本篇博文主要内容为 2026-11-18 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 (2026-11-18)
今日共更新862篇论文,其中:
- 自然语言处理共103篇(Computation and Language (cs.CL))
- 人工智能共259篇(Artificial Intelligence (cs.AI))
- 计算机视觉共208篇(Computer Vision and Pattern Recognition (cs.CV))
- 机器学习共212篇(Machine Learning (cs.LG))
- 多智能体系统共22篇(Multiagent Systems (cs.MA))
- 信息检索共20篇(Information Retrieval (cs.IR))
- 人机交互共26篇(Human-Computer Interaction (cs.HC))
多智能体系统
[MA-0] Capability Composition in Heterogeneous Teams via Barrier-Guided Diffusion Sampling
【速读】:该论文旨在解决异构团队在缺乏协作示范数据的情况下,如何生成符合物理约束的协同任务轨迹问题。由于真实协作示范成本高昂且难以获取,尤其对于未曾执行过的任务,传统基于示范学习的生成模型无法直接应用。其解决方案的关键在于:将团队协同的物理条件(如空间可达性、动作兼容性等)建模为对团队轨迹的显式约束,并在仅使用单个智能体独立示范训练的生成模型采样过程中强制施加这些约束。通过理论证明,该方法可生成所有满足约束的团队轨迹,而无需依赖任何协作示范;其中,约束决定了可生成轨迹的可行性,示范数据则决定其概率分布。数值实验和多机器人平台实证表明,该方法能高效生成符合约束的协同轨迹,且支持动态成员变更而无需重新训练,显著提升了生成式协同规划的灵活性与可扩展性。
链接: https://arxiv.org/abs/2611.23149
作者: Diana Laura Morales,Ruoyu Lin,Magnus Egerstedt,Fabio Pasqualetti
机构: University of California, Irvine (加州大学欧文分校); University of North Carolina at Chapel Hill (北卡罗来纳大学教堂山分校)
类目: Robotics (cs.RO); Machine Learning (cs.LG); Multiagent Systems (cs.MA); Systems and Control (eess.SY)
备注:
Abstract:The agents in a heterogeneous team can cooperate to accomplish tasks that exceed their individual capabilities. The team trajectories for a cooperative task can be generated by a generative model, provided that the model has been trained on demonstrations of the agents cooperating. Yet, demonstrations of cooperation are costly and may not exist for tasks that the team has never performed. The conditions under which the agents can combine their capabilities, however, are physical properties of the team and are known before any demonstration is collected. We encode the conditions as constraints on the team trajectories, and we enforce the constraints while sampling from generative models trained on demonstrations of each agent alone. We prove that the proposed method can generate every team trajectory that satisfies the constraints, although the demonstrations contain no cooperation. Hence the constraints determine which team trajectories can be generated, and the demonstrations determine how likely each team trajectory is. Numerical results on two cooperative tasks show the effectiveness of the method, which generates cooperative team trajectories that satisfy the constraints and requires no retraining when agents join or leave the team. Finally, we implement the method on a multi-robot testbed.
[MA-1] Values Shape What Self-Improving Coding Agents Build
【速读】:该论文旨在解决自改进型编码代理(self-improving coding agent)在任务执行过程中缺乏对行为过程的可控性问题,尤其在开放性任务中,传统指令难以有效引导其行为模式与产出质量。其核心挑战在于:如何通过设计可持久化的价值导向机制,使代理不仅提升性能,还能生成符合特定价值观的解决方案。解决方案的关键在于引入基于舒瓦茨基本人类价值理论(Schwartz’s theory of basic human values)的短期价值声明,并将其嵌入代理的持久指令文件中,以系统性地影响其自我改进路径。研究发现,不同价值类型显著影响代理的表现与产出特征:开放性价值(openness to change)可提升模型在未见数据上的性能,而保守性价值(conservation)则导致表现劣于无价值设定的情况;同时,价值取向直接影响生成算法的规模、复杂度和实现策略。此外,尽管代理自主修订自身价值会削弱价值间的差异,且从AI文献中提炼的价值未能带来可观测优势,但经过人工精心设计的价值仍能有效塑造代理的构建行为与性能表现,表明人类监督在价值设计中的不可替代性。
链接: https://arxiv.org/abs/2611.23067
作者: Jason Liang,Dylan Waldner,Risto Miikkulainen
机构: Cognizant AI Lab(认知智能实验室)
类目: Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
备注:
Abstract:Self-improving agents are typically optimized for the results they produce, with less attention to how they produce them. Their instructions give designers limited ways to shape how they work, especially when the task is open-ended. This paper studies whether a short value statement in an agent’s persistent instruction file changes not only how well the agent performs but also what it builds. Four higher-order value types in Schwartz’s theory of basic human values were evaluated: openness to change, conservation, self-enhancement, and self-transcendence. Across diverse coding tasks, the values were held fixed within each run over multiple iterations of self-improvement. Three findings emerged: (1) Values matter for performance across a diverse set of tasks. The openness values increase performance on held-out data, whereas the opposite conservation values result in scores below those of an agent given no values. (2) Values change what agents build, producing solution algorithms that differ in size, complexity and approach. (3) Human oversight is still needed for designing values. Allowing the agent to revise its own values narrows the differences between sets of values, and values distilled from the AI literature provide no detectable gain over having no values. Properly designed values can thus influence what a self-improving coding agent builds and, where the task permits, how well it performs.
[MA-2] Rag HarnessDebugger: Jev-Guided Agent ic RAG in Contract Inference Repair Auditing
【速读】:该论文旨在解决生成式 AI(Generative AI)在使用检索增强生成(RAG)系统时,因缺乏可解释性与可审计性而导致的“修复失效”问题:即系统虽能自主进行证据补充、策略测试与修复决策,但其修复过程难以验证有效性,传统指标如准确率可能掩盖深层错误。其核心挑战在于如何有效识别并调试修复路径中的失败模式,尤其是在高复杂度合同文本(如非披露协议)中,系统未能完成应有修复却未被察觉。解决方案的关键是提出 RagHarnessDebugger 框架,通过基于 Jev(Justification Evidence Validation)引导的可追踪执行流水线,完整记录每一次修复决策,并构建审计层以验证每条修复轨迹的完整性、验证逻辑的匹配性以及接受规则的可达性。该框架结合反事实重放(counterfactual replay)与标注错误分析,实现对修复路径的端到端回溯,揭示了开发集与复现集中分别高达77%和76%的可用但未被实际执行的修复机会。实验表明,经调试优化后的系统在开发集与复现集上分别提升准确率3.6与3.7个百分点,并显著改善全合同访问基准得分。本研究的核心价值不仅在于定位修复失败的根本原因,更在于为 RAG 系统提供可解释的判断透明性机制,支持持续迭代与决策可解释性建设。
链接: https://arxiv.org/abs/2611.22991
作者: Moriya Dechtiar,Daniel Martin Katz,Michael J. Bommarito II,Hongming Wang,Sylvain Jaume
机构: Illinois Tech - Chicago Kent College of Law, USA; CLTDS, Bucerius Law School, Germany; Yong Pung How School of Law, Singapore Management University; CodeX - The Stanford Center for Legal Informatics, Stanford University; Bommarito Consulting, LLC; Harvard University, USA; Massachusetts Institute of Technology, USA
类目: Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA); Software Engineering (cs.SE)
备注:
