different velocity field. The resulting algorithm operates entirely in the log domain, incorporates reward-based steering at no additional conceptual cost, and corrects the generation on the fly, without retraining or fine-tuning the models. We validate the method on analytic Gaussian models, where it closely matches the ground-truth targets, and on diffusion models trained on MNIST, where it samples from non-trivial combinations of the models and enables effective steering.
[LG-17] Is It the Curriculum or Just the Easy Start? Benchmarking Unsupervised Environment Design on Quadrupeds
链接: https://arxiv.org/abs/2610.40364
作者: Jakub Gilewicz,Dominik Galus,Piotr Wojdas
类目: Machine Learning (cs.LG); Robotics (cs.RO)
*备注: 22 pages, 10 figures. Code, level space and evaluation suite: this https URL
Abstract:Unsupervised environment design (UED) methods such as PLR, ACCEL, SFL and PAIRED are compared mostly on 2D mazes, grid worlds and BipedalWalker, while terrain curricula for legged robots are hand-designed and evaluated against a few baselines. We present a benchmark that runs fourteen UED, domain-randomisation and robotics curricula under one protocol on five 3D quadrupeds in MuJoCo MJX: the closed-chain W01-TEK and the open-chain Unitree Go1 and Go2, Google Barkour vB and Boston Dynamics Spot. Levels are vectors of eight interpretable terrain parameters, and all curricula drive the same PPO student at matched environment-step budgets. The evaluation reports success on random and held-out structured terrains, interquartile means with stratified bootstrap intervals, capability envelopes and a level-by-level domination comparison. We compare methods within each robot and report, per level set, the top group of methods not significantly below the best. No method is in the top group on every robot and level set. Methods with easy or low-difficulty training levels, curated or not, fill the top groups on in-distribution random levels of Go1 and Barkour, and of W01-TEK apart from PAIRED+BiBC. The success-driven terrain curriculum from legged-robot practice is in a top group most often, domain randomisation on uniform levels in none, and PAIRED in none outside Spot, where the fourteen methods lie within 3.8 points on random levels of difficulty 0.25. On three robots the leader on in-distribution random levels is not the leader on held-out terrain. We collect the results into a per-robot guide to choosing a curriculum.
[LG-18] Your LR Schedule May Not Be Principled: Effective Learning Rate Scheduling in Weight-Normalized Training
链接: https://arxiv.org/abs/2610.40362
作者: Qifan Yu,Xinyu Ma,Xingyan Bin,Di He
类目: Machine Learning (cs.LG)
*备注:
Abstract:Weight-normalized (WN) training is emerging as a compelling alternative to conventional weight decay (WD), with recent studies reporting substantial pretraining loss gains. We show that these gains are driven primarily by the induced effective learning rate (ELR) trajectory rather than by normalization alone. Matching ELR trajectories largely collapses the training loss dynamics of WN and WD. Schedule design should therefore target ELR rather than the nominal learning rate (LR) alone. Building on this principle, we propose AIR (Affine Inverse-square-Root decay), which combines a theoretically derived 1/\sqrtt decay with an affine adjustment to satisfy prescribed peak and terminal LRs over a finite training horizon. We further identify late-stage ELR scheduling as a key source of its gains, because a shared late schedule attenuates the effect of earlier LR choices. Experiments on 128 B-token Mixture-of-Experts pretraining demonstrate that AIR outperforms cosine and other candidate schedules in training loss, downstream performance, and hyperparameter robustness, reducing terminal training loss by approximately 0.01 relative to cosine at matched cumulative LR. Applying the same ELR trajectory to WD reproduces the corresponding training dynamics, showing that the benefits extend beyond fixed-norm training and establishing ELR as the principled scheduling object.
[LG-19] Rimuru: Unified Memory Management for Modern LLM Serving via Elastic Expert Offloading
链接: https://arxiv.org/abs/2610.40305
作者: Zhiyuan Fang,Dingyang Li,Jianfeng Zhong,Bin Wang,Weixi Zhu,Cunshu Ni,Zicong Hong,Wuhui Chen,Zibin Zheng
类目: Operating Systems (cs.OS); Machine Learning (cs.LG)
*备注:
Abstract:Modern LLMs strain device memory through large model weights and attention states, particularly in resource-constrained scenarios. Existing systems separately provision memory for model weights and attention states, imposing fixed fixed boundaries that cannot adapt to the changing memory demands of workloads. Under heavy workloads, insufficient state capacity limits request concurrency and reduces throughput; under light workloads, idle memory in one region cannot be reallocated to another, wasting device memory. Meanwhile, modern LLM architectures introduce heterogeneous tensors with diverse layouts and lifetimes, further complicating memory management. These problems motivate unified memory management for heterogeneous tensors. However, a shared pool alone does not make memory directly reusable across tensors with different allocation granularities and execution layouts. This paper presents Rimuru, a modern LLM serving system that unifies memory management through elastic expert offloading. Rimuru establishes a shared page pool to enables elastic memory sharing across heterogeneous tensors. To accommodate different allocation granularities, Rimuru introduces memory views that allow each tensor type to interpret the shared memory space using its own page size and page table. Furthermore, Rimuru combines huge-page alignment with a multi-level page allocator that splits and merges pages, enabling cross-type capacity reuse. A page-table-aware fused MoE kernel directly accesses non-contiguous expert weights, avoiding copy overhead. Building on these mechanisms, Rimuru employs a workload-aware adjuster that dynamically reallocates pages across tensor types according to runtime demands. Implemented on top of vLLM, Rimuru remains compatible with modern LLMs. Evaluations show that Rimuru achieves up to 19.03x baseline throughput and reduces TTFT by 98.1% on average.
[LG-20] Causal-prior-constrained deep learning for rapid-intensification warning of western North Pacific tropical cyclones
链接: https://arxiv.org/abs/2610.40304
作者: Xinrui Li,Yijun Jin,Yifan Dong,Shoujuan Shu
类目: Machine Learning (cs.LG); Atmospheric and Oceanic Physics (physics.ao-ph)
*备注: 33 pages, including 15 pages of supplementary information; 5 main figures and 6 supplementary figures
Abstract:Rapid intensification (RI) is one of the most dangerous and unpredictable phases of tropical cyclones, and its environmental and structural drivers are difficult to identify and separate in observations. Here we introduce CausalVisionRI, a causal-prior-constrained deep-learning system for 24-h RI warning in the western North Pacific. It integrates infrared remote sensing imagery, atmospheric and oceanic fields, and an observationally discovered causal graph into a unified model. The graph and eight sign-verified effects on 24-h intensity change provide an auditable causal prior for training. Discovered from 522 cyclones in 2000-2017, it identifies causal pathways linking atmospheric and oceanic conditions, the cyclone’s dynamical and structural state, and its 24-h intensity change. On a forward holdout covering 2020-2024 and 3,004 records, the five-member ensemble achieves a Brier skill score (BSS) of 0.259 relative to a fixed development climatology, significantly exceeding four reference systems; the three LDA baselines reach at most 0.105. It also attains higher detection rates than a published knowledge-block scheme and a reconstructed contrastive-learning scheme at matched false-detection rates under our USA 30-kt standard and the published RITCF standard. Ablations show that the graph pathway adds 0.032 BSS across the holdout and 0.063 over 2020-2021, with a significant contribution in one ensemble member. A deep-learning model constrained by observationally discovered causal structure thus offers a new route to physically informed RI warning.
[LG-21] DolphinRouting: Cluster-Level Binary Routing for Disaggregated LLM Inference Serving
链接: https://arxiv.org/abs/2610.40292
作者: Yishuo Yuan,Shaoyu Yang,Chunrong Fang
类目: Machine Learning (cs.LG)
*备注:
Abstract:Prefill-decode (PD) disaggregation isolates the two phases of large language model serving, but request-level path selection must account for prefill work, cache reuse, and decode demand. We present DolphinRouting, which learns this choice from end-to-end SLO feedback under a given P/D resource allocation. Requests are clustered by raw prompt length, cache-adjusted input lengths on both worker pools, and output-length information. Offline policy search learns one binary action per cluster. Online feature extraction, cluster assignment, and table lookup select local or remote prefill. We evaluate DolphinRouting on a real PD-disaggregated cluster using three public benchmarks, a mixed workload, and real data. Across the mixed workload and real data, DolphinRouting with k -means reduces P90 TTFT by 5.7–7.9% and P90 TPOT by 2.6–10.1% relative to Dynamo.
[LG-22] From MS/MS evidence to adaptive spectroscopic characterization: Reinforcement learning guided by candidate molecular structures
链接: https://arxiv.org/abs/2610.40264
作者: Bowen Gao,Lei Zhu,Yiying Wang,Wenjie Yu
类目: Machine Learning (cs.LG)
*备注: 17 pages, 5 figures, 2 tables. Supporting information: 10 pages, 3 figures and 7 tables, provided as an ancillary PDF
Abstract:Small-molecule identification from tandem mass spectrometry (MS/MS) may require complementary measurements. Herein, we propose a structure-informed deep Q-network (SI-DQN) that uses existing MS/MS evidence to decide whether further characterization is needed, which measurement to request, and when to stop. The policy combines spectral representations and candidate rankings with score-weighted descriptors of the candidates’ chemical features and structural differences. The policy was evaluated on 79,462 held-out samples with simulated MS/MS, infrared (IR), and nuclear magnetic resonance (NMR) spectra from the Multimodal Spectroscopic Dataset. Evaluation used fixed hard candidate sets containing the correct structure and equal normalized acquisition costs. Averaged across three training runs, SI-DQN increased Top-1 accuracy from 80.10% to 93.78% for positive-ion MS/MS starts and from 70.44% to 93.31% for negative-ion starts. The corresponding mean additional measurement counts were only 0.227 and 0.340 per sample, respectively. Complete acquisition achieved 97.72% accuracy with four additions, whereas SI-DQN reduced the additional measurement count by 94.3% and 91.5%. Relative to a rule-based policy, SI-DQN reduced the composite loss combining terminal ranking error and acquisition cost by 19.3% and 25.5%. Adding structural information to DQN further improved accuracy while reducing requests across all three training runs. At comparable ranking uncertainty, SI-DQN increased request rates relative to DQN more strongly in groups with high structural disagreement, where larger reductions in composite loss were also observed. Together, these results support using the ambiguity among candidate structures after MS/MS analysis to guide further characterization, providing a quantitative basis for balancing identification performance against subsequent measurement effort.
[LG-23] LLM -guided Planar Loco-Manipulation for Quadruped Robots
链接: https://arxiv.org/abs/2610.40248
作者: Georges Jetti,Michael Khayyat,Jean-Pierre Sleiman,Stefano Arrigoni,Francesco Braghin
类目: Robotics (cs.RO); Machine Learning (cs.LG)
*备注: 20 pages, 9 figures, 15 tables. Includes appendix
Abstract:Robotic navigation typically assumes a collision-free route to the goal exists; in cluttered environments with movable obstacles, however, reaching the goal may require physical interaction. Classical solutions encode these behaviors through geometric rules and task-specific logic, while learning-based approaches often constrain the decision space to low-dimensional motion commands or predefined action sets. We instead consider a broader decision space, jointly selecting the skill, relevant object, and geometric target. Our language-conditioned task planner, fine-tuned with LoRA on synthetic data, maps natural-language tasks and a structured scene representation to variable-length sequences of parameterized actions, executed by low-level reinforcement-learning policies. The planner learns the geometric relations underlying interactive navigation and outperforms a classical planner and a prompted frontier model on interactive navigation tasks. Comparisons with models trained from scratch indicate that composition does not require pretraining when the representation is structured, whereas pretraining yields more reliable geometric targets. At the low level, we introduce an end-to-end whole-body planar pushing policy that achieves a 93.3% success rate in simulation, versus 83.5% for a hierarchical baseline, while reducing completion time by more than 60%. On hardware, the hierarchy yields the correct skill composition in 44/45 trials and end-to-end success in 34/45. Project website: this https URL
[LG-24] DYNAS: A Dynamic Network Penetration Testing Framework Using Deep Hierarchical Reinforcement Learning
链接: https://arxiv.org/abs/2610.40214
作者: Md. Rakibul Hassan,Majedul Islam,Md. Ashraful Alam,Muhammad Iqbal Hossain,Swakkhar Shatabda
类目: Cryptography and Security (cs.CR); Machine Learning (cs.LG); Networking and Internet Architecture (cs.NI)
*备注: Accepted at the NSS 2026: 20th International Conference on Network and System Security (submission version)
Abstract:As network scale and uncertainty complicate automated penetration testing, reinforcement learning (RL) environments must support controlled evaluation under defensive and non-stationary conditions. We present DYNAS (Dynamic Network Attack Simulator), which combines parameterized large-scale scenario generation with stateful intrusion detection responses, explicit scan false-positive and false-negative noise, temporary service unavailability, and duration-based action timeouts. We further introduce Pen-DHRL, a recurrent manager–worker policy that selects persistent subgoals using IDS-conditioned global context. We compare Pen-DHRL with GNN, Invariant, and Attention baselines across four scenario variants and multiple training regimes. Under IDS-disabled training followed by IDS-enabled evaluation, Pen-DHRL obtains approximately 64% and 24% higher reward per step on Bridge and Dynamic, respectively, and approximately 26% and 30% more captured hosts on VarA and VarB than the next-best baseline. With all defensive dynamics active from the beginning of training, Pen-DHRL achieves positive mean episode returns in all four scenarios, whereas the flat baselines produce negative mean returns in all 12 architecture–scenario combinations; staged curriculum training produces positive mean returns across all 16 evaluated combinations.
[LG-25] Closing the Log Gap: Minimax-Optimal Regret for Episodic RL
链接: https://arxiv.org/abs/2610.40165
作者: Guy Zamir,Victor Boone,Matthew Zurek,Yudong Chen
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:In this paper, we study the regret minimization problem in episodic tabular MDPs. Our algorithm achieves the first expected regret bound scaling as \sqrtH^3 S A K for sufficiently large K , where H is the episode length, K is the number of episodes, and S and A are the numbers of states and actions, respectively. Our method is the first to match the minimax lower bound \emphwithout extra polylog factors. We introduce a variant of UCBVI which does not maintain strictly optimistic estimates, and uses MOSS-style exploration bonuses whose logarithmic terms adapt to both visit counts and estimated variances. A key challenge is controlling the propagation of errors caused by under-optimism, which we accomplish in an occupancy-weighted way, avoiding logarithmic overhead that would be incurred by uniform-over-rows concentration. Our analysis develops time-uniform and variance-aware concentration bounds that may be of independent interest. As a corollary we also develop an algorithm for (\varepsilon, \delta) -PAC best-policy identification, which is the first to match the \Omega\big(H^3 SA\log(\frac1\delta)\varepsilon^-2\big) minimax lower bound up to constants in some regimes.
[LG-26] Mode Dynamics: A Framework for Neural Network Learning
链接: https://arxiv.org/abs/2610.40143
作者: Hossein Honarvar
类目: Machine Learning (cs.LG)
*备注: 32 pages, 17 figures, 5 tables
Abstract:We introduce Mode Dynamics, a framework that connects coordinated parameter motion during neural network learning to changes in model outputs. Starting from a recorded parameter trajectory, we decompose its updates into collective modes and map each mode to its state-dependent output effect. The framework organizes this link through three mechanisms. Decomposition identifies collective modes in parameter motion and their time-varying amplitudes. Reconstruction tests whether the effects induced by those modes recover local and accumulated output change. Concentration quantifies how behavioral contribution is distributed across modes. Controlled scalar regression and next-token experiments directly test these mechanisms. They show that small sets of leading modes reconstruct behavior, induced effects depend on the training path, and modal complexity varies with task and data. We then evaluate the framework on held-out image and language-model outputs using CIFAR-100 autoencoding and a public Pythia-70M pretraining trajectory. Across these settings, the results connect compact modal subspaces of recorded parameter motion to evolving behavior.
[LG-27] A Deep Learning Framework for Quantile Demand Forecasting in E-commerce with Doubly Censored Data
链接: https://arxiv.org/abs/2610.40138
作者: Ziwen Ye,Jiawen Wei,Teng Song,Chen Chen,Guangrui Ma
类目: Machine Learning (cs.LG)
*备注:
Abstract:In recent years, a growing number of online retail companies have embraced deep learning models for demand forecasting, leveraging their scalability in processing large-scale datasets. Accurate demand forecasts enable improved planning and allow operational flexibility to adjust sales accordingly. Moreover, quantile estimation has gained widespread application as it offers valuable uncertainty quantification, thereby enhancing the robustness of decision-making processes. However, limited research has addressed the effective quantification of uncertainty when target labels are ambiguous. In retail, actual sales can be capped by stock and boosted by temporary promotions, whereas the target is underlying demand. Such scenarios are characterized by double censoring and no deep-learning pipeline yet exists for such data. In this paper, we introduce a straightforward framework for learning quantiles from doubly-censored observations and propose a tailored evaluation metric. Our approach is supported by theoretical guarantees, and we empirically demonstrate that our metric correlates well with the oracle calibration measure. Experimental results show that our estimator surpasses existing baselines. Finally, inventory simulations and real-world applications confirm the practical utility of our method in retail operations.
[LG-28] Does System Size Matter in Uncertainty Quantification for Machine Learning Interatomic Potentials?
链接: https://arxiv.org/abs/2610.40126
作者: Chayaphol Lortaraprasert,Nontawat Charoenphakdee,Wenwen Li,Yong-Bin Zhuang,Yuta Tsuboi
类目: Machine Learning (cs.LG); Materials Science (cond-mat.mtrl-sci)
*备注:
Abstract:Machine-learning interatomic potentials (MLIPs) commonly use uncertainty estimates to flag out-of-distribution (OOD) configurations and select structures for active learning. A key question is whether the resulting structure-level uncertainty scores remain comparable across different system sizes. A common practice is to aggregate per-atom uncertainties by their mean or maximum and compare the resulting score against a fixed threshold, implicitly assuming such comparability. We test this assumption across several in-distribution (ID)/OOD pairs by characterizing both aggregations as a function of atom count. The two aggregations exhibit distinct size dependencies. Mean aggregation remains size-independent for homogeneous systems but loses sensitivity to defects in large cells. In contrast, maximum aggregation grows with system size for ID systems while remaining approximately flat for defect-containing systems, narrowing the ID-OOD gap in a manner consistent with extreme-value statistics. Although highly uncertain atoms in large structures may indicate novel local environments, retraining on these structures yields little improvement, in contrast to retraining on OOD structures constructed to lie outside the training distribution. Our results reveal a practical limitation of size-unaware uncertainty criteria and motivate size-aware calibration for more reliable OOD detection in large-scale atomistic simulations.
[LG-29] D-MOD: Diffusion mode estimation for accelerated MRI
链接: https://arxiv.org/abs/2610.40110
作者: Md Kawsar Ahmed,Mathews Jacob
类目: Machine Learning (cs.LG)
*备注:
Abstract:Diffusion models are powerful priors for accelerated MRI reconstruction, but most current methods are posterior samplers, which return a random draw from the posterior. A reproducible and high-quality reconstruction often requires averaging many samples, resulting in a high computational cost. We introduce D-MOD, a fast algorithm that instead provides a posterior mode. It uses a minorize-maximize algorithm on a lower bound of the smoothed log-posterior, which provably increases the objective. Each iteration draws S noisy samples of the current estimate at the current noise level, denoises them with a pretrained diffusion model, and combines their average with the measured k-space through a conjugate-gradient (CG) solve. We anneal the noise levels, as the distance between the bound and its target decays quadratically with noise. On fastMRI brain (T1, T2, FLAIR), a single D-MOD reconstruction exceeds, in PSNR, the 10-sample finite-MMSE estimates of DPS, DAPS and SNORE-P and stays within 0.31 dB of the strongest baselines (DiffPIR, DDS). This approach thus provides a fast deterministic reconstruction algorithm (11-73 times faster than MMSE estimates), which is useful in a clinical setting.
[LG-30] Coupled Guidance for Flow Matching NEURIPS2026
链接: https://arxiv.org/abs/2610.40109
作者: Mateo Clémente,Léo Maxime Brunswic,Rui Heng Yang,Amir Rasouli
类目: Machine Learning (cs.LG); Robotics (cs.RO)
*备注: Accepted at NeurIPS 2026. 25 pages
Abstract:Guidance methods are essential tools for steering the generation process in diffusion models, improving the generalizability of stochastic robotic controllers and enabling high-quality controlled image synthesis. Despite strong parallels between flow-matching and diffusion, integrating guidance into flow-matching models remains quite challenging. Recent work has proposed general guidance frameworks for flow-matching, however, they are direct adaptations of diffusion guidance methods and can prove unstable. In this work, we conduct a theoretical analysis of guidance in flow-matching and show that a naive translation of diffusion-based guidance to flow-matching breaks the coupling distribution used during training, resulting in weaker control over sampling accuracy. Building on this insight, we propose Coupled Guidance, a flow-matching-specific guidance mechanism derived from a geometric perspective that leverages the coupling structure inherent to flow-matching training. To validate the effectiveness of our approach, we conduct extensive experimentation on a custom robotic collision avoidance task using ManiSkill3, four robotic dexterous manipulation tasks from the Adroit benchmark, and three image generation tasks, namely inpainting, super-resolution, and deblurring, using the ImageNet-128 and CelebA-HQ-256 datasets. Our findings show that our framework improves sample quality and guidance stability across all tasks, without imposing any additional overhead.
[LG-31] FIDE: Fiber-Informed Differentiable Experimental Design for Reference-Relative Law Dynamics
链接: https://arxiv.org/abs/2610.40009
作者: Pietro Zanotta,Ján Drgoňa
类目: Machine Learning (cs.LG)
*备注:
Abstract:Aggregate measurements generally leave an evolving population law underdetermined, so different measurement systems can be equally adequate for a scientific task while constraining different aspects of the law. We study how to resolve this residual design freedom in baseline-preserving settings, where an existing dynamical model remains the working baseline outside quantities explicitly valued by the task. \textbfFiber-\textbfInformed \textbfDifferentiable \textbfExperimental Design (\textbfFIDE) completes each candidate’s aggregate constraints by information projection relative to a common frozen reference, then ranks scientifically adequate designs by the dynamical revision required to realize the resulting completed paths. The resulting \emphFull action is the minimum mean-squared velocity correction needed to realize a completed path; equivalently, its square root is the sharp reference-relative expectation-rate discrepancy over a gradient-normalized Sobolev class. This defines a conditional minimum-revision principle for baseline-preserving design rather than a generic truth-recovery objective. Experiments show that this preference changes sensor selection, can help or harm withheld downstream predictions depending on baseline reliability, can be used prospectively from aggregate predictions, and remains computationally usable for a 32 -dimensional configuration law.
[LG-32] Enhancing Atrial Fibrillation Detection Through Hyperparameter Optimization
链接: https://arxiv.org/abs/2610.40001
作者: Sara Gjorgjieva,Ilinka Ivanoska,Marjan Gusev
类目: Machine Learning (cs.LG)
*备注: 11 pages, 3 figures, 6 tables. Presented at the ICT Innovation 2025 Conference, Ohrid, North Macedonia. Accepted for publish by Springer in the Communications in Computer and Information Science Series (CCIS). In press
Abstract:This study explores the role of hyperparameter optimization (HPO) in enhancing the detection of Atrial Fibrillation (AF) using both machine learning and deep learning models. We evaluated a range of standard classifiers - Decision Tree, Random Forest, and XGBoost classifiers, alongside deep learning architectures including Recurrent Neural Networks, Deep Neural Networks, and Siamese Neural Networks. To fine-tune these models, we applied four HPO methods: Genetic Algorithm, Hyperband, Optuna, and SMAC3. The experiments were conducted using inter-patient and intra- patient data splits to assess model generalization and robustness in the presence of class imbalance, which is reflective of real-world AF occurrence. Our results, based on PhysioNet electrocardiogram datasets, showed that the Hyperband-optimized XGBoost model outperformed all others, achieving a peak F1 score of 96.95%. These findings highlight the critical role of suitable HPO strategies in enhancing the accuracy and robustness of AF classification outcomes in clinical settings.
[LG-33] A physics-informed uncertainty-aware surrogate for p-n junction diodes
链接: https://arxiv.org/abs/2610.39993
作者: Suchita Kulkarni
类目: Machine Learning (cs.LG); Mesoscale and Nanoscale Physics (cond-mat.mes-hall)
*备注: Comments welcome
Abstract:We develop an uncertainty-aware physics-informed surrogate for a one-dimensional p - n junction diode, conditioned on device length, doping and bias. Using the internal state of the device as predicted by the surrogate we develop a test-suite of quantities of interest including the terminal current, depletion width, peak field and stored charge. The formalism and the quantities are can be extended to asymmetric junction and two-dimensional diodes as shown by means of benchmark analysis. For the one-dimensional symmetric junction surrogate, we obtain 0.11% median L_1 error on the electrostatic potential, and 1.5% and 0.61% on the electron and hole densities. Finally, we develop an uncertainty analysis which determines the validity of surrogate answer within the user-specified tolerance.
[LG-34] Observability-Aligned Evidential Masked Pretraining for Reliable VRU Motion Prediction
链接: https://arxiv.org/abs/2610.39958
作者: Shang Gao,Franz Götz-Hahn,Konrad Doll,Bernhard Sick
类目: Machine Learning (cs.LG)
*备注:
Abstract:Vulnerable road user (VRU) prediction must operate under partial observability and distribution shift, so reliable uncertainty matters alongside accuracy. Masked pretraining is attractive in this setting because it naturally exposes the model to missing content during self-supervision, but existing time-series masked autoencoding methods are typically deterministic and do not constrain how confidence should change with observability. We introduce eCoMAE, an evidential masked pretraining method for multivariate time series that explicitly aligns predictive evidence with what is observed. eCoMAE combines block-wise time-step masking with a Normal-Inverse-Gamma decoder and three observability-aware mechanisms: uncertainty-weighted mean consistency on positions visible in both a masked view and a reference pass, evidence suppression on masked positions, and representation consistency across stochastic masks. On IMPTC benchmark for VRU intention prediction under leave-one-environment-out weather shift, eCoMAE improves the strongest baseline from 0.7971 to 0.8068 in OOD accuracy and from 0.0718 to 0.0648 in OOD AURC. Additional evaluation on pedestrian trajectory forecasting across IMPTC, Waymo, nuScenes, and inD further yields the best fine-tuning MSE on all four datasets. These results show that observability-aligned evidential pretraining improves reliability under shift and remains effective beyond the primary benchmark.
[LG-35] H2R: Hierarchical Hint-Based Reward for Mathematical Reasoning via Reinforcement Learning
链接: https://arxiv.org/abs/2610.39934
作者: He Tu,Siye Li,Hongyi Wen
类目: Machine Learning (cs.LG)
*备注:
Abstract:Reinforcement Learning with Verifiable Rewards has shown great promise in enhancing mathematical reasoning in large language models. However, existing approaches provide limited intermediate supervision for complex, multi-step reasoning tasks. To bridge this gap, we propose \textbfHierarchical Hint-based Reward, a framework that leverages the intrinsic sub-question structure of mathematical problems as structured intermediate conditioning signals during RLVR training. Unlike chain-of-thought prompting, these hierarchical hints serve as logically grounded intermediate milestones that effectively reduce the reasoning horizon toward the final answer. To validate this framework, we introduce \textbfHiMath, a benchmark of hierarchically structured mathematical problems. Across diverse LLMs, hierarchical hints consistently improve reasoning performance even without task-specific training. When integrated into RL training via H ^2 R, they boost Qwen2.5-Math-7B accuracy on HiMath from 30.4% to 39.1% using hierarchical hints, and up to 45.6% with Oracle Hierarchical Hints. This pronounced improvement demonstrates that incorporating reliable intermediate checkpoints into reward design is essential for sustaining complex mathematical reasoning.
[LG-36] Low-Bit GDN Recurrent-State Quantization: Global Gradient-aware Allocation and Pure-Write Sliding Windows
链接: https://arxiv.org/abs/2610.39931
作者: Qinghang Huang,Xuchu Huang,Xunzhao Yin
类目: Machine Learning (cs.LG)
*备注: 24 pages, 13 figures
Abstract:Linear attention stores historical information in a recurrent state of fixed size, providing a compact memory representation for long-context inference. Reducing state precision further lowers per-request storage requirements, but quantized writeback errors enter subsequent recurrences, degrade prediction quality on long-reasoning tasks, and can be accompanied by repetitive generation and excessively long outputs. We propose Global Gradient-aware Allocation (GGA) and the Pure-Write Sliding Window (PW) to address precision allocation and recent-information maintenance. GGA combines candidate quantization residuals with gradient sensitivity to future prediction loss to allocate head bit widths under a model-wide mixed-precision budget. PW independently maintains recent pure writes as FP8 vector pairs, propagates them through the full GDN state transition, and merges them into the low-bit historical state only after they leave the window. Across seven evaluations on Qwen3.5-4B/9B, GGA allocation alone improves over STEPQuant@4 by 19.48/29.84 percentage points. Adding PW-S8 yields seven-task macro-averages of 69.83%/78.63%, improving over STEPQuant@4 by 38.39/41.21 percentage points, respectively. Both precision reallocation and window maintenance improve task scores while mitigating excessively long outputs: at @6, GGA reduces mean generation length, averaged equally across tasks, by 10.45%/14.65% relative to STEPQuant; adding PW-S8 to GGA@4 further reduces mean generation length by 32.32%/36.77%. Comparisons at similar storage budgets further show that using additional capacity to maintain recent writes yields higher task scores and shorter mean outputs than solely increasing the bit width of the accumulated state.
[LG-37] Data-driven Random-feature Approximation of Koopman Generators of Nonlinear Continuous-time Systems
链接: https://arxiv.org/abs/2610.39897
作者: Ana Čukarska,Maximilian Gollwitzer,Viola Muning Li,Vladyslav Fediukov,Zahra Monfared,Felix Dietrich
类目: Machine Learning (cs.LG); Numerical Analysis (math.NA)
*备注:
Abstract:Learning continuous-time dynamical systems from data is an active area of research. The Koopman formalism is often used here, because it is a flexible approach that is informative about the system’s fundamental properties. The formalism represents dynamical systems through linear operators acting on a suitable function space. A practical difficulty arising in the numerical approximation of Koopman operators is the choice of a ``dictionary’', meaning an effective finite-dimensional representation of the function space. In this paper, we focus on learning dynamical systems in continuous-time, and so we consider numerical approximations of the Koopman generator instead of the Koopman operator. Our main contribution is a data-driven random-feature method to obtain a parsimonious and effective function representation for Galerkin approximation methods such as gEDMD. The representation we construct is data-informed, which allows us to improve upon data-agnostic methods. We empirically demonstrate the suitability of our method for forecasting tasks and system analysis. The numerical experiments show that our approach is accurate, and faster than iterative optimization. Further, we prove convergence of our approximation of the vector field and the resulting trajectories. In addition to the numerical results, we also make theoretical contributions, including a general convergence criterion for the semigroup generated by the gEDMD approximation and a counterexample showing that gEDMD does not generally preserve exponential boundedness. With our contribution, we advance the frontier of efficiently learning continuous-time dynamics from data, towards interpretable and mathematically robust modelling based on the Koopman framework.
[LG-38] ACFormer: Transformer-Based Generative Model for Target-Conditioned De Novo PROTAC Design
链接: https://arxiv.org/abs/2610.39891
作者: Yossra Gharbi,Rocío Mercado
类目: Machine Learning (cs.LG)
*备注:
Abstract:Targeted protein degradation (TPD) represents a fundamental shift in drug discovery, moving beyond traditional target inhibition toward the event-driven removal of disease-causing proteins. A prominent strategy for TPD is the use of proteolysis-targeting chimeras (PROTACs), heterobifunctional molecules that selectively route a protein of interest (POI) to the ubiquitin-proteasome system for degradation. While PROTAC design still relies heavily on trial-and-error synthesis and experimental testing, deep generative models offer a promising means to explore PROTAC chemical space in silico and generate novel molecules. However, many existing generative methods formulate PROTAC design primarily as a linker-generation problem, with POI-binding (warhead) and E3-ligase-binding components fixed in advance. This restricts exploration of the full PROTAC chemical space and often leaves the POI implicit through the predefined warhead, rather than using protein-level information as an explicit conditioning signal during generation. To address these limitations, we propose TACFormer: a transformer-based generative model for target-aware de novo PROTAC design. The model conditions a GPT-style molecular decoder on pretrained ESM-2 representations of POI sequences, allowing POI-level information to steer full-molecule generation. Our results show that TACFormer generates highly novel and chemically diverse PROTACs while showing consistent POI-dependent shifts in generated chemical space. In particular, molecules generated under specific target conditions preferentially populate chemical neighborhoods associated with POIs in the held-out dataset, while extrapolating to novel chemotypes. By natively integrating biological context into the generative process, TACFormer moves PROTAC design toward a more biology-informed exploration of degrader chemical space.
[LG-39] DWISE: Denoising Workload-Informed Scheduling Engine
链接: https://arxiv.org/abs/2610.39852
作者: Lie Li,Wen Li,Junxiao Shen,Jindong Gu,Guosheng Hu
类目: Machine Learning (cs.LG)
*备注: 17 pages, 11 figures, 17 tables
Abstract:Unlike autoregressive large language models (AR-LLMs), diffusion LLMs (dLLMs) generate text by iteratively denoising many tokens in parallel, offering a promising path to efficient inference. Existing serving frameworks, however, schedule dLLM requests with policies designed for AR models, namely first-come-first-served (FCFS) admission and prefill priority. We find that both policies are ill-suited to dLLMs, wasting much of their speed. FCFS ignores that a dLLM request’s decoding cost is highly heterogeneous and unknown at arrival, causing severe head-of-line blocking. Prefill priority, widely used in AR-LLM serving to keep time-to-first-token low, sharply cuts dLLM efficiency, as rounds that run only prompts leave most of the batch idle. We present DWISE, a Denoising Workload-Informed Scheduling Engine for efficient dLLM serving. DWISE schedules requests by their remaining work, measured in a block-level unit shared by prefill and denoising, so both phases share a round. To predict the unknown denoising workload, DWISE combines a lightweight prompt-based predictor at admission with a single refinement from the first denoising round’s hidden states, which capture model-specific information absent from the prompt and require no additional model execution. Implemented in SGLang and evaluated on five models from three dLLM families under bursty workloads, DWISE reduces mean end-to-end latency over SGLang by 5.4 \times on average and up to 14.9 \times while preserving task accuracy. The code will be released publicly.
[LG-40] Possibility-based Inference: A Deterministic Framework for Robust and Reliable Uncertainty Quantification in Decision-Making
链接: https://arxiv.org/abs/2610.39821
作者: Seyedfakhredin Musavishavazi,Maryam Alipour
类目: Machine Learning (cs.LG)
*备注:
Abstract:Decision-making under uncertainty remains a key issue in Machine Learning (ML), particularly in the deployment of safety-critical systems, where reliable quantification of system behavior is essential for evaluating behavioral impacts and strengthening confidence in ML models. This contribution, influenced by fuzzy set theory and grounded in Bayesian statistics, addresses challenges related to data precision and the performance-accuracy trade-off in Bayesian inference (BI). To this end, we introduce a possibility-based contrastive grade of prediction as an alternative quantification method to existing latent posterior-based approaches for parameter estimation (approximation) and model selection. Unlike conventional Bayesian inference, which represents uncertainty within the probabilistic framework, the proposed model adopts a possibility-based representation that allows uncertain and potentially conflicting information to be characterized more flexibly. Rather than focusing on isolated aspects of a problem, the resulting mathematical framework provides a systematic approach to data mining while mitigating overly confident, yet implausible predictions generated by highly specialized deep learning architectures.