Abstract:Agentic systems using retrieval-augmented generation (RAG) must repair their own answers by gathering more evidence, testing alternative policies and deciding if, when, and what repairs are needed. Yet how do we know if a system’s repairs are working when metrics like accuracy can be misleading? We present RagHarnessDebugger, a framework that records every repair decision in a Jev-guided harness for contract inference and audits the record to facilitate complete debugging of all decisions taken during a RAG repair journey. RagHarnessDebugger checks that each trace is intact, that each validation applies to the candidate it is counted for, and which answers the acceptance rule can reach. We tested our approach on ContractNLI non-disclosure agreements with two reader models and a replication set of separate contracts. Our debugger’s audit layer exposed failure patterns that were hidden under aggregate metrics, policies acceptance cost of repairs vs regressions and policy acceptance gap of 77% and 76% of available but unrealised repairs on development and replication sets. Our audit, together with a labeled error analysis and replay mechanism, allowed following the execution path all the way to the terminal decision - uncovering where missing repairs became lost. Our revised harness improved accuracy over the initial BM25-only baseline by 3.6 points on development contracts and 3.7 points on the replication set, and even improved our full contract access reference score by 1.1 on development and 2.4 on replication set. While our work here demonstrates how the debugger can conduct decision root-cause analysis for missed repairs, the true value lies with providing this debugger for judgment visibility serving continuous improvement of RAG decisions and their explainability. Keywords: agentic RAG, AI evidence repair, harness explainability, counterfactual replay, Jev, ContractNLI Subjects: Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA); Software Engineering (cs.SE) Cite as: arXiv:2611.22991 [cs.AI] (or arXiv:2611.22991v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2611.22991 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[MA-3] Virtual Double Oracle: Faster Convergence by Leverag ing Strategies That Were Watching All Along NEURIPS2026
【速读】:该论文旨在解决双人零和博弈中近似纳什均衡求解时,传统在线双奥多尔(Online Double Oracle, ODO)等方法在策略种群扩展过程中因重初始化导致的效率损失问题。其核心挑战在于:每次新增策略时,基于乘法权重更新(Multiplicative Weights Update, MWU)的方法将新策略重新初始化为均匀先验,从而抹去其历史表现信息,造成学习过程中的“遗忘”与低效探索。本文提出虚拟双奥多尔(Virtual Double Oracle, VDO),其关键创新在于利用一个双线性恒等式——即某策略的完整事后损失历史可坍缩为其对对手时间平均策略的期望收益——实现对新策略的最优初始权重设定,使其从诞生之初即具备合理的权重分配。该设计使VDO在策略种群规模为 k 时达到 O(k/T) 的可剥削性收敛率,优于原有方法。进一步地,折扣型虚拟双奥多尔(Discounted VDO, dVDO)通过引入衰减因子调节历史偏好,缓解重初始化带来的过度集中问题。实验结果表明,在矩阵博弈、扑克及Goofspiel等场景中,适当调整折扣参数的dVDO显著降低了有限时域下的可剥削性,并提升了所发现策略种群的质量。
链接: https://arxiv.org/abs/2611.22806
作者: Ruilan Wang,Francisco Aristi Reina,Katerina Papadaki
机构: London School of Economics (伦敦政治经济学院)
类目: Machine Learning (cs.LG); Computer Science and Game Theory (cs.GT); Multiagent Systems (cs.MA)
备注: Accepted at NeurIPS 2026. 25 pages
Abstract:Double Oracle (DO) and PSRO approximate Nash equilibria in two-player zero-sum games by iteratively expanding a strategy population with best responses supplied by an oracle. Online Double Oracle (ODO) interleaves multiplicative weights updates (MWU) with population discovery and obtains an O(\sqrtk\ln k/T) rate, where T is the number of learning rounds and k is the larger of the two players’ final population sizes. We identify the reinitialisation step as an underexplored design choice: at each expansion, MWU-based methods reset newly discovered strategies to a uniform prior. We introduce Virtual Double Oracle (VDO), which exploits a bilinear identity (a strategy’s full hindsight loss history collapses to its utility against the time-averaged opponent) to initialise each new strategy with the MWU weight it would have had from the start. We prove that VDO achieves O(\sqrtk/T) exploitability over the discovered population as it grows, provided the cumulative cost of adding strategies is at most linear in k , a condition supported by our empirical measurements. Discounted VDO (dVDO) moderates historical preferences to reduce over-concentration at reinitialisation. Experiments across matrix games, poker, and Goofspiel show that, with a suitable discount, dVDO can substantially reduce finite-horizon exploitability relative to both VDO and uniform-reset ODO, and can improve the discovered strategy populations.
[MA-4] Guarded Swarms in the Loop: A ROS 2/PX4 Digital Twin for Hybrid Digital-Analog Coordination of UAV/UGV Swarms
【速读】:该论文旨在解决在对抗性环境中运行的无人飞行器与地面车辆集群(UAV/UGV)因过度依赖软件栈而面临物理安全风险的问题。传统软件系统易受攻击,其通信网络构成潜在攻击面,难以保障任务执行过程中的本质安全。为此,论文提出“受保护集群”(Guarded Swarms)架构,其核心创新在于将数字协同层与独立的模拟安全层相结合:数字协同层基于基于话题的通信空间佩特里网(Topic-Based Communication Space Petri Nets, TB-CSPN),实现多模态感知、人工授权到车辆任务分配的逻辑链路;而模拟安全层则以每辆车的虚拟安全板(Virtual Safety Board)形式存在,作为独立于协同层的硬件抽象层,可对执行指令进行强制否决。解决方案的关键在于通过数字孪生系统在ROS 2、PX4飞控软件在环仿真(SITL)与Gazebo仿真环境下,首次实现了该架构的可执行实例化,并验证了其三个关键负向保证(即:无冲突证据驱动的任务、无时间不一致证据的任务、无未授权证据的任务)在60次反事实测试中均成立;同时证明威胁确认范围受限于最窄感知模态,且安全层能独立于协同层完全发挥其防护能力。在故障注入实验中,当安全门由独立机制控制时,可有效防止所有由伪造任务或被攻陷工作节点引发的碰撞事件,而若否决权仍由受损节点持有,则无法实现任何防护。此外,该数字孪生系统揭示了原架构规范未覆盖的实现级隐患,并明确划定了仿真可验证属性与必须通过硬件实测才能确认的属性之间的界限。
链接: https://arxiv.org/abs/2611.22773
作者: Uwe M. Borghoff,Paolo Bottoni,Luca D’Aurizio,Remo Pareschi
机构: University of the Bundeswehr Munich (慕尼黑联邦国防军大学); Sapienza University of Rome (罗马第一大学); University of Molise (莫利塞大学)
类目: Cryptography and Security (cs.CR); Multiagent Systems (cs.MA)
备注: 10 pages, 1 figure, 7 tables. Submitted to ICMCIS 2027
Abstract:Swarms of unmanned aerial and ground vehicles (UAV/UGV) operating in contested environments cannot rely on software alone for physical safety, since the software stack and its networks are part of the attack surface. The Guarded Swarms architecture addresses this by pairing a digital coordination layer, built on Topic-Based Communication Space Petri Nets (TB-CSPN), with an independent analog safety layer that can veto unsafe actuation commands. So far, the architecture has been specified and argued for, but not executed. We present the first executable instantiation of Guarded Swarms as a digital twin built on ROS 2, the PX4 autopilot in software-in-the-loop mode, and Gazebo, with two UAVs and one UGV in a coastal counter-UAS scenario. The TB-CSPN transition chain, from multi-modal sensing through human authorization to vehicle tasking, is mapped one-to-one onto ROS 2 nodes, and the analog layer is modeled as a per-vehicle Virtual Safety Board that gates every command. Eight groups of experiments show that the three negative guarantees of the architecture (no mission from conflicting, temporally incoherent, or unauthorized evidence) hold at three distinct stages of the chain in all 60 counterfactual runs; that threat confirmation is bounded by the narrowest sensing modality; that the safety layer exercises its full modeled envelope independently of the coordination layer; and, under fault injection, that it prevents every collision caused by a forged mission or a compromised worker once its veto is enforced by an independent gate, whereas a veto left to the compromised worker prevents none. The twin also exposes implementation-level hazards that the architectural specification does not address, and it draws an explicit line between the properties that simulation can validate and those that only hardware can.