[LG-41] Optimal Contextual Pricing in Non-Stationary Environments
链接: https://arxiv.org/abs/2610.39803
作者: Yiding Feng,Alireza Rezaeimoghadam
类目: Machine Learning (cs.LG)
*备注:
Abstract:We study contextual dynamic pricing in a non-stationary environment. In each round, a seller observes a buyer’s context, posts a price, and receives only binary purchase feedback. The buyer’s value is linear in the context and an unobserved feature vector that may change over time, subject to a known total variation budget V . We measure dynamic regret against the first-best benchmark that extracts each buyer’s full value. For identical contexts, we develop a deterministic algorithm that combines binary-search exploration, lower-envelope pricing, and periodic and inconsistency-triggered restarts. It achieves regret O((\sqrtTV+1)\log T) . With heterogeneous contexts, a new geometric difficulty arises: small feature changes can leave a relaxed feasible set narrow but far from the current feature vector. We address this difficulty through a multiscale shallow-pricing algorithm that maintains relaxed feasible sets and enclosing ellipsoids at several variation scales. Larger scales provide reliable localization certificates, while the smallest active scale determines the exploitation price. The algorithm achieves dynamic regret O((d^2+d^3/2\sqrtTV)\log^2 T) , where d is the context dimension. For fixed d , both guarantees match the lower bound \Omega(\sqrtTV+\log\log T) up to logarithmic factors in T , establishing near-optimality among deterministic algorithms for every V\in[0,T] , including the stationary case. Subjects: Machine Learning (cs.LG) Cite as: arXiv:2610.39803 [cs.LG] (or arXiv:2610.39803v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.39803 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-42] AI at a Turning Point: From ChatGPT through Neural Networks to Neuromorphic AI
链接: https://arxiv.org/abs/2610.39796
作者: Volker Kempe
类目: Machine Learning (cs.LG); Neural and Evolutionary Computing (cs.NE)
*备注: 88 pages, 20 figures
Abstract:The tension between the established technologies of connectionist AI and their neuromorphic challengers is attracting growing attention, above all because of the enormous energy demand of the former and the low power consumption of the latter. Developing and running large AI models consumes gigawatt- to terawatt-hours of energy, while the human brain operates on a continuous 20 watts. Spiking neural networks (SNNs) counter this energy demand with a frugal approach and bring elements of self-organization into the methodological repertoire of AI: local learning instead of global error backpropagation, event-driven computing, and time as a carrier of information. They do not replace today’s AI but extend it; in hybrid systems each approach can be deployed where its strengths lie. This monograph explains the field in a generally accessible yet mathematically rigorous way. Using the text-generation module of ChatGPT as an example, the principle, elegance, and computational cost of transformer-based AI are presented and compared with the coding, learning, and network-design methods of SNNs. Throughout, relevant mechanisms of the human nervous system serve as model and reference point. Topics include spike-timing-dependent plasticity, population and temporal coding, the Neural Engineering Framework, neuromorphic hardware from Loihi 2 to commercial edge processors, and the prospects for classical and neuromorphic AI to complement and extend each other.
[LG-43] Diagnosing Optimizer-Landscape Interaction via Representation in Subset Selection Multitasking: Evidence from Piano Fingering CEC2026
链接: https://arxiv.org/abs/2610.39779
作者: Ananda Phan Iman,Tingyang Wei,Chang Wook Ahn
类目: Neural and Evolutionary Computing (cs.NE); Machine Learning (cs.LG)
*备注: Accepted paper in WCCI/CEC 2026
Abstract:Continuous relaxations are commonly used as an encoding scheme for evolutionary multitasking optimization (EMTO), allowing heterogeneous tasks to be optimized within a unified search space. However, these continuous representations fundamentally reshape the induced fitness landscape, and their impact on optimization behavior and knowledge transfer remains underexplored. In this work, we bridge the gap by analyzing the interaction between the representation-induced fitness landscape and optimizer behavior through fitness landscape analysis. Using piano fingering estimation as a representative case of the subset selection problem, we introduce several encoding variants and investigate how their induced landscapes influence search dynamics and solution quality under EMTO. Our analysis indicates that transfer efficacy is strongly associated with changes in representation-induced landscape characteristics, rather than algorithmic factors alone, highlighting the representation design as the key factor of effective knowledge transfer in multitasking optimization.
[LG-44] Surface Conditioning for Zero-Shot Multivariate Correction in a Marine Biogeochemistry Emulator
链接: https://arxiv.org/abs/2610.39777
作者: Gabriela Martínez Balbontín(MOi, PEQUAN),Anastase Charantonis(ARCHES),Dominique Béréziat(PEQUAN),Stefano Ciavatta(MOi)
类目: Machine Learning (cs.LG)
*备注:
Abstract:Bias correction in process-based models usually means iterative data assimilation with explicit covariance modelling. We propose a lightweight alternative based on a neural emulator of marine biogeochemistry (BGC), showing that the conditioning pathway of its autoencoder can serve as an observation injection point. The decoder is conditioned on surface variables via Feature-wise Linear Modulation (FiLM), originally designed to incorporate physical boundary forcings. We extend these inputs to include surface chlorophyll-a and replace the simulated field with satellite observations at inference time. The surface substitution propagates through the water column, improving agreement with depth-resolved observations at all depths, and spreads to other output variables in a pattern consistent with expected trophic relationships.
[LG-45] Activation Probes Are Reliable With a Handful of Positive Examples NEURIPS2025
链接: https://arxiv.org/abs/2610.39766
作者: Riya Tyagi,Stefan Heimersheim
类目: Machine Learning (cs.LG)
*备注: Presented at the Mechanistic Interpretability workshop at NeurIPS 2025
Abstract:Efforts to monitor advanced AI for rare misalignments face a data challenge: abundant aligned examples but only a handful of misaligned ones. We test activation probes in this “few vs. thousands” regime on spam and honesty detection tasks. For our tasks, training with many negative examples is on average more positive-sample-efficient than balanced training for small numbers (1-10) of positive samples. We also find that LLM upsampling can provide a performance boost equivalent to roughly doubling the number of real positive samples, though excessive upsampling hurts performance. Finally, we show a positive scaling trend, where larger models are more positive-sample-efficient to probe. Our findings suggest we should leverage the large number of negative samples available to amplify the signal from rare but critical misalignment examples.
[LG-46] Going deep reaching further: What can Boltzmann machines efficiently represent?
链接: https://arxiv.org/abs/2610.39731
作者: Tobias Reinhart,Gemma De les Coves
类目: Machine Learning (cs.LG); Disordered Systems and Neural Networks (cond-mat.dis-nn); Mathematical Physics (math-ph)
*备注: 34 pages of main text + 32 pages of appendices, 14 figures, 2 tables
Abstract:While universal approximation theorems theoretically underpin the representation capabilities of machine learning models, they have two shortcomings: They do not focus on efficient representations and they tend to be insensitive to depth and other heuristically important properties. In this work we show that these two shortcomings are intertwined. We prove that machine learning models which are deeper or have higher-order interactions can efficiently represent a larger set of functions, i.e. have larger efficient reach. It follows that deep Boltzmann machines (DBMs) have maximal efficient reach, restricted Boltzmann machines (RBMs) sub-maximal, and k-body generalizations of RBMs form a hierarchy between the two, even though all are universal. More generally, (i) we provide a full characterization of machine learning models with maximal efficient reach; (ii) we derive separations between deep and shallow models for three representation tasks; and (iii) we prove that the efficient reach of shallow models constitutes an infinite hierarchy indexed by the interaction order. The proofs draw on different methodologies, namely on (i) a recent characterization of universal spin models; (ii) computational complexity bounds for spin models; and (iii) communication complexity bounds applied to shallow models. This work shows that universal machine learning models can strongly differ in their efficient representation capabilities, and provides a theoretical foundation for heuristic observations.
[LG-47] Grounding Predictive Reliability through Masked Observability
链接: https://arxiv.org/abs/2610.39728
作者: Shang Gao,Bernhard Sick,Franz Götz-Hahn
类目: Machine Learning (cs.LG)
*备注:
Abstract:An uncertainty estimate is meaningful only if its semantics can be tied to observable predictive behavior. Normal-Inverse-Gamma (NIG) evidential regression provides a structured decomposition of predictive variability into data-scale and predictive-mean components, but marginal likelihood does not identify their allocation. We address this semantic ambiguity in masked representation learning. MORSE (Masked Observability for Reliable Structured Encoding) predicts the same hidden target under controlled changes in visible context and uses normalized cross-context variation to ground predictive-mean uncertainty, while a Gaussian objective learns the conditional mean and data scale. The normalized variation plays the role of the NIG factor 1/\nu , which the marginal leaves unidentified; this yields a policy-relative observability variance that measures dependence on visible evidence, distinct from variation induced by resampling the training set. On a Gaussian-process benchmark with a contextual reference built from analytic Bayes conditional means, MORSE has the lowest log-variance error among compared evidential, ensemble, and dropout estimators in all nine tested settings, while the estimator closest to training-set variation perturbs the model rather than its context in every setting, so the two references separate the estimators empirically. MORSE amortizes observability estimation into one forward pass per observation state. On a heteroscedastic pulse benchmark, its predictor’s cross-context variance approaches a large-sample reference as training data increase. MORSE also improves low-label transfer on PTB-XL and in-distribution trajectory prediction on IMPTC.
[LG-48] Implicit Bias of (Stochastic) Gradient Descent on Deep Linear Diagonal Networks
链接: https://arxiv.org/abs/2610.39717
作者: Wiebke Bartolomaeus,Holger Rauhut
类目: Machine Learning (cs.LG); Optimization and Control (math.OC)
*备注: 38 pages, 14 figures
Abstract:Modern machine learning models are often overparameterized, i.e., have more parameters than training data. This typically implies that the loss function admits multiple minimizers so that training methods (and their hyperparameter choices) have significant influence on the computed solution. We study this so-called implicit bias phenomenon in the context of training deep linear diagonal networks. Previous work has shown that the gradient flow for learning such networks converges to (almost) sparse solutions when initialized small. For an arbitrary number of layers, we extend this analysis to the practically more realistic (stochastic) gradient descent iterations. Our approach identifies (stochastic) gradient descent as a perturbed (stochastic) mirror gradient descent. This allows us to establish convergence and characterize the limiting solution, relating it to weighted \ell_1 -minimization. Moreover, we provide sufficient conditions on the step sizes and quantify the influence of the step sizes on the implicit bias.
[LG-49] Smartphone Indoor Localization Enhanced by Hierarchical Multi-Sensor Context Recognition
链接: https://arxiv.org/abs/2610.39691
作者: Wenhui Feng,Muhan Weng,Jianyu Liu,Xiansheng Yang,Hao Zhang,Liang Wang,Yuan Zhuang
类目: Machine Learning (cs.LG); Robotics (cs.RO)
*备注:
Abstract:Smartphone indoor localization requires low-cost, robust positioning across scene transitions and phone-use variations. Pedestrian dead reckoning (PDR) enables infrastructure-free deployment but accumulates errors from heading drift, structural transitions, and carrying-mode changes. Existing methods often treat contextual cues as a single event stream or local correction signal, limiting their ability to address distinct localization errors. This paper presents a framework enhanced by hierarchical multi-sensor context recognition. A jointly data- and model-driven recognition model predicts environment, scene, and activity probabilities, which are converted into scene-event and action-context segments. Scene-event segments guide scene-constrained PDR correction through corridor heading constraints, landmark attraction, fixed-lag backtracking, and vertical-transition handling. Action-context segments drive action-aware RoNIN-based position compensation for carrying-mode-dependent drift. Experiments on cross-floor and multi-action trajectories against LiDAR-SLAM references show that the proposed framework outperforms PDR and representative learned inertial odometry baselines, demonstrating that hierarchical semantics provide reliable, localization-specific correction cues.
[LG-50] A New AFIB Detection Method with Deep Neural Networks
链接: https://arxiv.org/abs/2610.39665
作者: Sara Gjorgjieva,Ana Angjelevska,Dimitar Trajanov,Marjan Gusev
类目: Machine Learning (cs.LG)
*备注: 10 pages, 7 tables, 3 formulas
Abstract:Early detection of atrial fibrillation (AFIB), characterized by irregular heart rhythms, can potentially save thousands of lives annually [20]. Numerous studies have attempted to address this challenge using various algorithms and methodologies. This paper introduces a novel and robust method that enhances the precision of AFIB detection. Neural Networks (NN) are widely recognized for their ability to identify complex electrocardiogram (ECG) patterns, facilitating the accurate detection of irregular heart rhythms. This study explores multiple conventional machine learning algorithms alongside neural network approaches to develop a robust AFIB detection system. To evaluate the effective- ness and robustness of the proposed method, the MIT BIH arrhythmia database (MITDB) and the Long-term Atrial Fibrillation Database (LTAFDB), both available on the Physionet site. Our approach presents a novel feature engineering technique, employing specific features such as a time series of the differences in successive heart rates (dBPM) rather than traditional inter-beat intervals. The model analyzes inputs labeled into five distinct classes, using only clean intervals identified through a new labeling method for training while treating others as outliers. This strategy ensures a more accurate and reliable AFIB detection process. The best-performing model, a Deep Neural Network (DNN), achieved an F1 score of 95.18% for binary atrial fibrillation classification on a benchmark dataset distinct from the training set. Models trained using our approach have demonstrated superior performance than previously published models.
[LG-51] Improving the Intrinsic Interpretability of Differentiable Decision Trees through Euclidean Feature-Concentration Regularisation
链接: https://arxiv.org/abs/2610.39596
作者: Qiqi Yang
类目: Machine Learning (cs.LG)
*备注: 17 pages, 3 figures, 5 tables. Based on the author’s MSc dissertation, University of Birmingham
Abstract:Differentiable decision trees (DDTs) support end-to-end optimisation and explicit decision paths, but dense oblique splits can obscure the role of individual input features. We propose the Euclidean-Concentrated Differentiable Decision Tree (EU-DDT), which regularises the relative absolute feature weights at each internal node using their normalised squared Euclidean distance from the uniform distribution. This encourages dominance by a small number of features while retaining probabilistic soft routing. Across five matched random seeds, EU-DDT reduces the mean effective number of features from 8.94 to 1.54 on Wine and from 45.28 to 9.06 on Digits. Mean test accuracy changes from 100.0% to 93.9% on Wine and from 92.3% to 94.6% on Digits, illustrating the dataset-dependent trade-off between concentration and predictive performance. We also use EU-DDT to parameterise a PPO actor with an MLP critic. Two random-seed runs per environment show a consistent qualitative increase in concentration while preserving effective task performance. In representative CartPole-v1 and MiniPacman runs, EU-DDT reaches 100% final evaluation success, while mean node concentration increases from 0.1272 to 0.7179 and from 0.0437 to 0.7237, respectively. These results support EU regularisation as a mechanism for reducing node-level feature complexity; the reinforcement-learning evidence remains preliminary because only two seeds are evaluated.
[LG-52] rajeform: Embedding Crowd Data
链接: https://arxiv.org/abs/2610.39563
作者: Marilena Lemonari,Panayiotis Charalambous,Yiorgos Chrysanthou
类目: Graphics (cs.GR); Machine Learning (cs.LG)
*备注: 14 pages, 17 figures, 5 tables
Abstract:Crowd simulation aims to generate agent behaviours that are rich, natural, and appropriate to their surrounding environment. Recent advances in Machine Learning have enabled such behavioural patterns to be learned directly from trajectory data, allowing simulation models to capture increasingly complex crowd dynamics. However, the capabilities of these systems remain strongly dependent on the type and diversity of the data used during training. When evaluating how well crowd simulators generalise to real-world datasets, most existing metrics focus primarily on trajectory-level differences rather than behaviour-level similarity. In this paper, we present Trajeform, a framework that trains a reconstruction VAE (Variational Autoencoder) on real trajectory data to learn a meaningful continuous embedding that preserves behavioural characteristics of real crowds. The trained encoder can then be applied to synthetic trajectories generated by simulators and used to compare the resulting embedding distributions. Additionally, the learned embedding can be used to analyse supplementary data, such as trajectories corresponding to a simulator’s principal behaviours or captured data from controlled environments (e.g., cycling), providing insight into their ability to span the real behavioural embedding space. The framework compares data from three simulators (CCP, modCCP, and RVO) against ground-truth datasets and is evaluated quantitatively and qualitatively using latent-space analysis, distributional metrics, and trajectory generation through sampling from the aggregated posterior distribution. The results indicate that the learned latent space captures meaningful trajectory dependencies found in the real data, making the framework useful both as an additional measure of synthetic data believability and as a tool for informing the training of future trajectory models.
[LG-53] Quantifying the Relay-Coverag e Trade-off in Mobile Multihop Flooding
链接: https://arxiv.org/abs/2610.39561
作者: Johannes F. Loevenich,Thies Möhlenhof,Roberto Rigolin F. Lopes
类目: Networking and Internet Architecture (cs.NI); Machine Learning (cs.LG)
*备注:
Abstract:Network-wide flooding in mobile multihop networks selects relays from locally maintained topology information. Mobility makes this information outdated, leaving two basic responses: add redundant relays or refresh the topology and re-select relays more frequently. How these mechanisms trade reliability for overhead, however, is not well quantified. We quantify this trade-off on a public trace of 157 vehicles using Optimized Link State Routing version 2 (OLSRv2) and exact minimum-cost relay sets. Twofold coverage increases median relay-cover lifetime by only 2.3x, rather than about an order of magnitude as predicted by an independent-hazard model. The limitation is structural: 43.6 % of two-hop neighbours have only one possible relay, and newly appearing neighbours invalidate existing covers. Moreover, the delivery gain of a second relay falls from 0.061 to 0.016 when relay selection uses instantaneous link state, showing that redundancy mainly compensates for outdated topology information. Relay coverage and topology refresh therefore form a measurable redundancy-freshness trade-off rather than independent robustness mechanisms. In our trace, twofold coverage has lower overhead below about 100 bit/s of application flooding per node, whereas fresher topology information has lower overhead above it. This makes relay redundancy a load-dependent design choice rather than a fixed heuristic robustness setting.
[LG-54] LatentWeave: Test-Time Latent Optimization for Multimodal Reasoning
链接: https://arxiv.org/abs/2610.39557
作者: Fanci Meng,Kun Wang,Guibin Zhang,Xinlei Yu,Jiaming Zhang,Ming Li,Yang Wang
类目: Machine Learning (cs.LG)
*备注:
Abstract:Recent advances in test-time optimization have shown that the multimodal reasoning performance of large multimodal language models (MLLMs) can be improved during inference without modifying backbone model parameters. However, existing approaches primarily operate at the prompt or output level, leaving the latent reasoning process largely uncontrolled and insufficient to prevent error accumulation during autoregressive generation. To address this limitation, we propose LatentWeave, a test-time framework that treats multimodal reasoning as a controllable latent trajectory. LatentWeave performs structured Latent Weaving, progressively refining intermediate hidden states segment-wise to correct emerging deviations, and further introduces Reasoning Anchors via lightweight WEAVE tokens to stabilize reasoning structure and guide answer-oriented generation. These mechanisms are designed to enable fine-grained, step-aware intervention in the latent reasoning process while keeping the main MLLM backbone parameters fixed. Extensive experiments across diverse multimodal reasoning benchmarks report improvements over the evaluated baselines; the contribution of latent optimization, anchor training, and additional test-time computation is analyzed separately in the ablation and efficiency studies. Our code is available at this https URL.
[LG-55] Explainable and Adaptive Machine Learning for Fraud Risk Detection in Bangladeshs Digital Banking Ecosystem
链接: https://arxiv.org/abs/2610.39556
作者: Md Najmus Swaqeeb
类目: Machine Learning (cs.LG)
*备注: 23 pages, 6 figures, 5 tables. Presented at the 11th Annual Banking Conference 2026, Bangladesh Institute of Bank Management (BIBM), Dhaka, Bangladesh, 28 October 2026
Abstract:The rapid expansion of digital banking and mobile financial services creates growing challenges for fraud detection, particularly under severe class imbalance and operational constraints. This study presents an explainable and adaptive machine-learning framework integrating chronological evaluation, leakage-conscious feature engineering, probability calibration, SHAP-based explainability, and adaptive risk prioritization. Experiments use the synthetic PaySim mobile-money benchmark and compare Logistic Regression, Random Forest, XGBoost, and LightGBM. XGBoost achieved the strongest fixed-threshold performance, with PR-AUC of 0.5658, ROC-AUC of 0.9068, and F1-score of 0.5369. Platt calibration reduced the test Brier score from 0.1141 to 0.0245, improving probability reliability. SHAP analysis identified temporal, destination-account, transaction-type, and amount features as important model drivers. An adaptive top-1% policy reduced false positives from 192 to 117, but recall declined from 0.4236 to 0.2122, demonstrating a clear alert-capacity trade-off. The results provide a reproducible methodological framework rather than evidence of deployment performance in Bangladesh, since PaySim is synthetic. Validation using governed real-world financial data is therefore an important direction for future research.
[LG-56] Last-Iterate Convergence for Constrained MDPs: From Dual Descent to Primal-Dual
链接: https://arxiv.org/abs/2610.39524
作者: Michael Lu,Max Qiushi Lin,Mo Chen,Sharan Vaswani
类目: Machine Learning (cs.LG)
*备注:
Abstract:We study policy optimization for infinite-horizon, discounted constrained Markov decision processes. While primal-dual ( \textttPD ) methods are widely used in practice, their theoretical guarantees typically hold for an impractical mixture policy. In contrast, dual descent ( \textttDD ) methods output a single policy (“last iterate”) with guarantees on its performance. However, such methods require iteratively solving entropy-regularized MDPs in an inner-loop, limiting their practical effectiveness. Moreover, existing analyses rely on strong coverage assumptions that are difficult to satisfy in practice. Consequently, we first provide a new analysis of \textttDD methods that does not require coverage assumptions. We then employ entropy-regularized natural policy gradient and dual regularization to eliminate the inner loop, resulting in an efficient \textttPD algorithm. We develop a novel tracking analysis which shows that, compared to existing methods, the proposed algorithm achieves improved last-iterate convergence guarantees. Furthermore, our theoretical framework allows us to analyze a truncated bilevel method that interpolates between \textttDD and \textttPD methods. Finally, we generalize our analysis to handle linear function approximation. Compared to prior last-iterate methods, we prove stronger guarantees under less restrictive assumptions.
[LG-57] Interpretable by Design Tabular Prediction via In-Context Local Linear Learning
链接: https://arxiv.org/abs/2610.39506
作者: Ran Eisenberg,Steven Berger,Ofir Lindenbaum
类目: Machine Learning (cs.LG)
*备注:
Abstract:In-context learning (ICL) lets tabular foundation models make predictions without dataset-specific training or fine-tuning. These models achieve strong predictive accuracy, but interpreting how individual features contribute to the prediction remains difficult. In biology, for example, feature contributions can help identify biomarkers or risk genes. Post-hoc interpretability methods assign feature importance scores, but these scores do not indicate what the model actually uses and can lead to explanation errors. We introduce TICL ^3 (Tabular In-Context Local Linear Learning), an interpretable-by-design predictor that uses a tabular foundation model to guide a simple linear prediction rule. TICL ^3 first learns a shared linear model to capture dataset-level feature effects. Then, for each query, it retrieves similar training samples using representations from the foundation model and fits a second linear model to the residuals in that neighborhood. The shared model captures broad effects, while the local model corrects errors among similar samples. Their combined coefficients determine each feature contribution to the final query-specific linear score, so the same model makes and explains the prediction. Across 386 classification benchmark-dataset evaluations, TICL ^3 trails TabICLv2 by 1.1 percentage points and outperforms CatBoost in 266 evaluations. It improves average support F1 over MAPLE by 4.7 points and achieves the best performance on both real-data selected-feature utility metrics.
[LG-58] opology Survives Shuffling: Robust Long-Context Memory Access Prediction for Graph Analytics
链接: https://arxiv.org/abs/2610.39477
作者: Dongyan Sun,Neelesh Gupta,Rajgopal Kannan,Viktor K. Prasanna
类目: Machine Learning (cs.LG)
*备注:
Abstract:Graph analytics generates irregular, topology-dependent memory-access streams via indirect traversals, limiting data prefetching. Accurate prediction enables timely prefetches that hide memory latency. However, ML predictors couple short histories with fixed horizons, fitting the training trace’s node layout rather than learning transferable graph patterns. We propose LoCoMAP, a long-context representation-learning framework for memory-access prediction with independently controlled observation history and prediction horizon. We evaluate Connected Components (CC), PageRank (PR), and Single-Source Shortest Paths (SSSP) on six synthetic graph families, with Erdos-Renyi (ER) and Grid as random and deterministic controls. We test two interventions: same-family regeneration changes connectivity, while node shuffling disrupts layout but preserves connectivity. Across 18 graph-workload pairs, extending history from 8 to 128 accesses improves F1 by 7.8%, while extending prediction horizon from 8 to 512 improves F1 by 88.3%. For CC, regeneration is less disruptive than shuffling: without retraining, long context retains 95.4% of original F1 and outperforms short context in six families. Under full shuffle, long context improves CC F1 by 33.6% across ER, Barabasi-Albert, Copying Model, and Stochastic Block Model (SBM), with a 2.06x gain on SBM, but not on Watts-Strogatz or Grid, where near-zero delta mass shifts into heavy tails beyond the predictor’s range. Long context outperforms short context on six real-world CC graphs. At full shuffle, long-context PR retains 57.6% of original F1 and averages 11.0% above short context. Both SSSP models retain about 30% of original F1 under full shuffle as distance updates access vertex data in a new order. Within these bounds, long-context predictors are robust across related graphs and layouts without retraining, enabling a generalizable graph prefetcher.
[LG-59] CBE: A Complexity-adaptive Boundary-aware Ensemble Method for Imbalanced Data
链接: https://arxiv.org/abs/2610.39466
作者: Xun Wei
类目: Machine Learning (cs.LG)
*备注: 12 pages, 5 figures
Abstract:In many real-world classification tasks, datasets exhibit class imbalance, which can degrade the generalization performance of conventional learning algorithms. Data complexity has recently been recognized as a more objective indicator of learning difficulty than the imbalance ratio alone, because class overlap and complex decision boundaries often impair classifiers more severely than imbalance itself. Samples near the decision boundary are inherently difficult to classify, which has motivated many researchers to focus on such instances to improve classifier performance. However, most under-sampling approaches fail to differentiate their sampling of boundary samples across datasets of varying complexity. To overcome this limitation, we propose a complexity-adaptive boundary-aware ensemble (CBE) framework that links the sampling weight distribution to data complexity and dynamically adjusts the boundary preference of under-sampling accordingly. We also introduce an overall complexity measure that combines metrics at multiple levels and use it to divide imbalanced datasets into three categories. Extensive experiments on 24 benchmark datasets from UCI and KEEL show that the proposed method outperforms many existing methods in terms of F-measure and G-mean on Easy and Median datasets, without degrading performance on hard cases.
[LG-60] Achieving Minimax-Optimal First-Order and Best-of-Both-Worlds Regret Simultaneously in Multi-Armed Bandits
链接: https://arxiv.org/abs/2610.39457
作者: Mengxiao Zhang
类目: Machine Learning (cs.LG)
*备注:
Abstract:We study first-order and best-of-both-worlds regret guarantees in multi-armed bandits. Pogodin and Lattimore (2019) proposed an algorithm that combines minimax-optimal worst-case regret with a first-order guarantee whose loss-dependent term is \mathcalO(\sqrtK L^\star\log T) , where T denotes the horizon, L^\star denotes the smallest expected cumulative loss among fixed arms, and K denotes the number of arms. However, a logarithmic gap remains between this loss-dependent term and the optimal first-order scale. In this paper, we remove this logarithmic factor by proposing an efficient and parameter-free algorithm that achieves the minimax optimal first-order regret guarantee \mathcalO(\sqrtK L^\star) . Technically, our algorithm is built upon FTRL with a Tsallis entropy regularizer (Zimmert and Seldin,2021), with a learning rate adapted to the learner’s cumulative loss and a sampling distribution determined by a scalar fixed point. This construction controls the combined estimation and stability cost in terms of the learner’s loss. Moreover, we show that the same algorithm achieves a best-of-both-worlds guarantee, with \mathcalO(\sum_i\ne i^\star\frac\log T\Delta_i) regret for i.i.d. losses with a unique best arm i^\star , where \Delta_i denotes the suboptimality gap of arm i .
[LG-61] Smooth Action Chunking Via Movement-Primitive Representations for Imitation Learning
链接: https://arxiv.org/abs/2610.39451
作者: Shota Nomura,Takayuki Osa,Yasuhiro Kato,Kazuki Ota,Naoki Shitanda,Tatsuya Harada
类目: Machine Learning (cs.LG); Robotics (cs.RO)
*备注:
Abstract:Imitation learning has become a widely adopted approach for vision-based robot policies, owing to its conceptual simplicity, scalability, and ability to leverage large-scale datasets. As these policies have grown more capable, action chunking has become their standard interface. At each decision step the policy outputs several future actions at once. Backbones from diffusion policies to large vision-language-action models all share it. Whatever backbone produces the chunk, its temporal structure is learned only implicitly. The commanded trajectory is therefore jerky both within a chunk and between successive chunks. On real hardware this appears as visible vibration and as repeated saturation of the follower’s velocity limit. We address this issue by incorporating a movement-primitive-based trajectory representation into action chunking. Specifically, a trajectory segment is represented as a linear combination of smooth basis functions, yielding smooth trajectories by construction while leaving the policy architecture, training objective, and inference procedure unchanged. The same representation also yields a closed-form, inference-time correction for the second source of jerk. We condition the predicted coefficients so that the new chunk passes through a via-point at the previous chunk’s last commanded action. This correction needs no retraining and, unlike asynchronous Real-Time Chunking, applies to policies queried in one shot. We instantiate it on a one-shot (ACT) and an iterative (DiT) backbone and evaluate in simulation and on a real bimanual ALOHA platform. The representation lowers within-chunk jerk by an order of magnitude on the one-shot backbone and by 2.4 times on the iterative one, with task success statistically non-inferior. The conditioning then removes 73-84% of the boundary discontinuity, at negligible per-step cost.
[LG-62] End-to-End Guarantees for Masked Diffusion Models under Conditional Mixing NEURIPS2026
链接: https://arxiv.org/abs/2610.39444
作者: Fareed Sheriff
类目: Machine Learning (cs.LG)
*备注: Accepted at the NeurIPS 2026 Workshop on Diffusion Language Models: Foundations, Efficiency, and Reasoning (DiffuLM). 26 pages, 2 figures, 6 tables
Abstract:Masked diffusion models train a denoiser on masked cross-entropy and generate by unmasking many positions in parallel. Existing MDM sampling bounds assume an exact denoiser or leave its error uncontrolled while training bounds cover only the masking distribution on which the model was trained. We present an end-to-end guarantee for a learned denoiser under conditional mixing (CM): stationary sources (e.g. floored order- r Markov and two-sided-Doeblin hidden-Markov sources) for which revealing distant visible tokens moves a masked token’s posterior on average exponentially little in the distance with correlation length \tau_c . We show the empirical risk minimizer trained under any masking distribution attains per-token excess \varepsilon’ on that distribution from \varepsilon’^-\mathcalO(\tau_c\log|\mathcalV|) training sequences, independent of sequence length N – one sample budget serves every decoding schedule when each schedule’s denoiser is trained under its own weighting. We further show that the exponent’s order \Theta(\tau_c\log|\mathcalV|) is minimax-optimal up to constants; more efficient rates require added structure. With this, we bound the end-to-end KL as \mathrmKL(\pi\Vert p_\mathrmalg)\leq \varepsilon+N\varepsilon’ under a dilated schedule of T=\mathcalO(\tau_c\log(N|\mathcalV|/\varepsilon)) denoising steps versus \Omega(N) steps for a fixed random order on an explicit CM source. Finally, we show that block-autoregressive samplers chain across blocks with no added error and that a \Lambda -looped order- r message-passing class on order- r sources achieves polynomial sample complexity, with approximation error improving per loop until saturation at \Lambda r\approx \tau_c\log(1/\varepsilon’) . Experiments on released and retrained models support our results’ qualitative predictions.
[LG-63] Softmax Routing Dynamics in Mixture-of-Experts Models: Collapse and Recovery
链接: https://arxiv.org/abs/2610.39442
作者: Shengze Xu,Maolin Deng,Tieyong Zeng
类目: Machine Learning (cs.LG)
*备注:
Abstract:Softmax routing in mixture-of-experts models can concentrate probability on a subset of experts, suppressing access to the others. We study how this concentration forms, why it persists, and how suppressed probabilities recover. For fixed experts, a stationary-loss barrier identifies the face supporting the optimal routed output. In a common-target square-loss model with freely trainable expert outputs, slow adaptation tracks fixed-expert dynamics on finite horizons, while every fixed positive adaptation rate yields an interior routing limit. An exact normal-velocity decomposition separates complementary-expert covariance from relative gradient advantage. Under positive two-sided power bounds along recovery paths, uniform over initial masses, the resulting coefficient’s order of vanishing gives logarithmic or power-law recovery times. Integrations of constructed fields reproduce these orders. An exact discrete log-odds identity relates neural updates to threshold crossing. In paired Transformer-MoE training, removing a class-conditional router-gradient contrast delays confirmed recovery and reduces accumulated log-odds on fixed tokens, both with and without capacity constraints.
[LG-64] LinQuant: Functional Invariances for Low-Precision Linear Recurrences
链接: https://arxiv.org/abs/2610.39418
作者: Neehal Tumma,Philipp Nazari,Rohan Cherukuri,Daniela Rus,T. Konstantin Rusch
类目: Machine Learning (cs.LG)
*备注: 35 pages, 16 figures. Code: this https URL
Abstract:Linear recurrences are emerging as practical alternatives to softmax attention in large language models, appearing in both standalone and hybrid architectures. However, despite their rapid emergence, low-precision inference remains much less developed for these models than for softmax attention. During prefill, chunkwise-parallel algorithms cast much of the recurrence as dense matrix multiplications, suggesting a natural path to low-precision execution. Simply quantizing these operations in their native form, however, can be fragile. In this work, we identify functional invariances that preserve the full-precision computation while making the operands more amenable to quantization. Building on this observation, we introduce LinQuant, a framework for low-precision recurrent prefill based on seven such invariances, including rotations, diagonal rescalings, and alternative placements of recurrent decay and write factors. We derive when these transformations remain valid across common recurrence families spanning linear, gated, and delta-rule updates, and show that they provide complementary degrees of freedom for improving low-precision accuracy. Finally, we translate this framework to production by augmenting SSDi8, an existing low-precision prefill pipeline for Mamba-2, with the applicable invariances. LinQuant improves perplexity at matched latency on Mamba-2 1.3B, 2.7B, and 8B and Nemotron-H 8B, demonstrating the practical utility of invariance-based quantization for recurrent models.
[LG-65] Accuracy is not compatibility: two Evo 2 checkpoints hinge on one FP8 matrix multiplication
链接: https://arxiv.org/abs/2610.39415
作者: Xun Ding(Independent researcher),Ziyuan Qin(Emory University)
类目: Machine Learning (cs.LG)
*备注: 18 pages, 3 figures, 1 table. Data: this https URL ; code: this https URL
Abstract:FP8 formats specify how values are stored, not how products are summed; this gap can break checkpoints. Evo 2, a DNA language model, documents a Hopper FP8 requirement for its 1B and 40B checkpoints. On two Blackwell (sm_120) GPUs both fail the developers’ forward test (the 7B passes); on an RTX 5070 Ti the 40B’s next-base accuracy is 25.7%, near guessing. Greater accuracy breaks them, as BioNeMo saw for the 40B on Hopper: natively, 99.996% of one projection’s outputs are correctly rounded (H100: about 30%), and fp32 accumulation of the same operands fails. Our H100 FP8 GEMM emulator, mmaemu (bit-exact on captured 1B projections), localises the dependence to the 1B’s first projection (of 21), where exact sums fail but pass when shrunk by the H100’s 1.4% (measured in-sample), not 2.1%. The 40B fails with only its first projection native. Emulating only that projection, weights unchanged, raises 40B BRCA1 variant-classification AUROC from 0.434 to 0.869 (H100: 0.868) on the RTX 5070 Ti, despite a narrow forward-test miss, and 0.481 to 0.867 on an RTX PRO 6000. We tested one model family but not training origin; releases should specify reference arithmetic.
[LG-66] Dobb: Shard-Local Preconditioning in Block-Diagonal Eigenbases
链接: https://arxiv.org/abs/2610.39412
作者: Shi-Yuk Fong,Shuhua Yu,Jiawei Zhang
类目: Machine Learning (cs.LG); Distributed, Parallel, and Cluster Computing (cs.DC); Optimization and Control (math.OC)
*备注: 33 pages, 3 figures
Abstract:Training language models whose parameters and optimizer state exceed a single node’s memory requires sharding these tensors across devices. Under tensor parallelism (TP) or fully sharded data parallelism (FSDP), matrix-aware optimizers such as Muon introduce communication to perform global matrix operations across shards. We propose Dobb (Distributed Optimizer with Block-diagonal Eigenbases), a matrix-aware optimizer designed to compute updates locally within parameter shards. Dobb approximates the input-activation second moment with diagonal blocks and tracks a complete basis within each block. It projects momentum into these bases, applies sign normalization to the coefficients, and reconstructs the update. When each block lies within one shard under TP or FSDP, Dobb applies its matrix update locally, requiring no communication beyond data-parallel synchronization of the activation second moments. We benchmark Dobb against Dion, a distributed matrix-aware optimizer designed to retain the benefits of orthonormalized updates while reducing communication. On GPT-2 XL (1.56B parameters), Dobb with 16 blocks reaches a validation loss of 2.476, compared with 2.511 for Dion at rank fraction 1/16. Under FSDP on 32 H100 GPUs, Dobb reduces the amortized optimizer-call time from 124.1 ms for Dion to 9.2 ms.
[LG-67] More Is Better? Exploring Latent Correspondence in Federated Hypergraph Learning
链接: https://arxiv.org/abs/2610.39362
作者: Bo Yan,Siming Zhang,Siqi Li,Hang Yu
类目: Machine Learning (cs.LG); Distributed, Parallel, and Cluster Computing (cs.DC)
*备注:
Abstract:Federated hypergraph learning with latent cross-client hyperedge correspondence poses fundamental identifiability and computational challenges, as the same local release can correspond to multiple global hypergraph structures. We characterize the tractability of recovering these latent correspondences, showing tractable pairwise assignment but generally intractable multi-client hyperedge matching. Building on this characterization, we derive a cross-entropy decomposition that reveals when local prediction can outperform predictors fitted with additional cross-client information. Motivated by this principle, we introduce Federated Adaptive Spectral Training with Representation Alignment (FedASTRA), which learns correspondence-invariant diffusion representations through adaptive heat diffusion and masked latent prediction while communicating model parameters only. We establish its asymptotic optimality with respect to the permitted local information and prove global population-risk convergence under explicit conditions on recovery, representation, and update alignment. Experiments on eight real-world datasets demonstrate the effectiveness of FedASTRA against identifier-free federated graph and hypergraph methods. On ogbn-papers100M, FedASTRA reaches the performance-communication Pareto frontier, achieving 50.07 Accuracy and 29.73 Macro-F1 while communicating only 59.60 MiB per round.
[LG-68] MulClip: A table-free activation function with monotonic and folding regimes for reservoir computing on microcontrollers
链接: https://arxiv.org/abs/2610.39343
作者: Guren Matsumura,Masato Takemoto,Shinzo Koyama
类目: Machine Learning (cs.LG); Neural and Evolutionary Computing (cs.NE)
*备注: 15 pages (7 pages main text, 8 pages appendices), 2 figures
Abstract:This letter proposes MulClip, a lightweight activation function for reservoir computing (RC) on microcontrollers (MCUs). In a ring reservoir, a representative sparse RC topology, multiply-accumulate operations are few, and evaluating tanh accounted for 75% of the state-update time on a Cortex-M4. MulClip introduces curvature into the response using only comparison, absolute value, multiplication, and subtraction, and places its clipping threshold not at the extremum but at the point where the response returns to zero after passing the extremum. As a result, a single function has both a monotonic regime and a nonmonotonic folding regime that increases the expressiveness of a single unit. Which regime is used is determined by the usual hyperparameter search, e.g., over the input scale, so no task-specific threshold tuning is required. The state-update time measured on hardware was 0.34 times that of tanh and shorter than that of any of the 11 evaluated tanh approximations. In PC-based evaluation, third-order MulClip achieved a 22% lower NMSE than tanh on NARMA-20, where the pre-activations remained in the monotonic regime, and on 8-bit parity it raised the accuracy from 0.76-0.78 for the monotonic-clip variant without folding to 0.97-0.98. For these two tasks, on-device inference yielded nearly the same values as on the PC.
[LG-69] Class-aware Latent Attention Replay (CLAR): Class Incremental Hazard Diagnosis for Resilient AI in the Manufacturing Industrial Internet
链接: https://arxiv.org/abs/2610.39323
作者: Avranshu Mahanty,Ran Jin,Yingyan Zeng
类目: Machine Learning (cs.LG)
*备注:
Abstract:Artificial Intelligence (AI) systems provide critical computation services in the Manufacturing Industrial Internet (MII), and their resilience depends on timely diagnosis of hazards across the data, AI pipeline, and cyber-physical layers. During operation, new hazard types keep emerging with only a few samples each, hazards co-occur, and the diagnosis model runs on memory-constrained computation nodes. Replay-based class-incremental learning (CIL) can learn new hazard classes while revisiting a small memory of past samples, but existing methods assume single-label, balanced streams, store high-dimensional raw samples, and select them independently of the classifier. We propose Class-aware Latent Attention Replay (CLAR), which combines a dual buffer that stores mostly compact latent representations with a few raw samples, an Attention Leverage Score (ALS) that upper-bounds the classifier logit and guides class-balanced reservoir sampling, and an exponential-moving-average teacher that slows latent-space drift. For an idealized buffer model, we prove that the buffer is class-balanced and follows an ALS-weighted class distribution, and that below a critical per-class capacity, ALS-guided replay has a lower replay-gradient error than class-balanced uniform replay. In the case study, On hazard diagnosis data from an MII testbed, CLAR achieves the highest mean average weighted F1 score among the evaluated replay methods with the same buffer size
[LG-70] BEVLight: Generalizable Traffic Signal Control from UAV Birds-Eye Views
链接: https://arxiv.org/abs/2610.39320
作者: Maonan Wang,Yuxin Cai,Junming Liu,Zhengyan Huang,Yi Yu,Ding Wang,Yifeng Zhang,Yirong Chen
类目: ystems and Control (eess.SY); Machine Learning (cs.LG)
*备注: 27 pages, 15 figures
Abstract:Learning-based traffic signal control (TSC) has achieved promising performance, but most methods rely on structured traffic states that must be estimated from sensor observations in real deployments. Visual TSC learns signal decisions directly from images, yet existing methods depend on fixed roadside cameras and remain tied to specific intersection configurations and visual domains. UAVs offer on-demand bird’s-eye-view (BEV) observations that jointly capture intersection topology and traffic dynamics, making them a promising sensing platform for visual TSC. We present BEVLight, a generalizable TSC framework that directly maps BEV observations to signal actions. BEVLight grounds visual features onto individual lanes and composes them along the junction topology into movement and phase representations, allowing a shared policy to score candidate phases across heterogeneous intersection layouts and signal plans. To train the policy, we develop a web-based virtual teleoperation platform and collect over 5K human control decisions together with their underlying traffic states. Simulator-provided traffic states first supervise lane- and movement-level representations, after which human phase selections and rankings supervise the phase scorer. Re-rendering the recorded trajectories under diverse appearances preserves both forms of supervision and encourages robustness to visual changes. Closed-loop experiments demonstrate zero-shot generalization across traffic demands, signal plans, and intersections: BEVLight reduces average waiting time by 7–9% relative to perception-to-control visual baselines and matches or outperforms structured-state RL controllers that receive ground-truth traffic states, achieving the lowest waiting time in 12 of 16 scenarios. An offline study further shows that BEVLight transfers to real UAV observations without target-domain adaptation.
[LG-71] axoMFM: A Taxonomy-Aware Foundation Model for the Human Gut Microbiome
链接: https://arxiv.org/abs/2610.39316
作者: Haochen Xue,Anastasia Godneva,Yulong Li,Qi Cao,Zhaodong Wu,Kaiyu Liang,Jionglong Su,Eran Segal,Imran Razzak
类目: Machine Learning (cs.LG)
*备注: 19 pages, 4 figures
Abstract:Existing gut microbiome foundation models typically use flat taxon representations, limiting their ability to exploit known taxonomic hierarchies. We introduce TaxoMFM, a microbiome foundation model that integrates taxonomic structure into self-supervised learning. TaxoMFM jointly learns abundance patterns across taxonomic ranks and relative proportions within groups through taxonomy-aware encoding and hierarchical compositional reconstruction. We further construct TaxoGut-HPP, comprising 100,806 gut microbiome profiles from 231 source cohorts and covering 40 classification and 8 regression tasks. TaxoMFM achieves competitive performance across disease, gut health, host trait, and liver ultrasound phenotype prediction. In cross-cohort zero-shot transfer, where target cohorts are excluded from both pretraining and downstream adaptation, TaxoMFM achieves AUROCs of 0.838, 0.736, and 0.722 for colorectal cancer, type 2 diabetes, and inflammatory bowel disease, respectively. Further analyses show that TaxoMFM recovers community changes across taxonomic scales, while preserving community information and producing consistent representations under database updates and species-vocabulary shifts. Our code is available at this https URL.
[LG-72] Multi-Solution Inverse Design of Pixelated Patch Antennas Using an Autoregressive Transformer
链接: https://arxiv.org/abs/2610.39300
作者: Young-Mok Yun,Seung-Won Shin,Jaeyong Chung,Kyung-Sik Choi
类目: Machine Learning (cs.LG)
*备注: 16 pages, 28 figures, 7 tables
Abstract:Deep-learning-based inverse design has emerged as a promising framework for pixelated patch antenna synthesis by enabling broader exploration of discrete geometry spaces and reducing design time compared with conventional template-based methods. Existing methods, including surrogate-assisted metaheuristic optimization, tandem inverse networks, and diffusion models, have advanced the field but still face practical limitations. Metaheuristic methods such as binary particle swarm optimization (BPSO) are sensitive to initialization and search hyperparameters and are often time-consuming, tandem networks have difficulty capturing the one-to-many nature of inverse antenna design, and diffusion models may require excessive complexity and data for moderate-scale binary layout synthesis. In this paper, an autoregressive (AR) transformer is presented for the inverse design of 10 x 10 pixelated patch antennas. Its sequential discrete generation process enables the synthesis of multiple feasible structures for a given target response. The pretrained model is further fine-tuned for dual-band design using step-shaped target responses with sharp spectral transitions, allowing intuitive band specification without manually constructing detailed S11 curves. Compared with BPSO, our proposed method generates valid and diverse antenna candidates more efficiently. Fabricated and measured examples confirm that distinct pixelated layouts can realize the desired antenna responses, thereby demonstrating the multi-solution capability of the proposed framework.
[LG-73] Strongly Adaptive Online Learning Achieves Uniform Discounted Regret
链接: https://arxiv.org/abs/2610.39292
作者: Sifan Yang,Wenhao Yang,Lijun Zhang
类目: Machine Learning (cs.LG)
*备注:
Abstract:Discounted regret measures the performance of an online learner by assigning exponentially decaying weights to past losses, reflecting the greater importance of recent losses in non-stationary online learning. While previous work obtains uniform discounted regret bounds over a continuous range of discount factors using a specialized meta-algorithm, its results are limited to convex functions. In this paper, we demonstrate that strongly adaptive regret bounds can directly imply uniform discounted regret guarantees for convex, strongly convex, and exp-concave loss functions. Our key observation is that discounted regret against a fixed comparator admits an exact decomposition into a convex combination of interval regrets. For convex functions, this reduction yields an O(\sqrt\log T/(1-\lambda)) discounted regret bound for all discount factors \lambda\in(0,1) , matching the previous rate over a broader range of discount factors. For strongly convex and exp-concave functions, we provide regret bounds of O\left(\log T\left(1+\log\frac11-\lambda\right)\right) .
[LG-74] IMRouter: Imbalance-aware Temporal Pattern Routing for LLM -based Time Series Forecasting
链接: https://arxiv.org/abs/2610.39246
作者: Xin Yang,Cai Xu,Haishun Chen,Ying Liu,Jie Zhao,Wei Zhao,Ziyu Guan
类目: Machine Learning (cs.LG)
*备注:
Abstract:Accurate time series forecasting (TSF) is essential for Web applications, such as traffic monitoring and online resource management. Recent advances in leveraging pre-trained large language models (LLMs) have shown promising performance in TSF. However, existing LLM-based TSF methods largely optimize the overall forecasting objective while overlooking temporal pattern imbalance (TPI), where different temporal patterns occur at highly unequal frequencies. Our empirical analysis reveals that TPI is prevalent across widely used TSF benchmarks, with substantially higher forecasting errors on less frequent patterns. To address this issue, we propose IMRouter (Imbalance-aware Temporal Pattern Routing), a novel framework that mitigates TPI-induced biases in both temporal representation learning and cross-modal alignment. Specifically, IMRouter introduces Pattern-Adaptive Temporal Representation Learning (PATR), which leverages gradient conflict and local gradient density to guide expert specialization. By penalizing incompatible expert assignments with density-aware weighting, PATR reduces the dominance of frequent optimization signals in router training. Furthermore, Routing-Guided Semantic Alignment (RGSA) transfers the pattern-specific knowledge learned by the router to cross-modal alignment through mutual information maximization, preserving distinctive temporal information across different patterns. Extensive experiments on benchmark datasets demonstrate the effectiveness of IMRouter, particularly in improving forecasting performance on less frequent temporal patterns.
[LG-75] Event2Reward: Learning Robotic Rewards from Progress Relations
链接: https://arxiv.org/abs/2610.39212
作者: Junnan Nie,Zitian Fan,Jiachen Zhang,Junyi Lao,Likui Zhang,Chenghao Liu,Xiaoyang Lu,Yunbo Wang,Yixue Zhang,Linbo Zhai,Jiayi Li,Mengzhe Wang,Xichang Wu,Bin Wang,Chang Liu,Tianle Zhang,Songfang Huang
类目: Robotics (cs.RO); Machine Learning (cs.LG)
*备注: 44 pages, including references and appendices
Abstract:Learning task progress from robot demonstrations provides a practical way to construct dense shaping rewards for robotic reinforcement learning. A common approach uses normalized demonstration time as progress labels, implicitly assuming monotonically increasing task progress. During policy exploration, however, actual task progress can stall or even regress. Such models may mistake regressions for progress, assigning positive shaping rewards to actions that cause progress loss. Real suboptimal and failed executions offer corrective experience, but assigning absolute progress values to their states is difficult. Our key observation is that stalls, regressions, and recoveries establish local relative progress relations even when absolute progress is unknown. We introduce Event2Reward, which uses normalized temporal labels to anchor the progress scale and these within-trajectory relations to constrain local changes, combining absolute and relative progress supervision. Its progress predictions serve as potentials for reward shaping. We curate an event-annotated dataset from approximately 100K raw real-robot execution records, covering successful, suboptimal, and failed executions, without requiring framewise absolute progress labels. Under the same training budgets, Event2Reward improves average policy success by 18.16 percentage points over a baseline using only sparse task rewards across 40 tasks in four LIBERO suites. On real robots, average success after 30 minutes of wall-clock training reaches 60.0% on three precise insertion tasks and 90.0% on three object manipulation tasks, exceeding the sparse-reward baseline under the same budget by 36.7 and 23.3 percentage points, respectively. Ablation studies further show that absolute progress supervision and relative progress relations are complementary, with joint supervision yielding better downstream policy learning.
[LG-76] Sparse Autoencoders as Implicit Diversification Regularisers for Cross-Sectional Cryptocurrency Returns
链接: https://arxiv.org/abs/2610.39205
作者: Weijia Han,Lisha Qu
类目: Machine Learning (cs.LG); Portfolio Management (q-fin.PM); Statistical Finance (q-fin.ST)
*备注: 8 pages, 4 figures. Accepted at the 7th ACM International Conference on AI in Finance (ICAIF '26)
Abstract:Factor models are central to asset pricing and portfolio construction, but neural factors are often difficult to interpret, and their effect on portfolio risk is unclear. A sparse autoencoder (SAE) may address both concerns by describing daily returns across many assets with only a few active features. We ask whether this sparse representation produces stable financial factors and better-diversified portfolios. We study cryptocurrency returns, use equities as a secondary test, and compare the SAE with common benchmarks in a time-ordered evaluation. We find consistently lower volatility in SAE portfolios and link this result to lower correlations among the selected assets. However, the SAE does not outperform a simple contrarian strategy on risk-adjusted return, and its factor identities are not stable across training runs. Our contribution is to show that sparse coding can support portfolio diversification, while clearly separating this risk effect from claims about alpha or stable factor identities.
[LG-77] Pair-Dependent Judge Errors Change the Minimax Rate of Semantic Entropy
链接: https://arxiv.org/abs/2610.39196
作者: Jiamiao Liu,Dewen Qiao,Yu Zhang,Xuetao Chen
类目: Information Theory (cs.IT); Machine Learning (cs.LG)
*备注: 47 pages, 5 figures, 3 tables. Code: this https URL
Abstract:Semantic entropy, the entropy of the meanings among a language model’s sampled answers, is estimated from the verdicts of a judge that declares pairs of answers equivalent or not, and automatic judges err in ways that depend on the answers they compare. The existing minimax theory covers only judges whose false-equivalence probability is the same for all pairs. We model a known judge that declares two answers with meanings a and b equivalent with probability q+\Delta M_ab , for a symmetric matrix M and q bounded away from 0 and 1 , and show that for three meanings the rate is set by tree statistics on four answers, while on an open set of channels the consistency threshold is set by trees on five. For every channel in an explicit open class, the minimax squared risk is of order N^-1/3 with N=n^4\Delta^6 on polynomial weak windows on which n^5\Delta^8 exceeds a multiple of \log n , and of order \min\1,N^-1/3\ on every polynomial weak window when the tree functions on at most four vertices are injective. The exponent comes from a cubic degeneracy at the equitable point. On an open set of channels in the class, the four-vertex tree functions take equal values at two distributions with different entropies. For such a channel, on every polynomial weak window, the minimax risk is of the order of the sum of \min\1,N^-1/3\ and a term that decays exponentially in n^5\Delta^8 , so no estimator is consistent while n^5\Delta^8 stays bounded, even as N\to\infty . For \Delta=n^-\beta with \beta1/2 , channels of the class with injective four-vertex tree functions allow consistent estimation exactly for \beta2/3 , and the channels of this open set exactly for \beta5/8 , at which the risk exponent drops from 1/12 to 0 . Computer-assisted certificates show that both kinds of channels occur on open sets.
[LG-78] Analyzing Condensation on Patterned Surfaces via Deep Learning
链接: https://arxiv.org/abs/2610.39195
作者: Tae E. Cooper,Michael Gee,Sheila Whitman,Jemma Dengel,Anthony Katselas,Duygu Dengiz,Frank Keutsch,Joanna Aizenberg
类目: Machine Learning (cs.LG)
*备注: 13 pages, 4 figures
Abstract:Dropwise condensation is key to power and water systems. State-of-the-art condensation surfaces are often inhomogeneous, either patterned by design or random defects. Inhomogeneous surface features dictate the rates of each stage of dropwise condensation-nucleation, growth, coalescence, and shedding. Characterizing these rates requires the precise detection of thousands of droplets. Yet, existing segmentation methods are unable to detect droplets on inhomogeneous surfaces. We developed TileSAM3, a universal droplet segmentation tool built on the Segment Anything Model 3 (SAM3). Hierarchical tiling with per-tile inference followed by post-processing extended SAM3’s detection cap and raised the effective resolution of small droplets. We created two datasets: High Quality Droplet Dataset (N=365,826; HQDD) with homogeneous surfaces and Low Quality Droplet Dataset (N=63,577; LQDD) with patterned, defect-laden surfaces under adverse lighting. TileSAM3 closely reproduced thresholding annotations on HQDD, surpassed it over sixfold on LQDD (Average Precision 0.57 versus 0.088), and outperformed other deep learning models on all metrics evaluated. Combined with KD-tree tracking, TileSAM3 identified localized nucleation and coalescence in narrow active bands on a stiffness-patterned surface. TileSAM3 enables analysis of understudied inhomogeneous substrates with poor lighting and videography which facilitates faster, lower cost research of condensation materials.
[LG-79] ACAM-J-MARS: Multi-Atlas Riemannian Spectral Representations for Decoding Advanced Meditation States from fMRI
链接: https://arxiv.org/abs/2610.39171
作者: Jian Wang,Winson F.Z. Yang,Wugeng Zheng,Matthew D. Sacchet
类目: Machine Learning (cs.LG)
*备注: 4 pages; 4 figures
Abstract:Inferring jhana-type advanced concentration absorption meditation (ACAM-J) states from fMRI data is challenging due to the spatially distributed nature of meditation-related effects, substantial inter-individual variability that can exceed within-subject state differences, and limited sample sizes. Conventional machine-learning approaches often formulate this problem as unimodal, Euclidean, atlas-based classification, thereby neglecting the non-Euclidean geometry of functional connectivity matrices and underrepresenting temporal information in functional magnetic resonance imaging (fMRI); these limitations can lead to suboptimal classification performance. We propose ACAM-J-MARS (\textbfMulti-\textbfAtlas \textbfRiemannian \textbfSpectral representation), which computes functional connectivity using multiple atlases, represents symmetric positive definite (SPD) connectivity matrices on a Riemannian manifold, projects them into a log-Euclidean tangent space, and augments connectivity-derived features with low-frequency BOLD spectral energy to better resolve subtle state differences under heterogeneous variability. Evaluated on 204 segments from 22 practitioners, ACAM-J-MARS improves performance across multiple backbone models, achieving a balanced accuracy of 0.757 for early versus late ACAM-J classification and 0.545 for three-class classification, corresponding to a 10.8-point absolute improvement over the strongest baseline. These findings suggest that the proposed representation improves the decoding of advanced meditative brain states and may enable downstream applications such as closed-loop neurofeedback and real-time psychophysiological brain-computer interfaces.
[LG-80] Learning Erasing Equal-Length Relational Pattern Languages in the Limit from Positive Data over Binary Alphabets
链接: https://arxiv.org/abs/2610.39156
作者: S. Mahmoud Mousawi,Sandra Zilles
类目: Logic in Computer Science (cs.LO); Formal Languages and Automata Theory (cs.FL); Machine Learning (cs.LG)
*备注:
Abstract:Relational pattern languages generalize Angluin’s pattern languages by requiring the strings substituted for repeated occurrences of a variable to stand in a given relation, rather than to be identical. We study erasing pattern languages under the equal-length relation, where all occurrences of a variable must be replaced by strings of the same length. Over alphabets with at least three letters, this class is known to be learnable in the limit from positive data. We show that over a binary alphabet it is not, which resolves an open question from earlier work. Our proof reveals that non-learnability already holds for the restricted subclass of languages generated by patterns that have exactly three maximal substrings of variables. We then show that patterns with at most two maximal substrings of variables are learnable in the limit from positive data for every alphabet of size at least two. If the target pattern is known to have at most one maximal substring of variables, this can even be achieved by an iterative learner, which updates its hypothesis using only the current example and its previous hypothesis.
[LG-81] DICE: Decentralized Incentive-driven Federated Learning for Efficient Scalable Customization of Pretrained Models in Web3
链接: https://arxiv.org/abs/2610.39132
作者: Hong Lin,Zhongle Xie,Lidan Shou,Ke Chen,Gang Chen
类目: Cryptography and Security (cs.CR); Distributed, Parallel, and Cluster Computing (cs.DC); Machine Learning (cs.LG)
*备注:
Abstract:Customizing Pretrained Models (PMs) for downstream tasks demands task-specific data, which burgeoning Web3 applications provide abundantly. Blockchain-powered federated learning (BCFL) offers a secure and privacy-preserving paradigm to customize PMs among self-interested and untrusted Web3 participants. Yet, existing BCFL algorithms face severe efficiency and scalability challenges when applied to large PMs, due to their substantial computational and memory requirements and the hardware constraints of many participants. This paper introduces DICE, a novel BCFL algorithm that addresses the efficiency and scalability challenges of customizing large PMs in Web3 via two sets of synergistic techniques: (i) a dynamic progressive learning scheme and a model slice verification scheme to enable a wider range of honest participants, including those unable to deploy the full model, to engage in customizing PMs, while preserving model utility; (ii) a tree-structured committee consensus protocol and an efficiency-driven incentive scheme to address the dominant bottleneck that hinders existing BCFL algorithms from scaling to a larger number of data holders for customizing data-hungry large PMs, while maintaining security. Theoretical analysis and extensive experimental results on customizing vision and language PMs demonstrate DICE’s efficiency, effectiveness, and security.
[LG-82] Uncertainty-Guided Cross-Modal Alignment for Diffusion-Based IoT Intrusion Detection NEURIPS
链接: https://arxiv.org/abs/2610.39117
作者: Mahdi Arab Loodaricheh,Anita Raja,Mohammad Hossein Manshaei
类目: Machine Learning (cs.LG)
*备注: 7 pages, 1 figure, 2 tables. Accepted at NeurIPS DynaFront 2026
Abstract:Diffusion models evolve noisy states toward a learned data distribution, so their sampling behavior depends on the geometry of the representation in which denoising occurs. This dependence is especially important for multimodal data: independently trained encoders can map synchronized views to incompatible latent regions, while uniform alignment can let an uncertain view distort the shared geometry. To address this gap, we introduce a novel reliability-guided alignment rule that uses normalized VAE posterior variance to control directional cross-modal influence before diffusion training. We study this issue in minority-attack generation for imbalanced Internet-of-Things intrusion detection. Tabular flow attributes, a DeepInsight image, and a packet sequence are encoded by modality-specific variational autoencoders. Per-modality normalized VAE posterior variances provide bounded, sample-specific weighting signals in an asymmetric stop-gradient alignment objective. We use these quantities as heuristic encoder-derived reliability proxies rather than calibrated probabilistic uncertainties. The full modality-specific latent means obtained after alignment fine-tuning are standardized and modeled jointly by a class-conditional EDM; a second-order sampler then generates paired attack-conditioned latent blocks for the three decoders. In controlled experiments on TON-IoT, uncertainty-aware alignment obtains the highest mean precision, recall, density, and coverage among three diffusion variants. Augmenting only the training split yields ROC-AUC 0.9515, PR-AUC 0.8954, balanced accuracy 0.8395, and attack-class F1 0.7940 on an untouched real test set. Among the two variants with matched alignment fine-tuning, posterior-variance-based weighting improves both tabular generative fidelity and downstream detection relative to uniform weighting.
[LG-83] Closed Gates Leak: Decomposing Adjacent-Layer Cancellation in Gated Language Models
链接: https://arxiv.org/abs/2610.39068
作者: Harvey Dam,Ganesh Gopalakrishnan
类目: Machine Learning (cs.LG)
*备注:
Abstract:In transformer language models with gated multilayer perceptrons (MLPs), adjacent layers write partly against each other into the residual stream. Previous work interpreted this cancellation as layers proposing features and the next layer selectively rejecting the inappropriate ones. We decompose the cancellation by removing the previous layer’s write from the input of the next MLP: what the next layer writes along that write splits into what it writes anyway and how much less its MLP writes because the write is present, which we call the damping, and we trace the damping to the neurons that carry it. In gated MLPs the damping is carried by neurons whose gates are closed. SiLU and GELU are slightly negative below zero, so a closed gate still passes a small negative multiple of what the neuron reads, and a neuron whose output weight points along its input weight then subtracts a fraction of the direction it read. Summed over such neurons, closed gates contract the average direction a gated MLP reads, and the previous layer’s write is not singled out: the same layer’s write at another token is damped as much or more, and the eigenvalue signature from which that work concluded that the correction is selective appears for another token’s write as well. Over training the damping grows while what the next layer writes anyway stays flat, and cancellation begins when the damping overtakes it; annealing the negative side of the gate activation away in continued pretraining removes the damping and the cancellation. Adjacent-layer cancellation is therefore weak evidence about which features a model rejects: it is produced by a contraction that gated MLPs apply, on average over directions, to whatever they read.
[LG-84] BASILISC: Bayesian Alignment of Siamese-Integrated Latents for Inferring Single-cell Correspondence
链接: https://arxiv.org/abs/2610.39050
作者: Sepideh Saran,Mahsa Ghanbari,Uwe Ohler
类目: Machine Learning (cs.LG)
*备注: 12 pages main text, 3 figures, 1 table; supplementary material with 10 figures and 10 tables (38 pages total). Code to be released on publication. Currently under review
Abstract:Single-cell assays study the genome one cell at a time. Unimodal assays measure one molecular layer, the newer multimodal assays measure several. Most data are single-modality, so integrating them relies on models trained on multimodal assays, where the cell correspondence across modalities is known. These models return a matching, but rarely a confidence measure. We present BASILISC (Bayesian Alignment of Siamese-Integrated Latents for Inferring Single-cell Correspondence), a framework for matching single-cell representations with uncertainty on the correspondence itself. A BASILISC model is a probabilistic siamese network over the two modalities with a predictive distribution over the correspondence for a cross-modal pair of cells. We instantiate it at two levels: at the cell-type level the question is whether two cells share a cell type, at the cell level which cell is the partner in the other modality. To our knowledge, this is the first uncertainty score on a learned cross-modal correspondence evaluated against a known ground-truth pairing. BASILISC learns a similarity across modalities that reflects cell lineage: an unseen cell type is matched to its lineage relative, matching still works on such held-out types (balanced accuracy 0.86 on unseen against 0.93 on seen types), and the predictive uncertainty rises on these out-of-distribution cells. The uncertainty score also selects high-quality matches for downstream analysis: discarding the least certain fifth of the pairs raises accuracy from 0.93 to 0.98. At the cell level, a deep ensemble of the deterministic matcher reaches 0.062 matching accuracy on the benchmark test set, above the competition winner; on the Bayesian matcher’s assignment, the posterior flags the wrong matches that cross a cell-type boundary better than a deterministic read of the same scores (AUROC 0.68 against 0.64).
[LG-85] Exponential-Nemytskii Neural Operators: Learning Field Evolution Across Step Sizes
链接: https://arxiv.org/abs/2610.39041
作者: Jingye Li,Jinglai Li
类目: Machine Learning (cs.LG); Numerical Analysis (math.NA)
*备注: 22 pages, 1 figure
Abstract:In variable-step operator learning, agreement at observed step sizes does not by itself determine predictions at other temporal resolutions. We propose an operator whose step-size dependence is prescribed by a first-order Duhamel approximation of semilinear dynamics. It combines exact linear propagation with a local nonlinear function of the field and its derivatives, allowing evaluation below the observation interval without supervision densely sampled across step sizes. Under suitable conditions and at fixed spatial resolution, we prove that agreement between models at two step sizes controls their prediction differences throughout a prescribed range. We also give conditions for local identifiability of both components from two step sizes, supporting interpretation of the learned structure. A latent extension applies the same construction to learned fields. We evaluate the base model on two synthetic problems and the extension on two real datasets against several baselines.
[LG-86] FairSplit: Decomposing the Embedding Space for Fair Classification NEURIPS2026
链接: https://arxiv.org/abs/2610.39031
作者: Andrii Shkabrii,Anna Beer,Ylli Sadikaj,Claudia Plant
类目: Machine Learning (cs.LG)
*备注: Accepted at NeurIPS 2026
Abstract:Fair classification faces a twofold challenge: maintaining high classification accuracy while limiting reliance on sensitive information. Our method FairSplit achieves this by learning a decomposition of the embedding space into a Fair Space used for prediction and a Sensitive Space that isolates sensitive information. During training, the model must simultaneously learn the classification task and separate fair from sensitive information. The Sensitive Space gives sensitive attributes a destination, so the Fair Space can be used for prediction without relying on them. This decomposition is guided via a learnable mask and the Hilbert-Schmidt Independence Criterion (HSIC) as a measure of statistical dependence. We provide theoretical guarantees showing that the demographic parity gap is upper-bounded by both (i) the learned dimensionality of the Fair Space and (ii) the HSIC between the Fair Space and sensitive attributes. FairSplit effectively handles multiclass tasks and non-binary sensitive attributes, overcoming key limitations of existing fair classification methods. Experiments on established fairness benchmarks show that FairSplit achieves competitive accuracy-fairness trade-offs. Code is available at this https URL.
[LG-87] MCCL: Training Pytorch models across a fleet of Macs at Skywalker Sound
链接: https://arxiv.org/abs/2610.39029
作者: Kevin Sasso,Chris Fiegel,Francis Aitken,Nicolas Tsingos,Nicholas Tong
类目: Distributed, Parallel, and Cluster Computing (cs.DC); Machine Learning (cs.LG)
*备注:
Abstract:Skywalker Sound, a premier audio post-production facility, operates more than 200 M4 Max Mac Studios on a 10Gb Ethernet network. When not in use by artists, these machines sit idle, representing massive untapped unified memory and compute capacity. While these fleets are excellent candidates for distributed machine learning alongside existing NVIDIA GPU clusters, PyTorch lacks an accelerated distributed backend for Apple’s Metal Performance Shaders (MPS). To bridge this gap, we introduce the Metal Communication Collective Library (MCCL), the first open-source MPS backend for multi-node PyTorch training over switched Ethernet. We evaluate its performance and scalability across 24 nodes, demonstrating that large all-reduce operations reach 98-99.7% of TCP goodput. Training a 113.5M-parameter Vision Transformer (ViT) scales from 198 to 4,568 samples/s, achieving 95.9% parallel efficiency. Benchmarked against an NVIDIA A100 GPU, one M4 Max Mac equals roughly 1.7x the throughput of FP32 matrix multiplication, and 5.5 Macs match an A100 using TF32. Harnessing 100 idle Macs for 12 hours nightly offsets approximately 200,000 annually in equivalent cloud compute costs. Our results demonstrate that existing Mac infrastructure can be seamlessly repurposed for high performance distributed training without abandoning the standard PyTorch codebase.
[LG-88] Self-Supervised Doppler-Guided RF Odometry NEURIPS2026
链接: https://arxiv.org/abs/2610.39005
作者: Kunzhe Song,Jialuo Du,Jingkai Lin,Huacheng Zeng
类目: Robotics (cs.RO); Machine Learning (cs.LG)
*备注: Accepted at NeurIPS 2026
Abstract:Radio-frequency (RF) sensing enables robot ego-motion estimation in darkness, smoke, and fog, yet RF odometry remains challenging because multipath and moving reflectors make cross-frame matching highly ambiguous. Existing methods often rely on auxiliary inertial sensors or ground-truth pose supervision. We present RFWalk, a self-supervised, RF-only odometry framework that models dense correspondences across range-azimuth frames as random walks on a space-time graph and learns them from unlabeled sequences through temporal cycle consistency. To rule out correspondences that are cycle-consistent but physically implausible, we introduce a Doppler consistency loss that constrains correspondence-induced motion using radial velocity measured by the RF sensor itself. Ego-motion is then recovered through robust \mathrmSE(2) registration that suppresses uncertain correspondences and those arising from independently moving objects. Without pose labels or auxiliary sensors, RFWalk remains competitive with supervised baselines in static scenes and reduces absolute trajectory error by 58% in dynamic scenes with moving pedestrians.
[LG-89] he Utility of Physics-Informed Machine Learning in Solar Power Production Forecasting
链接: https://arxiv.org/abs/2610.38976
作者: Peter Pavlík,Franz M. Rohrhofer,Bernhard C. Geiger
类目: Machine Learning (cs.LG)
*备注: Submitted to Solar Energy
Abstract:Physics-informed machine learning (PIML) and hybrid modeling have attracted significant attention as approaches for combining physical knowledge with the flexibility of deep learning. In this work, we investigate whether these approaches deliver their commonly claimed benefits for large-scale photovoltaic (PV) power forecasting. We develop two hybrid forecasting architectures that integrate the ADR physical PV efficiency model with the TSMixer neural forecasting architecture through feature augmentation and residual learning. The proposed models are evaluated on a dataset comprising over 3 million samples from 3,713 PV installations across 33 European countries. We assess convergence speed, forecasting accuracy, transferability across geographical regions, and data efficiency. Although the physics-informed models seem to exhibit slightly faster convergence during training, they do not achieve consistent improvements in test-set accuracy, generalization, or performance under limited training data compared to the purely data-driven baseline. These findings suggest that, for modern deep forecasting architectures trained on large datasets, explicitly incorporating the ADR physical model provides limited practical benefit for day-ahead PV forecasting.