[MA-5] Groups of AI agents underuse resources despite shared evidence of increased capacity
【速读】:该论文旨在解决在资源受限环境下,多个生成式 AI (Generative AI) 代理因信息不对称导致资源利用不充分的问题。当系统容量发生变化(如某服务恢复)时,若代理无法及时获取可用性更新,仍可能持续尝试已失效的资源,造成任务延迟或失败。其解决方案的关键在于通过共享真实可用性的证据(如成功请求记录),使代理能够动态调整行为,从而更有效地分配任务。实验表明,仅告知恢复可能性不足以促使代理改变行为,但展示实际成功的请求记录可显著降低任务超时率;进一步发现,引入差异化标识符(如不同编号)可有效引导代理分散请求,即便整体选择频率不变,也能显著提升资源利用率和任务完成量,揭示了在透明化资源状态基础上,设计合理的分布式协调机制对优化群体性能至关重要。
链接: https://arxiv.org/abs/2611.22632
作者: Takahiro Ezaki,Naoto Imura,Katsuhiro Nishinari
机构: 未知
类目: Multiagent Systems (cs.MA)
备注: 47 pages including Supplementary Information; 7 main figures and 4 supplementary figures
Abstract:When AI agents share limited resources, such as computing services or public infrastructure, their individual choices affect how much work they complete as a group. Can sharing evidence of what is available let a group make full use of its resources after capacity changes? We tested this with language-model agents choosing between two simulated servers. One server recovered while its status report still listed it as unavailable. Each server had limited capacity per round, so rejected jobs required another attempt before the deadline. Telling agents that recovery was possible reduced missed deadlines. In controlled tests with Anthropic’s Claude Haiku 5.5, no agent tried the recovered server after a rule reminder, yet all chose it when shown a successful request record. After sharing, all five models tested left capacity unused. Keeping the success record fixed, we gave agents numbers without instructions, using each value equally often but giving group members either identical or distinct numbers. With OpenAI’s GPT-6.1 Sol or Google’s Gemini 3.5 Flash-Lite, distinct numbers spread simultaneous requests across servers and increased job completions even though overall choice frequencies barely changed. Telling AI agents which resources are available leaves a second design challenge: helping them spread demand across those resources.
[MA-6] Surfacing Disagreement Beats Resolving It: A Gated Ablation Study of Provenance-Weighted Memory for Multi-Agent Debate
【速读】:该论文旨在解决多智能体辩论(multi-agent debate)中共享内存机制默认消除不一致性的合理性问题。传统做法通过检索返回单一胜出事实来快速化解分歧,但这种“隐性调和”可能掩盖真实矛盾,影响模型对对抗性攻击的鲁棒性。其核心解决方案在于:在DebateGraph框架中,四个后端共享同一合约(read/write/snapshot),在固定预算下可独立评估五种来自智能体-记忆文献的机制,关键突破是保留矛盾并将其排序后的冲突集传递至辩论过程,而非强制合并为单一结果。实验表明,这一策略在39次运行中显著提升F1得分(+0.340,95% CI [+0.267, +0.414]),并将对抗攻击成功率从122个污染样本中的22例降至7例(McNemar精确检验p=0.0003),且成本不变。该效果的实现依赖于修复了原有评估框架中的瓶颈——冲突集仅基于检索到的边构建,而预设阈值导致四分之三的预置矛盾无法被检索,初始测试仅获得+0.084的增益。经验证,该效应在两种检索底座及HotpotQA文档上均具方向性重复性。其余机制均未存活,其中两个甚至劣于无干预状态:一旦矛盾进入提示,写操作信任门控与路由器溯源特征均造成准确率下降(F1分别损失-0.075和-0.096)。信任门控结构失效的根本原因在于:污染对象始终是图中已存在的实体,其与真实事实的匹配度高达53%,导致模型对谎言的“接地性”评分反而高于首次提及的真实信息。研究团队开源了评估框架、攻击数据集与日志,并主张应摒弃“沉默调和记忆”的默认假设。
链接: https://arxiv.org/abs/2611.22461
作者: Ziaullah Khan,Abdul Rehman Khalid,Shrestha Nischal Lal,Nheng Vanchhay,Md Ariful Islam Mozumder,Hee-Cheol Kim
机构: Inje University (仁川大学); Department of Digital Anti-Aging Healthcare (数字抗衰老健康科学系); Department of Computer Engineering (计算机工程系)
类目: Multiagent Systems (cs.MA); Computation and Language (cs.CL)
备注:
Abstract:Shared memory for multi-agent debate is usually built to resolve inconsistency: retrieval returns one winning fact and the disagreement disappears before any agent sees it. We ask whether that is correct. In DebateGraph, four backends share one contract (read/write/snapshot), so five mechanisms from the agent-memory literature can be removed independently under a fixed budget, each gated on a pre-declared threshold. Across 39 runs one survives: preserving contradictions and passing the ranked conflict set to the debate, rather than collapsing it to a winner, improves F1 by +0.340 (95% CI [+0.267, +0.414]) and cuts attack success from 22 to 7 of 122 poisoned items (McNemar exact p=0.0003), at equal cost. Reaching it meant fixing a bottleneck in our harness: conflict sets are built only over retrieved edges, and our pre-declared cap left three quarters of planted contradictions unretrievable, first measuring the mechanism at +0.084. The effect replicates on two retrieval substrates and directionally on HotpotQA documents. No other mechanism survives, and two are worse than inert: once contradictions reach the prompt, a write trust-gate and the router’s provenance features each cost accuracy (-0.075 and -0.096 F1). The gate fails structurally: poisoned objects are always entities the graph already contains, against 53% of gold, so groundedness scores the lie above a truthful first mention. We release the harness, attack set, and logs, and argue that silently reconciling memory is the assumption worth abandoning.
[MA-7] Bad Apples in the Kitchen: How LLM Teams Detect Correct and Calibrate Against Their Misaligned Members
【速读】:该论文旨在解决多智能体生成式人工智能(Multi-agent Generative AI)系统中,当团队成员在行为上出现偏离而语言仍保持合作时所引发的隐蔽性风险问题。传统方法依赖于回合间对话审查以实现协作,但在实时具身协作场景中,对话与行动并行发生,导致无法及时识别和响应潜在的“沉默型破坏者”——即言行不一的异常智能体。为探究此问题,研究构建了一个定制化的四人《Overcooked》环境,引入三种基于理论的非对齐智能体,分别针对三个关键联合行动层级施加压力:言语-行为不一致、角色入侵以及整体进度干扰。通过一个3×4×2的实验设计(共720个实验回合),将系统鲁棒性分解为检测、纠正与校准三个顺序性关卡。结果显示,尽管绝大多数异常行为最终可被识别,但有效恢复仅在话语中同时包含责任归属(attribution)与情境诊断(diagnosis)时才可能发生;直接进行源头归因提示虽提升了最复杂情况下的表现,却使无异常情况下的基准性能下降了37.5%。因此,研究揭示:系统的鲁棒性并非单纯依赖高敏感度的检测机制,而更取决于团队对话是否能在单一语句中整合责任归属与情境分析,提示应聚焦于精准化地提升对不对齐信号的感知能力,而非普遍提高敏感度。
链接: https://arxiv.org/abs/2611.22342
作者: Dongseok Heo,Junghwan Kim,Bongwon Suh
机构: Seoul National University (首尔国立大学); Seoul, Republic of Korea (首尔,大韩民国)
类目: Computation and Language (cs.CL); Multiagent Systems (cs.MA)
备注:
Abstract:What protects an LLM team when one teammate quietly goes wrong? Most multi-agent LLM work assumes cooperative partners and inspects chat between turns before acting. Real-time embodied cooperation does not allow this pause: chat runs alongside action, so a teammate whose words stay cooperative while behavior diverges can cause damage faster than the team can name what is happening. We study this case by injecting three theory-grounded misaligned agents into a custom four-player Overcooked environment, each stressing a different joint-action layer (speech-action mismatch, role invasion, and aggregate-progress disruption), and decompose robustness into three sequential gates of detection, correction, and calibration through a 3x4x2 design (n=720 episodes). Nearly all trolls are eventually named, but recovery follows only when attribution and situational diagnosis appear in the same utterance; direct source-attribution prompting improves the hardest case only by degrading the troll-free baseline by 37.5%. Robustness, then, depends less on detection alone than on whether team chat combines attribution with diagnosis in a single utterance, suggesting that sensitivity to misalignment should be targeted rather than uniformly raised.