[LG-90] Communication-Efficient Clustered Federated Learning in Non-IID Wireless Networks
链接: https://arxiv.org/abs/2610.38970
作者: Martin Isaksson,Rickard Cöster,Šarūnas Girdzijauskas,Seif Haridi
类目: Machine Learning (cs.LG)
*备注: 64 pages (25 main text + appendices)
Abstract:Current and emerging mobile networks face increasing privacy, scalability, and performance challenges as distributed learning becomes essential for advanced 6G services. These challenges are amplified by the presence of non-IID data, naturally arising in heterogeneous wireless environments, that degrade the performance of standard Federated Learning (FL) approaches. To address non-IID data across clients, we propose a Clustered Federated Learning (CFL) framework that assigns clients to specialized models during training. Data-domain clustering initializes model-based clustering, enabling identification and exploitation of client similarities without sharing raw data while reducing communication cost. Motivated by mmWave beam selection, we evaluate our methods on a wireless simulation dataset (DeepMIMO) and complement it with two non-IID benchmarks (FEMNIST, Rotated CIFAR-10) to isolate different non-IID characteristics. Our best configurations improve accuracy over IFCA by 4.8 pp on DeepMIMO and 10.6 pp on FEMNIST while reducing downlink data volume by 83.1 % on FEMNIST compared to IFCA. We demonstrate that different clustering strategies suit different non-IID characteristics. These results show that our CFL methods enable resource-efficient, and high-performance distributed learning in future 6G networks. Comments: 64 pages (25 main text + appendices) Subjects: Machine Learning (cs.LG) ACMclasses: I.2.6; I.2.11; I.5.3; C.2.1 Cite as: arXiv:2610.38970 [cs.LG] (or arXiv:2610.38970v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.38970 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-91] Planned Restlessness: Delayed Meta Bandits with Budgets
链接: https://arxiv.org/abs/2610.38955
作者: Sudipto Guha
类目: Machine Learning (cs.LG); Data Structures and Algorithms (cs.DS)
*备注:
Abstract:We study episodic finite-horizon Bayesian bandits, or meta-bandits, with independent arms. Each episode draws fresh arm qualities from unknown recurring priors. Observations arrive after known delays, and plays obey hard per-episode caps and total budgets. A play earns reward and informs both the current quality and the recurring prior: the learner must explore, use, and generalize. We formulate a common planning optimization over learning trajectories. Its LP relaxation values experiments through current reward, resource consumption and continuation: the value of future decisions from the learning state produced. Dual prices separate the arms into local programs and certify every admissible policy, including hierarchical Thompson sampling. Execution enforces the hard constraints while preserving the promised learning transitions or accounting for their loss. The framework also represents risk-sensitive objectives and prices purchased observations and buyout options. Delayed reports create planned restlessness: an arm’s information changes while it is not played, through reports of its own earlier plays arriving on a known calendar. For stationary priors, finite reward alphabets, bounded report lag and nonbinding budgets, posterior sampling over the recurring priors bounds the price of not knowing them by O(\sqrtN) over N episodes, with other parameters fixed. For one undelayed T -step episode with arms capped at T/X plays each, we obtain the LP-relative approximation factor 1+(2\pi X)^-1/2+O(X^-1) , improving previous factors, with a matching lower bound for the leading term. Subjects: Machine Learning (cs.LG); Data Structures and Algorithms (cs.DS) Cite as: arXiv:2610.38955 [cs.LG] (or arXiv:2610.38955v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.38955 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-92] Structural-Test Geometry as an Inductive Bias for Partially Observable World Models
链接: https://arxiv.org/abs/2610.38903
作者: Todd Y. Zhou
类目: Machine Learning (cs.LG)
*备注:
Abstract:Predictive world-model objectives need not organize latent histories according to distinctions required for later control. With privileged simulator access during training, should response-derived structural information be predicted, decoded, or imposed directly on latent geometry–and how do these choices compare with structured belief distillation? We define posterior-marginalized structural tests whose response distributions induce a pseudometric over histories, then train quotient-preserving (QP) world models by aligning shared latent Euclidean distances to that metric. Across a fixed six-task diagnostic suite, the largest differences appear under held-out structural variation: QP improves planning success by 26.1 points [21.5,30.7] over metric decoding and 6.4 points [3.8,9.0] over structured belief distillation. On GridWorld, metric decoding accurately recovers held-out structural distances ( \rho=.91 ) while its native latent geometry remains more distorted than QP’s (paired IQD difference .22 [.20,.24] ), showing that relational recoverability and direct geometric alignment can diverge.
[LG-93] Geometric Time Score Estimators: Bridging Self-Predictive Representations and Generative Occupancy Models For Reinforcement Learning
链接: https://arxiv.org/abs/2610.38901
作者: Aravind Venugopal,Grace Liu,Jeff Schneider,Benjamin Eysenbach
类目: Machine Learning (cs.LG)
*备注:
Abstract:In reinforcement learning (RL), temporal reasoning is essential for estimating the density of future outcomes under a policy’s occupancy measure. While self-predictive representations have been a dominant paradigm for occupancy density estimation, we find that they yield surprisingly poor estimates. On the other hand, generative occupancy models can produce remarkably accurate samples from the occupancy measure. Yet, they cannot easily recover occupancy densities, limiting their use for control despite their expressivity. In this paper, we propose Geometric Time Score Estimators (GTSE), a framework that bridges the two paradigms by learning occupancy density estimators along probability paths. We analyze how GTSE implicitly induces better occupancy estimators through improved representation learning and show how to directly extract self-predictive representations from this framework. Through experiments across several locomotion and manipulation tasks, we show that GTSE improves performance over state-of-the-art methods by 2\times to 6\times in goal-conditioned RL settings, and that the representations learned using our framework transfer effectively to downstream RL tasks. Website: this https URL
[LG-94] Discovery of Diverse and Realistic Financial Tail-Risk Using Generative Flow Networks
链接: https://arxiv.org/abs/2610.38838
作者: Adir Serruya,Gilad Katz
类目: Machine Learning (cs.LG)
*备注:
Abstract:Accurate modeling of rare, high-impact financial events is essential for robust risk management. Sampling-based techniques such as Sequential Markov Chain Monte Carlo (SMCMC) generate diverse scenarios but inefficiently explore high-risk regions, while Deep Reinforcement Learning (DRL) converges on narrow sets of scenarios, limiting diversity. We present GRID, which uses Generative Flow Networks (GFlowNets) – samplers that draw objects in proportion to a reward – to systematically generate diverse and realistic financial tail-risk scenarios. Building on continuous GFlowNets, GRID introduces a flexible mixture-policy architecture with variable-specific distribution families for high-dimensional, sequential macro-financial data, together with plausibility constraints that keep generated trajectories realistic. A central practical advantage is robustness to reward-proxy imperfection: because GFlowNets sample proportionally rather than maximize, they remain broadly exploratory even when the oracle’s out-of-sample accuracy is low – an unavoidable reality in financial forecasting. On three real-world crises, GRID improves scenario diversity, realism, and tail-risk coverage over RL and MCMC baselines, and ablations over oracle architectures and distribution families yield practical insights into continuous GFlowNet behavior under noisy rewards.
[LG-95] Causal Attention on CPU: A Cache-Aware and Load-Balanced Scheduling Kernel for Prefill
链接: https://arxiv.org/abs/2610.38809
作者: Kwun Fung Lau,Frost Mitchell,Kanya Mo,Emani Ashok
类目: Machine Learning (cs.LG)
*备注: Preprint
Abstract:Agentic AI drives significant productivity gains across diverse domains, but concurrent long-context prefills from multiple agents can saturate GPU compute resources. Although CPUs offer a viable alternative, existing CPU attention kernels fall short in adapting to multi-tier cache hierarchies and lack effective thread scheduling. To address this, we introduce CACPU, a high-performance attention kernel featuring tiered cache blocking and an in-kernel task dispatching mechanism designed for causal full-sequence prefilling. CACPU achieves up to a 2.01x speedup over PyTorch SDPA, FlexAttention, and SGLang attention kernels. Integrated into SGLang, it reduces Time to First Token (TTFT) of Llama-3.1-8B by up to 1.74x while preserving model accuracy.
[LG-96] rustworthy Distributed Machine Learning in Cloud-Edge-IoT Systems: A Heterogeneity-Aware Survey of Security Privacy and Robustness
链接: https://arxiv.org/abs/2610.38793
作者: Long Dang,Ali Alfatemi,Kaiqi Xiong,Mohammed Aledhari,Abdellah Chehri,Shawqi Al-Maliki,Mohamed Rahouti
类目: Cryptography and Security (cs.CR); Distributed, Parallel, and Cluster Computing (cs.DC); Machine Learning (cs.LG)
*备注:
Abstract:Distributed machine learning (DML), including conventional distributed training and federated learning (FL), enables collaborative model development across cloud-edge-IoT systems without fully centralizing raw data. However, distributing data, computation, updates, and orchestration increases the attack surface and introduces coupled challenges across privacy, robustness, reliability, fairness, and governance. This survey presents a security-centered, heterogeneity-aware roadmap for trustworthy DML, with FL treated as a prominent high-risk case. We first characterize the assets, adversaries, trust assumptions, and coordination patterns that define DML/FL security boundaries. We then analyze how statistical, system, and trust heterogeneity, including non-IID data, resource asymmetry, intermittent participation, and cross-silo/cross-device trust differences, affect convergence, fairness, attack detectability, and defense reliability. Next, we synthesize attack-defense interactions across poisoning, backdoors, Byzantine manipulation, privacy leakage, model extraction, network/orchestration attacks, Sybil behavior, and strategic participation. We further compare layered defenses, including robust aggregation, drift-aware optimization, differential privacy, secure aggregation, homomorphic encryption, secure multiparty computation, trusted execution environments, watermarking, provenance, authenticated communication, and monitoring. Finally, we discuss deployment, evaluation, governance, and open research guidelines, emphasizing reproducible, non-independent and identically distributed (non-IID) partitions, attack-aware benchmarks, privacy budgets, attack success rate, client-level disparity, latency, communication, memory, energy/carbon overhead, and auditable lifecycle artifacts.
[LG-97] Loss is Not Behavior: A Unified Output-Space Analysis of Gradient-Based Machine Unlearning NEURIPS2026
链接: https://arxiv.org/abs/2610.38784
作者: Xingjian Zhao,Mohammad Mohammadi Amiri,Malik Magdon-Ismail
类目: Machine Learning (cs.LG)
*备注: Accepted at NeurIPS 2026
Abstract:Gradient-based machine unlearning methods typically ``unlearn’’ by optimizing forget and retain losses, but similar losses do not necessarily imply similar output-space behavior. Prior work has shown that high forget loss alone does not certify that unlearning has produced retraining-like output behavior. We therefore ask: when a gradient-based unlearner differs from the model that would have been learned on only the retain data, what structure does the resulting output-space gap have? We derive closed-form expressions for this gap in linear regression, with an extension to nonlinear models via a generalized Gauss-Newton and local linearity approximations. Our analysis puts various gradient-based methods under a unified framework parametrized by geometry, reweighting, and projection. Two leverage matrices emerge that capture the forget-retain entanglement under the retrained model geometry: forget self-leverage and retain cross-leverage. We show both theoretically and empirically that these two leverage matrices drive the impact of the forget residual on the forget and retain outputs, and existing methods account for this geometry only partially. Our framework yields precomputable leverage diagnostics for auditing forget sets to identify what type of gradient-based unlearning is likely to be successful.
[LG-98] Skillful and realistic precipitation forecasting with reinforced diffusion
链接: https://arxiv.org/abs/2610.38775
作者: Ozan Sener,Anand Avati,David Hafner,Jierui Lin,Laura Manduchi,Xiaoying Pang,Langwen Huang,Torsten Hoefler,Russ Schumacher,Vladlen Koltun
类目: Machine Learning (cs.LG); Atmospheric and Oceanic Physics (physics.ao-ph)
*备注:
Abstract:Accurate precipitation forecasting remains elusive as physics-based models struggle to resolve mesoscale convection and machine learning often yields physically inconsistent, blurred results that consistently underestimate extreme events. We present FitzRoy, the first system that simultaneously achieves state-of-the-art predictive skill and physical realism over the continental United States. Through a reinforced diffusion framework, where a conditional discrete diffusion model is refined via reinforcement learning, FitzRoy effectively resolves the long-standing trade-off between pointwise accuracy and structural fidelity. Across lead times up to 24 hours, FitzRoy consistently matches or exceeds state-of-the-art models while preserving structural realism and capturing extreme events. Its low-latency inference further enables deployment in real-time operational settings. In case studies such as the 2025 Texas Hill Country floods and the 2026 Missouri floods, FitzRoy signals record-breaking accumulations that existing models largely missed. By bridging the gap between realism and skill, FitzRoy establishes a new frontier for high-fidelity precipitation forecasting.
[LG-99] Auditable Hierarchical Inference with Multimodal Evidence and Bayesian Active Inquiry
链接: https://arxiv.org/abs/2610.38774
作者: Yucheng Liu,Everett X. Wang
类目: Machine Learning (cs.LG)
*备注:
Abstract:Most multimodal classifiers assume that all useful evidence is available upfront, whereas clinical inquiry is sequential. We present a synthetic evaluation framework separating objective multimodal inference from active inquiry. An objective model maps synthetic tongue, bilateral multipressure pulse, and acoustic observations to eight syndrome families. A provenance-bearing Knowledge Core and training-only answer model support Bayesian narrowing over 33 fine terms using 25 typed questions, while seven safety questions remain outside the diagnostic posterior. Weak shared symptom factors limit fine-term influence on objective phenotypes, and an objective-only probe tests leakage. In a preregistered 600-case synthetic evaluation, objective family accuracy was 87.0%. Cost-sensitive expected information gain (cost-EIG) raised family accuracy to 95.5% and exact fine-term accuracy to 42.5%, versus 92.7% and 33.0% for an unoptimized fixed-order baseline. Selective family accuracy reached 96.9% at 95.3% coverage. However, cost-EIG asked more questions on average than fixed order (9.795 versus 9.352). Missing-audio routing retained 94.0% family accuracy, and a workflow invariant enforced abstention when all objective modalities were absent. Permuting fine-term answer likelihoods reduced family accuracy to 78.7%. Five of six frozen checks passed, but the question-efficiency hypothesis failed. Comparisons are descriptive because only aggregate, not paired case-level, outputs were retained. These results establish a reproducible synthetic workflow, not clinical diagnostic or treatment validity. Subjects: Machine Learning (cs.LG) Cite as: arXiv:2610.38774 [cs.LG] (or arXiv:2610.38774v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.38774 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-100] Rennala NSGD: Asynchronous Stochastic Optimization Beyond Euclidean Geometry NEURIPS2026
链接: https://arxiv.org/abs/2610.38773
作者: Igor Sokolov,Alexander Tyurin,Peter Richtárik
类目: Machine Learning (cs.LG); Distributed, Parallel, and Cluster Computing (cs.DC); Optimization and Control (math.OC)
*备注: Accepted at NeurIPS 2026. 111 pages, 2 algorithms, 3 tables
Abstract:The empirical success of matrix-aware optimizers, such as Muon (Jordan et al., 2024) and Shampoo (arXiv:1802.09568), highlights the potential of non-Euclidean geometry for deep learning, including large language model (LLM) training. Despite this progress, the theoretical understanding of stochastic optimization in general normed spaces remains limited, particularly in distributed settings. We analyze minibatch stochastic gradient descent (SGD) for smooth nonconvex optimization with respect to an arbitrary norm on a finite-dimensional space. Using the framework of \kappa -regular normed spaces (arXiv:0809.0813), we derive high-probability oracle complexity bounds under a light-tail condition stated directly in the dual norm induced by the update norm. The bounds make the dependence on geometry explicit through the regularity constant \kappa of the dual space. In the Euclidean case, our bound recovers the standard minibatch SGD rate up to logarithmic factors. For some non-Euclidean geometries and noise models, it is sharper than bounds obtained by converting a Euclidean noise assumption, reducing a linear dimension dependence to a logarithmic one. Building on this framework, we introduce Rennala NSGD, a semi-asynchronous distributed algorithm, and establish wall-clock time complexity guarantees under fixed, heterogeneous worker computation times. In the Euclidean case ( \kappa=1 ), our bound matches the parameter dependence of classical Rennala SGD up to logarithmic factors.
[LG-101] Alpha-TCM: A Nine-Layer Framework for TCM-Guided Multimodal Inference and Structured-Prior Auditing
链接: https://arxiv.org/abs/2610.38754
作者: Yucheng Liu,Everett X. Wang
类目: Machine Learning (cs.LG)
*备注:
Abstract:Alpha-TCM makes two contributions under a central thesis: auditable systems should separate domain knowledge, case evidence, probabilistic inference, and downstream authority. First, it proposes a nine-layer TCM-oriented architecture covering semantic definitions, multimodal acquisition, quality control, provenance-bearing knowledge, probabilistic inference, longitudinal updating, constrained decisions, language interaction, and human governance. Each layer specifies its information contract, failure boundary, dependencies, and validation status. This is a system-design proposal, not a claim of complete clinical implementation. Second, a controlled synthetic prototype instantiates Layers 1-5 to audit the knowledge-evidence interface. Four conditional-Gaussian encoders process synthetic tongue images, pulse waveforms, inquiry fields, and text, while a frozen bridge predicts a five-dimensional synthetic outcome. A versioned contextual prior is compiled independently of the renderer. Fusion removes duplicated reference-prior information before adding context. Equal-information controls, prior regularization, graded contamination, and structural corruption test contextual benefits, Bayesian specificity, and conflict detection. Across ten paired seeds and four training sizes, contextual knowledge improved mean absolute error over marginal committee fusion at 600 and 1,200 cases, but not over equal-information neural controls. Fully incorrect context harmed performance at every size, while prior effects were inconsistent and conflict scores did not track case-level damage. The results support contextual structure in low-data synthetic settings, not a specifically Bayesian advantage. The prototype establishes no claims about TCM theory or clinical, diagnostic, therapeutic, causal, longitudinal, or transfer validity. Subjects: Machine Learning (cs.LG) Cite as: arXiv:2610.38754 [cs.LG] (or arXiv:2610.38754v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.38754 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-102] Adversarial Self-Distillation
链接: https://arxiv.org/abs/2610.38732
作者: Mohammad Sadegh Akhondzadeh,Soroush H. Zargarbashi,Aleksandar Bojchevski
类目: Machine Learning (cs.LG)
*备注:
Abstract:Even though current safety-aligned language models are able to refuse harmful requests, their guard can still be bypassed by adversarial perturbations. As in other domains, adversarial training is considered a standard defense. However, due to the autoregressive nature of language generation, supervised finetuning on template refusals does not necessarily teach the model to refuse. Evidence of this is that the adversarially trained model still complies when the attacker injects a compliant prefill. We propose Adversarial Self-Distillation (ASD), which extends a model’s own safe behavior to the space of adversarial perturbations. The teacher (the model itself given the clean prompt) can refuse the harmful request by default, or with the help of an additional instruction in its context. ASD distills this knowledge to the student (which is the same model under attack) at the prefixes that the student visits (due to on-policy sampling). These potentially include compliant openings. ASD achieves lower attack success than state-of-the-art defenses with little loss in multiple-choice accuracy, and it becomes able to recover a refusal even after a compliant opening.
[LG-103] On the Generalization Gap of Streaming Federated Learning with Markovian Data
链接: https://arxiv.org/abs/2610.38686
作者: Masoud Kavian,Milad Sefidgaran,Youssef Attia El Hili,Malik Tiomoko,Abdellatif Zaidi
类目: Machine Learning (cs.LG); Information Theory (cs.IT); Machine Learning (stat.ML)
*备注: 78 Pages, 5 Figures
Abstract:Streaming federated learning (SFL) allows clients to train a shared model from continuously arriving data without retaining large local datasets. Such streams often exhibit temporal dependence, yet its effect on generalization remains poorly understood. In this paper we study the expected generalization gap of the aggregated model when client data follow Markov processes, using information-theoretic and rate-distortion techniques. Our framework accommodates heterogeneous Markovian dynamics across clients and distinguishes temporal dependence from information acquired during local learning. We establish bounds under cross-entropy loss, specializing them to Linear Next-Token Prediction (LNTP) and Self-Attention Next-Token Prediction (SANTP). These results quantify the effects of client count, communication rounds, sample size, data dimension, mixing time, and context length. In particular, they reveal a communication–generalization tradeoff: although frequent aggregation may help optimization, it can weaken generalization guarantees. They also suggest a potential advantage of distributed over centralized streaming learning under a fixed total data budget. Temporal dependence enters through the square root of the mixing time. By analyzing the history-state process, we obtain refined bounds with improved context-length dependence under suitable mixing conditions. Experiments on GPT-style language-model pretraining with TinyStories and real-world time-series forecasting illustrate how collaboration and communication frequency influence generalization under temporal dependence
[LG-104] ExplainCC: Explaining Correlated Clusters by Node Labels CIKM2026
链接: https://arxiv.org/abs/2610.38667
作者: Nadeesha Shyami Kumari Herath Mudiyanselage,Florian Adriaens,Nikolaj Tatti
类目: Machine Learning (cs.LG); Data Structures and Algorithms (cs.DS)
*备注: Accepted to CIKM 2026
Abstract:An increasingly important requirement for clustering algorithms is that decisions and outcomes need to be clearly understandable for human users. We address this requirement for the Correlation Clustering (CC) problem, in case the objects to be clustered are accompanied by labels. In CC one seeks to partition a set of objects into clusters based on pairwise similarity and dissimilarity information. Our proposed model only considers clusters that can be explained by a label tree, which is essentially a decision tree that partitions the objects according to labels. The goal is to find an explainable partition of the data that minimizes the CC objective function. We formalize this problem from a theoretical viewpoint, and prove several hardness of approximation results, also when constraining the size of the tree. Our proposed greedy method finds locally optimal splits, and works recursively in a top-down approach. To efficiently compute the best split, we carefully maintain certain counters. An experimental evaluation demonstrates its effectiveness. The proposed algorithm processes graphs with over million edges in a few minutes, can find the ground truth in synthetic data, and is able to produce interpretable clusters in real world networks.
[LG-105] Logical Guidance of Stochastic Interpolants
链接: https://arxiv.org/abs/2610.38648
作者: Gerrit Gerhartz,Francesco Alesiani,Jonathan Warrell,Henrik Christiansen,Matheus Ferraz,Mathias Niepert
类目: Machine Learning (cs.LG)
*备注:
Abstract:Guidance and composition of generative models have applications in visual design, chemistry, and biology. Logical guidance composes conditional scores exactly for diffusion models. However, Flow Matching (FM) and Stochastic Interpolants (SI) learn velocity fields. We show that negation and exclusive disjunction admit exact conditional velocity composition on any interpolant, whereas conjunction, and therefore any formula containing one, is exact on Gaussian paths but in general incurs a drift-interaction error, so composed velocities need not sample the logical target. LOGSI solves this by using score-based re-equilibration as our principled mechanism and falling back to the Gaussian-path velocity rule as an approximate transport predictor. In non-Gaussian bridge problems, generation alternates transport steps with score re-equilibration steps. On image trans fer (EMNIST to CMNIST, CelebA) and QM9 source-to-target molecular editing with logical attribute constraints, LOGSI improves rule conformity at the same network-evaluation budget over composed-velocity baselines, while also improving fidelity on the image tasks. More broadly, LOGSI applies wherever stochastic interpolants need guidance and logical combinations of constraints are natural.
[LG-106] CFKS: Continuum-Feature Kernel PDE Solvers
链接: https://arxiv.org/abs/2610.38644
作者: Vikas Dwivedi,Balaji Srinivasan,Monica Sigovan,Bruno Sixou
类目: Machine Learning (cs.LG)
*备注:
Abstract:Neural-network-based PDE solvers require architectural choices, particularly the number of layers and neurons per layer. Fixing a single Gaussian radial basis function (RBF) feature layer and taking the number of features to the continuum limit removes the need to prescribe a finite feature count or select individual feature centers, thereby reducing representation arbitrariness. Motivated by this principle, we introduce Continuum-Feature Kernel Solver (CFKS), a deterministic framework for PDEs obtained by analytically marginalizing a bounded continuum of Gaussian RBF features. CFKS works directly with the resulting continuum-feature mean kernel and its operator-applied counterparts, parameterized by prescribed Gaussian widths and bounded center distributions. For linear PDEs, CFKS leads to symmetric operator-kernel collocation systems with null-space enforcement of physical constraints. Nonlinear PDEs are addressed through linearized initialization and nonlinear least-squares residual minimization using analytical kernel derivatives and Jacobians. Residual-guided collocation enrichment and sequential space–time-slab marching provide additional resolution strategies, while prescribed multiscale kernel refinement helps resolve localized nonlinear structures. Finite-feature experiments confirm convergence of empirical value and operator Gram matrices toward their continuum counterparts. Numerical studies include oscillatory and multiscale ODEs, irregular-domain Poisson and biharmonic problems, singular perturbations, transient transport equations, and viscous Burgers’ equation. The results establish CFKS as a direct computational connection between finite Gaussian-RBF representations and continuum-feature kernel PDE solvers. Codes: this https URL
[LG-107] wo-Stage Clipping: Linear-Time Approximate and Exact Mixed-Norm Projections for Transformers
链接: https://arxiv.org/abs/2610.38598
作者: Michel Barlaud,Jean-Paul Marmorat
类目: Machine Learning (cs.LG)
*备注:
Abstract:Large Language Models (LLMs) based on Transformer architectures have achieved remarkable performance across artificial intelligence tasks, but their computational cost remains a major bottleneck. Structured sparsification provides a promising approach to reducing this cost. In particular, \ell_1,\infty projections can induce column-wise sparsity in model matrices, but existing proximal methods (Prox) typically require \mathcalO(nm\log(nm)) operations for matrices in \mathbbR^n\times m , limiting their computational and energy efficiency. In this paper, we introduce Two-Stage Clipping (TSC), an efficient structured sparsification operator inspired by bilevel optimization. Rather than solving a bilevel optimization problem in the strict sense, TSC decomposes the computation into two successive stages. First, group-level quantities are projected to allocate the global sparsity budget and identify active groups. Second, the resulting values are used as group-wise clipping thresholds for the matrix coefficients. For square attention matrices ( n \times n ), the complexity of our TSC method is O(n^2) , which is linear in the number of matrix coefficients. We establish a theoretical bound on the discrepancy between TSC and the exact proximal projection and show on synthetic data that the additional Euclidean approximation cost remains small in practice. Extensive benchmarks demonstrate substantial runtime improvements over proximal baselines and validate the theoretical norm identities underlying the proposed construction. Subjects: Machine Learning (cs.LG) Cite as: arXiv:2610.38598 [cs.LG] (or arXiv:2610.38598v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.38598 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-108] Robust Active Learning for More Efficient and Reliable Paths to Optimal Design
链接: https://arxiv.org/abs/2610.38582
作者: Hayriye Pehlivan Solak,Francesco Di Fiore,Jeroen Wackers,Riccardo Pellegrini,Andrea Serani,Matteo Diez,Laura Mainini
类目: Computational Engineering, Finance, and Science (cs.CE); Machine Learning (cs.LG)
*备注: Global Conference on Naval Architecture and Ocean Engineering (G-NAOE) 2026
Abstract:Simulation-driven design optimization requires active learning strategies that efficiently combine models with different computational costs and levels of fidelity. We introduce Multifidelity Cost-Aware Screening (MF-CAS), which enables the joint selection of sampling location and fidelity using a surrogate that provides decomposed uncertainty intervals rather than a complete predictive probability distribution. MF-CAS exploits the contribution of each fidelity to the high-fidelity uncertainty together with its computational cost to determine whether a candidate should first be evaluated at low fidelity or directly progressed towards high fidelity. The method is assessed on the constrained 14-dimensional hydrodynamic optimization of the DTMB 5415 hull under a deliberately limited computational budget. MF-CAS achieves an approximately 9.0% resistance reduction, compared with 6.7% for the best baseline optimisation method, demonstrating that incorporating fidelity-resolved uncertainty directly into the joint location-fidelity decision can substantially improve multifidelity active learning efficiency.
[LG-109] Dual DeltaNet: Value-Side Memory Reorganization for Linear Attention
链接: https://arxiv.org/abs/2610.38578
作者: Yixiao Qian,Yaxin Zhang,Song Chen,Jiaxu Liu,Shengze Cai,Chao Xu
类目: Machine Learning (cs.LG)
*备注:
Abstract:Linear attention and state space models (SSMs) offer efficient alternatives to softmax attention by maintaining a fixed-size recurrent state, but this design requires careful management to preserve useful historical associations and limit interference. The delta rule helps reduce interference through correction of the value associated with each key, but it does not explicitly account for how related value features are distributed across the state. We observe that a state \boldsymbolS\in\mathbbR^d_k\times d_v can be viewed as a bank of d_k addresses storing d_v -dimensional values. While a key-side read \boldsymbolS^\top\boldsymbolk retrieves the value associated with \boldsymbolk , a value-side read \boldsymbolS\boldsymbole aggregates addresses whose stored value features match the selector \boldsymbole . Based on this observation, we propose Dual DeltaNet (DDN), which extends Kimi Delta Attention (KDA) with a two-sided delta rule. The key-side delta rule reads with the current key to correct its associated value, while the complementary value-side delta rule reads with selected features of the current value to reorganize the addresses it occupies. We derive a chunkwise parallel training algorithm and a rank-2 recurrence that increases arithmetic intensity during inference. Compared with KDA, GDN-2, and Mamba-3, DDN achieves the strongest overall performance across language modeling, zero-shot downstream tasks, and retrieval, while incurring modest training and inference overhead relative to KDA. Code: this https URL
[LG-110] REVBRAIN: Real-time Probabilistic EEG Forecasting Using Residual Vector Quantization
链接: https://arxiv.org/abs/2610.38558
作者: Hanna Pankka,Jaakko Lehtinen,Timo Roine
类目: Machine Learning (cs.LG); Signal Processing (eess.SP); Neurons and Cognition (q-bio.NC)
*备注:
Abstract:Real-time forecasting of resting-state EEG signals is required by applications like brain-state-dependent stimulation, where computer generated actions are coupled with ongoing neural oscillations. Recently, the deep learning model WaveNet was proposed for the forecasting task. It shows superior accuracy compared to the prevailing method, the linear autoregressive (AR) model, and provides a built-in uncertainty measure due to its probabilistic modeling framework. However, inference using WaveNet is too slow for real-time use due to its sample-by-sample autoregressive operation: a single forecast requires evaluating the model tens or hundreds of times depending on the length of the prediction window. In this paper, we present REVBRAIN, a probabilistic deep learning method that produces forecasts in real time by generating the forecast over the whole prediction horizon at once. This is achieved by tokenizing the output space using residual vector quantization (RVQ) and generating forecasts over the items in the tokenized space. We tested REVBRAIN for forecasting 200 ms of EEG time series in the alpha-frequency band (8-13 Hz) and compared our results against WaveNet model. We found that REVBRAIN achieved real-time inference on a laptop GPU and an over 50\times speedup compared to WaveNet, while reaching superior prediction accuracy. By achieving real-time inference speed, REVBRAIN enables incorporating accurate and probabilistic EEG forecasting into BCI applications, opening up new ways for investigating the functioning of the human brain. Subjects: Machine Learning (cs.LG); Signal Processing (eess.SP); Neurons and Cognition (q-bio.NC) Cite as: arXiv:2610.38558 [cs.LG] (or arXiv:2610.38558v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.38558 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Hanna Pankka [view email] [v1] Wed, 28 Oct 2026 09:53:55 UTC (285 KB)
[LG-111] Beyond Reachability: When Edit Order Preserves Couplings in Graph Flow Matching
链接: https://arxiv.org/abs/2610.38486
作者: Zilin Li,Zixing Song
类目: Information Theory (cs.IT); Machine Learning (cs.LG)
*备注: 14 pages. Theoretical preprint
Abstract:Shortest-path optimality does not guarantee that a memoryless graph generator preserves a prescribed source-target correspondence. We study whether reordering legal shortest edits can retain this correspondence without changing the pairing, mandatory edits, or sampling budget. Under a common Poisson clock, we establish an exact criterion for endpoint coupling recoverability: the source-blind Markov projection attains the unavoidable clock-error floor if and only if routes to distinct targets do not share graph states. Routes to the same target may merge without sacrificing endpoint fidelity. This characterization leads to an exact finite-state routing formulation and a legality-preserving scheduling procedure with recovery certificates. A marker-free construction demonstrates how shared edge structure can retain source distinctions, while a capacity-constrained family exhibits a sharp separation in paired error between edit orders with identical marginal error. Finally, a target-conditioned aliasing bound connects structural route design to the terminal fidelity of a learned graph-only generator. These results characterize when legal edit ordering can preserve a prescribed coupling under fixed operational costs.
[LG-112] Recurrent Bipartite Graph Neural Networks for QC-MDPC Error Decoding in McEliece Cryptosystems
链接: https://arxiv.org/abs/2610.38479
作者: Parth Rajeshkumar Tandel,Elisabeth Fink
类目: Cryptography and Security (cs.CR); Machine Learning (cs.LG)
*备注: 19 pages
Abstract:As the advent of large-scale quantum computing threatens classical encryption, Quasi-Cyclic Moderate Density Parity-Check (QC-MDPC) codes have emerged as a foundational element of Post-Quantum Cryptography (PQC). However, decoding these relatively dense codes using classical Belief Propagation algorithms often leads to failures caused by short graph cycles that stall the decoder. In cryptographic applications, these failures are severe security vulnerabilities that leak secret key information. To address this, this paper proposes a Recurrent Bipartite Graph Neural Network (R-BGNN) as proof of concept. By integrating Gated Recurrent Units (GRUs) into the message-passing framework and employing a multi-topology curriculum training strategy, the R-BGNN actively learns to break limit-cycle oscillations. Evaluated across 20 unseen matrix topologies, the proposed architecture significantly outperforms classical Bit-Flipping and Min-Sum baselines, reducing the Decoding Failure Rate (DFR) by 63.7% at moderate noise levels ( t=10 ) while converging in fewer iterations.
[LG-113] Continual Adaptation Across Modalities: Learning and Forgetting in Unified Multimodal Models
链接: https://arxiv.org/abs/2610.38436
作者: Xinrui Wang,Zhihao Chen,Jia Tang,Alexandra Gomez-Villa,Bartlomiej Twardowski,Joost van de Weijer,Songcan Chen
类目: Machine Learning (cs.LG)
*备注:
Abstract:Unified multimodal models (UMMs), which can process and generate diverse modalities within a single framework, have recently emerged as a powerful paradigm. However, when these models undergo continual learning on multiple tasks across different modalities, their learning and forgetting dynamics remain poorly understood. In this paper, we systematically investigate how architectural design and adaptation method shape intra-modal learning and cross-modal interference between the generation and understanding sides. Specifically, we report three findings. First, the severity of cross-modal interference decreases with the degree of modality-specific parameter isolation. Second, a within-model asymmetry: the generation side of a UMM is markedly more fragile than its understanding side. Third, what post-training adds is the first to be forgotten: purely post-training capabilities such as personalized generation collapse by 84% to 90%, while pretraining capabilities remain stable. Together, these results clarify how parameter isolation and post-training fragility shape UMMs’ learning and forgetting dynamics, offering a starting point for developing continually evolving UMMs.
[LG-114] Revisiting Temporal Link Prediction through Structural Proximity
链接: https://arxiv.org/abs/2610.38385
作者: Yuzhang Fei,Yanping Zheng,Lu Yi,Zhewei Wei
类目: ocial and Information Networks (cs.SI); Machine Learning (cs.LG)
*备注:
Abstract:Existing temporal link prediction methods typically rely on learned temporal representations to capture evolving interaction patterns. In this paper, we show that a classical structural proximity measure can match or outperform these methods without training or learned temporal sequence modeling. We revisit Personalized PageRank (PPR) on historical interaction graphs, using simple sliding-window retention or exponential time decay. Surprisingly, across 14 datasets, Time-decay PPR ranks first on nine and second on four among the compared methods, including datasets with few repeated edges. To understand this result, we examine the historical paths connecting source nodes to previously unseen destinations. We find that such paths contain useful predictive information, and that PPR outperforms a strong neural baseline on queries where historical structure clearly distinguishes the true destination from negative candidates. These results show that strong performance on current benchmarks does not necessarily require complex temporal models. More fundamentally, they raise the question of how much predictive value learned temporal modeling adds beyond historical structural proximity. Our findings motivate reassessing progress in temporal link prediction with strong classical baselines.
[LG-115] CSO: Refining Robotic Policies via Skill Distribution Alignment and Skill-Grained Optimization NEURIPS2026
链接: https://arxiv.org/abs/2610.38357
作者: Zhiyuan Xiang,Xiang Deng,Wanjie Tao,Xu Chen,Zhao Jielun,Weili Guan
类目: Robotics (cs.RO); Machine Learning (cs.LG)
*备注: Accepted at NeurIPS 2026. 31 pages, including appendices and checklist
Abstract:Discretizing continuous actions into discrete skills using methods like VQ-VAE has emerged as a powerful paradigm for robotic manipulation. However, quantization errors yield a suboptimal distribution for the skill prior, degrading its performance. While reinforcement learning offers a path for refinement, its direct application faces unstable encoder updates and a granularity dilemma in importance sampling. To address these challenges, we introduce Cascaded Skills Optimization (CSO), a two-stage post-training framework. First, Rejection-Sampling Supervised Fine-tuning aligns the model’s observation-to-skill mapping with the distribution of successful online trajectories via supervised fine-tuning. Second, with the observation encoder frozen, Skills Policy Optimization computes an independent, clipped importance ratio for each complete skill, enabling stable and efficient updates. With a lightweight QueST-style policy, CSO raises overall LIBERO success from 80.1% to 97.5%, improves performance on RoboTwin2.0, and achieves 61.0% success across five physical single-arm tasks. Additional comparisons with FAST, discrete BEAST, and fixed-buffer IQL further validate the effectiveness of our two-stage post-training strategy.
[LG-116] Learning a Prior-Data Fitted Network for Graph Foundation Models via Structural Causal Model Priors
链接: https://arxiv.org/abs/2610.38335
作者: Ziwei Zhang,Tianyin Liao,Chunyu Hu,Ge Lan,Nuo Xu,Yi Huang,Qingyun Sun,Xingxuan Zhang,Peng Cui,Jianxin Li
类目: Machine Learning (cs.LG)
*备注:
Abstract:Graph Foundation Models (GFMs) have recently emerged as a promising paradigm for learning transferable representations across different graphs and domains. However, existing GFMs face notable limitations, including limited pre-training scale or reliance on aligning with textual information in Large Language Models (LLMs). To overcome these limitations, recent studies have drawn inspiration from prior-data fitted networks (PFNs), which learn from large synthetic datasets and support multi-domain prediction backed by causal learning. However, existing graph PFNs encounter severe difficulties in modeling causal relationships within and across node and edge sets, generating diverse graph structures, and designing suitable architectures that can learn from complex synthetic datasets. To address these challenges, we propose GSCM-PFN (Graph SCM-based Prior-Data Fitted Network), a graph PFN pre-trained on synthetic data generated from an SCM-based prior that jointly models nodes and graph topology. Specifically, we (1) design a joint feature-edge-label structural causal model (SCM) to guide synthetic graph generation and capture causal relationships within and across node and edge sets; (2) introduce prior-guided node-edge dependency modeling to instantiate diverse graph generation strategies, and prove that our SCM-based generation prior is more expressive than existing priors; and (3) propose a feature-permutation-invariant dual-path graph network with theoretical guarantees for learning from diverse synthetic graphs. Experiments on 23 real-world benchmarks show that GSCM-PFN outperforms both conventional GFMs and PFN-based GFMs, without utilizing real-world graph data for pre-training. Code: this https URL
[LG-117] Optimal Regret for Communication-Efficient Decentralized Online Learning
链接: https://arxiv.org/abs/2610.38334
作者: Sifan Yang,Wenhao Yang,Wei Jiang,Yuanyu Wan,Lijun Zhang
类目: Machine Learning (cs.LG)
*备注:
Abstract:In this paper, we investigate decentralized online convex optimization with compressed communication, where n learners over a network jointly minimize a sequence of global loss functions based only on their local losses and compressed messages from their neighbors. The best known regret bounds for this problem are O(\omega^-1/2n\rho^-1\sqrtT\ln n) and O(\omega^-1n\rho^-2\ln n\ln T) for convex and strongly convex loss functions, respectively, where \rho\in(0,1) is the spectral gap of the communication matrix and \omega\in(0,1] is the compression quality factor. Compared with the corresponding lower bounds of \Omega(\omega^-1/2n\rho^-1/4\sqrtT) and \Omega(\omega^-1n\rho^-1/2\ln T) , these upper bounds exhibit polynomial gaps in their dependence on \rho^-1 . To close these gaps, we first develop Ascent-gossip, which combines an online compression strategy with a Chebyshev recursion and attains a faster consensus rate than existing compressed gossip methods. Furthermore, by adopting Ascent-gossip, we propose an online algorithm that achieves optimal regret bounds of \widetilde O(\omega^-1/2n\rho^-1/4\sqrtT) and \widetilde O(\omega^-1n\rho^-1/2\ln T) for convex and strongly convex loss functions, respectively. Up to logarithmic factors in n and \rho^-1 , these results match the lower bounds in terms of T , n , \rho , and \omega .
[LG-118] Multi-Token Prediction Done Right
链接: https://arxiv.org/abs/2610.38331
作者: Jayden Teoh
类目: Machine Learning (cs.LG)
*备注: Preprint
Abstract:Multi-Token Prediction (MTP) is now a common ingredient in language-model pre-training, as it improves downstream performance and enables faster inference through speculative decoding. We identify a flaw in standard MTP training, in which a residual component of the MTP gradient interferes with both next-token and multi-token learning, increasingly so with scale. Specifically, the MTP cross-entropy gradient can be decomposed exactly into a KL self-distillation term that pulls each MTP head toward the model’s own next-token prediction (NTP) distribution and improves speculative decoding, and a residual that has a low signal-to-noise ratio and buries the useful KL signal. We introduce the KL gradient-recovery trick: when two distributions share a linear output head, the gradient of their KL divergence can be recovered exactly by combining their cross-entropy gradients through straight-through gradient routing. Unlike minimizing the KL directly, this trick requires no vocabulary-sized logit matrices and is compatible with standard fused cross-entropy kernels. This enables Soft Multi-Token Prediction (Soft-MTP), a one-line code change to MTP that removes the noisy residual and backpropagates only the KL self-distillation signal by reusing the existing NTP and MTP cross-entropy computations. Soft-MTP supports both DeepSeek- and Meta-style MTP, adds no parameters, and incurs no measurable overhead. From 167M to 1.2B parameters, Soft-MTP improves over standard MTP with lower NTP perplexity, faster speculative decoding and higher zero-shot accuracy at every scale, lower MTP perplexity from 689M onward, and stronger math and code reasoning after fine-tuning at 1.2B. Soft-MTP also matches, if not exceeds, direct KL divergence minimization on most metrics. Soft-MTP is, in effect, a free lunch for MTP optimization. Our code can be found at this https URL.
[LG-119] A Witnessing Duality for ReLU Network Depth
链接: https://arxiv.org/abs/2610.38281
作者: Dimitrios Myrisiotis
类目: Computational Complexity (cs.CC); Machine Learning (cs.LG); Logic in Computer Science (cs.LO)
*备注: 158 pages, 1 figure, 1 table
Abstract:A depth lower bound for ReLU networks asserts that every shallower network errs on some input; we ask when a weak theory can prove such a bound so that an error can be computed. For the exact computation of \max(x_1,\ldots,x_n) , recent work shows two hidden layers suffice with unrestricted rational weights. By contrast, Haase, Hertrich and Loho (ICLR 2023) proved that integer weights require depth logarithmic in n , and Averkov, Hojny and Merkert (ICLR 2025) extended the bound to nonbias weights whose denominators are powers of a fixed integer not divisible by a prime q . With integer weights the bound is sharp: the classical balanced construction has depth \lceil\log_2 n\rceil . For each fixed q and depth cap K , we prove this known bound in \mathbfVTC^0 , the bounded-arithmetic theory of uniform constant-depth threshold computation, together with a refuter that is itself such a computation and whose correctness the theory proves. Given a network with at most K hidden layers, nonbias weights whose reduced denominators avoid q , arbitrary rational biases and nq^K inputs, the refuter returns a rational input of polynomial length on which the network errs. The proof makes the obstruction finite: an iterated intersection product of piecewise-linear functions is divisible by q for the network but not for the maximum, so one of finitely many test points exposes the error. By witnessing, provable refutation corresponds to a provably correct refuter, and provable existence of a certified network to a provably correct constructor. On the construction side, we introduce certificates that label each maximum gadget by the inputs it aggregates; with them the theory verifies the balanced construction, which meets the refuter at the sharp threshold n\le2^K , and a rational family of polynomial description length and depth O(\log n/\log\log n) built from known depth-two blocks.
[LG-120] NAGO: Noise-Aware Generative Operator via Flow Matching NEURIPS2026
链接: https://arxiv.org/abs/2610.38273
作者: Guanyu Chen,Shengze Xu,Pengwei Liu,Xingyu Ren,Pengkai Wang,Antonios Armaou,Tieyong Zeng,Dong Ni
类目: Machine Learning (cs.LG)
*备注: Accepted at NeurIPS 2026. 49 pages including appendices and checklist. Code: this https URL
Abstract:We introduce NAGO, a noise-aware generative operator for efficient sampling from invariant measures of stochastic partial differential equations (SPDEs) in function space. Our central idea is to treat the source measure as a design variable rather than a fixed Gaussian default. We establish admissibility conditions for well-defined infinite-dimensional flows and show that the finite-band source–target covariance ratio controls the spectral stiffness of the marginal velocity field. This analysis yields an admissibility-constrained spectral-matching rule: choose a source that is sufficiently rough for well-posedness, then minimize covariance mismatch over the resolved Fourier modes. On three representative stochastic-field benchmarks, this source design improves few-step spectral accuracy and several complementary distributional and physical statistics. In a controlled Matern-field ablation, it reduces the accurate RK4 budget from about 30 steps for Gaussian functional flow matching to 3–5 steps, a 6 – 10\times reduction in neural ODE evaluations. These gains are strongest for stationary multiscale fields and depend on the metric, resolution, and guidance protocol.
[LG-121] A Hybrid Physics-informed holomorphic neural networks Method
链接: https://arxiv.org/abs/2610.38269
作者: Zhijie Su,Qiaolin He,Zhiping Mao,Shiquan Zhang
类目: Machine Learning (cs.LG); Numerical Analysis (math.NA)
*备注: 34 pages, 14 figures
Abstract:Physics-informed holomorphic neural networks (PIHNNs) exploit complex-analytic representations to satisfy homogeneous PDEs by construction, but their extension to nonhomogeneous problems is nontrivial. We propose a hybrid physics-informed holomorphic neural network (HPIHNN) framework based on a decomposition of the solution into particular and homogeneous components. The particular component is computed by a spectral method on a simple extended domain, while the homogeneous component is approximated by a PIHNN trained only on the boundary with modified boundary data. The resulting formulation avoids meshing complex geometries and retains the dimensionality-reduction advantage of holomorphic representations. The method is applied to nonhomogeneous Poisson, Stokes, biharmonic, linear elasticity, and Helmholtz equations. Numerical experiments covering complex geometries, corner singularities, and nearly incompressible regimes demonstrate that the proposed method accurately resolves the corresponding solutions and, in the considered tests, generally achieves lower errors and shorter training times than standard PINNs. The results indicate that coupling spectral particular solutions with holomorphic neural approximations provides an efficient approach to nonhomogeneous PDEs on complex domains.
[LG-122] ParaTTT: Parallelizing Test-Time Training Layers with Deep Memories
链接: https://arxiv.org/abs/2610.38255
作者: Noel Loo,Neehal Tumma,T. Konstantin Rusch,Mathias Lechner,Alexander Amini,Daniela Rus
类目: Machine Learning (cs.LG)
*备注: 34 pages, 7 figures. Project repository: this https URL
Abstract:Test-time training (TTT) layers replace the linear fast-weight memory of modern recurrent architectures with a small neural network trained online by gradient descent. This nonlinear memory is more expressive than the linear memories of Gated DeltaNet and Mamba-2, but is difficult to parallelize: existing implementations such as TTT-MLP rely on sequential, block-wise weight updates that run several times slower than chunkwise-parallel linear recurrences. Moreover, their block-wise execution only approximates the token-by-token online recurrence, whereas linear recurrences admit an exact chunkwise-parallel evaluation. We introduce ParaTTT (PTTT), an algorithm for parallelizing arbitrarily deep nonlinear memories by iteratively refining estimates of the inner-loop gradient trajectory. Starting from a guess of the hidden-layer gradients, each refinement consists of three chunkwise-parallel passes: a forward pass that evaluates the resulting feature trajectory, an exact gated-delta-rule pass that computes the readout trajectory and its feature gradients, and a backward pass that propagates these gradients into a new estimate. The exact online gradient descent recurrence is a fixed point of the iteration, while decoding requires no refinement iterations. We prove geometric convergence for a two-layer variant under sufficient contraction conditions. In practice, two iterations suffice. With two refinement steps at 4K context, PTTT evaluates the inner-loop trajectory 32\times faster than TTT-MLP with the block approximation at matched error, while the full 340M model trains 4.9\times faster. Models trained at 340M and 1.3B parameters match or exceed Gated DeltaNet and Mamba-2 and remain competitive with same-size Transformers on language-modeling perplexity and zero-shot accuracy, with better recall than the linear recurrences.
[LG-123] ransportable Offline Reinforcement Learning in Presence of Unmeasured Confounding
链接: https://arxiv.org/abs/2610.38253
作者: Junyi Liao,Hangqi Ren,Zhuoran Yang
类目: Machine Learning (cs.LG)
*备注:
Abstract:Offline reinforcement learning seeks to learn decision policies from previously collected data, but deployment environments may differ from the domains in which the data were collected, and logged variables may omit confounders that affected both historical actions and outcomes. We study offline policy learning under the joint challenges of domain discrepancy and unmeasured confounding, where multiple heterogeneous source datasets provide complementary information about a target domain. We develop a causally adjusted transportability framework combining transportability theory with instrumental variables, negative controls, and proxies to identify target-domain causal quantities. For confounded contextual bandits, identification reduces to systems of bridge-function integral equations, enabling minimax estimation from finite samples. Confidence sets for the identified quantities are then used for pessimistic policy optimization, yielding CAT-CCB. We extend the framework to episodic confounded Markov decision processes, where complementary sources identify target reward and transition mechanisms under structural and completeness conditions. Under Gaussian dynamics and linear function approximation, this yields CAT-CMDP. We establish high-probability finite-sample suboptimality guarantees that quantify the effects of approximation, estimation, source-to-target shift, offline coverage, and sample size. For bounded finite-dimensional linear classes, the contextual-bandit guarantee recovers the standard \widetilde\mathcal O(N^-1/2) rate. Numerical experiments support the effectiveness of the framework for learning target-domain policies from heterogeneous confounded data. Together, the results provide a principled approach when no single source dataset suffices for causal identification and policy optimization.
[LG-124] Nearly Minimax Optimal Regret for Adversarial Linear Bandits with Strong Corruptions
链接: https://arxiv.org/abs/2610.38243
作者: Qiwei Di,Quanquan Gu
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注: 39 pages, 1 table
Abstract:We study adversarial linear bandits with strong corruption, in which an adaptive adversary chooses the reward vectors and a strong adversary corrupts each observed reward after seeing the learner’s action, subject to a known total corruption budget C . Strong corruption breaks the importance-weighted reward estimates used by adversarial bandit algorithms: by corrupting only the rounds in which a rarely sampled action is drawn, the adversary amplifies each unit of corruption by the inverse of its sampling probability. The best known bound for this problem, \tilde O(\sqrtd^3T + dC) by Liu et al. (2024), is a factor \sqrtd above the corruption-free rate in its leading term, and whether \tilde O(d\sqrtT + dC) regret is achievable was left open. We propose CRW-FTRL, which runs follow-the-regularized-leader with a radius-weighted entropy regularizer and estimates rewards with a normalized kernel estimator whose sensitivity to corruption is governed by per-action radii instead of inverse sampling probabilities. The radii are nondecreasing and are chosen jointly with the sampling distribution through a monotone log-determinant barrier problem, so that a negative linear term in the regularizer cancels both the corruption bias and the estimation error of every comparator. For K actions in \mathbbR^d , CRW-FTRL achieves \tilde O(\sqrtdT\log K + dC) regret with high probability against an adaptive adversary, which matches the best known corruption-free rate and the \Omega(dC) lower bound up to logarithmic factors. Applied to a discretization of a general compact action set, it yields \tilde O(d\sqrtT + dC) regret, which settles the statistical rate of corrupted adversarial linear bandits under strong corruption.
[LG-125] Variational Consequence-Driven Offline Reinforcement Learning NEURIPS2026
链接: https://arxiv.org/abs/2610.38238
作者: Wen Jiang,Ke Jiang,Xiaoyang Tan
类目: Machine Learning (cs.LG)
*备注: accepted by NeurIPS 2026
Abstract:Offline reinforcement learning (RL) typically mitigates distribution shift by imposing divergence constraints on the action distributions of the learned and behavior policies. However, this paradigm cannot ensure the underlying dynamic consistency structure: even minor action deviations can lead to drastically different transitions, inadvertently driving the agent away from the offline dataset. To address this, we explore policy constraints through their induced state-consequence distributions rather than relying on pure action-distribution matching. Nevertheless, directly enforcing such consequence-aware constraints is computationally intractable without access to the true underlying dynamics model. To bypass this issue, we propose Structural Behavior Regularization (StBR), which replaces the intractable objective with a closed-form latent divergence. Structuring this latent space to reflect transition dynamics renders the latent divergence a theoretically bounded surrogate for the true consequence-distribution divergence. By regularizing the learned policy with this latent divergence, StBR adaptively tightens the behavior constraint in dynamics-sensitive regions, effectively preventing deviations outside the offline dataset. Empirically, we demonstrate that StBR achieves superior performance across diverse offline RL benchmarks.
[LG-126] he Endgame of FSDP Communication: Layout-Aware Collectives on Copy Engines RDMA and NVSwitch
链接: https://arxiv.org/abs/2610.38228
作者: Zejian Wang,Yifan Zhao,Kaihui Ren,Xiaoling Zhou,Yubin Wang,Chenhao Guan,Lin Zhang,Yufeng Li,Zhikang Mo,Zhuoru Ma,Yingtong Xiong,Jihu Guo,Yiqiao Liao,Chengru Song
类目: Distributed, Parallel, and Cluster Computing (cs.DC); Machine Learning (cs.LG); Performance (cs.PF)
*备注: 18 pages, 13 figures, 9 tables
Abstract:Fully Sharded Data Parallelism (FSDP2) maintains parameters and gradients as per-parameter tensors, whereas its collective path operates on flat, rank-major buffers. Converting between these layouts copies all-gather outputs into parameter storage and packs gradients into full-group fp32 buffers. Layout conversion alone accounts for 6.4% of step time in Llama-3.1 70B training at 8K tokens per GPU. Communication kernels impose a second cost: a default NCCL reduce-scatter can slow a concurrent GEMM that would otherwise complete in a single wave by 69%. We present TIDE, a set of zero-copy collectives built on public NCCL 2.31+ APIs. TIDE preserves FSDP2’s per-parameter storage and compute kernels while folding layout placement into collective addressing. TIDE-AG delivers shards directly to parameter storage using NVSwitch multicast and paced recursive-doubling RDMA, with no SMs used for transport or placement. TIDE-RS consumes per-parameter bf16 gradients directly and reduces them into fp32 shards (a mixed-precision mode that NCCL does not provide) using copy engines and short-lived thread blocks (CTAs) for node-local accumulation and NVSwitch for cross-node reduction. A two-phase pipeline separates gradient reads from network progress; Slot-Reuse optionally stages gradients in vacated all-gather slots. On 64 GPUs, TIDE reduces step time by up to 27.1% relative to Stock FSDP2 and 6.0% relative to FSDP2+SymmMem across Llama-3.1 70B, Wan2.2, and Qwen3-30B-A3B. It reduces reduce-scatter SM time by factors of 12.5-14.5 and cuts interference with concurrent compute by up to an order of magnitude. Its reduce-scatter buffer remains nearly constant from 16 to 64 GPUs at a fixed per-GPU output size, and its peak memory is up to 4.7 GiB below FSDP2+SymmMem’s and, with Slot-Reuse on Wan2.2, below Stock FSDP2’s.
[LG-127] HotPOT: Operator Transformer with Heterogeneous Physics Experts for Large-Scale PDE Pre-training
链接: https://arxiv.org/abs/2610.38218
作者: Jiayi Li,Guancheng Zhou,Yisi Luo,Ruihua Chen,Penghao Jiang,Deyu Meng,Flora D. Salim,Piotr Koniusz
类目: Machine Learning (cs.LG)
*备注:
Abstract:Large-scale PDE pre-training trains a single model on the data from multiple partial differential equations (PDEs). However, adding heterogeneous PDEs with different governing equations, physical variables, and variable interactions does not always improve predictive performance across tasks. Existing PDE mixture-of-experts models mitigate cross-task interference through learned routing, but typically use homogeneous expert architectures, making it difficult to directly reflect differences in state evolution across equations. We introduce HotPOT, an operator transformer with heterogeneous physics experts for large-scale PDE pre-training. HotPOT uses a shared encoder to capture information common across PDEs, while each PDE family uses a neural-numerical expert that pairs a learnable predictor with a fixed numerical update containing no trainable parameters. Each predictor outputs the quantities required by its paired numerical update, which then computes the next state. This design allows different PDEs to share a common encoder while retaining their own learnable predictors and numerical updates. Experiments across five PDE families and six settings show that HotPOT outperforms MoE-POT in all 24 matched-scale comparisons before and after fine-tuning, reducing rollout error by 50.7% on average. Notably, the 17M model also surpasses substantially larger MoE-POT models, and the performance gains extend to zero-shot prediction under unseen physical conditions. Code is available at this https URL.
[LG-128] RainCheck: Harnessing Historical Weather Data with Machine Learning for Rainfall Forecasting in Bangladesh
链接: https://arxiv.org/abs/2610.38211
作者: Shahran Rahman Alve,Muhammad Zawad Mahmud,Samiha Islam,Al Jubayer Pial
类目: Machine Learning (cs.LG)
*备注: 11 Pages
Abstract:Machine-learning rainfall forecasts are often evaluated by shuffling station records into random training and test sets. For rainfall, this lets a model see what happened at neighbouring stations on the very day it is asked to forecast, and it hides the fact that a seasonal lookup table already explains most of the variance. We quantify both effects on 618,650 station-days from 35 Bangladesh Meteorological Department stations (1948-2014). We compare random, temporal (train =1999, test 2005-2014) and spatio-temporal (unseen stations and unseen years) evaluation, calendar-only versus antecedent-rainfall features, and six learners against climatology, persistence and SARIMA. Three findings stand out. First, a station-month climatology alone classifies high-rainfall months with 88.4% accuracy (AUC 0.950), and no learner improved on it by more than 0.8 points under temporal evaluation. Second, a calendar-only random forest reached a next-day wet-day AUC of 0.889 under a random split, higher than any model with real predictors under temporal evaluation, yet fell to 0.835, below climatology (0.859), on future years. Third, models with lagged rainfall add genuine but modest skill for next-day forecasts: a Brier skill score of 0.074 (95% CI 0.064-0.088) over climatology, rising to 0.091 when station and year inputs are removed and to 0.131 when other stations’ same-day observations are added. A GRU sequence model matches but does not exceed gradient boosting, and skill disappears beyond a two-day lead. For monthly totals, no learner beat climatology significantly. We release a leakage-aware protocol and code so that future station-based studies report skill against climatology on held-out years.
[LG-129] Progressive Coupling Method for Diffusion Posterior Sampling in Ptychography
链接: https://arxiv.org/abs/2610.38207
作者: Yihan Qiu,Mingtao Huang,Tianyuan Li,Shudong Li,Yuan Shen
类目: Machine Learning (cs.LG)
*备注:
Abstract:Ptychography is a computational imaging method that reconstructs a sample from overlapping diffraction measurements. This technique relies on these overlapping measurements to provide the redundancy information needed for robust reconstruction, but dense scanning increases acquisition time and data volume. Reducing scan overlaps can improve measurement efficiency but weakens the physical constraints available for reconstruction. Diffusion models provide strong learned prior information, offering a promising approach to compensate for the loss of measurement redundancy. However, effectively utilizing this prior information under weak measurement constraints remains a challenge. We propose a progressive coupling posterior sampling (PCPS) method, which allows for a broader exploration of the solution space under high uncertainty while maintaining stronger diffusion-guided consistency as the reconstruction becomes more refined. Experiments on real X-ray ptychographic data at various scan overlaps demonstrate that PCPS consistently improves reconstruction accuracy compared to traditional iterative methods and existing diffusion posterior samplers, especially at low overlaps. These results show that diffusion priors can reduce the measurement redundancy required for accurate ptychography reconstruction.
[LG-130] Averag ed Annealed Markov Chains Achieve the Algorithmic Threshold for Tensor PCA Part II: Langevin Dynamics
链接: https://arxiv.org/abs/2610.38206
作者: Aniket Das,Tselil Schramm
类目: Machine Learning (cs.LG); Data Structures and Algorithms (cs.DS); Probability (math.PR); Statistics Theory (math.ST)
*备注: Companion paper to arXiv:2610.16705 (Part I). 71 pages, 1 figure
Abstract:We consider the tensor principal components analysis (tensor PCA) problem on the sphere, in which we are asked to recover an unknown vector in \mathcalS_N=\x\in\mathbbR^N:\lVert x\rVert^2=N\ from a noisy observation of its k -th tensor power, for a fixed k\ge3 . Tensor PCA exhibits a well-studied information-computation gap: global algorithms such as spectral methods and sum-of-squares work above an algorithmic threshold, below which we know matching lower bounds (against these and some other restricted models of computation). In contrast, local algorithms such as Markov chains were only known to succeed from a random initialization at a polynomially larger signal-to-noise ratio, and were believed to be inferior to global algorithms for tensor PCA. Continuing the program initiated in Part I [arXiv:2610.16705], we show that spherical Langevin dynamics, when supplemented with annealing and ergodic averaging, achieves polynomial-time strong recovery above the algorithmic threshold from a uniformly random initialization. Together with Part I, our results demonstrate that local Markov chains can succeed all the way down to the algorithmic threshold from an uninformative cold start.
[LG-131] Embodied at Sea: Imitation Learning for Multi-Task Underwater Manipulation
链接: https://arxiv.org/abs/2610.38199
作者: Greta Perez-Haiek(1),Luca Macesanu(1),Kevin Angers(1),Mohammed Imam(1),Timothy Gambin(2),Christopher Clark(1) ((1) New York University, (2) University of Malta)
类目: Robotics (cs.RO); Machine Learning (cs.LG)
*备注: 8 pages, 8 figures. For associated files, see this https URL
Abstract:Behavior Cloning has demonstrated promising generalization capabilities for robotic manipulation in terrestrial environments. However, marine visual distortions and dynamic currents create a severe domain shift that remains a key bottleneck in adapting generalist architectures to underwater field manipulation. To address this, 1616 episodes of underwater manipulation data were collected in calm and turbulent waters and were used to train 14 generalist Vision-Language-Action (VLA) models (DynamicVLA) with varying proprioception observation configurations. Across three tasks on 1260 gradable on-hardware rollouts in a pool, 10 models outperformed the baseline, with the highest-performing model achieving a 93.3% overall success rate, including partial successes. Results are also introduced from a case study evaluating the performance of a single-task Action Chunking Transformer (ACT) and a generalist VLA (SmolVLA) policy for two tasks in marine archaeology and conservation: grasping and mooring. Experiments are performed on an ROV on and around the WWII wreck Coralita X127 and at Manoel Island Marina, Malta. Across 748 collected episodes and 508 gradable on-hardware rollouts, the ACT policy substantially outperformed SmolVLA (47.9% vs. 7.9% pooled success rate), earning a 66.7% success rate on the most successful task at the marine archaeological site. To enable further contributions to embodied intelligence in marine field robotics, the BLUEMBODIED KIT used to collect data, train models, and deploy policies on the ROV has been open sourced.
[LG-132] GearUp: DAG-aware Acceleration for Serverless Inference Serving via Slack-guided Pre-loading
链接: https://arxiv.org/abs/2610.38196
作者: Qi Qi,Yifan Sui,Hanfei Yu,Hong Zhang,Hao Wang
类目: Distributed, Parallel, and Cluster Computing (cs.DC); Machine Learning (cs.LG); Operating Systems (cs.OS)
*备注:
Abstract:Serverless platforms increasingly host machine learning (ML) inference workflows, modeled as directed acyclic graphs (DAGs) of interdependent functions. Beyond container startup, each ML node must load libraries, model weights, and CUDA kernels before it can run. These costs accumulate along dependent DAG paths and place initialization on the end-to-end critical path. Existing systems optimize one half of this problem. DAG-agnostic pre-loading prepares nodes the request may never reach, wasting host and GPU memory. DAG-aware serving routes and scales whole functions, leaving the loading stages inside a node untouched. This paper presents GearUp, a serverless DAG inference serving system that removes ML-node loading from the end-to-end critical path at low resource cost. GearUp profiles the loading stages of each node and schedules them inside the slack time the DAG exposes, hiding initialization behind upstream execution. It reuses initialized instances within and across requests to avoid redundant cold starts. A probabilistic, traffic-adaptive keep-alive policy sets how long idle instances are retained, balancing memory cost against cold-start risk under bursty load. On the AMBER Alert and FINRA-ML workflows with NVIDIA L40S GPUs, GearUp reduces median end-to-end latency by 64.5% and 75.9% against SMIless, for median speedups of 3.1x and 10.0x over cold execution. Integrated GPU and host memory allocation fall by up to 83.9% and 94.6% against eager pre-loading in InstaInfer. GearUp achieves 4.2x strong scaling on four workers.
[LG-133] In-Context Learning for Multi-Fidelity Multiscale Model Closure and Cross-Operator Out-Of-Distribution Generalization
链接: https://arxiv.org/abs/2610.38179
作者: Xinyuan Li,Matthew Zahr,Panos Stinis,Zecheng Zhang
类目: Numerical Analysis (math.NA); Machine Learning (cs.LG)
*备注:
Abstract:Multi-fidelity learning uses low-fidelity information (coarse-scale PDE models) together with limited high-fidelity data (fine-scale solutions) to obtain high-accuracy PDE solutions. It is a data-driven approach to downscale reduced-order models to solve PDEs efficiently and accurately. Standard multi-fidelity learning focuses on one operator or PDE; given a new PDE, retraining using additional coarse-fine solution pairs may be required, which is costly. In this work, we propose an in-context multi-fidelity learning framework for learning multiple multi-fidelity operators with one shared model. The model is trained using paired low- and high-fidelity examples as demonstrations from multiple operators, together with a new coarse-scale solution query, to predict the corresponding fine-scale solution. Once trained, given a new PDE operator, a few (fewer than five) coarse-fine demonstration pairs, and a low-cost coarse query, the model predicts the corresponding fine-scale solution without fine-tuning. When the target operator appears in the training dataset but has a different prediction goal, such as predicting beyond the training time interval, no additional demonstrations are needed. Instead, the model uses training coarse-fine pairs from that PDE and information from other training operators. Testing then only requires generating a low-cost coarse-scale solution query, without fine-tuning. Notably, our smallest model contains only 255,364 trainable parameters, corresponding to approximately 456 parameters per training operator for 560 operators.
[LG-134] A Gradient based Adversarial Augmentation Framework for Mitigating Unlearnable Datasets
链接: https://arxiv.org/abs/2610.38173
作者: Long Dang,Satpal Sura,Kaiqi Xiong
类目: Machine Learning (cs.LG)
*备注: Accepted at IEEE DSC 2026. 8 pages
Abstract:Unlearnable data methods (ULD) protect public data from unauthorized model training by adding imperceptible perturbations that create a learning shortcut. A deep neural network can exploit the shortcut learning path to minimize training loss without learning generalizable features. Intuitively, finding adversarial versions of perturbed images that can substantially increase the model’s loss can force the model to learn generalizable features. Existing data augmentation based attacks rely on heuristic search strategies/random sampling over a curated augmentation set to identify adversarial augmentation operations, an approach that often does not scale when the number of candidate operations is large. To address the challenge of a large search space, we propose \techname, a gradient based adversarial augmentation framework that learns an augmentation policy via stochastic gradient ascent. \techname leverages a differentiable data augmentation pipeline to efficiently search a set of up to 18 augmentation techniques, from which a grid search selects a protection-specific subset. The augmentation subset is trained to maximize a surrogate model’s classification loss. We introduce a perceptual-plus-pixel regularization to learn operation parameters that preserve image semantics. Under a controlled ablation, \techname consistently outperforms an equivalent random sampling baseline (80.0% vs. 77.29% clean test accuracy on EMIN-protected CIFAR-10 for ResNet-18 and 83.28% vs. 66.45% on NTGA-protected CIFAR-10 for MobileNet-V2), demonstrating that gradient guidance drives the clean test accuracy recoverability rather than the expanded search space. \techname offers an effective and scalable adversarial augmentation framework for stress-testing ULD methods, a critical tool for developing robust ULD methods.
[LG-135] B2P-Corr: Batch-to-Population Gradient Estimators for Non-Decomposable Correlation Losses NEURIPS2026
链接: https://arxiv.org/abs/2610.38171
作者: Jingquan Yan,Yuwei Miao,Peiran Yu,Junzhou Huang
类目: Machine Learning (cs.LG)
*备注: Accepted by NeurIPS 2026
Abstract:Correlation metrics such as the Pearson Correlation Coefficient (PCC) and Concordance Correlation Coefficient (CCC) are standard regression metrics and learning objectives, but their non-decomposable nature poses a challenge for stochastic optimization. Since these metrics depend on dataset-level moments (mean, variance, covariance), naive mini-batch training with local statistics yields biased, high-variance gradients whose expectation does not align with the population gradient. To bridge this batch-to-population'' (B2P) gap, we propose \textbfB2P-Corr, a novel framework that transforms correlation objectives into decomposable pseudo-losses driven by lightweight global moment sketches’'. We introduce two efficient variants: \textbfB2P-EMA, which utilizes exponential moving averages and stop-gradient operators to ensure asymptotic gradient tracking with O(B) cost ( B is batch size); and \textbfB2P-U, which uses sampled cross-batch U-statistic estimators and a reservoir-based stabilization with an O(B) implementation. Theoretically, we quantify the O(B^-1) bias of naive mini-batch optimization, prove a moving-target tracking result for B2P-EMA under a fast-sketch / slow-parameter regime, and provide an idealized control-variate analysis for B2P-U. Experiments on two synthetic and three real-world datasets demonstrate B2P-Corr achieves stable optimization and higher correlation than naive mini-batch optimization with negligible overhead.
[LG-136] Replaying the Same Gradients at a Different varepsilon: A Registered Test of the Fixed-Trajectory Account of Adams varepsilon for One Rare Row
链接: https://arxiv.org/abs/2610.38169
作者: Lngel Kyo(A2O Labs)
类目: Machine Learning (cs.LG)
*备注: 24 pages, 2 figures
Abstract:Adam divides every update by the square root of a running average of squared gradients plus a small constant, epsilon. For an output-embedding row whose token is rare, which we call the probe row, raising epsilon can be read as shrinking that row’s steps by arithmetic alone. We test that reading directly. While a training run proceeds, we also divide the probe row’s update by a different epsilon, using the run’s own moment estimates; this replay is exact arithmetic on the gradients the run actually saw, so it holds the run’s trajectory fixed. A rule for when the replay counts as wrong was written down before the runs. We train a small causal transformer in a standard configuration. When epsilon rises by four orders of magnitude, the replay counts as wrong on each of the three seeds registered for the test: the probe row keeps 8.9–11.4 times more of its displacement than predicted with the main, sampled-negative head, and 5.0–7.7 times more with a full-softmax head scored as a second instrument. At three orders of magnitude, tested with the main head only, it does not, by narrow margins; a further two seeds, registered after this result had been committed and scored on their own, follow the same pattern. In separate real runs, raising epsilon on the probe row alone is enough to reproduce the row’s change and raising it on every other parameter alone also moves the row part of the way — readings about the runs that do not rescue the replay; scaling the row’s learning rate instead, with the denominator untouched, is mispredicted by the replay in the same direction. This is one embedding row of one model family on one corpus; we make no claim about rates, about rows in general (other rows were read only as a panel on one trajectory) or about how the effect changes with model size. Comments: 24 pages, 2 figures Subjects: Machine Learning (cs.LG) Cite as: arXiv:2610.38169 [cs.LG] (or arXiv:2610.38169v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2610.38169 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-137] Floor It Then Fill It: Simulating Creator Ecosystems for Fair Video-Content Recommendation Systems at Billion User Scale NEURIPS
链接: https://arxiv.org/abs/2610.38155
作者: Sohini Roychowdhury,Long Sha,Sunand Agarwal,Payal Kamboj,Shuyang Chen,Junlin Zhang,Scott Chen,Hui Zhang
类目: Machine Learning (cs.LG)
*备注: 9 pages, 5 figures, Neurips Workshop 2026
Abstract:Engagement-optimized content recommendation systems exhibit a well-documented rich-get-richer dynamic that structurally disadvantages emerging creators on large-scale platforms. Evaluating debiasing interventions at scale demands faithful simulation of creator-viewer ecosystems, yet the fidelity, diversity, and validity of such simulators remain underexplored. We present a grounded user-simulation framework that models a heterogeneous population of \sim42 K creators and their coupled viewer interactions over 12~weeks, and use it to evaluate equitable view-allocation policies formulated as a budgeted multi-armed bandit (MAB). The simulator captures three behavioral dimensions: (i)~\textbfcreator diversity, modeled via a clustering approach grounded in real platform creator-level distributions; (ii)~\textbfbehavioral fidelity, captured through coupled creator-content state machines that represent growth trajectories, engagement accumulation, and graduation dynamics; and (iii)~\textbfpopulation-level validity, measured by calibration against observed platform metrics. We evaluate simulative-configurations that vary content-viewership target definitions, reward functions, and a small-creator reserve lever to maximize creator-content recommendation equity. A floor-protected per content-view-target with a view-parity-gap reward guarantees 90% of view targets being met by creators every week, grows small-creator organic reach \sim+33% per week, and achieves high distribution stability (set of creators with funded-views Jaccard = 0.819 ). We analyze the Sim2Real gap and propose validation protocols, contributing evidence for when simulation-derived fairness conclusions can be trusted.
[LG-138] MASS: Multi-mask Selective Self-Distillation for Leakage-Resilient Reasoning
链接: https://arxiv.org/abs/2610.38154
作者: Xiuqi Liu,Oveys Delafrooz Noroozi,Upamanyu Madhow
类目: Machine Learning (cs.LG)
*备注:
Abstract:On-policy self-distillation enables the same model to act as student and teacher, with dense token-level supervision obtained by the teacher being given access to privileged information. However, conditioning the teacher on a single reference can concentrate its predictions on one reasoning trajectory and limit transfer to the student. We propose Multi-Mask Selective Self-Distillation (MASS), which approximates the unavailable marginal over diverse reference realizations using multiple independently masked views of one verified solution. MASS aggregates these complementary teacher predictions, selectively transfers reliable token-level corrections, and adaptively focuses masking on influential reference spans. Across mathematical reasoning benchmarks and model families, MASS consistently outperforms strong self-distillation baselines.