[MA-8] OAK-Mind: Zero-Training Mental Health Change Inference from Passive Sensing via Knowledge-Grounded Multi-Agent LLM s
【速读】:该论文旨在解决从可穿戴设备与智能手机传感器采集的被动感知数据中推断心理健康状态时,传统监督学习模型依赖标注数据且缺乏预测可解释性的问题。其核心解决方案是提出一种零样本(zero-shot)训练、多智能体(multi-agent)、基于知识(knowledge-grounded)的大语言模型(LLM)推理框架——OAK-Mind。该框架将传感数据按睡眠、身体活动、位置移动性和手机使用等行为维度进行组织,并通过基于心理健康文献的领域特定智能体协同分析纵向行为变化,综合评估结果并结合量表化症状评分,实现相对于个体基线的动态心理状态变化推断,同时生成基于观测行为模式与领域知识的人类可读推理依据。实验表明,OAK-Mind在无需任务特定微调的情况下,性能可媲美甚至超越依赖额外标注数据的最优监督模型,且显著优于直接提示(direct prompting)策略。消融实验进一步验证了行为解读与症状评分融合、跨领域智能体分工及引入领域知识的关键作用。研究结果凸显了基于知识的多智能体LLM在无需训练条件下实现心理健康变化推断的巨大潜力。
链接: https://arxiv.org/abs/2611.22248
作者: Yishu Ji,Zhihan Jiang,Will Ke Wang,Yuang Fan,Anind K. Dey,Xuhai “Orson” Xu
机构: Paul G. Allen School of Computer Science and Engineering, University of Washington, Seattle, United States(华盛顿大学计算机科学与工程学院); Biomedical Informatics, Columbia University, New York, New York, United States(哥伦比亚大学生物医学信息学系); Information School, University of Washington, Seattle, Washington, United States(华盛顿大学信息学院)
类目: Multiagent Systems (cs.MA); Artificial Intelligence (cs.AI); Human-Computer Interaction (cs.HC); Machine Learning (cs.LG)
备注: Under review at ACM IMWUT. 30 pages, 4 figures and 6 tables
Abstract:Wearable and smartphone sensors capture behavioral signals associated with mental health, yet inferring mental health outcomes from these signals has traditionally relied on supervised models requiring labeled data and offering limited insight into their predictions. Large language models (LLMs) offer new opportunities to reason over behavioral evidence, but existing LLM-based approaches to passive sensing have largely continued to rely on task-specific fine-tuning or sensor-specific pretraining. In this paper, we introduce \textbfOAK-Mind, a zer\textbfO-training, multi-\textbfAgent, \textbfKnowledge-grounded LLM pipeline for inferring mental health changes from passive sensing data. The pipeline organizes sensing data into behavioral dimensions including sleep, physical activity, location mobility, and smartphone use, and orchestrates specialized agents grounded in mental health literature to interpret longitudinal behavioral changes. It synthesizes their assessments and applies questionnaire-based symptom rating to infer changes relative to an individual’s baseline assessment with human-readable rationale grounded in observed behavioral patterns and domain knowledge. Across datasets and prediction tasks, while without task-specific training, OAK-Mind achieves comparable or better performance than the most predictive supervised baselines trained on additional labeled samples. It also demonstrate consistent improvement upon direct prompting across four tested LLM backbones. Ablation analyses support the benefits of combining behavioral interpretation with symptom rating, distributing analysis across domain-specific agents, and incorporating domain knowledge. Our findings highlight the potential of knowledge-grounded, multi-agent LLMs for training-free inference of mental health changes from passive sensing data.
[MA-9] Multi-Agent Memory Inconsistency: Where Factual Conflicts Disappear in LLM Agents
【速读】:该论文旨在解决多智能体系统(multi-agent system)在分布式证据收集过程中因各智能体拥有独立上下文而引发的记忆不一致问题,即不同智能体检索到的事实可能无法被主导智能体完整获取,从而影响联合结论的准确性。其解决方案的关键在于通过构建基准测试与证据流分析工具,针对哈维实验室(Harvey LAB)文档审查任务中的受控事实冲突,系统性地评估多智能体协作中的信息一致性。研究发现,在同等生成成本下,启用多智能体委派使冲突主张的检索率从50.2%提升至68.7%,但冲突检测率却从16.2%下降至13.2%;尤其当冲突信息被分置于不同智能体时,仅有35.2%的检索对能同时报告两方主张,且仅当单一智能体读取全部主张时,冲突检测性能优于单智能体执行。因此,智能体向主导者的信息交接(worker-to-lead handoffs)成为提升多智能体大语言模型(LLM)中冲突检测能力的核心优化目标。
链接: https://arxiv.org/abs/2611.22245
作者: Hejia Zhang,Zhongming Yu,Yimin Ding,Wei-Ting Chen,Hengjia Yu,Wentao Ni,Yijun Liu,Cong Xu,Xin Chen,Matrix Yao,Fan Zhao,Jishen Zhao
机构: UC San Diego(加州大学圣地亚哥分校); Intel(英特尔); Hewlett Packard Enterprise(惠普企业)
类目: Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
备注:
Abstract:Multi-agent delegation lets LLM systems distribute evidence collection across workers with separate contexts. This creates a consistency challenge: facts retrieved by workers may be missing from the evidence available to the lead for a joint conclusion. We introduce a benchmark and evidence-flow profiler for studying multi-agent memory inconsistency through controlled factual conflicts in Harvey LAB document-review tasks. Pooled across three models with comparable generation spending, requiring delegation increases retrieval of both conflicting claims from 50.2% to 68.7% relative to single-agent execution, yet conflict detection falls from 16.2% to 13.2%. Within this setting, workers report both claims for only 35.2% of retrieved pairs, and pooled matched comparisons show higher detection than single-agent execution when one worker reads both claims but lower detection when the claims are split across workers. Worker-to-lead handoffs are a central target for improving conflict detection in multi-agent LLMs.
[MA-10] Emergence and Collapse of Leader-Follower Flocks through Phantom Targets: Surprise Nullification in Bayesian and Inverse Bayesian Inference
【速读】:该论文旨在解决在贝叶斯(Bayesian)与逆向贝叶斯(Inverse Bayesian, BIB)推断中,当预测误差较大时如何处理“意外”(surprise)的问题,特别是探讨在缺乏显式对齐规则或指定领导者的情况下,个体间如何通过动态的信念与似然方差调整形成并维持群体性集体行为(如领导者-追随者蜂群结构)。其核心解决方案的关键在于引入惊喜消减机制(surprise nullification),即在出现大预测误差时,通过以相同因子同时扩大信念方差(belief variance)和似然方差(likelihood variance),保持两者方差比不变,从而维持相对精度的同时显著提升信念不确定性。这一机制使得个体在感知到异常信息时仍能保留一定的学习能力(低学习率),但通过调节方差比例,诱导出“链接断裂”(link cut)现象——即一个代理几乎停止采纳另一代理的观测信息,并转向一个脱离现实的“幽灵目标”(phantom target)移动,进而吸引其他个体形成临时性领导-追随结构。这种结构在链接恢复或幽灵目标吸引力平衡时发生崩溃。研究发现,尽管个体预测本身可能严重偏离真实状态,但通过惊喜消减所引发的动态方差调整,反而促成了群体秩序的涌现。因此,该模型表明,惊喜消减通过在保持相对精度的同时降低绝对精度,实现了群体结构的持续存在与最终崩溃之间的动态平衡,为个体错误预测如何催生集体有序行为提供了机制解释。
链接: https://arxiv.org/abs/2611.22190
作者: Shuji Shinohara,Daiki Morita,Yoshihiro Nakajima,Takeshi Takano,Masakazu Higuchi,Ung-il Chung,Yukio-Pegio Gunji
机构: 未知
类目: Multiagent Systems (cs.MA); Adaptation and Self-Organizing Systems (nlin.AO)
备注:
Abstract:In Bayesian and inverse Bayesian (BIB) inference, a large prediction error can trigger surprise nullification, which expands belief and likelihood variances by a common factor, leaving the updated variance ratio the same as in standard Bayesian updating. This can increase belief uncertainty while keeping the learning rate low. In the simulations, three agents used BIB inference to predict one another’s positions and moved toward those predictions, repeatedly forming leader-follower flocks that subsequently collapsed despite having no alignment rules or designated leaders. Three leadership criteria (front position, heading stability, and flock initiation) identified the same agent in 89% of eligible flocks, with different agents serving as leaders across flocks. Flocks formed when an agent nearly stopped incorporating another’s observations (link cut) and moved straight toward a predicted position detached from reality (phantom target), thereby attracting followers. Link cuts arose when two successive resets assigned a very small and then an ordinary likelihood variance, sharply increasing the likelihood-to-belief variance ratio. Flocks collapsed upon link recovery or when returning phantom targets balanced the leader’s forward and backward attractions. Without nullification, cuts persisted throughout the analysis period; expanding only belief variance caused recovery within a few steps. Expanding likelihood variance as well allowed cuts to persist while belief uncertainty grew, enabling later resets and recovery. Thus, nullification preserves relative precision, allowing cuts and flocks to persist, while reducing absolute precision to enable recovery and collapse, reconciling these seemingly opposing processes through a single operation. These results suggest that, in this model, individually inaccurate predictions can give rise to collective order.
[MA-11] he Only Edge Was Memorized in Training: LLM Trading Agents Around Earnings and Bad News
【速读】:该论文旨在解决生成式 AI (Generative AI) 交易代理在真实市场情境下是否能有效捕捉事件驱动型投资机会的问题,特别是针对已知时间的财报公告前及重大负面消息后的股价异常波动。现有研究存在局限:一方面,多数对生成式 AI 的评估基于新闻文本回放,但新闻与股价反应时间高度同步,难以体现真正的前瞻性;另一方面,相关研究或仅分析公告后文档、或采用月度反转策略,无法验证代理在交易成本约束下是否具备实际盈利能力。本文的关键解决方案在于构建一个“无前瞻”(look-ahead-free)的基准测试框架,严格在事件发生前(598次财报公告前)和事件发生后(409次大型美股负面下跌后)进行独立评估,且所有输入均在决策收盘时截断,确保信息时效性。实验对比了多轮推理代理与单次调用、简单规则、逻辑回归模型,并考察是否包含股票名称的影响。结果表明,尽管使用了来自四家供应商的五种大语言模型,所有代理均未能超越简单的“买入所有”策略,其表现主要依赖训练数据中的记忆效应。例如,基于 GPT 的模型在财报公告方向预测上仅分别达到 59.2% 和 53.3% 的准确率,且未显著优于随机猜测。因此,论文强调,生成式交易代理的开发必须以“简单贝叶斯赌注”为基准进行盲测,并在训练截止日期之后验证其泛化能力,否则易陷入虚假性能幻觉。
链接: https://arxiv.org/abs/2611.22177
作者: Jackson Wang
机构: 未知
类目: Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
备注: 21 pages, 9 figures, 12 tables. Original empirical study of LLM trading agents before scheduled earnings announcements and after bad-news drops, run on the rta-bench-1 benchmark ( arXiv:2611.19530 )
Abstract:Large language model (LLM) trading agents are tested on news, but news reaches stock prices within minutes, and a benchmark that replays headlines at their publication time finds no alpha in trading on them. Money can then remain in two places, before news whose date is known, such as earnings announcements, and after bad news, if the market’s fall went too far. However, studies of LLMs and earnings read documents published with or after the announcement, and studies of reversal sort monthly portfolios, so neither tells whether an agent can trade these moments after costs. In this paper, we test LLM agents at both moments on the same look-ahead-free benchmark, before 598 earnings announcements and after 409 bad-news drops of large US stocks. Every input is cut at the decision close, staged agents are compared with single calls, simple rules and a logistic regression, and every decision is made with and without the stock’s name. Across five models from four vendors, no agent beat the simple bets. Before earnings, the agents mostly bought and earned what buying before every announcement earned, and after bad news, a Gemini-based agent called almost every fall an overreaction while no agent picked better dips than buying them all. The only gains came from what the models had memorized in training. Given only the ticker and the dates, a GPT-based model called the direction of 59.2% of the announcements before its training cutoff and 53.3% after it. Agent builders should therefore compare an event agent with the simple bet and test it blind after the training cutoff.
[MA-12] Addressing Exogenous Variability in Cooperative Multi-Agent Reinforcement Learning NEURIPS2026
【速读】:该论文旨在解决合作式多智能体强化学习(Cooperative Multi-Agent Reinforcement Learning, MARL)在面临外生可变性(exogenous variability)时的性能退化问题,尤其当对手策略或环境状态转移机制发生不可控的动态变化时,传统方法难以适应。为此,论文提出一种新的形式化框架——外生部分可观测马尔可夫决策过程(Exogenous Dec-POMDP, ED-POMDP),其中独立于智能体团队选择的“外生状态”(episode-level exogenous regime)决定系统的转移动态,并将全局状态划分为两类:一类是当前团队动作直接响应的状态分量,另一类则不直接受当前动作影响。这一划分揭示了“中介影响”(mediated influence)机制:尽管当前动作无法直接改变非响应型状态分量,但可通过影响响应型分量间接调控其后续演化路径。针对此机制,论文提出LEICA算法,采用中心化训练、去中心化执行(CTDE)兼容的架构,通过学习历史依赖的上下文表征来分别建模两类状态分量,并引入基于个体影响权重的反事实内在奖励(influence-weighted counterfactual intrinsic rewards)以引导策略更新。在包含对手策略切换的SMAX基准测试中,LEICA在训练性能和对未见对手策略的泛化能力方面均显著优于现有基线方法,验证了在存在外生训练-测试状态转移差异下,利用影响加权机制进行策略塑造的有效性。
链接: https://arxiv.org/abs/2611.22067
作者: Seongmin Kim,Woohyeon Byeon,Jiwon Jeon,Seungyul Han,Youngchul Sung
机构: KAIST(韩国科学技术院); UNIST(韩国蔚山科学技术院)
类目: Multiagent Systems (cs.MA); Machine Learning (cs.LG)
备注: 28 pages. Includes supplementary material. Accepted at NeurIPS 2026
Abstract:Cooperative multi-agent reinforcement learning (MARL) often fails under exogenous variability such as shifts in opponents or environment regimes that cannot be controlled by the team but alter the transition dynamics. We formalize this challenge as Exogenous Dec-POMDP (ED-POMDP), in which an episode-level exogenous regime, selected independently of the team, indexes the dynamics, and the global state is partitioned into components that are directly responsive to the current team action and components that are not directly action-responsive at the current step. This formulation exposes mediated influence: while team actions do not directly update the latter at the current step, they can shape their later evolution through induced changes in the former. Based on this, we propose LEICA, a CTDE-compatible algorithm that learns history-conditioned context representations for both components and shapes policy updates using agent-wise, influence-weighted counterfactual intrinsic rewards. Across SMAX benchmarks with opponent strategy shifts, LEICA consistently improves both training performance and generalization to unseen opponents’ strategies over existing baselines, supporting the usefulness of influence-weighted shaping under exogenous train-test regime shifts.