[LG-139] On Worst-Case Guarantees for Graph-Based Nearest-Neighbor Search
链接: https://arxiv.org/abs/2610.38133
作者: Majid Daliri,Mohammadreza Daneshvaramoli,Purna Dutta,Edo Liberty,Cameron Musco,Christopher Musco,Torsten Suel
类目: Data Structures and Algorithms (cs.DS); Machine Learning (cs.LG)
*备注:
Abstract:Graph-based nearest neighbor search algorithms like HNSW, DiskANN, NSG, and others have surged in popularity in recent years due to their excellent empirical performance in comparison to space-partitioning methods like IVF data structures and Locality Sensitive Hashing (LSH). However, in contrast to LSH and related methods, obtaining strong worst-case theoretical guarantees for graph-based search has proven challenging. In this paper, we prove a strong lower bound that helps explain the lack of progress towards such guarantees. We show that no standard graph-based method (any that runs a variant of greedy search on a flat or hierarchical graph) can achieve sublinear query time for approximate nearest neighbor search in Euclidean space, even for an arbitrarily large approximation factor. On the other hand, we prove that sublinear query time is possible if we allow for Steiner points in the search graph. We obtain bounds comparable to LSH, showing e.g., that graph-based search with Steiner points can solve c -approximate nearest neighbor search in Euclidean space with query time \tildeO(n^1/(c-1)) . We complement this result with initial experiments exploring how Steiner points can be incorporated into practical graph-based methods.
[LG-140] A Separation Principle for Adversarial Attacks and Defense of Graph Neural Networks
链接: https://arxiv.org/abs/2610.38127
作者: Farid Mafi,Saber Jafarpour,Mohammad Pirani
类目: Machine Learning (cs.LG)
*备注:
Abstract:Adversarial attacks on graph neural networks (GNNs) have been studied as a model-centric optimization problem driven by gradients, learned policies, or meta-learning, with the graph treated as a passive substrate. This view overlooks a defining feature of GNNs: their predictions are produced by structured message passing along the graph, so their adversarial vulnerabilities are encoded as much in the geometry of the graph as in the learned weights. We develop a structure-driven framework for feature attacks in which the attack is derived from the spectral properties of the normalized adjacency operator. Its central result is a separation principle: under a natural relaxation of the attack objective, the optimal feature perturbation is rank one and factorizes into a graph-dependent component (degree-proportional spatial allocation, depending only on the graph) and a weight-dependent component (top singular vector of the composed weights, depending only on the weights). Node degree thus emerges as the fundamental graph quantity governing adversarial vulnerability to feature perturbations. The resulting attack, which we call SepAttack, requires no gradient computation, carries analytical guarantees, achieves competitive or superior performance against established baselines such as Nettack and IG-Attack with up to \sim70\times speedup in the targeted setting, and transfers near-perfectly across architectures on continuous-feature graphs because its graph-dependent component is architecture-invariant by construction. We further derive a separation-principle adversarial training procedure (SepAT) directly from the closed-form worst-case attack and report robustness gains on binary-feature benchmarks.
[LG-141] GTV: An Interpretable Framework for UAV Fault Detection and Traceability via Spatiotemporal Graph Learning and Large Language Models
链接: https://arxiv.org/abs/2610.38106
作者: Zixuan Wang,Changbo Shi,Ji Liu
类目: Machine Learning (cs.LG)
*备注: 21 pages, 8 figures, 4 tables. Submitted to IEEE Sensors Journal
Abstract:Fixed-wing unmanned aerial vehicles experience operating-condition transitions, asynchronous sensor sampling, and control-system compensation that can obscure fault signatures. Existing temporal detection methods have limited capacity to represent coupled faults across subsystems, track their evolution, and constrain language-model diagnoses with quantitative evidence. This paper proposes TGTV, an interpretable fault detection and traceability framework that integrates spatiotemporal graph learning with a large language model. A temporal convolutional network and a graph attention network with weighted fused adjacency extract multiscale temporal patterns and spatial dependencies among sensors. A hybrid temporal Transformer and variational autoencoder then evaluates window-level anomalies using prediction errors and reconstruction probabilities. Continuous-time Shapley attribution identifies critical sensors and high-contribution intervals while preserving the temporal continuity of fault evolution. A structured evidence package combines detection results, graph relationships, temporal attribution, sensor attribution, and operating context. An evidence-constrained diagnosis protocol uses this package to generate flight-level conclusions with traceable evidence identifiers. Cross-flight holdout experiments on the public ALFA fixed-wing flight dataset yield a window-level area under the precision-recall curve of 0.976, accuracy of 0.973, recall of 0.902, and F1 score of 0.920. Flight-level fault diagnosis achieves recall of 0.943 and an F1 score of 0.944. These results demonstrate accurate anomaly detection and quantitatively supported fault-tracing cues, providing a basis for interpretable post-flight fault review and analysis.
[LG-142] Same Label Different Interface: Beyond Action-Space Labels in Robot Learning
链接: https://arxiv.org/abs/2610.38104
作者: Arshia Sangwan,Zhongyang Zhang
类目: Robotics (cs.RO); Machine Learning (cs.LG); Systems and Control (eess.SY)
*备注:
Abstract:Robot-learning studies often compare position and velocity actions as though each label identified a single experimental treatment, yet the label alone does not determine what the robot executes. Before affecting robot motion, bounded policy output passes through filtering, physical scaling, reference construction, feedback, timing, and effort limits. We study how these omitted properties affect attribution under one shared learner and training budget. Across three contact-rich Factory tasks (PegInsert, GearMesh, and NutThread), we compare five joint-space interfaces, with ten independently trained policies per condition. We first remain within the same position-feedback law and vary the executed interface: substituting only the physical action scale as the directly modified component changes sample-median evaluation success by 0.645-1.000 across tasks, while removing an exponential moving average produces task-dependent changes. We then move to a cross-law comparison and choose a velocity scale that exactly matches the position interface’s local raw-torque command gain. This calibration raises the sample median on GearMesh and NutThread, with no median change on PegInsert. Yet clear residual sample-median gaps remain on PegInsert and GearMesh, while the NutThread interval includes zero. Thus, even this explicit equality leaves feedback and held-action dynamics unmatched and does not establish position-velocity trained-outcome equivalence. Action-interface comparisons should therefore report the executed pathway and state the physical quantity they match.
[LG-143] CADENCE: Distributional Next-Transaction Forecasting for Business Cash-Flow Underwriting
链接: https://arxiv.org/abs/2610.38084
作者: Alexander Wu,Jingchao Huang
类目: Machine Learning (cs.LG); Risk Management (q-fin.RM)
*备注: 19 pages, 9 figures
Abstract:Recurring cash flows are the backbone of cash-flow underwriting: they separate durable revenue from one-off deposits, expose standing obligations, and ultimately determine ability to pay. For small businesses, recurrence is both indispensable and hard to estimate, because much of their income arrives as ambiguous transfers, most payment streams are short, and schedules follow calendar conventions that frequency-domain methods treat as noise. We present CADENCE, Calendar-Aware Distributional Event Network for Cash-Flow Estimation, a compact causal Transformer that treats each transaction as a token and predicts the amount and a full, nonparametric timing distribution of every cash flow’s next payment. Inheriting transaction semantics from an external labeler and a frozen text encoder keeps CADENCE compact yet general across the long tail of small-business counterparties, while predicting a calibrated distribution rather than an opaque embedding makes its outputs interpretable enough to feed existing underwriting models and power credit decisions. Ablations confirm the role of each design choice, with counterparty semantics driving recall on short histories, calendar inputs driving timing accuracy, and no-event supervision driving precision. Against the production FFT baseline, CADENCE improves one-day timing accuracy by 93.3% and next-event F1 by 6.7%, while producing calibrated distributions that serve directly as a confidence gate for downstream decisions. In an end-to-end credit evaluation on production applications, replacing FFT recurrence features with CADENCE features raises KS by 4.8%, raises the approval rate at a fixed loss rate by 18.2%, and lowers the loss rate at a fixed approval rate by 13.1%.
[LG-144] Progressive-Observation Physical-Layer Fingerprinting for Efficient Open-Set Ethernet Authentication
链接: https://arxiv.org/abs/2610.38074
作者: Chenming Zhang,Jiaxiang Geng,Aiqun Hu,Azzedine Boukerche,Peng Sun
类目: Cryptography and Security (cs.CR); Machine Learning (cs.LG); Networking and Internet Architecture (cs.NI); Signal Processing (eess.SP)
*备注: 6 pages, 5 figures, 2 tables. Submitted to IEEE ICC 2027
Abstract:Physical-layer fingerprinting (PLF) has emerged as a promising authentication approach for Internet of Things devices. However, most PLF systems acquire and process a fixed-length waveform for every authentication query, burdening resource-constrained edge nodes. To address this issue, we propose a progressive-observation PLF method that adapts the waveform length to each query. Specifically, a lightweight convolutional neural network (CNN) first rejects clearly unknown devices from a short observation. Spectral matching then accepts clearly registered devices from a slightly longer observation. Only ambiguous queries proceed to the same matching on the full-length waveform, reducing both data acquisition and computation. Because fingerprints drift across days, we also design an adaptive calibration algorithm to avoid retraining the CNN. The algorithm realigns the features and decision thresholds using a few current-day references from registered devices. We evaluate the method with 100BASE-TX Ethernet devices of two controller types on held-out days. Compared with fixed-length processing, it keeps the false rejection and false acceptance rates equal or lower for both controller types on both days. It also reduces the waveform data per query by 75.1% and the CPU processing time on a Raspberry Pi 5 by 73.7% on average. With the same monitoring bandwidth, an edge node can therefore serve about four times as many authentication queries.
[LG-145] Integral Probability Boundaries NEURIPS2026
链接: https://arxiv.org/abs/2610.38036
作者: Bogdan Kulynych,Shahab Asoodeh
类目: Machine Learning (cs.LG); Information Theory (cs.IT); Machine Learning (stat.ML)
*备注: NeurIPS 2026
Abstract:Bounds based on Integral Probability Metrics (IPMs) are ubiquitous in machine learning and statistics, appearing, e.g., in generalization, alignment, fairness, and privacy. IPMs reduce the discrepancy between two distributions to a single worst-case gap between expectations. This compression is costly: for rare events, safety failures, or under-represented groups, IPM-based bounds are loose or vacuous. We introduce a non-linear generalization of IPMs, which we term Integral Probability Boundaries (IPBs). Unlike IPMs, IPBs provide bounds that are tight at the extremes. We derive their basic properties and present algorithms for estimating them in practically relevant settings. We demonstrate the usage of IPBs in both theoretical and practical applications for alignment, judge-free performance estimation, and fairness, where they enable us to obtain sharp bounds on performance and fairness gaps, e.g., 15 p.p. more precise estimates on fairness gaps for rare groups in a realistic generative modeling task, compared to the state of the art.
[LG-146] Learning from Relative Action Credit for Escaping Local Minima
链接: https://arxiv.org/abs/2610.38034
作者: Giovanni Minelli,Giulio Turrisi,Lorenzo Amatucci,Victor Barasuol,Claudio Semini
类目: Machine Learning (cs.LG); Robotics (cs.RO)
*备注:
Abstract:Reinforcement learning holds the promise of robots that learn manipulation skills from interaction alone, without pretrained priors or demonstration data. In practice, on-policy methods such as PPO routinely stall on long-horizon, contact-rich tasks: dense reward shaping carries the policy through stages, yet stitching them together and discovering sparsely rewarded events rarely succeeds. We trace this failure to the missing action discrimination in state-value credit assignment: a state-value critic judges progress only by episode outcomes, so when the policy converges to a partial solution all outcomes coincide, and improvement stops because no action can prove better than another. We therefore extend PPO with an action conditioned value critic which, can compare the executed action with alternatives sampled at the same state and aid the learning signal. Combined with Generalized Advantage Estimation and a bound on exploration noise, we show how the extended algorithm escapes the plateau and completes precise peg insertion within a training budget where PPO with identical exploration fails.
[LG-147] Detecting Brain Cancer Progression Through Signatures of Wearable Accelerometry Data NEURIPS2026
链接: https://arxiv.org/abs/2610.38021
作者: Rishit Das,Barbora Barancikova,Matt Williams
类目: Machine Learning (cs.LG)
*备注: 13 pages, 7 figures, 4 tables. Accepted at NeurIPS 2026 TS-LIMITS Workshop
Abstract:Glioblastoma and other high-grade gliomas (HGG) have a poor prognosis and are associated with progressive functional decline. The standard of care to assess progressive disease is contrast-enhanced MRI, which is expensive, claustrophobic, and performed only once every 2-3 months. Wearable accelerometers offer a non-invasive alternative for continuously monitoring patient function between clinic visits. However, accelerometry data streams are typically noisy and highly patient-specific, making conventional time series methods and summary statistics poorly suited to detecting clinically meaningful change. This work aims to classify stable versus progressive disease states by applying rough path signatures for feature extraction from wrist-worn accelerometer data. We convert accelerometer recordings from HGG patients enrolled in the BrainWear study into walking-path segments and use signature kernel maximum mean discrepancy (SigMMD) to quantify change in patients’ walking behaviour. We overcome the challenge of inter-patient heterogeneity by adjusting for the natural variation in a patient’s walking during stable tumour periods. This patient-specific normalisation substantially improves separation, yielding a candidate biomarker of tumour progression. Our results provide proof-of-concept evidence that signature-based analysis of wearable accelerometry can distinguish stable from progressive disease states, motivating larger longitudinal studies of continuous wearable monitoring as a digital biomarker in neuro-oncology.
[LG-148] Preserving Safety During Sequential Fine-Tuning: A Factorial and Geometric Analysis of Replay
链接: https://arxiv.org/abs/2610.38001
作者: George Ruellan,Diana Benavides-Prado
类目: Machine Learning (cs.LG)
*备注:
Abstract:Benign fine-tuning can erode safety alignment in large language models, and sequential adaptation substantially compounds this risk. Prior work mitigates this erosion by replaying harmful-prompt refusals and benign-compliance examples, while other work instead matches base-model logits via distillation. However, the individual contributions of supervised exposure and distillation to safety behavior and refusal geometry during sequential adaptation remain unclear. We factor these two strategies apart and study their behavioral and geometric effects during sequential LoRA fine-tuning of Mistral-7B and Llama-3-8B across diverse task sequences (code generation, German-language summarization, and science question-answering). Behaviorally, supervised replay exposure yields the largest reduction in HarmBench attack success rate (ASR) on Mistral, averaging 51.1 points below base, but at a substantial over-refusal cost, averaging 53.4 points above base on OR-Bench-hard. Logit distillation, without any supervised replay loss, largely prevents safety erosion at a fraction of this cost: an average 5.4-point ASR reduction against only a 5.1-point rise in over-refusal. Llama shows the same asymmetry between the two strategies. Combining exposure and distillation moderates exposure’s over-refusal while leaving downstream task scores largely unaffected. Geometrically, the two models diverge: on Mistral, distillation preserves the base refusal direction more closely than exposure does, whereas on Llama, refusal behavior after exposure largely survives ablation of the extracted refusal direction. This dissociation between models indicates that preserving the base refusal direction is not a reliable proxy for behavioral safety.
[LG-149] Deep learning for non-transvenous ICD screening: tailored neural network architectures extended experiments and explainability
链接: https://arxiv.org/abs/2610.37993
作者: Bartosz Morawski,Samuel J. Ward,Alex Touze,Lin-Thiri Toon,Paul R. Roberts,John Paisey,Alain B. Zemkoho
类目: Machine Learning (cs.LG)
*备注:
Abstract:This paper studies two non-transvenous implantable cardioverter-defibrillators (ICDs), namely the Subcutaneous ICD (S-ICD) and the Extravascular ICD (EV-ICD). Both avoid the complications of transvenous leads but are prone to oversensing, which can trigger inappropriate shocks associated with increased mortality. Oversensing risk is predicted by the ratio of T- or P-wave amplitude to R-wave amplitude (T:R, P:R), and these ratios vary over a 24-hour period. Yet S-ICD screening uses a single 10-second electrocardiogram (ECG) segment, and the EV-ICD has no formal screening protocol whatsoever. We present a 24-hour ambulatory screening framework for both devices which includes the first T:R analysis in an EV-ICD cohort and the first full evaluation of P:R prediction. We introduce AmpRatioNet, a family of 1D convolutional neural network architectures operating directly on the ECG signal, and benchmark it against the best existing approach - based on 2D phase space reconstruction images. Evaluation uses 4,605 clinician-annotated 10-second segments from 89 patients, with patient-level 10-fold cross-validation and bootstrap confidence intervals. AmpRatioNet-SE demonstrates superior performance on every metric in each setting with no device- or ratio-specific redesign. An explainability analysis against interpretable classical peak labelling algorithms shows that AmpRatioNet-SE’s increase in error on oversensing-prone segments is up to 14 times smaller than theirs. Segment-level predictions are aggregated into a 24-hour timeline, extending S-ICD screening beyond 10 seconds and offering a plausible EV-ICD screening pathway pending device-level validation.
[LG-150] Predicting Sparse Observables in Parameterized Unsteady Flows Using Coarse-Mesh Graph Neural Networks
链接: https://arxiv.org/abs/2610.37986
作者: Brener Ramos,William R. Wolf,Tarcísio C. D. Oliveira,Scott T. M. Dawson
类目: Machine Learning (cs.LG); Fluid Dynamics (physics.flu-dyn)
*备注:
Abstract:Full-order surrogates for high-dimensional nonlinear flows are often unnecessary when applications require only selected degrees of freedom. We propose reduced-order models that bypass full-state reconstruction by deploying graph neural networks (GNNs) directly on coarse meshes. Applied to parameterized 2D/3D confined cylinder flows and an unevenly heated cavity over Reynolds/Rayleigh-number ranges, the models generalize to unseen conditions and accurately predict sparse probe locations and integrated quantities. Coarse-mesh GNNs match or outperform finer-mesh surrogates while reducing training time and memory, even for 3D chaotic dynamics. A dynamics-informed adaptive sampling strategy guided by frequency-spectrum variations reduces maximum errors by an order of magnitude versus uniform sampling at negligible cost. Coarse-mesh training also makes noise-augmented multi-step unrolling practical, improving long-horizon autoregressive predictions. The approach outperforms parametric DMD, sparse MLPs, U-Nets, and Fourier neural operators at lower cost, and enables system identification and state estimation from sparse observations. These results show that coarse-mesh representations, adaptive sampling, and noise-augmented unrolled training form an efficient framework for reduced-order modeling of parameterized unsteady flows without solver-resolution reconstruction.
[LG-151] Separability Is Not Alignability: Representations for Spectrum-to-Molecule Retrieval NEURIPS2026
链接: https://arxiv.org/abs/2610.37979
作者: Karol Musielinski,Fanurs C.E. Teh,Takahiko Koyama,Artur Dubrawski
类目: Machine Learning (cs.LG)
*备注: 22 pages, 3 figures. Accepted as a poster at the NeurIPS 2026 Workshop on Representations for the Physical Sciences (RPS). OpenReview: this https URL
Abstract:Identifying a small molecule from its tandem mass spectrum is limited by paired data. Foundation models offer a way around this scarcity: pretrain a spectrum encoder on unlabeled spectra and a molecule encoder on a large corpus of structures, then align the two frozen representations, as in vision-language and as the Platonic Representation Hypothesis suggests. We audit this route on MassSpecGym with spectrum encoders retrained free of test-fold contamination. Alignment needs a metric relating distinct molecules, but the objectives that produce spectrum encoders never see molecular structure: we prove that some predictive objectives fix molecular identity without fixing that metric. Moreover, we find that a candidate-based objective rewards a spectrum-independent prior over the benchmark’s decoys, which we isolate. What ultimately addresses data scarcity is not architectural: simulated spectra surpass frozen geometry.
[LG-152] Sometimes less is more: Investigating inductive biases of MLPs and Transformers on noisy k-parity features
链接: https://arxiv.org/abs/2610.37976
作者: Alasdair Paren
类目: Machine Learning (cs.LG)
*备注:
Abstract:We empirically investigate how MLPs and Transformers differ when trained on data sets containing one or more features encoded by noisy k-parity. Noisy k-parity gives us hard synthetic data sets in which we can control the predictivity and availability of every feature. Across a range of data distributions, data scales and model sizes, we find large differences in the training dynamics and inductive biases of the two architectures: Transformers are more robust to noisy features but struggle to learn complex, low-availability features, whereas MLPs show the opposite trend and improve steadily with more data. We also find that, under a fixed training budget, increasing the number of unique training examples can slow Transformer optimisation enough to reduce final accuracy, so that a smaller data set performs better. Training-accuracy results show this is an optimisation-speed effect; it shrinks, but does not vanish, when the budget is scaled with the data set size. By giving precise control over feature predictivity and availability, this testbed offers a clear starting point for understanding when and why Transformers and MLPs learn differently, and for designing data and training budgets that exploit those differences.
[LG-153] FORGE: Feasible Online Reduced-Order Reference Generation and Whole-Body Execution for Agile Humanoid Locomotion
链接: https://arxiv.org/abs/2610.37974
作者: Adrian B. Ghansah,Yujin An,Jeeseop Kim,Soon-Jo Chung
类目: Robotics (cs.RO); Machine Learning (cs.LG)
*备注: 8 pages, 6 figures. Accepted to the 2026 IEEE-RAS 25th International Conference on Humanoid Robots (Humanoids)
Abstract:This paper presents a two-stage learning framework for versatile and agile humanoid locomotion that couples online reduced-order model (ROM) reference generation with whole-body tracking control. In the first stage, we train a command-conditioned whole-body reference generator using single rigid body dynamics combined with full humanoid kinematics. In the second stage, the frozen ROM policy is rolled out online to generate references for training a whole-body tracking policy on the full-order dynamics model. Learning a kino-dynamically consistent reference generator in the simplified ROM space enables us to generate references over a broad command space and over multiple locomotion strategies using a single unified framework. We demonstrate the approach on a Unitree G1 humanoid robot, achieving longitudinal velocities from -2.0 m/s to 2.8 m/s, lateral velocities up to ±1.0 m/s, angular velocities up to ±3.0 rad/s, one-legged hopping, and continuous transitions across the commands, all using one unified reference policy and one whole-body tracking policy.
[LG-154] Application of PDE emulators for a multi-material multi-physics simulation problem ICML
链接: https://arxiv.org/abs/2610.37972
作者: Diane Oyen,Morgan Jones,Adithan Kathirgamaraju,Abu Bucker Siddik,Alexander Most,David Schodt,Christine Sweeney,Kyle Hickmann,Michael McCann,Jemma Stachelek,Alexius Wadell,Michal Kucer,Mahindra Rautela,Siddharth Mansingh,James Amarel,Gerd J. Kunde,Emily Casleton Taylor,Soumi De,Ayan Biswas,Aimee Hungerford,Nathan Debardeleben,Earl Lawrence
类目: Machine Learning (cs.LG)
*备注: Accepted at ICMLA 2026, Rochester, MI, USA
Abstract:Machine learning (ML) architectures are being developed for fast emulation of numerical simulators to accelerate both scientific discovery and engineering design. While these approaches are well-tested on popular PDE (partial differential equation) benchmarks, it is necessary to evaluate their performance on datasets that more accurately reflect industrial and engineering applications. We evaluate several computer vision-based architectures trained as surrogate models on a challenging computational physics task involving multiple materials, revealing significant implementation challenges for adapting certain pre-trained models, including transformer-based architectures. For such multi-physics multi-material datasets, our results underscore the importance of learning algorithms designed for sparse and imbalanced data, techniques for adapting spatial encodings and physical consistency metrics, in addition to methods for reducing the cost of generating sufficient training data.
[LG-155] Federated Learning for Semantic Communications over Multiple Access Channels
链接: https://arxiv.org/abs/2610.37951
作者: Yalin E. Sagduyu,Tugba Erpek,Yi Shi
类目: Information Theory (cs.IT); Distributed, Parallel, and Cluster Computing (cs.DC); Machine Learning (cs.LG); Networking and Internet Architecture (cs.NI); Signal Processing (eess.SP)
*备注:
Abstract:This paper presents a federated learning (FL) framework for semantic communication (SemCom) over the multiple access channel (MAC), where multiple distributed transmitters simultaneously convey semantic information to a common receiver using end-to-end learned semantic representations. Semantic feature extraction, compression, wireless transmission, and semantic inference are jointly optimized through FL-based SemCom. Unlike conventional SemCom systems that train users independently, simultaneous semantic transmissions over the MAC enable concurrent training of multiple transmitters within the same transmission interval, thereby reducing training overhead and channel usage. Federated semantic aggregation is incorporated with shared and personalized encoder components, allowing transmitters with limited local training data to benefit from semantic representations learned from transmitters possessing larger datasets while preserving local adaptation. Evaluated using the CIFAR-10 dataset over fading channels, the proposed approach maintains high semantic inference accuracy under shared wireless access and significantly improves performance under severe data imbalance. In the meantime, simultaneous semantic transmissions can significantly reduce training overhead while preserving robust semantic performance across distributed transmitters with heterogeneous local datasets. These results highlight the potential of combining FL and SemCom to support scalable and communication-efficient edge intelligence in future wireless networks.
[LG-156] Finite-Step Geometry of Marginal-Preserving Velocity Fields in Generative Flows
链接: https://arxiv.org/abs/2610.37936
作者: Ewsey Obzherin,Valentin Khrulkov
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注: 60 pages, including appendices
Abstract:Generative flows can admit different velocity fields that induce the same continuous marginal evolution but behave differently after numerical discretization. We study this finite-step non-equivalence while keeping the marginal path and numerical method fixed. Projected modified equations characterize the component of local integration error that changes the marginal law and give sufficient conditions for higher marginal convergence order. For affine fixed-shape distributions, we characterize exact finite-step reachability and derive spectral obstructions and sharp nearest-reachable Wasserstein errors. For diffusion probability-flow ODEs, we construct skew-score perturbations that preserve the exact marginal path under exact scores and quantify their leakage under score error. Controlled Gaussian and mixture experiments recover the predicted order, reachability, and obstruction phenomena.
[LG-157] Bi-Phase Training: Learning Efficiently From Scratch AACL
链接: https://arxiv.org/abs/2610.37934
作者: Abel Gurung,Joseph Campbell
类目: Machine Learning (cs.LG)
*备注: Findings of AACL-IJCNLP 2026
Abstract:Pre-training foundational models is computationally expensive as models scale. In this work, we introduce Bi-Phase Training (BPT), a parameter-efficient pre-training method that preserves the high-rank learning dynamics observed in full-parameter optimization with substantially reduced trainable parameters. BPT parameterizes each weight update as a composition of constrained high-rank left/right transformations via trainable diagonal matrices and low-rank adapters whose updates are periodically merged into the frozen base weights, resulting in a compounding expansion of accessible directions during training. Across language-model pre-training from scratch (Qwen2.5 models from 100M to 1.5B parameters and OLMo-2-1B) on FineWeb-Edu (10B and 100B tokens), BPT achieves validation perplexity close to full-parameter baselines while using far fewer trainable parameters; e.g., at the 1.5B scale, BPT uses 66% fewer trainable parameters with only a small validation-perplexity gap. Additionally, a comprehensive evaluation across diverse downstream tasks shows that BPT preserves much of the full-parameterized model’s downstream performance.
[LG-158] Generative modeling for fusion: a large-scale benchmark of stellarator fast-ion simulations
链接: https://arxiv.org/abs/2610.37910
作者: Misha Padidar,Teresa Huang,Mark Goldstein,Michael Czekanski,Jeffrey Soules
类目: Machine Learning (cs.LG); Plasma Physics (physics.plasm-ph)
*备注:
Abstract:Stellarators offer a promising magnetic confinement approach for achieving controlled nuclear fusion. However, stellarator design remains challenging, as evaluating configuration performance requires computationally expensive, high-fidelity simulations. In particular, key performance metrics derived from guiding center simulations are costly to include directly in the design optimization loop, and are almost exclusively used as diagnostics. Machine learning has the potential to advance fusion research by enabling rapid generation and evaluation of new stellarator configurations. Effectively applying ML for fusion demands large-scale datasets that capture both design diversity and high-fidelity simulation outputs. In this work, we introduce the Benchmark of Stellarator Simulations (BoSS), a dataset of stellarator designs paired with guiding center simulation results and trajectories. The BoSS is the first large-scale dataset of guiding center simulations, containing simulation data for 180,000 distinct magnetohydrodynamic equilibria spanning a wide range of magnetic geometries and confinement properties. To exemplify the utility of ML to accelerate stellarator design, we formulate and address two sets of modeling tasks: generative modeling for designing stellarators with good particle confinement, and surrogate modeling for predicting stellarator heat deposition patterns and particle trajectories. The tasks reveal the potential impact ML can have in the stellarator field and important opportunities to improve the applicability of ML models in the sciences. Our baselines generate diverse, performant stellarators and provide useful surrogate predictions, and point to long-horizon trajectory forecasting and out-of-distribution generation as key directions for future work. The BoSS can be accessed through this https URL.
[LG-159] SAC-GCL: Structure-Aware Compression for Graph Contrastive Learning
链接: https://arxiv.org/abs/2610.37898
作者: Muhammad Ifte khairul Islam,Tanvir Hossain,Esra Akbas
类目: Machine Learning (cs.LG); Social and Information Networks (cs.SI)
*备注:
Abstract:Graph Contrastive Learning (GCL) has achieved remarkable success in self-supervised representation learning by maximizing agreement between augmented views of a graph. Although augmented view generation is the central step in GCL, existing models rely on stochastic augmentation of the input graph, such as edge perturbation and node masking, which lack structural awareness and frequently destroy critical topology or introduce semantically inconsistent views. While recent adaptive methods improve upon random perturbations, they primarily operate on localized node or edge neighborhoods and fail to capture higher-order, macro-level structural organization crucial for graph classification. To address this limitation, we propose Structure-aware Compression for Graph Contrastive Learning (SAC-GCL), a novel framework that employs structural decomposition and graph compression as a principled view generation strategy. Moving beyond random augmentations, SAC-GCL leverages structural decomposition techniques to extract cohesive subgraphs and condense them into supernodes, constructing coarsened graph views that preserve local density and capture higher-level structural patterns while reducing noise and redundancy. We leverage multiple cohesive subgraph models to derive complementary views of the graph, each capturing distinct structural properties. To capture multi-resolution information, a hierarchical view encoder processes these compressed views by initializing super-node features with shared-weight cohesive subgraph embeddings. Our results from extensive experiments on six benchmark datasets demonstrate that structure-aware compression provides a powerful alternative to random perturbations, yielding more robust and discriminative graph representations for graph classification tasks.
[LG-160] Interpretable neural equation learning for high-dimensional materials data
链接: https://arxiv.org/abs/2610.37880
作者: Philip Jacobson,Prasad Iyer,Remi Dingreville,Saaketh Desai
类目: Machine Learning (cs.LG); Materials Science (cond-mat.mtrl-sci)
*备注:
Abstract:Equation learning seeks to discover compact, interpretable equations from data using various machine-learning techniques. However, few workflows discover and interpret learned equations in sparse, high-dimensional materials datasets. We present nn-EQL, an equation-learning framework that generates candidate equations through iterative training, pruning, and retraining of a neural network with customizable, user-specified neuron-level activation functions. Candidate equations are ranked using a minimum description length (MDL)-inspired score, balancing error and sparsity, and diagnosed to assess term-wise contributions, input-scaled sensitivities, and uncertainties. We demonstrate nn-EQL on standard AI-Feynman benchmarks as well as multiscale high-dimensional materials datasets such as finite-element simulations of heterogeneous materials, and high-throughput experimental thin-film deposition data. nn-EQL integrates with tabular data and high-dimensional data via standard encoder models, learning equations and providing insight into learned terms. Looking forward, integrating physically interpretable representations, multimodal and multi-fidelity data, and adaptive data acquisition could enable equation learning to formulate and test mechanistic hypotheses.
[LG-161] Improving Data-Constrained Scaling with Hyperparameter Transfer and Ensembling
链接: https://arxiv.org/abs/2610.37842
作者: Ethan Baron,Shmuel Berman,Anirudh Satheesh,Akshay Vegesna,Samip Dahal,Andrew Gordon Wilson
类目: Machine Learning (cs.LG)
*备注:
Abstract:As growing compute budgets outpace the growth of high-quality training data, large language models must be trained with both compute and data constraints in mind. Unfortunately, existing prescriptions for data-constrained scaling suffer from two important limitations. First, past works hold key hyperparameters such as weight decay fixed with scale, despite the fact that optimal hyperparameters vary wildly with model size and number of epochs. Second, past works neglect ensembling, a promising axis for scaling under data constraints. As a surprising consequence, prior prescriptions can lead to loss that increases with compute budget. We address both limitations by fitting power laws that prescribe how to transfer hyperparameters with scale and scaling laws that prescribe how to allocate compute with ensembling. Training on 100M tokens from FineWeb, our recipe reduces validation loss by over 0.15 nats compared to prior recipes, a 14% decrease in perplexity, and achieves monotonic improvements with compute through ensembling. These gains are achieved using prescriptions fit to smaller runs and validated at up to 12 \times larger compute budgets.
[LG-162] Depth through Recurrence: Towards Ultra-Efficient On-Device ASR
链接: https://arxiv.org/abs/2610.37833
作者: Chen Feng,Tianyi Xu,Ramchalam Kinattinkara Ramakrishnan,Yicheng Lin,Shaojie Zhuo,Zhaocong Yuan,Chenzheng Su,Xiaopeng Zhang
类目: Machine Learning (cs.LG)
*备注:
Abstract:Modern automatic speech recognition (ASR) systems have achieved remarkable accuracy by scaling model depth and capacity, but at the cost of increased memory, energy, and computational demands. On edge devices where ASR is often most needed, such as watches and glasses, deploying such large models is infeasible due to extremely constrained resources. This raises a fundamental question: can we achieve high representational power in deep ASR models without scaling up parameterization? In this work, we revisit the role of depth and identify layer wise representational dynamics, in which most layers learn functionally similar transformations. Motivated by this insight, we propose a block-recurrent ASR architecture that replaces parameterized depth with a small set of recurrent blocks, each consisting of weight-shared layers, thereby preserving effective depth while drastically reducing model size. Through representation-guided grouping and knowledge distillation, block-recurrent models retain the accuracy of large reference models while using an order of magnitude fewer parameters. Across Open ASR Leaderboard benchmarks, a model with only two shared blocks recovers 97.4% of the accuracy of large models on average. The compact backbone further enables efficient on-device personalization through lightweight MoE-LoRA adaptation, allowing user-specific models to match or surpass the accuracy of general-purpose ASR systems with significantly lower inference cost. By reusing weights across recurrent steps, the proposed architecture reduces memory-access overhead and inference energy consumption. Together, these results show that high quality ASR can be achieved without large parameterization, and that structured recurrence provides a principled path towards ultra-compressed, energy-efficient, and personalized speech recognition on the edge.
[LG-163] Anisotropic Trust Region Warping for Prior-Data Fitted Networks in Bayesian Optimisation NEURIPS
链接: https://arxiv.org/abs/2610.37737
作者: Samuel Younger,Tinkle Chugh,George De Ath
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注: Accepted at the 2026 NeurIPS Workshop on Optimization for Machine Learning (OPT). 18 pages, 12 figures
Abstract:Prior-Data Fitted Networks (PFNs) replace the hyperparameter refitting of Gaussian Process (GP) surrogates in Bayesian optimisation with a single forward pass, at the cost of a prior that is fixed at meta-training time. We show empirically that a PFN’s predictive log-likelihood falls increasingly behind that of a refitted GP as the objective function’s lengthscale grows beyond the mode of the PFN’s meta-training prior, and that this deficit carries over to optimisation performance. To correct the mismatch without retraining the network, we introduce Anisotropic Trust Region (ATR) warping. Per-dimension lengthscales are estimated by kernel target alignment, and an affine rescaling of the input space maps the objective onto the lengthscale the PFN expects, with a trust region around the anchor determining which observations are retained, leaving the PFN’s \mathcalO(d \cdot N^2) inference cost unchanged. On functions drawn from a Matérn-3/2 kernel, ATR warping improves optimisation performance of the PFN at both extremes of the lengthscale range. Our implementation is publicly available at this https URL.