[MA-13] Operating an Agent -Driven Dataset Factory: Discovery Construction and Public Release
【速读】:该论文旨在解决从公开数据中自动化构建可用于预测任务的数据集所面临的系统性挑战,核心问题在于如何实现从数据源发现、任务定义、数据集构建到发布分发的全流程连续化与可复现性。其解决方案的关键在于提出并部署“数据集工厂”(Dataset Factory)系统,通过受限语言模型(bounded language model)负责理解数据源、生成构建代码并验证任务语义,同时借助确定性软件(deterministic software)精确管理状态、工作节点、重试机制及发布流程,从而实现端到端的自动化数据集生产。实验表明,在超过81.53小时的运行周期内,该系统以约2.74美元/次的估算运营成本发布了14个数据集,涵盖计算资源消耗、推理开销及失败尝试等成本。尽管所有已发布版本均通过独立代理完成预测实验验证,但源审查仍发现一个任务的构建过滤逻辑与其声明的种群定义存在矛盾,且原始评审者未对后期记录的排除项与已发布选择规则进行一致性校验。这一发现凸显了在数据集生产流程中,任务语义的正确性需在系统执行与下游使用者验证之间进行独立检查,以保障数据集的可信度与可用性。
链接: https://arxiv.org/abs/2611.21957
作者: Eduardo Ariño de la Rubia,Szilard Pafka
机构: Central European University (中央欧洲大学); Epoch (埃波克)
类目: oftware Engineering (cs.SE); Multiagent Systems (cs.MA)
备注: 5 pages, 3 figures. Code and evidence: this https URL
Abstract:Turning public data into prediction datasets requires connecting source discovery, task definition, construction, and distribution. We deployed Dataset Factory to perform these steps continuously under a standing release policy. Bounded language-model tasks interpret sources, write construction code, and review task meaning; deterministic software manages state, workers, retries, and publication. Over 81.53 hours, the factory published 14 datasets at an estimated recurring operating expenditure of about \ 2.74 per publication. The accounting includes failed attempts, inference, workers, and hosting. Separate agents completed prediction experiments on all twelve evaluated releases. Source review nevertheless identified a task whose construction filters contradicted its population declaration. The original reviewer accepted later-record exclusions without reconciling them with the published selection rule. This experience documents an operating source-to-release workflow and shows why successful execution and consumer use need separate checks of task meaning.
[MA-14] Beyond 4: Improved Competitive Traveling Repairperson
【速读】:该论文旨在解决在线旅行维修员问题(Online Traveling Repairperson Problem, OTRP)中的竞争比优化问题,特别是在存在多个维修人员(repairpersons)从同一出发点(depot)出发的场景下,如何设计高效的在线算法以最小化最大响应时间。针对任意度量空间下的通用情况,提出了一种3.877-竞争性的确定性算法,显著优于此前最优的4-竞争性算法。对于零残余(zero-residual)这一特殊情形,进一步将竞争比改进至3+ϵ(对任意ϵ>0),实现了理论上的突破。其解决方案的关键在于设计一种基于动态调度与局部最优策略相结合的在线决策机制,通过精确控制服务顺序与路径规划,在不依赖未来信息的前提下实现近似最优的性能保障。此外,该方法还可扩展至相关的在线叫车接送问题(Online Dial-a-Ride Problem),在多种变体中保持优异的竞争性能。
链接: https://arxiv.org/abs/2611.21943
作者: Bibek Maharjan,Gokarna Sharma
机构: Kent State University (肯特州立大学)
类目: Data Structures and Algorithms (cs.DS); Discrete Mathematics (cs.DM); Multiagent Systems (cs.MA)
备注: 27 pages, 1 figure, 2 pseudocodes, 1 table
Abstract:With k\geq 1 repairpersons starting initially from a depot, we present a 3.877-competitive deterministic algorithm for the fundamental online traveling repairperson problem on an arbitrary metric. For the special class of zero-residual instances, we improve the ratio to 3+\epsilon for every \epsilon0 . We also extend these results to some variants of the related so-called online dial-a-ride problem. The state-of-the-art is the 4-competitive deterministic algorithm for both problems.
[MA-15] Coordination of Heterogeneous Teams with Uncertain Task Requirements
【速读】:该论文旨在解决异构智能体团队在任务需求未知的情况下协同完成任务的多智能体规划难题。核心挑战在于智能体在发现任务需求与执行任务能力上的差异,要求规划器同时协调任务发现、任务执行与任务调度。其解决方案的关键在于提出一种分层规划框架,通过图神经网络(Graph Neural Network, GNN)策略为智能体分配动作,并结合最短路径算法确定动作执行位置。该方法的核心是构建“任务-能力-行动图”(CAAT graph),以统一表征智能体、任务、能力与动作之间的复杂关系。实验表明,该学习型规划器能够更高效地联合优化任务发现与任务执行,在较短时间内完成任务;更重要的是,该策略在零样本迁移(zero-shot transfer)场景下表现优异,仅在室外地图上训练后即可直接应用于包含更多智能体、任务和能力的室内地图,无需微调,验证了其强泛化能力。
链接: https://arxiv.org/abs/2611.21828
作者: Jongann Lee,Melkior Ornik
机构: University of Illinois Urbana-Champaign (伊利诺伊大学厄本那-香槟分校)
类目: Multiagent Systems (cs.MA); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Coordinating a team of heterogeneous agents to complete tasks with unknown requirements is a challenging multi-agent planning problem. Agents differ in their ability to discover task requirements and in their ability to engage tasks, so the planner must coordinate task discovery, execution, and staging. This paper introduces a hierarchical planner that uses a graph neural network policy to assign actions to agents, then finds a path to the action’s location using a shortest-path algorithm. The central part of our approach is the CAAT graph, which is a graph representation of the planning context consisting of the agents, tasks, capabilities, and actions. We demonstrate that our learned planner can jointly coordinate a heterogeneous team to simultaneously discover task requirements and complete tasks in a shorter time compared to baselines that optimize task discovery and task engagement independently. Furthermore, we show that our policy can zero-shot plan on out-of-distribution settings by training only on an outdoor map then transferring without finetuning to an indoor map with more agents, tasks, and capabilities than seen during training.
[MA-16] Corrupting Ties: Investigating Robustness to Bias Injection Across Multi-Agent Network Topologies
【速读】:该论文旨在解决生成式AI(Generative AI)在高风险决策场景中,因多智能体系统内部组件间交互而引入的公平性风险难以通过孤立审计单个组件或整体输出发现的问题。其核心解决方案是提出一种受控偏见注入框架,通过替换特定智能体的响应为预设偏见内容,实现对网络拓扑结构与注入配置如何影响系统行为的干预式分析。关键在于设计了多层次的决策分歧度量指标,从个体智能体、网络层级到系统整体量化干预效果,并基于Qwen与Gemma系列大语言模型,在三种二分类任务上评估五种网络拓扑结构。研究发现,序列型网络表现出最大分歧,全连接与网格结构则相对较低;在星型网络中,中央节点的偏见注入影响显著大于边缘节点;且不同模型对偏见传播的响应存在差异,部分模型的决策分歧并不完全遵循注入偏见。这些结果揭示了网络结构在偏见传播中的决定性作用,推动了面向拓扑结构的多智能体系统公平性风险审计方法的发展。
链接: https://arxiv.org/abs/2611.21804
作者: Cléa Chataigner,Mina Arzaghi,Florian Carichon,Pablo Piantanida,Golnoosh Farnadi
机构: MILA(蒙特利尔学习系统研究所); McGill University(麦吉尔大学); Paris-Saclay University(巴黎萨克雷大学); ILLS - International Laboratory on Learning Systems(国际学习系统实验室); HEC Montréal(蒙特利尔高等商学院); CentraleSupélec(中央理工-高等电力学院); CNRS(法国国家科学研究中心)
类目: Computation and Language (cs.CL); Multiagent Systems (cs.MA)
备注:
Abstract:AI systems are increasingly deployed in high-stakes decision-making contexts, where interactions among components can introduce fairness risks that remain undetected when auditing components or collective outcomes in isolation. As LLM-based multi-agent systems become more prevalent, understanding how biased information propagates through agent interactions is essential for assessing their fairness. We introduce a controlled bias injection framework that replaces selected agent responses with predefined biased ones, enabling an intervention-based analysis of how network topology and injection configurations shape system behavior. We propose decision divergence metrics that quantify the effects of these interventions at the agent, network, and system levels, and evaluate five topologies on three binary classification tasks using LLMs from the Qwen and Gemma families. Sequential networks show the greatest divergence, whereas fully connected and mesh networks show less. Injection position also matters: in star networks, injecting the central agent has larger effects than injecting peripheral agents. Models further differ in whether divergence follows the injected bias. These findings highlight the role of network structure in the propagation of biased information and motivate topology-aware auditing of fairness risks in multi-agent systems.
[MA-17] Better Calibrated Better Planned: Uncertainty-Aware Multi-ASV Informative Path Planning for Oil Spill Monitoring
【速读】:该论文旨在解决在有限时间与能量预算下,利用自主水面航行器(ASV)舰队对异质油污污染场进行稀疏、噪声干扰下的高效监测问题。核心挑战在于如何通过信息感知路径规划(IPP)实现高精度的污染场重建,而传统方法多集中于规划策略设计,忽视了重建模型中不确定性校准的重要性。本文提出的关键解决方案是引入证据深度学习(Evidential Deep Learning, EDL),作为一种具备可校准不确定性的信息建模方法,并将其与在线共享不确定性地图的多智能体协同机制相结合,以支持鲁棒的不确定性感知路径规划。实验表明,在物理仿真油污场景下,基于闭环测试的五种规划器对比中,模型选择对最终重建误差的解释力高达80%,远超规划策略的影响(仅1%)。EDL在所有规划器下均取得最低重建误差和最高交并比(IoU),且具备最优的不确定性校准能力,推理速度达每步3.9毫秒。相比蒙特卡洛丢弃(MC Dropout)和深度集成(Deep Ensembles),EDL在仅50个单元/车辆的配置下即超越后者150个单元的表现;且当深度集成模型误差增大时其置信度反而上升,呈现过度自信现象,而EDL则保持稳定,证明了信息模型对自身规划所采集数据的鲁棒性是实现可靠不确定性感知IPP的必要条件。
链接: https://arxiv.org/abs/2611.21783
作者: Alejandro Casado-Pérez,Marija Popović,Daniel Gutiérrez-Reina,Sergio L. Toral.,Samuel Yanes
机构: Universidad de Sevilla(塞维利亚大学); MAVLab, Faculty of Aerospace Engineering, TU Delft(代尔夫特理工大学航空航天工程学院)
类目: Robotics (cs.RO); Machine Learning (cs.LG); Multiagent Systems (cs.MA)
备注: 11 pages, 7 figures, Submitted to IEEE Internet of Things Journal
Abstract:Monitoring oil spills with fleets of Autonomous Surface Vehicles (ASVs) requires mapping a heterogeneous pollutant field from sparse, noisy measurements under tight time and energy budgets. Informative Path Planning (IPP) addresses this problem by coupling a reconstruction model, which infers the field and its uncertainty, with a planner that decides where to sample next. Prior work has focused on planning strategies and has largely relied on Gaussian Processes (GPs), leaving the calibration of the model’s uncertainty underexplored. We study how uncertainty calibration conditions multi-vehicle IPP with a fleet of ASVs that coordinates online through a shared uncertainty map. We introduce Evidential Deep Learning (EDL) as an uncertainty-calibrated information model. We benchmark EDL, Monte Carlo Dropout and Deep Ensembles against a GP in closed loop with five planners on physics-based oil spill simulations. Across 60 maps, the choice of model dominates the outcome, explaining 80% of the variance of the final reconstruction error, against 1% for the planner among the deep models. EDL achieves the lowest error and the highest IoU under every planner, is the best calibrated model and runs at 3.9 ms per step. With 50 cells per vehicle it reconstructs the spill better than MC Dropout and Deep Ensembles with 150. Deep Ensembles become more confident as their error grows, whereas EDL is unaffected. Robustness of the information model to the data its own planner collects is therefore a requirement of uncertainty-aware IPP.
[MA-18] Consistency Is Not Correction: Separating Confidence Abstention and Recovery in LLM Agents
【速读】:该论文旨在解决生成式 AI(Generative AI)中基于置信度信号的动态推理机制有效性评估问题,具体聚焦于三类关键主张之间的逻辑独立性:答案排序能力( discrimination)、选择性预测支持( selective prediction)与纠错实用性( correction utility)。研究指出,前两者并不蕴含后者,且纠错实用性可被精确分解为“门控恢复”与“门控损害”之差。这一分解揭示了一个根本性限制:任何仅在基础模型错误时才触发的门控机制,其可纠正的不一致对数受初始错误数量的严格上限制约。当该上限低于配对测试所需的最小样本量时,实验无法获得显著正向结果,也无法区分有效机制与从未触发的无效门控。作者在两个模型规模下验证了这一边界约束——小模型套件中,扰动-一致性验证器虽达到0.990的AUROC(95%置信区间[0.95, 1.00]),但每轮决策需4.32倍调用次数且无实际替换行为;在840次受控实验中,所有GSM8K任务均未发生替换,导致比较测试失效;诊断集上最大可容忍误差为4,而测试所需至少6个,即便理想门控也无显著结果。研究进一步提出两项预检措施以避免资源浪费,并揭示排名处理、截断阈值设定及输出长度差异均会系统性影响核心指标。结论表明,当前文献中的主要瓶颈并非置信度信号的选择,而是测量分辨率不足。
链接: https://arxiv.org/abs/2611.21651
作者: Khushal Sethi
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
备注:
Abstract:A controller that resamples an answer, escalates to a larger model, or abstains is usually justified by a confidence signal. Three claims about such a signal do not follow from one another: it ranks answers (discrimination), it supports withholding some of them (selective prediction), and it changes an answer for the better (correction utility). We give an exact decomposition of the third into gated recoveries minus gated harms, which bounds whether a comparison can succeed before it is run: a gate that only replaces a proposal can act only where the base policy was already wrong, so the initial error count caps the discordant pairs any correct gate can produce. When that cap falls below the count a paired test needs, the experiment cannot return a positive result for a working mechanism, nor distinguish one from a gate that never fired. We measure all three targets on two model scales and find the bound binds. In a small-model suite a perturbation-agreement verifier reaches within-policy AUROC 0.990 (95% interval [0.95, 1.00]) while every episode outcome matches single-sample decoding at 4.32 times the call count. In a controlled 840-episode study no verifier replaced the proposal on any of the 60 GSM8K decisions, so that comparison tested nothing; the entire GSM8K gain across six controllers is one task, and it is the task whose identical temperature-zero repeats disagreed, making it the one answer the service does not reproduce. On the diagnostic set the gate could reach four errors while the test needed six, so even a perfect gate had no significant result available. We report two checks that would have caught this before any budget was spent, and show that tie handling in rankings, tied selection cutoffs, and unequal output limits each move a headline number in the same dataset. The binding constraint on this literature is measurement resolution, not the choice of signal.