[LG-164] hreshold Compatibility and Support-Selection Diagnostics in Neural Discovery of Partial Differential Equations
链接: https://arxiv.org/abs/2610.37726
作者: Zeen Song
类目: Machine Learning (cs.LG)
*备注: 19 pages, 6 figures, 4 tables; includes supplementary material
Abstract:Accurate regression need not yield the correct equation after support selection. For a median/sample-standard-deviation threshold in neural PDE discovery, we derive an exact two-signal retention boundary and a complete-vector perturbation bound. Exact advection-diffusion libraries lose diffusion despite coefficient recovery to numerical precision. True-support refits of forty saved sparse states reduce diffusion error without restoring all derivative accuracy. A fixed sequential threshold ridge comparator recovers exact support in sixty ideal controls but selects extra terms in every learned state. All-fold correct-support candidates are rejected by conditional validation; post hoc energy pooling preserves their higher minimum loss in all sixty states. Supplied-support estimates outperform unchanged selected equations in all 240 physical-residual comparisons on four unused analytic trajectories. With coefficients, observation points and weights fixed, exact replacement of the learned library and target reverses their learned-row ordering in 59/60 states. Weighting both losses also fails the specified Burgers diffusion-benefit criterion at every tested budget. These finite diagnostics distinguish threshold incompatibility, refit damage and score rejection; they establish neither automatic correction nor general recovery or predictive stability.
[LG-165] Learning Graph Beliefs for Active Radio Mapping With Decision-Certified Multi-Step Sensing
链接: https://arxiv.org/abs/2610.37717
作者: Wangqian Chen,Hao Wang,Yumeng Zhang,Chenghong Bian,Junting Chen,Jun Zhang
类目: Networking and Internet Architecture (cs.NI); Machine Learning (cs.LG)
*备注:
Abstract:Radio maps provide essential wireless propagation knowledge for network optimization and autonomous platforms. Acquiring dense measurements, however, is prohibitively costly, requiring active sampling to prioritize informative measurements. Existing active sampling methods typically rely on variances from prescribed spatial models or learned pointwise uncertainty, which may not fully capture the spatial correlation of different measurements. This paper proposes GraphBARC, a graph-structured active radio map construction framework that integrates learned spatial modeling with analytical Bayesian inference. A graph network learns correlated node and edge factors using only acquired measurements, inducing a joint full-map belief that supports analytical observation updates without retraining. Rather than immediate one-step acquisition, it adopts a nonmyopic sensing strategy that evaluates joint map-wide Bayes risk reduction over feasible multi-step routes, accounting for information redundancy among planned measurements. We reuse candidate responses and derive residual-based bounds directly on route-value differences, retaining shared response-error structure to avoid uniformly high-accuracy evaluation. An error-aware continuation bound screens unpromising prefixes, after which only responses needed for unresolved comparisons are refined. We further establish a computable decision-gap bound and show that certified exclusions remain valid during refinement. Experiments show that GraphBARC reduces mean root-mean-square error (RMSE) by over 16% relative to the evaluated baselines and core planning time by 56% compared with the fixed-accuracy, unpruned implementation.
[LG-166] Generation Not Selection: LLM Proposals for Certified Access-Control Policy Mining
链接: https://arxiv.org/abs/2610.37716
作者: Gautam Kumar(Indian Institute of Technology Kharagpur, India),Ravi Sundaram(Northeastern University, USA),Shamik Sural(Indian Institute of Technology Kharagpur, India)
类目: Cryptography and Security (cs.CR); Machine Learning (cs.LG)
*备注:
Abstract:Attribute-Based Access Control (ABAC) policy mining—recovering a compact, auditable policy from an access log—is posed as minimum set cover and solved greedily, which enjoys an s^(1+\ln m) guarantee. That guarantee is \emphconditional on a precondition the literature never states: the candidate pool handed to greedy must contain the optimal rules. Generation, not selection, is the binding constraint, and existing generators are purely \emphsyntactic—they surface a rule only when enough log records witness it, so they fail on precisely the rare exception rules that fill a heavy-tailed policy. We replace the generator with a language model proposing rules \emphsemantically, from attribute names, and keep the combinatorial selector to carry the guarantee: every proposal is filtered against the observed denials, so soundness is independent of the proposal mechanism and an unreliable proposer can only waste candidates, never grant access. Over five seeds on each of two enterprise domains, adding semantic proposal to a syntactic miner dominates it on \emphten of ten runs—strictly better on seven—with near-optimal policies ( 9.2 – 9.6 rules against s^=9 , versus 11.2 – 14.6 ) and false grants cut four- to sevenfold. Replacing the syntactic miner rather than augmenting it wins only where witnesses are scarce. An anonymization ablation isolates the mechanism: opaque attribute names cost two thirds of the model’s rule recovery while leaving the syntactic baselines provably unmoved, and cost nothing where witnesses are plentiful. Semantic knowledge and log evidence are substitutes.
[LG-167] Moment-Guided Optimizers for Noisy Deep Learning: MGO-SC as a Simple Consistency Gate for Adam-Style Updates
链接: https://arxiv.org/abs/2610.37712
作者: Alicem Koyun
类目: Machine Learning (cs.LG)
*备注: 8 pages, 4 figures, 3 tables. Accepted at the 2026 IEEE Global Emerging Technologies Conference (iGET), Irvine, CA, USA, Oct. 21-25, 2026
Abstract:Adaptive optimizers are widely used in deep learning, yet their behavior can degrade under heavy stochasticity and label noise where the sign and magnitude of gradients can be unstable. We study MGO-SC (Moment-Guided Optimizer with Sign-Consistency), a lightweight modification of Adam-style methods that derives a coordinate-wise consistency gate from the magnitude of a sign exponential moving average (EMA): coordinates whose gradient sign is stable are trusted, while coordinates whose sign fluctuates are down-weighted. We show that the naive form of this gate, which can only shrink updates, loses effective step size and underperforms when the learning rate is matched to the baseline. We therefore introduce a mean-preserving variant that reallocates trust across coordinates without changing the net per-tensor step magnitude. On CIFAR-10 with ResNet-18, under a matched learning rate and decoupled weight decay, mean-preserving MGO-SC consistently improves over its AdamW baseline across clean, symmetric, and asymmetric label noise (e.g., +1.4 points at 40% symmetric noise; paired bootstrap 95% CI excluding zero; Wilcoxon p0.01), while adding no second-order information or per-sample filtering. We report matched-seed effect sizes, nonparametric bootstrap confidence intervals, and paired significance tests, and we release code and raw logs.
[LG-168] wo-Colour Zig-Zag Numberlink on Rectangles: A Forbidden-Pattern Characterization a Constructive Polynomial-Time Algorithm and a Lean Verification
链接: https://arxiv.org/abs/2610.37691
作者: Zzyv Zzyzek
类目: Data Structures and Algorithms (cs.DS); Machine Learning (cs.LG)
*备注:
Abstract:In Zig-Zag Numberlink, pairs of endpoints in a grid must be joined by vertex-disjoint paths that together visit every cell. The problem is NP-complete when the number of pairs is un- bounded (Adcock et al.). For a single pair it is the Hamiltonian path problem in rectangular grid graphs, solved by Itai, Papadimitriou and Szwarcfiter. Two pairs on a rectangle form the paired 2-disjoint path cover problem. We settle the case of two pairs on rectangles with the endpoints anywhere in the grid. If both sides of the rectangle are at least 11, an instance is solvable if and only if no entry of a finite catalogue of forbidden patterns occurs. The catalogue consists of a parity condition, an alternation condition on the border, a handful of local corner and edge patterns, 65 listed corner configurations, and a rule that propagates forced routes out of the corners before testing alternation. If a side is at most 10, a frontier dynamic programming algorithm decides the instance in time linear in the long side. The characterization is construc- tive. Solutions are built by deleting empty lines, splitting the rectangle by straight cuts, and splicing the pieces back together, in time O((R+C) RC). Both directions of the characterization and the correctness of the dynamic programming algorithm are verified in Lean 4. The finite case analyses, which are window certificates for the corner configurations and a check of all irreducible instances with sides 11-22, are carried out by the Lean compiler via native decide. We describe the algorithm, the catalogue, the proof structure, the compromises made in the formalization, and implementations in Python, JavaScript and C that produce identical output.
[LG-169] Accelerated Dataflow: scaling inference in streaming data analytics with efficiency
链接: https://arxiv.org/abs/2610.22731
作者: RuiLong Jiang,Dustin Rhodes,Danny McCormick,Yuta Labur,Reza Rokni
类目: Distributed, Parallel, and Cluster Computing (cs.DC); Databases (cs.DB); Machine Learning (cs.LG)
*备注:
Abstract:Streaming data analytics helps businesses identify opportunities and spot risks in a timely fashion. As large language model (LLM) usage grows, people are exploring incorporating agents or LLMs in streaming data pipelines to help find patterns in datasets without supervision or interpret natural languages for content moderation. However, most streaming systems are currently optimized for data movement and perform poorly on GPU intensive workloads such as performing inference with LLMs. We introduce a system on Google Cloud Dataflow designed to tackle hybrid streaming machine learning workloads with co-designed autoscaling on heterogeneous worker pools and prefix-aware batching. This increases throughput 1.5x to 2x for streaming pipelines using LLMs while maintaining similar latency with up to 75% peak resource savings and 84% lower peak tail latency compared with existing streaming data processing systems.
[LG-170] Learning developmental dynamics from lineage context
链接: https://arxiv.org/abs/2610.40447
作者: Gokul Gowri,Nic Fishman,William N. Colgan,Minming Wang,Yusuke Kijima,Sachin A. Kumar,Aleksandr Zimin,Luke W. Koblan,Marinka Zitnik,Zachary D. Smith,Stephen Bates,Philippe Rigollet,Jonathan S. Weissman
类目: Quantitative Methods (q-bio.QM); Machine Learning (cs.LG)
*备注:
Abstract:A central goal of biology is to predict the dynamics of multicellular systems. Lineage context (the states of a cell’s sisters, cousins, etc.) could inform these predictions, but how best to leverage it remains unclear. Here we introduce lineage inpainting, a generative modeling approach that uses lineage context with tree-structured transformers and stochastic interpolants to model cell state dynamics across lineage trees. Applied to a lineage-traced transcriptional atlas of mouse embryogenesis, it recovers dynamics not resolved by gene expression alone, including multiple transcriptional convergence events where distinct lineages are drawn to a common state. Despite being fit on unperturbed development, lineage inpainting can generate hypotheses about perturbation responses by simulating counterfactuals. These simulations agree in many cases with observed developmental phenotypes, including a single-cell measurement of the Tbx5 knockout response. Our results show that lineage information provides a rich source of context for learning multicellular dynamics.
[LG-171] OCCAMS: A Multi-Task Learning-Based Investigation of the CMB Power Spectra
链接: https://arxiv.org/abs/2610.40415
作者: Ankit Mandal,Rahul Shah,Md. Wali Hossain,Somasri Sen,Anjan A Sen,Supratik Pal
类目: Cosmology and Nongalactic Astrophysics (astro-ph.CO); Instrumentation and Methods for Astrophysics (astro-ph.IM); Machine Learning (cs.LG)
*备注: 13 pages, 4 figures, 4 tables, plus appendices. Comments are welcome
Abstract:How many latent variables are needed to fit the cosmic microwave background (CMB) temperature and polarisation power spectra when a neural network is given no cosmological model, and to what extent do they map onto the standard cosmological parameters? We use multi-task learning with variational autoencoders trained on CLASS-generated spectra (TT, TE, EE) over 2\le\ell\le2500 . No cosmological model is built into the network, and the number of latent variables, L , is varied over L=5 , 6 and 7 . Their fits are compared to the Planck 2018 data using \hat\chi^2 , AIC, BIC and Bayesian evidence. Their correspondence with cosmological parameters is measured with normalised mutual information post-training. The loss weights are set within a single training run by uncertainty weighting, with the weight of the disentanglement term either fixed (semi-automated, SA) or learned (completely automated, CA). The reconstruction loss is split over multipole ranges in three configurations that separate the roles of temperature and polarisation and of the low- and high- \ell ranges. In four of the six combinations of configuration and strategy, the fit improves from L=5 to L=6 and not beyond, and L=6 is the best, or close to the best, in five of the six cases. So, six latent variables are needed, as many as the free parameters of the standard, model-dependent approach. SA gives clearer disentanglement, while CA needs no a priori choice of the disentanglement weight and has the higher evidence at L=5 and 6 : the two strategies are points on a trade-off between interpretability and fit to the data. Notably, the latent variables for \tau and A_s share almost no information in any model, consistent with large-scale polarisation separating the two, as in standard analyses. A deep-learning approach can thus give a compact, data-driven parametrisation of the CMB spectra.
[LG-172] Statistical Analysis of Synthetic Data with Verification: A Linear Regression Framework
链接: https://arxiv.org/abs/2610.40254
作者: Heejune Sheen,Harrison H. Zhou
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Statistics Theory (math.ST)
*备注:
Abstract:Synthetic data are widely used for training modern AI systems. However, their statistical understanding remains limited. It is well recognized in statistics that repeated self-training on synthetic data can amplify errors inherited from pre-trained models, leading to model collapse. Verification plays a crucial role in selecting high-quality synthetic data to help prevent model collapse and improve model performance. We study verified self-training in a linear regression framework, where a pre-trained estimator generates synthetic data with injected generation noise and a verifier selects samples according to a verification threshold. Post-training on the accepted samples updates either all coordinates (full fine-tuning) or only an active subset (partial fine-tuning). Our analysis reveals three potentially counterintuitive effects. We establish that tighter verification does not always lead to lower prediction risk. We also find that a suitable level of injected generation noise can reduce prediction risk, while excessive noise can increase it. Moreover, full and partial fine-tuning can respond differently to the same verification rule, since pre-training error in the frozen coordinates affects both sample selection and post-training risk. We analyze these effects by deriving finite sample risk bounds for four separate regimes determined by the relative scales of the verification threshold, generation noise, and pre-training error. The bounds precisely characterize how verification affects prediction risk across these regimes. In particular, we quantify how the generation noise, the effective accepted sample size, and the covariance of the accepted covariates affect prediction risk. The analysis extends to iterative verified full fine-tuning, where sufficiently tight verification reduces the initial error contribution at a geometric rate.
[LG-173] Confidence-Aware Speed-over-Ground Estimation from Doppler Spectra via Reconstruction-Based Representation Learning
链接: https://arxiv.org/abs/2610.40253
作者: Malte Stelzer,Marvin Sach,Timo Bartels,Dinu Mihailescu-Stoica,Sören Kohnert,Florian Poprawa,Marc Zeller,Tim Fingscheidt
类目: ignal Processing (eess.SP); Machine Learning (cs.LG)
*备注:
Abstract:Reliable speed-over-ground estimation from ground-facing Doppler spectra is essential for mobile platforms such as railway systems. Classical signal-processing approaches can provide accurate estimates under favorable conditions, but often degrade in challenging conditions. We propose confidence-aware speed-over-ground estimation trained on rail-domain data that estimates speed over ground and frame-level confidence from Doppler spectra with several novel contributions: First, we transfer the idea of exploiting a reconstruction task by an autoencoder for performance estimation from previous work to Doppler RaDAR. Second, we append a speed-over-ground estimation head operating on the autoencoder’s latent space features. Third, we omit the reconstruction decoder during inference but add a low-cost dedicated confidence head for frame-wise error-bound estimation. Fourth, we reveal how noise augmentation training can provide high robustness to the method. Experiments show that our approach improves speed estimation accuracy over classical baselines while providing informative confidence estimates that indicate the reliability of individual predictions, even under adverse signal conditions.
[LG-174] Conformal Annotation with Randomized Expert Review: Application to Online Cell-Type Labeling
链接: https://arxiv.org/abs/2610.40252
作者: Xuhe Zhang,Zhi Huang,Ying Jin
类目: Methodology (stat.ME); Machine Learning (cs.LG); Applications (stat.AP); Machine Learning (stat.ML)
*备注: 48 pages, 17 figures
Abstract:Artificial intelligence (AI) models can accelerate cell-type annotation in digital pathology, but safely scaling their use requires determining, with error control, which model-generated labels can be accepted without expert review. This is challenging in evolving workflows because cell morphology and type composition may change across tissue regions and slides, making one-time calibration unreliable. We propose Conformal Annotation with Randomized Expert review under Distribution shift (CARED), a distribution-free framework that leverages randomized expert audit to achieve long-run error control for AI-assisted annotation under distribution shift. For each incoming batch, CARED uniformly selects a subset of cases for expert review and uses the acquired labels to calibrate later decisions. Among unreviewed cases, CARED determines a subset of them to safely skip expert labeling, while suggesting candidate labels for the others for downstream resolution. We prove that this workflow provides long-run control of cumulative error over both the full stream and the cells outside of the audit sample, without stationarity or exchangeability assumptions. We further enhance the framework by introducing class-conditional guarantees to ensure error control within important but rare cell types, together with a utility module to target more automatic resolutions, smaller candidate sets, and greater biological coherence, without compromising validity. Experiments on pancreas, lung, and ovary annotation streams, together with a stream constructed from real expert annotations, show that CARED controls the cumulative errors while a substantial fraction of cells (up to 70%) bypass expert review.
[LG-175] Hidden Disparities in Ranking: Category-Aware Fairness via Distributional Stability
链接: https://arxiv.org/abs/2610.40182
作者: Gayane Taturyan,Charlotte Laclau,Stephan Clémençon
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:We study the problem of fair ranking in recommender systems. We provide a formal probabilistic framework in which ranking policies are distributions over permutations, and the utility of a user is the exposure given to the items they interact with, relative to that of an ideal ranking. We consider utility parity across user groups, as well as its category-aware refinement, conditional utility parity, which requires parity within each category of items. Following a recent view of fairness as distributional stability, we show that both criteria are equivalent to the stability of the utility under changes in the composition of user groups. Optimizing over mixtures of ranking policies leads to a second-order cone program, for which we provide generalization bounds. A numerical study validates our theoretical findings.
[LG-176] Fisher Contrastive Learning to Mitigate Feature Suppression: A Case Study in Acute Promyelocytic Leukemia Diagnosis
链接: https://arxiv.org/abs/2610.40162
作者: Shushan Wu,Yongkai Chen,Xilin Gong,Yufang Liu,Wenxuan Zhong,Ping Ma
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注: 31 pages, including supplementary material; 3 figures and 2 tables
Abstract:Self-supervised contrastive learning (SSCL) learns representations by contrasting augmented views of the same sample, but it can exhibit feature suppression, where dominant or easy-to-learn factors suppress other class-relevant signals, degrading the performance of representations, particularly in high-dimensional data with heterogeneous features. We introduce Fisher Contrastive Learning (FCL), which uses a regularized Fisher trace criterion to address feature suppression. FCL learns embeddings by maximizing the discriminative strength of Fisher discriminant functions, producing representations that preserve class-relevant information implied by the augmentation process. We prove that the representations learned by FCL could alleviate the feature suppression effects. We demonstrate FCL on simulated benchmarks designed to induce feature suppression and on a real-world biomedical application, pretraining on large white blood cell microscopy image dataset and transfer learning for diagnosis of acute promyelocytic leukemia in an external cohort. In all experiments, FCL outperforms the self-supervised baselines and yields more robust and discriminative representations.
[LG-177] Reinforcement Learning Under Runtime Confounding: Policy Learning and Off-Policy Evaluation
链接: https://arxiv.org/abs/2610.40056
作者: Raphael Kim,Michele Santacatterina,Ramin Zabih,Keith Barnatchez
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Methodology (stat.ME)
*备注:
Abstract:Reinforcement learning (RL) is increasingly used to train robust, adaptable systems across healthcare, e-commerce, and robotics. In these settings, legal, ethical, or logistical concerns may preclude certain key variables that drive decisions from being collected during deployment time. This gives rise to a setting known as runtime confounding: the logged data hold full observability, but deployment time lacks knowledge of certain variables, known as runtime confounders. Unlike single timepoint and non-adaptive settings, runtime confounding under reinforcement learning has the potential to propagate over time, as earlier decisions affect future states, observations, and action assignments, leading to biased learning. We address these problems with three contributions: we (1) develop policy learning algorithms under runtime-confounding, (2) develop semi-parametric efficient off-policy estimation procedures, and (3) provide diagnostics for understanding the price of runtime confounding.
[LG-178] Probabilistic Meshless Methods on Manifolds
链接: https://arxiv.org/abs/2610.39855
作者: Federico Pavesi,Noémie Jaquier,Antonio Candelieri
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:Probabilistic meshless methods solve partial differential equations (PDEs) by placing a Gaussian process prior on the solution and conditioning it on the PDE at a finite set of collocation nodes. The posterior provides both an approximation of the solution and a Bayesian quantification of its error. While these methods are well established on Euclidean domains, their extension to manifolds lacks, to our knowledge, any convergence analysis. We extend them to compact manifolds and study their convergence in two settings. First, for known manifolds whose Laplace-Beltrami eigenpairs are available, the prior can be evaluated exactly, but the PDE operator generally cannot be applied to it in closed form. We replace this operator by a standard generalized finite difference discretization on a point cloud sampled from the manifold, and show that the resulting method is consistent at the points of the cloud, at a rate governed by their fill distance. Second, for implicit manifolds, known only through a point cloud, both the prior and the operator must be approximated. We propose a new generalized finite difference scheme with a free stencil radius, and show that it retains the consistency of the standard construction while controlling the growth of its coefficients, and that an appropriate radius yields uniform convergence of the discrete Gram matrix. In both settings, we derive error bounds for the posterior mean which separate the collocation error intrinsic to the method from the error due to approximating the operator and, for implicit manifolds, the prior. Experiments on the circle, the sphere and the flat torus confirm our theoretical claims. On real-world cortical thickness data, our approach attains the accuracy of established discretizations, while conditioning on fewer than a tenth of the observations and providing a credible interval at every vertex.
[LG-179] Adaptive Federated Optimization Across Smooth and Nonsmooth Regimes
链接: https://arxiv.org/abs/2610.39716
作者: Zimeng Wang,Ali Dadras,Alp Yurtsever
类目: Optimization and Control (math.OC); Machine Learning (cs.LG)
*备注:
Abstract:AdaGrad-style methods can simultaneously adapt to both the order and constants of smoothness in centralized optimization. However, existing extensions to federated optimization with heterogeneous client objectives and multiple local steps require the smoothness order to be specified in advance. We address this limitation with AdaFed, a federated method that adapts across several regimes under a single choice of parameters. In the deterministic setting, AdaFed achieves an O(1/T) convergence rate for both smooth nonconvex and smooth convex objectives, and an O(1/\sqrtT) rate for Lipschitz nonsmooth convex objectives up to logarithmic factors. Under a light-tailed stochastic-gradient noise assumption, these guarantees extend to the stochastic setting with additional logarithmic factors and noise-dependent terms. Experiments across the three problem classes further support our theoretical findings.
[LG-180] Bayes-Consistent Losses for Adaptive Inference in Classification
链接: https://arxiv.org/abs/2610.39709
作者: Shirley Xiaoqi Liu,Florence Regol,Antonios Valkanas,Patrick Rebeschini,Mark Coates
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注: 69 pages, 13 figures
Abstract:Adaptive inference systems balance accuracy and inference cost by routing inputs to models with varying capabilities. Although jointly training models and routers is empirically more sample-efficient than training them separately, principled losses for joint training remain underexplored. We address this gap in a classification setting by proposing cost-sensitive hinge losses for jointly training classifiers with single-stage or cascade routers. Single-stage routers directly select a model per input, whereas cascade routers sequentially decide at each stage whether to predict with the current model or defer to the next stage, which uses a more accurate but more costly model. For both settings, we prove linear calibration bounds establishing Bayes consistency: minimizing the hinge risk recovers the Bayes-optimal risk under the target cost-sensitive zero-one loss. Moreover, since routers can require more labeled data to train than classifiers, we propose a novel post-training prediction-powered inference (PPI) procedure that uses unlabeled data to refine a trained single-stage router. Experiments on synthetic and real datasets demonstrate that the proposed hinge losses are effective for joint training and that PPI can correct routing errors when labeled data is scarce.
[LG-181] ransformer-inspired architectures make machine-learned electron densities accurate and transferable enough to bypass self-consistency
链接: https://arxiv.org/abs/2610.39687
作者: Jeheon Woo
类目: Chemical Physics (physics.chem-ph); Machine Learning (cs.LG); Computational Physics (physics.comp-ph)
*备注: 36 pages, 4 figures, 2 tables; Extended Data (4 figures, 2 tables) and Supplementary Information included
Abstract:We introduce machine learning models that accurately predict the electron density, the central variable of density functional theory. They are built on Transformer-inspired architectures, which require little computation per parameter and can therefore be enlarged at low cost. The proposed models, SwinUNETR-V2 and its attention-free counterpart NeXtUNETR, achieve state-of-the-art accuracy on QM9, with normalized mean absolute errors in the density averaging 0.075% and 0.046%, respectively. The predicted densities are accurate enough to bypass the self-consistent iteration: a single Kohn-Sham step reproduces the self-consistent electronic structure. For NeXtUNETR, the median errors are 0.036 meV in the total energy, 0.65 meV/angstrom in the forces and 1.8 meV in the HOMO-LUMO gap. The forces not only recover the self-consistent minima in geometry optimization but also successfully drive ab initio molecular dynamics. Fine-tuned on small datasets, both networks also transfer accurately to systems containing elements absent from QM9 or larger than its molecules.
[LG-182] Formalizing the Intrinsic Dimension of Objective Landscapes by a Generalized Notion of Co-dimension
链接: https://arxiv.org/abs/2610.39674
作者: Ernesto Araya,Gitta Kutyniok,Weilin Li,Johannes Maly,Manjot Singh
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Metric Geometry (math.MG)
*备注:
Abstract:In this work, we formalize the concept of intrinsic dimension of objective landscapes introduced in Li et al. (2018), which measures the structural complexity of learning problems by training along randomly generated subspaces. We define the intrinsic dimension of a set \Theta\subset\mathbbR^p as the minimal dimension d_\mathrmint(\Theta) such that a random plane of dimension d_\mathrmint(\Theta) in \mathbbR^p intersects \Theta with probability one. We show that it is a generalized notion of co-dimension that depends on the global geometry of \Theta . We study mathematical properties of the intrinsic dimension and develop techniques to estimate the intrinsic dimension of sets \Theta that arise in learning problems. More specifically, we derive upper and/or lower bounds where \Theta is the set of optimal parameters in student-teacher settings with shallow and deep ReLU networks, including finite and infinite sample cases, and with and without regularization. We discuss how our theory leads to novel types of generalization bounds, thereby showing that the intrinsic dimension is a measure of learning complexity. Our results connect techniques from seemingly different areas of machine learning research such as classification of symmetries of ReLU networks and stochastic geometry.
[LG-183] Carbon-Aware Electric Vehicle Routing Problem with Decision-Focused Learning
链接: https://arxiv.org/abs/2610.39645
作者: Jaume Reixach,Wanru Chen,Jiaming Wu,Balázs Kulcsár
类目: Optimization and Control (math.OC); Machine Learning (cs.LG)
*备注:
Abstract:Electric vehicles produce no tailpipe emissions, but charging them produces emissions that depend on the time-varying generation mix of the electricity consumed. We study an Electric Vehicle Routing Problem with a carbon-aware objective that accounts for charging-related emissions alongside conventional routing costs. The problem determines routing, charging, and waiting decisions, allowing charging demand to be shifted toward periods with lower grid carbon intensity. Because these decisions depend on forecasts of future carbon intensity, we propose a decision-focused learning approach so that the training of the model employed for predicting the carbon intensity of the grid over time is solver-aware. In particular, we train a linear predictor using approximate SPO+ directions and employ a time-limited Large Neighborhood Search followed by exact postprocessing as the underlying solver. Experiments on realistic Germany-based instances show that explicitly considering charging-related emissions reduces realized emissions by 28.76% on average relative to carbon-blind optimization, although at the expense of higher non-carbon operating cost. Moreover, decision-focused training reduces realized total cost by 2.30% on average relative to conventional prediction-focused learning and performs better on 75.5% of the test instances, despite substantially lower pointwise predictive accuracy. Finally, we show that its average advantage remains positive across varied solver configurations and carbon-cost weights.
[LG-184] Learning Kolmogorov Backward Operators for Spectral Analysis and Forecasting: A Data-Driven Variational Galerkin Framework
链接: https://arxiv.org/abs/2610.39580
作者: Yuanchao Xu,Claire Valva,Ben Gao,Isao Ishikawa
类目: Dynamical Systems (math.DS); Machine Learning (cs.LG)
*备注:
Abstract:For a time-homogeneous Markov process, the fixed-lag solution operator ( (T_\tau f)(x)=\mathbb E[f(X_t+\tau)\mid X_t=x] ) of the Kolmogorov backward equation evolves observables linearly and deterministically, even when the dynamics are stochastic. We approximate this operator using a two-stage method where we first approximate this operator directly from data, without knowledge of the governing equation (e.g., Fourier Neural Operator), and then compress the Galerkin projection onto a finite-dimensional observable space that is fixed in advance or selected variationally. The error analysis traces the learning error of the estimator, the sampling error of the Galerkin inner products, and the invariance and reconstruction properties of the observable space to the computed spectra and forecasts. The sampling error enters only through a multiplicative factor. Near an isolated spectral cluster, the invariance defect of a subspace controls its distance from the target invariant subspace. The forecast error is controlled by the one-step error of the reduced model and the projection error of the observable; when the measure is invariant and the reduced model preserves constants and is stable on mean-zero observables, the bound is uniform in the horizon. For deterministic dynamics, a learning rate for Fourier neural operators (FNOs) gives one explicit bound for the learning error. Numerical experiments illustrate spectral approximation and forecasting, including long-rollout stability and a comparison with extended dynamic mode decomposition (EDMD) when fewer transition pairs than basis functions are observed.
[LG-185] Hidden Phase Transition in Single-Polymer Stretching under Elongational Flow
链接: https://arxiv.org/abs/2610.39575
作者: Aiqing Zhu,Fushuai Wang,Eleonore Vissol-Gaudin,Qianxiao Li,Beatrice W. Soh
类目: oft Condensed Matter (cond-mat.soft); Machine Learning (cs.LG); Computational Physics (physics.comp-ph)
*备注:
Abstract:Using machine learning, we identify a second transition in polymer chain stretching driven by planar elongational flow, beyond the well-known coil-stretch transition. We achieve this by developing a conditional extension of Universal and Identifiable Dissipative Dynamics (UIDD) that jointly learns reduced stochastic dynamics from microscopic trajectories across flow strengths and yields the corresponding stationary nonequilibrium free energy landscapes. As the flow strength increases, the evolving landscape exhibits the expected coil-stretch transition, and then a previously unrecognized dynamical transition. In the latter, the stretched states persist, but the distribution along a folding coordinate becomes bimodal when conditioned on small end-to-end separation. Large-deviation analysis of a simplified analytical chain model reproduces both transitions and shows that they are phase transitions in the long-chain limit. Single-molecule DNA experiments provide qualitative support for the dynamical transition through the predicted reversal of the folding response.
[LG-186] Diffusion models for the O(3) nonlinear sigma model
链接: https://arxiv.org/abs/2610.39568
作者: Daniel Schubring
类目: atistical Mechanics (cond-mat.stat-mech); Machine Learning (cs.LG); High Energy Physics - Lattice (hep-lat)
*备注: 29 pages, 7 figures
Abstract:A score-based diffusion model is trained on the two-dimensional classical Heisenberg model, the lattice representation of the O(3) nonlinear sigma model. Neural networks are trained on three different coarse-grainings of the same training set to implement a multiscale sampling scheme that modestly improves both the accuracy and computational cost of direct sampling on the finest lattice. Stein (or Schwinger-Dyson) diagnostics for the two-point correlation function are introduced to assess the quality of the learned score without relying on reverse diffusion sampling. These diagnostics are also incorporated into an additional training loss. We discuss several technical considerations relevant to implementing diffusion models on S^2 . These include a consideration of lifting the network input to S^3 , as in the CP^1 representation, and developing a boundary-layer approximation to the short-time spherical heat kernel that extends its usefulness to larger diffusion times.
[LG-187] abCUQ: Context-Uncertainty Quantification for Tabular Foundation Models
链接: https://arxiv.org/abs/2610.39558
作者: Yuebei Shi,Molei Liu,Zijian Guo
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:Tabular foundation models use in-context learning to estimate conditional expected outcomes and average treatment effects for specific groups of individuals. Using these estimates to guide scientific decisions requires reliable uncertainty quantification. The challenge is to account simultaneously for two sources of uncertainty: limited-context bias and context variation arising from the small labeled sample supplied at inference. Predictive spread does not directly quantify both sources of uncertainty, and we observe that the ordinary bootstrap may substantially overestimate context variation, whereas subsampling may underestimate it. We introduce TabCUQ, a perturbation-based framework that injects noise into observed outcomes and applies the foundation model to the perturbed contexts. TabCUQ estimates context variation using wild-bootstrap dispersion and accounts for limited-context bias using the maximum deviation between perturbed and original estimates. For a kernel-regression analogue whose estimation error admits an exact decomposition into these two components, we show that our framework provides asymptotically valid pointwise coverage. In simulations, TabCUQ substantially improves coverage, with average interval widths typically about twice those of an oracle bias-aware benchmark. Across seven real-world tasks, it also improves coverage of reference functions learned from much larger samples.
[LG-188] he Cost of Working Memory in Streaming Mean Estimation
链接: https://arxiv.org/abs/2610.39540
作者: Zeyu Wu,Zihao Wang,Minghua Deng
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:A streaming estimator must compute from the evidence that fits in memory. We determine when storing an accurate mean is easier than computing it. For Gaussian mean estimation over the Euclidean unit ball in \mathbbR^d , we establish the minimax rate under squared error with K persistent states, a known horizon, and a public time index: \min\1, K^-2/d+\fracdn (1+\frac\log(ed)1+(\log K)/d)\ . Unrestricted computation followed by K -level encoding has rate \min\1,d/n+K^-2/d\ . At a fixed sufficiently small error, statistically optimal sample size therefore requires order d\log d working bits, while order d bits suffice to store the answer. The lower bound tracks repeated state compression through an entropy potential obtained by Gaussian smoothing. A matching construction combines a multiscale modular counter with workspace reused across coordinate groups. In one dimension, it attains the terminal rate n^-1+K^-2 . Any fixed confidence requirement that the realized loss not exceed the zero baseline preserves these rates. Increasing confidence creates a further memory requirement: at fixed nontrivial risk in every fixed dimension, failure probability e^-\gamma n requires and permits order \log n working states for sufficiently small fixed \gamma0 . In the scalar problem, a fixed terminal alphabet suffices under the same protection. These bounds separate the memory needed to represent an answer from the memory needed to accumulate evidence and validate a decision.
[LG-189] Conic-Lifting for Learning-rate-free Convex Optimization
链接: https://arxiv.org/abs/2610.39531
作者: Julien Grand-Clément,Joon Kwon,Francesco Orabona
类目: Optimization and Control (math.OC); Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:A central topic in online and stochastic optimization is to design algorithms that do not require tuning of hyperparameters. Successful strategies in online learning include AdaGrad-style learning rates, Conic Blackwell algorithms, and parameter-free algorithms. However, these classes require different, often difficult, analyses. We develop a unifying family of Online Convex Optimization (OCO) algorithms that require no user-tuned learning rate on convex domains that may be unbounded. The construction lifts the domain to a cone by adding one scalar coordinate, an idea going back to Blackwell (1954). The lifted subgradients are orthogonal to the corresponding lifted decisions, leaving the comparator scale free to optimize in the regret analysis. Applying Follow-the-Regularized-Leader (FTRL) and Online Mirror Descent (OMD) on the cone recovers Conic Blackwell Algorithm (CBA) and Conic Blackwell Algorithm Plus (CBA ^+ ) on bounded domains in Hilbert spaces and extends them to finite-dimensional normed spaces with a strongly convex squared norm. For arbitrary nonempty closed convex domains in these spaces, a known gradient bound yields parameter-free regret guarantees. Surprisingly, on an unconstrained Hilbert space, the construction recovers the Krichevsky–Trofimov (KT) coin-betting algorithm exactly. We also give a parameter-free algorithm for unconstrained domains that requires no gradient bound.
[LG-190] Distilling Large-Scale Machine Learning Interatomic Potentials to Predict Long-Timescale Electrolyte Transport Properties
链接: https://arxiv.org/abs/2610.39513
作者: Ishan Amin,Yue Jian,Nitesh Kumar,Christine Zhang,Jianwei Lai,Casey S. Mezerkor,Aaron M. Melemed,Jonathan R. Thurston,Neil P. Dasgupta,J. David Bazak,Ethan J. Crumlin,Samuel Blau,Aditi S. Krishnapriyan
类目: Chemical Physics (physics.chem-ph); Materials Science (cond-mat.mtrl-sci); Soft Condensed Matter (cond-mat.soft); Machine Learning (cs.LG); Computational Physics (physics.comp-ph)
*备注:
Abstract:Accurately predicting the transport properties of liquid electrolytes is critical for next-generation battery design and requires force fields that are both quantum-mechanically accurate and fast enough for long-timescale simulation. Recent large-scale pretrained machine learning interatomic potentials (MLIPs) approach density-functional-theory accuracy, but remain too slow to access the necessary timescales to converge transport properties. We provide a knowledge distillation procedure to transfer the accuracy of a large-scale “teacher” MLIP into fast, system-specific “students”, without any additional ab initio data. Our distilled models run ~40x faster, reproduce teacher-predicted structural observables, and produce stable 20 ns trajectories in days, rather than the months the teacher would require. Across a diverse set of sodium- and lithium-ion electrolytes, student-predicted self-diffusivities, ionic conductivities, and shear viscosities track experiment and outperform classical force-field predictions, establishing distillation as a practical route to accurate, long-timescale simulations necessary for high-throughput electrolyte screening.
[LG-191] Satisficing Regret Minimization in Bandits with Knapsacks
链接: https://arxiv.org/abs/2610.39321
作者: Qing Feng,Tianyi Ma,Ruihao Zhu
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:We study satisficing regret minimization in bandits with knapsacks (BwK), where a learner sequentially selects randomized policies, i.e., distributions over arms, to obtain mean rewards above a satisficing threshold as frequently as possible while respecting finite resource budgets. The performance is measured by satisficing regret, the cumulative deficits of the selected policies’ mean rewards below this threshold. Resource constraints introduce two difficulties: an algorithm can incur linear satisficing regret even with small standard regret as deficits to the threshold cannot be compensated, and random consumption can cause early resource depletion even when policies are feasible in expectation. To address these challenges, we propose SELECTwK, a general algorithmic template that uses a BwK learning oracle algorithm with sub-linear expected (standard) regret upper bound to search for and test candidate satisficing policies (i.e., policies whose mean rewards exceed satisficing threshold and whose expected resource consumption is feasible over the time horizon). Our regret analysis, supported by an LP-duality argument, establishes that SELECTwK attains constant expected satisficing regret when a satisficing policy exists. We further introduce a stronger regret condition that bounds the expected sum of positive reward deficits to the optimal one at each time step, under which SELECTwK achieves a sharper constant bound together with a sub-linear bound in the time horizon. As a complement, SELECTwK still preserves the oracle’s sub-linear expected standard regret bound if the satisficing threshold is unattainable. Finally, we establish the oracle conditions and corresponding satisficing regret bound for existing finite-armed BwK and dynamic pricing algorithms, and conduct numerical experiments on dynamic pricing with limited inventory to demonstrate the empirical performance of SELECTwK.
[LG-192] Zero-Complement Extrapolation: Parameter Recovery and Classification in Severely Imbalanced Logistic Regression under Rank-Deficient Regimes
链接: https://arxiv.org/abs/2610.39194
作者: Chongsun Park
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Methodology (stat.ME)
*备注: Submitted to a peer-reviewed journal; 16 pages, 1 figure, 4 tables
Abstract:Parameter estimation and classification in binary logistic regression deteriorate severely when the number of minority class observations ( n_1 ) is smaller than the feature dimension ( p ). Under such rank-deficient rarity ( n_1 p ), maximum likelihood estimators suffer from quasi-complete separation, penalized likelihood methods exhibit severe intercept bias, and convex minority oversampling techniques such as SMOTE collapse into an (n_1 - 1) -dimensional affine subspace. In this paper, we propose a novel \textitSelf-Calibrated Zero-Complement estimator that shifts the paradigm from minority interpolation to majority fringe extrapolation. Grounded in Stein’s Lemma, we prove that the conditional expectation shift of the abundant majority class ( Y=0 ) induced by logistic truncation uniquely identifies the true hyperplane normal \boldsymbol\beta after Mahalanobis whitening. We further establish that extrapolating synthetic minority instances outward from the high-quantile density fringe of the majority distribution along the conjugate direction preserves full-rank covariance regularity ( \mathrmrank = p ). To accommodate both semi-supervised environments (with an unlabeled background mean) and strictly supervised subsamples, we introduce a hybrid estimator incorporating the intrinsic third-moment skewness of the majority class alongside a location-shift-invariant zero-boundary intercept calibration. Extensive Monte Carlo simulations and real-world benchmark applications (Wisconsin Diagnostic Breast Cancer and Banknote Authentication) under extreme rarity ( n_1 \in \3, 5\ ) demonstrate that our approach prevents parameter divergence, recovers the oracle hyperplane with high fidelity, and substantially outperforms Firth’s regression and SMOTE in F1-score and estimation stability.
[LG-193] PhysicsNeMo-Mesh: GPU-Native Meshes and Simulation Cases for Physics AI
链接: https://arxiv.org/abs/2610.39072
作者: Peter D. Sharpe,Corey Adams,Negin Sobhani,Sanjay Choudhry
类目: Computational Physics (physics.comp-ph); Computational Geometry (cs.CG); Machine Learning (cs.LG)
*备注:
Abstract:PhysicsNeMo-Mesh is the mesh module of the open-source PhysicsNeMo library. It provides mesh processing similar to PyVista, but on the GPU, in batches, and inside PyTorch’s automatic differentiation. Like xarray for unstructured meshes, it attaches semantics to data: fields on points, cells, and whole meshes have names and shapes, and a simulation case keeps the complete PDE problem specification, including boundary conditions and their prescribed values. It addresses three problems that plague machine learning for physical simulation. First, GPU utilization is bottlenecked by I/O, because samples are very large relative to the compute per forward pass. Second, because traditional preprocessing is expensive, it is necessarily done once, during curation, rather than differently for each model and workflow at the point of use. Finally, advanced equivariant and mesh-native architectures require information that curation ordinarily discards. Fast GPU mesh processing makes it practical to preprocess at the point of use: on one GB200 GPU, mesh operations run 6.2\text–1,068\times faster than VTK with 72 threads, and loading a whole case is 8.0\text–88\times faster than reading its VTK files ( 8.0\times when both read the same fields). Overlapped with training, this preprocessing costs 2.1% of throughput for a 127,\textms Transolver step. So one dataset serves many models and workflows: in PhysicsNeMo’s public recipe, one stored DrivAerML dataset feeds Transolver, GeoTransolver, and GLOBE. PhysicsNeMo-Mesh combines the speed of an intermediate array store with the flexibility of an archival format such as VTK. When shapes and wall conditions both vary, per-boundary conditions cut a heat-conduction error from 87% to 6\text–9% . And by keeping field ranks and named boundaries, it retains the semantic structure that next-generation architectures will need.
[LG-194] abular foundation models are not a shortcut to calibrated clinical risk prediction: a pre-registered simulation against the true risk
链接: https://arxiv.org/abs/2610.39065
作者: Ebrahim Khaled Ebrahim
类目: Methodology (stat.ME); Machine Learning (cs.LG)
*备注: 37 pages, 7 figures, 1 table; Supplementary Information (55 pages) as an ancillary file. Pre-registered on OSF Registries (doi: https://doi.org/10.17605/OSF.IO/8HMTZ ). Code: doi: https://doi.org/10.5281/zenodo.23737378%3B benchmark data: doi: https://doi.org/10.5281/zenodo.23737416
Abstract:Tabular foundation models are pretrained networks that predict risk from a table without fitting a model. Whether their risks are correct (calibrated) has been judged on real data, where true risk is unknown. In a pre-registered simulation with known true risks, we compared TabPFN-3.5, TabICLv2 and Mitra-v2 with Firth logistic regression with intercept correction (FLIC) in 58 scenarios and 1,000 random designs, then on four clinical data sets. Foundation models were not a shortcut to calibration. Their calibration slope matched FLIC’s within the pre-registered margin in 36 or 37 of 48 core scenarios and was better only at n = 200 with a 3% event rate. Even 3,000 patients were often not enough: with FLIC, TabPFN-3.5 or Mitra-v2, 63%-83% of samples gave risks close to the truth in average, intercept and slope at event rates of 10% and 30%, and 35%-50% at 3%. TabICLv2 did worse (8%-39% and 1%-3%). Across the random designs, foundation models had a lower true calibration error than FLIC beyond the margin in 25%-38% and a higher one in none, though small losses were predicted at n = 3000. On real data, no model resisted a change in event rate between populations; recalibration on 500 local patients largely repaired it.
[LG-195] Modular Bayesian Foundation Models: Bayesian time series encoders for OOD detection routing and forecasting
链接: https://arxiv.org/abs/2610.39043
作者: Mame Diarra Toure,Quang Hung Pham,Mohamed Ryad Zemouri,David Stephens,Luc Vouligny,Martin Gagnon
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Applications (stat.AP); Computation (stat.CO)
*备注: 44 pages, 15 figures
Abstract:Industrial systems continuously generate time series that vary in sampling frequency, scale, dynamics, and operating regime. Consequently a single time-series foundation model may not be able to represent all regimes effectively and retraining a large model for every new dataset or regime shift is computationally expensive. Modular foundation models address this issue by reusing specialised encoders trained on diverse datasets and by allowing the expansion of capabilities through the addition of new experts. This approach requires compact experts to minimize storage cost and a proper selection mechanism of experts for unseen data. We propose a modular foundation model framework that combines end-to-end Bayesian encoder training with uncertainty based routing. Using Singular Bayesian Neural Networks (SBNNs), we build a collection of compact Bayesian encoders whose epistemic uncertainty guides label-free selection. At the selected ranks, our Bayesian collection uses 63.8% less storage than its deterministic counterpart. We evaluate this framework on time-series data through distribution shift, encoder selection, and probabilistic forecasting. Across seven incoming datasets, uncertainty-based routing is competitive with distance-based routing in terms of forecast MSE, while reducing average ECE from 0.086 to 0.072 and average NLL from 0.907 to 0.836. It also increases PICP90 from 82.28% to 89.03%. In a separate model-family comparison, we evaluate the SBNN pipeline against full-rank Bayes by Backprop, deep ensembles, and MC Dropout. The SBNN pipeline achieves an average MSE of 0.390, comparable with 0.396 for deep ensembles, 0.389 for MC dropout, and 0.443 for the full-rank Bayesian model. It uses 63.9% fewer parameters than MC dropout, 81.9% fewer than the full-rank model, and 92.8% fewer than the deep ensemble.
[LG-196] Optimization-based lane-changing trajectory planning in mixed traffic with embedded input convex neural networks
链接: https://arxiv.org/abs/2610.38939
作者: Yuanxiang Yang,Yu Liu,Fabricio Oliveira,Yu Han,Claudio Roncoli
类目: Optimization and Control (math.OC); Machine Learning (cs.LG)
*备注:
Abstract:Lane-changing (LC) maneuvers are highly challenging vehicle interactions, with substantial implications for traffic safety and efficiency. Developing effective LC strategies for connected and automated vehicles in mixed traffic is complicated by the uncertainty associated with surrounding-vehicle behavior and the computational burden of embedding realistic prediction models within optimization schemes. To capture these complex interactions while retaining tractability, this study proposes a data-driven optimization framework that embeds neural network (NN) surrogates for safety and lateral comfort within a Bézier-curve-based trajectory planner. Specifically, the safety cost during the LC process is quantified via a surrogate composed of an input convex recurrent neural network (ICRNN) encoding the planned trajectory, a fully input convex neural network (FICNN) embedding the LC duration, and a partially input convex neural network (PICNN) decoder fusing them with historical traffic states; separately, another FICNN surrogate approximates the optimal lateral comfort cost as a function of the LC duration. Both surrogates are embedded via linear programming (LP) formulations, avoiding the binary variables required by conventional mixed-integer programming (MIP) approaches for NN-embedded optimization. We further theoretically establish their exactness at optimality and extend the framework to problems with additional analytical objective terms and multiple surrogates. Experiments on a hybrid dataset combining real-world and simulated trajectories demonstrate that the proposed method achieves surrogate accuracy competitive with unconstrained baselines while substantially improving optimization efficiency over conventional MIPs. Closed-loop simulations further show that the proposed planners markedly reduce the weighted planning cost and consistently outperform baseline methods.
[LG-197] Memorization Capacity of Invariant Convolutional Neural Nets
链接: https://arxiv.org/abs/2610.38938
作者: Sjoerd Dirksen,Patrick Finke,Paul Geuchen
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Statistics Theory (math.ST)
*备注:
Abstract:Neural networks are powerful memorizers which are often able to interpolate their training data even when the labels have been randomized. While there exists an extensive literature on the memorization capacity of fully connected networks, in practice more specific architectures with problem-dependent properties are preferred, such as convolutional networks for image classification. In this paper we consider convolutional networks which are invariant with respect to a given arbitrary finite group of orthogonal transformations, which may include translations, reflections, and/or rotations of images and invariances of higher order tensors. We prove upper bounds on the size of an invariant convolutional neural network with threshold activations that is required to interpolate a given dataset with two classes. Our bounds are formulated in terms of a complexity measure that takes the invariant structure into account and is instance-specific, i.e., it is related to the mutual arrangement of the classes in the given dataset and their geometric properties rather than the number of samples. Our proof is constructive: for given input data, instances of an interpolating network can be computed via a randomized algorithm with high probability. In numerical experiments we investigate the practical effectiveness of this algorithm.
[LG-198] UnEZ: A Fully Unsupervised Self-Templating Neural Network for Redshift and Line Strength Estimation
链接: https://arxiv.org/abs/2610.38816
作者: Asha Abd,Karl Glazebrook,Colin Jacobs
类目: Instrumentation and Methods for Astrophysics (astro-ph.IM); Astrophysics of Galaxies (astro-ph.GA); Machine Learning (cs.LG)
*备注: 22 pages, 17 figures. Submitted to MNRAS
Abstract:As spectroscopic surveys grow, estimating galaxy properties such as metallicity, star formation rate, and redshift becomes computationally cumbersome. Methods of estimating such properties rely on template spectra or curated redshift catalogs, which fail to represent the diversity of objects seen in large-scale surveys. We present UnEZ, a fully unsupervised, physics-informed machine learning model that can predict spectroscopic redshifts and line strengths with no prior redshift or line strength labels. An encoder predicts emission and absorption line amplitudes/widths, and a fixed deterministic decoder constructs a unique template, estimated redshift, redshift confidence curve, and spectral model. UnEZ was tested on the Sloan Digital Sky Survey (SDSS) Data Release 18 due to its large dataset and thoroughly vetted redshift catalog which serves as an independent benchmark. Despite the lack of templates, line strength labels, or redshift labels, UnEZ’s performance competes with modern redshift estimation algorithms. At SNR 10, UnEZ achieves an outlier fraction of 0.459% and a normalized median absolute deviation (NMAD) of 1.237 \times 10^-4 . At SNR 3, UnEZ achieves an outlier fraction of 3.89% and an NMAD of 1.471 \times 10^-4 . Furthermore, we demonstrate how UnEZ’s line strength and redshift predictions can be utilized for various purposes such as constructing template spectra required in traditional redshift estimation algorithms. UnEZ illustrates how a fully unsupervised neural network can allow redshift, among other galaxy properties, to emerge naturally from the spectra themselves, providing a template-free, label-free approach that scales to the demands of future surveys.
[LG-199] Pinpointing Pivot Points: The DECIDE Framework for Data Analysis Decision Importances
链接: https://arxiv.org/abs/2610.38797
作者: Zachary T. Rewolinski,Jan Simson,Frauke Kreuter,Christoph Kern,Bin Yu
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Methodology (stat.ME)
*备注:
Abstract:Human judgment calls made throughout the data science life cycle can change the conclusions drawn from data. Independent analysts often make different choices when given the same data, leading to often unacceptable variations in results and contributing to the ongoing reproducibility crisis across scientific disciplines. The rapid adoption of agentic tools threatens to amplify this crisis. Several approaches have been developed to address this variation by following the same strategy: repeating an analysis under different combinations of analytical decisions and inspecting the distribution of outcomes. Examples include multiverse analysis, specification curve analysis, and vibration of effects. More broadly, the Predictability-Computability-Stability (PCS) framework assesses how documented judgment calls across the data science life cycle affect conclusions, with predictability serving as a reality-check that a model captures signal. Each of these approaches results in a “multiverse” object. While the distribution of outcomes reflects the overall variability of conclusions, estimating the influence of each individual decision requires additional tools. A common heuristic is to train a model to predict the outcome from these decisions and treat its coefficients or feature importances as proxies for each decision’s influence. Although this strategy may work in certain cases, it assumes the model is correctly specified and accurately predicts the outcome, which we show is not always the case. In this work, we introduce DECIDE, a model-free framework that quantifies each decision’s importance from the actual outcomes across the plausible decision space, avoiding issues of predictability by design. We demonstrate the effectiveness of DECIDE by applying it to real-world multiverse analyses. We further provide what is, to our knowledge, the first formal definition of the multiverse.
[LG-200] Streaming Speech Recognition on a Microcontroller: a 65.7M-Parameter Transducer in 2 and 4 Bits on the ESP32-P4 KR
链接: https://arxiv.org/abs/2610.38739
作者: Billal Boumedine
类目: Audio and Speech Processing (eess.AS); Hardware Architecture (cs.AR); Machine Learning (cs.LG)
*备注: 18 pages, 2 figures, 11 tables, 1 algorithm. Code: this https URL ; weights: this https URL
Abstract:We run Kroko, a 65.7M-parameter streaming Zipformer2 transducer for English, on the ESP32-P4, a microcontroller with two 400 MHz RISC-V cores, a 128-bit SIMD extension (PIE) and 32 MB of PSRAM, but no neural accelerator. Without any training, the encoder’s 290 matrices are stored with 2 or 4 bits per weight in Hadamard-rotated groups: the width of each matrix comes from its measured sensitivity, and each matrix is first corrected toward the original model’s output, then rounded with GPTQ and a search over the group norms. The released 71 MB of int8 weights become a 26.65 MB file. Activations are int16 between the matrix products and int8 into them. A C engine with hand-written PIE kernels processes each 1.28 s chunk of audio in 0.90 s on both cores, beam search included, and its PC build computes the same bits as the chip, so accuracy is measured on whole test sets. LibriSpeech WER is 4.34% on test-clean and 10.26% on test-other, against 3.94% and 9.17% for the released model; on FLEURS and Common Voice the gap grows to 2.3 and 1.1 points. A GPU simulator shows that the integer arithmetic costs nothing and that the whole gap comes from the weights, and an ablation on all of test-clean shows that GPTQ does most of the work. We also report the port’s speed history, from 51.3 s to 0.62 s of encoder time per chunk, each design step measured again in one build, an attempt to reach 16 MB that stopped at 19.63 MB, and what did not work. We did not measure energy. The inference engine, the firmware and the 2/4-bit weights are available at this https URL.
[LG-201] A goodness of fit approach to select the optimal visit sequence for nonparametric data synthesis
链接: https://arxiv.org/abs/2610.38559
作者: R{é}my Chapelle(CESP, CB, EVDG),Bruno Falissard(CESP),Mohammed Sedki(CESP)
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:Synthetic data are a promising way to maintain privacy while disseminating research data, thereby enabling compliance with the open science principles. Popular methods for generating synthetic data include sequential synthesis, where each variable is generated conditionally on the preceding ones in a given order. This order, also called the visit sequence of the synthesizer, is known to affect the quality of synthetic data. This raises the question of the criteria that should guide its choice. In this work, we present a novel method to perform such selection based on a goodness of fit criterion, and show that it outperforms alternative structure-based methods, both in terms of synthesis quality and time complexity. The proposed method also has the advantage of being applicable to any type of data and any class of conditional distribution models.
[LG-202] Capacity of the Hopfield-Hebbian dynamics with correlated patterns
链接: https://arxiv.org/abs/2610.38437
作者: Mihailo Stojnic
类目: Machine Learning (stat.ML); Disordered Systems and Neural Networks (cond-mat.dis-nn); Information Theory (cs.IT); Machine Learning (cs.LG); Probability (math.PR)
*备注:
Abstract:Allowing for small fraction of errors in retrieved stored binary patterns, Hopfield-Hebbian neural network (NN) associative memory dynamics [14] achieves linearly scaled capacity. The famous proportionality constant \alpha_c\approx 0.14 is obtained in a large dimensional scenario as the ratio of the number of reliably stored patterns m and the patterns’ length n under the assumption that uniformly randomly chosen binary patterns are \emphindependent among themselves. While such an idealized scenario is now well understood, more practical scenarios with correlated/noisy datasets remaining challenging even for classical statistical mechanics replica approaches. We consider precisely such practical settings where each \empharchetype pattern is replaced with M correlated \emphexamples with parameter r governing the quality of correlation. We develop a powerful Random Duality Theory (RDT) based framework to analyze correlated data and already on the first level of RDT lifting obtain capacity characterizations that remarkably well match behavior exhibited through simulated dynamics. Moreover, characterizations explicitly feature correlation parameters M and r , which allows to precisely determine their effects on phase-transitioning behavior of the underlying dynamics. Our results are also a well-fitting complement to the existing literature. Namely, to circumvent challenging statistical mechanics analyses, an alternative combination of spectral and replica methods is considered in [4]. We confirm that [4]'s elegant spectral characterizations are excellent approximations in the low correlated regime. At the same time, we also close the gap in the regimes with higher correlation levels where spectral methods deviate from the practical dynamics performance. Subjects: Machine Learning (stat.ML); Disordered Systems and Neural Networks (cond-mat.dis-nn); Information Theory (cs.IT); Machine Learning (cs.LG); Probability (math.PR) Cite as: arXiv:2610.38437 [stat.ML] (or arXiv:2610.38437v1 [stat.ML] for this version) https://doi.org/10.48550/arXiv.2610.38437 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-203] Barren Plateaus in Quantum Linear Combinations: Interference and Lie-Algebraic Control
链接: https://arxiv.org/abs/2610.38193
作者: Amir Koutahi,Pouya Kananian,Hans-Arno Jacobsen
类目: Quantum Physics (quant-ph); Emerging Technologies (cs.ET); Machine Learning (cs.LG)
*备注:
Abstract:A coherent sum of quantum circuits introduces interference that can alter the training signal. We distinguish a single linear combination of unitaries (LCU) from a circuit that repeatedly alternates branch operations and ancilla mixing. For a single-pass LCU, independent branches with zero matrix means cannot erase the differentiated branch’s gradient variance: the total variance is at least the branch variance times the fourth power of its coefficient magnitude. This bound holds at any gate position and does not require the circuit unitaries to be Haar-random. Beyond this lower bound, we derive exact loss and gradient formulas that quantify the additional contributions from interference between branches. One of these terms depends on relative coefficient phases and vanishes when the branch ensemble is invariant under multiplication by i \mathbbI ; a four-qubit matchgate example shows that it can be nonzero. For repeated interleaving with full special-unitary ancilla control, we determine the generated Lie algebra from the associative algebra of the branches, including the available central phase directions. This control result predicts concentration only with additional assumptions on circuit sampling. Matchgate examples exhibit polynomial single-pass LCU variances and exponential concentration under joint Haar-random sampling. We also derive coefficient-selection rules and bounds for postselection, measurement error, and approximate moments.
[LG-204] MiniMMD: Minimization of MMD for Sampling Inference and Integration
链接: https://arxiv.org/abs/2610.38122
作者: Zonghao Chen
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:Many problems in computational statistics and machine learning can be formulated as approximating a target distribution by minimizing a discrepancy over a restricted class of candidate distributions. This minimum-discrepancy viewpoint appears, for example, in numerical integration, sampling, generative modelling, and statistical inference, but these instances are often studied separately in the literature. This thesis develops the first unified treatment of these many problems through the minimization of the maximum mean discrepancy (MMD), a kernel-based discrepancy between probability measures. The central theme is that MMD is not only a statistically meaningful distance between probability distributions, but also a tractable optimization objective to design practical and efficient algorithms. This thesis summarizes the recent theoretical and empirical advancements of such minimum-MMD based algorithms.
[LG-205] Neural Compression of Long ADMM Trajectory for Multiparametric Quadratic Program NEURIPS2026
链接: https://arxiv.org/abs/2610.38079
作者: Liang Wu,Honghui Zheng,Bo Yang,Xu Yang,Yilin Mo,Ján Drgoňa
类目: Optimization and Control (math.OC); Machine Learning (cs.LG); Systems and Control (eess.SY)
*备注: Accepted at NeurIPS 2026. Code: this https URL
Abstract:Solving large-scale multiparametric quadratic programs (mpQPs) in real time often requires thousands of iterative optimization updates, creating a major computational bottleneck in model predictive control and learning-enabled systems. This paper proposes TQPNet, a self-supervised neural compression framework that learns to compress long ADMM optimization trajectories into a compact neural warm start followed by a few refinement iterations. We derive a low-dimensional ADMM scheme operating on the primal-dual variables (x,\beta) and show that its iteration admits an equivalent ReLU-layer representation, termed ADMM(ReLU). Building on this insight, TQPNet combines a multilayer perceptron with ReLU activations (MLP(ReLU)) and a small number of unrolled ADMM(ReLU) refinement layers. The network is trained offline using an optimality-condition-informed residual loss, eliminating the need for solver-generated labels. The proposed residual-based framework enables both self-supervised training and online solution certification through adaptive ADMM(ReLU) refinement. Experiments on large-scale mpQPs and real-time MPC applications demonstrate that TQPNet achieves high accuracy, strong generalization, and substantially faster inference than conventional optimization-based methods.
[LG-206] Conformal Predictive Distributions for Contextual Order Fulfillm ent Optimization
链接: https://arxiv.org/abs/2610.38046
作者: Tinghan Ye,Amira Hijazi,Pascal Van Hentenryck
类目: Optimization and Control (math.OC); Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:Ship-to-home fulfillment requires a retailer to select a fulfillment location and carrier before the delivery time is known. These decisions often rely on static transit-time estimates, which cannot represent how delivery-time uncertainty varies across orders and fulfillment options. This paper constructs conformal predictive distributions (CPDs) using Conformal Predictive Systems (CPS) for regression and the cross Venn-Abers predictor (CVAP) for multiclass classification, and uses them in a contextual optimization model for fulfillment decisions. The model is solved by contextual sample-average approximation (C-SAA). The regret of each fulfillment decision, its expected cost above that of the best decision, is bounded by forecast error, scenario-evaluation error, and the candidate-generation gap. The forecast term depends on how well calibration residuals or class probabilities represent the current order. On more than five million proprietary order records, CVAP raises exact-day accuracy from 49.5% for the rule-based estimate to 64.1%. Paired with a decision rule that selects a quantile weighting late errors more heavily, the split CPS (SCPS) gives a point estimate that raises late-delivery detection from 3.8% to 79.4%. Residual-based CPS distributions also meet or slightly exceed nominal interval-coverage targets. In decision experiments with a this http URL simulator over six test dates, C-SAA with CVAP lowers mean total cost by 7.1% and C-SAA with the Mondrian CPS (MCPS) lowers it by 6.7%, relative to C-SAA with the original nonconformal forecast. Both cut the late-delivery rate from 5.7% to below 4.0%.
[LG-207] Efficient Deep Gaussian Process Surrogates via Mini-Batch Latent MAP Estimation
链接: https://arxiv.org/abs/2610.38038
作者: Chun-Yi Chang,Chih-Li Sung
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注: 35 pages including references, 11 pages appendix, 5 figures
Abstract:Deep Gaussian processes (DGPs) are flexible surrogate models for computer experiments with nonstationary response surfaces, but inference over their latent representations remains computationally challenging. Existing methods typically rely on variational inference or Markov chain Monte Carlo (MCMC), which impose variational-family assumptions or require costly posterior sampling. We propose the latent maximum-a-posteriori deep Gaussian process (LaMP-DGP), which uses gradient-based optimization to obtain MAP estimates of the latent representations and enables efficient probabilistic prediction through linked-GP calculations. To scale LaMP-DGP to large datasets, we develop a mini-batch optimization algorithm and, under a polynomial mini-batch bias-decay condition, establish a stationarity bound for latent-state optimization that motivates a growing batch-size schedule. Across numerical experiments and real-data analyses, LaMP-DGP provides consistently competitive predictive performance relative to doubly stochastic variational inference, stochastic gradient Hamiltonian Monte Carlo, and fully Bayesian Vecchia DGP inference. Computationally, LaMP-DGP is substantially faster than the Vecchia MCMC approach while remaining broadly competitive with the stochastic-gradient alternatives. These results position LaMP-DGP as a competitive and computationally efficient alternative for DGP inference. An R package, “LaMPDGP”, is available for implementing the proposed method.
[LG-208] Finite-step statistical mechanics of scale-invariant learning: spectral equations of state and curvature transport
链接: https://arxiv.org/abs/2610.38012
作者: Alexander Itin
类目: atistical Mechanics (cond-mat.stat-mech); Disordered Systems and Neural Networks (cond-mat.dis-nn); Machine Learning (cs.LG)
*备注: 45 pages, 6 figures, 9 tables
Abstract:Stationary learning admits an ideal-gas description under restrictive noise assumptions. In neural networks whose gradient noise departs from these assumptions, however, an entropy-scaling relation resembling a Maxwell relation remains approximately valid. We ask which thermodynamic-looking relations follow from scale symmetry rather than equilibrium, and how departures from the ideal-gas description can be quantified, for discrete stochastic gradient dynamics with memory. An exact rescaling of trajectories explains the entropy relation without assuming thermal equilibrium. Within a local harmonic approximation, self-adjusting weight norms yield a spectral equation of state whose virial coefficients are set by moments of the curvature spectrum, and coupled parameter groups can sustain stationary probability currents. Our central result beyond this approximation is an exact curvature-transport identity. Together with a finite-step loss balance, it decomposes corrections to the harmonic fluctuation relation into measurable curvature changes, finite-step remainders, gradient-memory correlations and finite-window drift terms. A related displacement-weighted curvature change provides a diagnostic of the arrow of time; its mean and second moment give a lower bound on the relative entropy between forward and reversed state pairs. We also calculate the activation-boundary contribution to curvature for rectified teacher-student networks with Gaussian inputs. Tests in sphere models, teacher-student networks and convolutional networks on CIFAR-10 and CIFAR-100 support the transport description and identify limits of the harmonic and cubic-remainder approximations. These results separate symmetry-based relations from equilibrium assumptions and provide quantitative diagnostics of nonideal fluctuations in learning dynamics.
[LG-209] New Multiple Descent Phenomena Induced by Heterogeneous Data
链接: https://arxiv.org/abs/2610.38002
作者: Mark-Anthony Moisescu-Pareja,Elvis Dohmatob
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:Modern machine learning models are often trained on large datasets assembled from many different sources, whose statistical properties can differ substantially. Understanding how this heterogeneity affects generalization performance is therefore important for explaining how model performance changes as model size increases. In this work, we study this question in linear random-feature models and show that mixtures of datasets with different covariance structures can give rise to a range of multiple-descent phenomena. We derive exact asymptotic expressions for the test error of each data source in the high-dimensional regime, and show how heterogeneity in covariance scale and spectral structure can introduce additional, group-dependent interpolation thresholds as model size increases. These phenomena arise even with noiseless labels and linear activations, showing that neither label noise nor activation nonlinearity is necessary for multiple-descent. Our results therefore provide a different perspective from classical explanations of multiple descent. More broadly, our results have potential implications for algorithmic fairness by showing that model scaling can affect different data groups unevenly, amplifying disparities in their generalization errors. We validate our theoretical results on synthetic Gaussian mixtures and real-world datasets, including MNIST and CIFAR-10.
[LG-210] Machine learning applications in the heliosphere
链接: https://arxiv.org/abs/2610.37927
作者: Vishal Upendran,Banafsheh Ferdousi
类目: olar and Stellar Astrophysics (astro-ph.SR); Machine Learning (cs.LG); Space Physics (physics.space-ph)
*备注: Review chapter submitted to the book Machine Learning Techniques for Astrophysics and Cosmology by Spring Singapore
Abstract:Space weather is the study of the near-Earth impact of the processes on the Sun. In a this era of technology, space weather emerges as a crucial area of study wherein the state of the near-Earth conditions need to be modeled and forecasted. Crucially, the advent of machine learning provides us with opportunities to study this subject in a rather comprehensive manner. In this chapter, we treat the the processes arising from Sun to the impact on the ground as a set of coupled systems, which can be studied in isolation or in a coupled manner. We define these individual systems in terms of their state vector \vecS , which provides us with an opportunity to define the physical variables determining the system. This also allows us to consider the systems in terms of their observables, obtained through the action of operators as \vecO[\vecS] . In such a formalism, the machine learning algorithms serve as \it learned operators, defined generally as inference ( \mathcalI ), propagation ( \mathcalP ), and emulation ( \mathcalE ). Through such a formalism, we present how machine learning helps advance our understanding of the physical systems under study, while also accelerating the development of operational models for space weather forecasting.
[LG-211] Finite-Sample Optimal Conformal Prediction Sets under Covariate Shift
链接: https://arxiv.org/abs/2610.37866
作者: Zhongyang Li
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Combinatorics (math.CO); Probability (math.PR); Statistics Theory (math.ST)
*备注:
Abstract:We study conformal prediction under covariate shift with a known covariate density ratio, seeking prediction sets with minimum expected length under the target distribution. For every calibration sample size, we prove that the rule combining the inverse conditional-density score with randomized weighted conformal calibration minimizes this expected length among all rules valid over the full weighted-exchangeability class. We approximate the oracle using quantile and kernel histograms, including forest-weighted estimators, and select estimators and tuning parameters by resampling to estimate calibrated target length. On a bounded response domain, we bound excess length linearly in L^1 density error at fixed calibration size, without requiring a uniform positive lower bound on the true density. We also bound selection-pool and resampling errors. Under explicit growth conditions, the selected procedure approaches the same oracle while the calibration size remains fixed. Experiments on 36 known-density configurations and four public regression datasets compare coverage, prediction-set length, and runtime with existing methods. In targeted low-coverage and strong-shift comparisons, inverse-density selection yields shorter mean sets in most cases and lower mean runtime than the evaluated weighted baselines, with some differences in achieved coverage. Subjects: Machine Learning (stat.ML); Machine Learning (cs.LG); Combinatorics (math.CO); Probability (math.PR); Statistics Theory (math.ST) Cite as: arXiv:2610.37866 [stat.ML] (or arXiv:2610.37866v1 [stat.ML] for this version) https://doi.org/10.48550/arXiv.2610.37866 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Zhongyang Li [view email] [v1] Tue, 27 Oct 2026 19:11:25 UTC (149 KB)
[LG-212] An Ω(T3/4) lower bound for projection-free online learning with linear minimization oracles
链接: https://arxiv.org/abs/2610.37845
作者: Julien Weibel,Pierre Gaillard,Wouter Koolen,Adrien Taylor
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:This work considers projection-free online learning algorithms with linear minimization oracles for handling the decision set. For Online Frank–Wolfe (OFW), regret upper bounds were already known for specific OFW tunings [Hazan and Kale, 2012, Hazan, 2016] which scaled as T^3/4 (where T is the horizon length). However, it remained an open question whether the regret of all OFW tunings scaled as T^3/4 . We answer that question, proving a universal worst-case regret lower bound \Omega(T^3/4 m^-1/4) for any projection-free algorithm that makes at most m linear minimization oracle calls per round. The analysis suggests an extended OFW to resort to multiple linear optimizations per round, for which we prove a regret upper bound O(\max \ T^3/4 m^-1/4, \sqrtT ) . Subjects: Machine Learning (stat.ML); Machine Learning (cs.LG) Cite as: arXiv:2610.37845 [stat.ML] (or arXiv:2610.37845v1 [stat.ML] for this version) https://doi.org/10.48550/arXiv.2610.37845 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-213] Investigating the Early Signatures of Feature Learning in Deep Linear Networks
链接: https://arxiv.org/abs/2610.37834
作者: Shaunak Bhandarkar,Boris Hanin
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注: 92 pages, 7 figures, 16 pages of main text
Abstract:Deep neural networks learn data-dependent features, a capacity widely credited for their empirical success. However, explicitly characterizing what features emerge – and when they improve generalization – is challenging, since training is nonlinear and non-convex and the data is often anisotropic. Here, we give a closed-form account of feature learning for a deep linear network in a mean-field parameterization, trained by one gradient step from initialization on high-dimensional data carrying a low-dimensional (spiked) signal. In the infinite-width mean-field limit, one step updates each layer’s feature kernel by a target-aligned rank-one spike, amplified by both the learning rate and the layer depth. We next assess how these features impact prediction through kernel ridge regression (KRR), finding that KRR on the learned feature kernel yields scalar inflation of the KRR estimator at initialization – an “anti-shrinkage” mechanism that reinflates task-relevant directions suppressed by ridge, while also amplifying noise. To study when this inflation improves generalization, in the overparameterized proportional limit we derive closed-form expressions for the bias, variance, and generalization errors of these estimators in terms of key macroscopic parameters – notably the spike strength and spike-target alignment. Crucially, we find that a one-step generalization advantage vanishes for isotropic inputs and appears when both the spike strength and an effective signal-to-noise ratio are sufficiently large, demonstrating that a successful first gradient step both requires and benefits from low-dimensional task structure.
附件下载