[MA-19] FieldGuide: A Corpus Map for Token-Efficient Agent ic Document Exploration
【速读】:该论文旨在解决记者在探索大规模文档集合时面临的效率瓶颈问题,尤其是在任务范围广泛且关键子问题尚不明确的情况下,传统生成式AI(Generative AI)方法因频繁调用大语言模型而导致高消耗的令牌使用量与长时间延迟。其解决方案的关键在于提出FieldGuide——一种面向智能体的语料库地图(corpus map),通过将文档集合组织为层次化主题结构,并提供语料定向、范围搜索和限定阅读等工具,实现对文档集合的高效导航与精准访问。实验结果表明,相较于直接访问原始文件和检索增强生成(RAG),FieldGuide在五项新闻调查任务中分别减少了27%的令牌消耗和41%的运行时间,相较RAG也分别减少38%的令牌使用和21%的耗时,同时保持了可接受的准确性水平。这表明,通过语料库定向引导可显著降低生成式AI在信息探索中的资源开销,而无需牺牲核心输出质量,同时也揭示了未来研究需进一步探讨受限访问对跨文档广度发现能力的影响。
链接: https://arxiv.org/abs/2611.21647
作者: Nick Hagar
机构: University of Minnesota (明尼苏达大学)
类目: Computation and Language (cs.CL); Multiagent Systems (cs.MA)
备注: Presented at Computation + Journalism Symposium 2026
Abstract:Generative AI agents can help journalists explore large document collections. But exploratory workflows can require substantial token use and time, especially when the task is broad and the relevant subquestions are not yet known. This paper introduces FieldGuide, an agent-facing corpus map that organizes a document collection into a topic hierarchy and exposes tools for corpus orientation, scoped search, and bounded reading. I evaluate FieldGuide on five journalistic document-exploration tasks using 38 Chicago Police Department public-records documents, comparing it with raw-file access and retrieval-augmented generation. Across the five tasks, FieldGuide used 27% fewer tokens and 41% less elapsed time than raw-file access; relative to RAG, it used 38% fewer tokens and 21% less elapsed time. An accuracy audit identified four minor issues in FieldGuide’s responses, comparable to the other conditions. In blind ratings of generated reporting leads, FieldGuide’s mean ratings were higher than RAG’s and lower than the raw-file condition’s. These descriptive results suggest that corpus orientation may reduce runtime token use and elapsed time without severe accuracy failures. They also motivate further study of whether selective access limits broad, cross-document discovery.
[MA-20] he Defense Rests: A Courtroom-Inspired Multi-Agent LLM Framework for Automated Safety Case Generation
【速读】:该论文旨在解决复杂且耗时的手动构建安全论证(safety case)问题,尤其是在机器学习(ML)赋能的自动驾驶系统等关键任务系统中,传统人工方式难以高效、一致地生成可验证的安全论证。其核心解决方案是提出一个基于大型语言模型(LLM)的多智能体框架eTrustWare,通过法庭类比的形式对安全论证过程进行形式化建模:安全分析师扮演“检察官”角色,识别潜在危害与安全关切;保证论证开发者作为“辩护律师”构建并证明安全论点;协调者充当“法官”以管理多方交互;评审委员会则如“陪审团”负责评估论证的完整性与说服力。该框架利用多个协同的LLM智能体实现安全论点的自动生成、证据支持、逻辑推理与交叉审查,显著提升了安全论证构建的自动化水平与质量。实验结果表明,基于协同LLM智能体的系统在生成和评审多类自动驾驶系统的安全论证方面具有可行性与有效性。
链接: https://arxiv.org/abs/2611.21618
作者: Gerhard Yu,Amanuel Aknaw,Amir Ali Mohaghegh,Yong Song(Alex)Cao,Alvine B. Belle
机构: 未知
类目: oftware Engineering (cs.SE); Multiagent Systems (cs.MA)
备注: 10 pages
Abstract:Safety cases are widely used to demonstrate that mission-critical systems, such as autonomous vehicles and unmanned aerial systems, can operate safely after deployment. However, manually constructing safety cases is complex, time-consuming, and labor-intensive, which calls for the proposal of automated approaches to support their creation. Inspired by the legal nature of safety case construction, we formalize the safety case creation problem through a courtroom analogy, where stakeholders collaboratively propose, examine, criticize, review, improve and validate safety claims and their supporting evidence. In this analogy, the safety analyst acts as a prosecutor by identifying hazards and safety concerns, while the assurance case developer serves as a defense attorney by constructing and substantiating safety arguments. An orchestrator acts as a judge by coordinating stakeholder interactions, and a review board serves as a jury by assessing argument completeness and persuasiveness. Based on this formalization, we propose eTrustWare, a Large Language Model (LLM)-based multi-agent framework for automated safety case generation and review. Our experiments on several safety cases of ML-enabled autonomous driving systems demonstrate the feasibility of coordinated LLM-based agents for generating safety cases.
[MA-21] Cycles of Arbitrary Period in Multidimensional Heterogeneous Hegselmann-Krause Dynamics
【速读】:该论文旨在解决异质型赫格塞尔曼-克拉豪斯(Hegselmann-Krause)意见动态系统中多维意见与范数定义信心邻域下的周期性行为问题。其核心问题是:在何种条件下,此类系统能够产生非平凡的周期轨迹?研究发现,对于任意维度 $ d \geq 3 $、任意 $ \ell_q $ 范数(其中 $ q \in [1, \infty] \setminus {2} $)以及任意整数 $ m \geq 2 $,存在一个由五个代理人组成的系统,可实现最小周期为 $ m $ 的周期性轨迹,且不同周期可在相同信心半径下共存。关键解决方案在于构造一种几何结构:四个代理保持静止,第五个代理在其凸包内沿线段移动,同时所有代理间均满足严格的信心约束条件,且静止代理可共享一个正信心半径。该构造依赖于一个几何准则,该准则可推广至更广泛的范数类,但该准则在低维(一维和二维)下无论采用何种信心范数均无法满足。此外,研究还证明,在任何维度和任意范数下,少于五个代理的异质型赫格塞尔曼-克拉豪斯系统不可能存在非平凡周期轨迹;而最多三个代理的系统则必然以指数速度收敛。
链接: https://arxiv.org/abs/2611.21656
作者: Anton V. Proskurnikov,Iryna Zabarianska
机构: Politecnico di Torino (都灵理工大学); Moscow Institute of Physics and Technology (莫斯科物理技术研究所)
类目: Physics and Society (physics.soc-ph); Multiagent Systems (cs.MA); Social and Information Networks (cs.SI); Systems and Control (eess.SY); Dynamical Systems (math.DS); Optimization and Control (math.OC)
备注:
Abstract:We study periodic behavior in heterogeneous Hegselmann-Krause opinion dynamics with multidimensional opinions and norm-defined confidence neighborhoods. We prove that, for every dimension d\ge3 , every \ell_q norm with q\in[1,\infty]\setminus\2\ , and every integer m\ge2 , a system of five agents admits a periodic trajectory of minimal period m ; moreover, different periods can coexist for fixed confidence radii. Four agents remain stationary, while the fifth moves along a line segment contained in their convex hull. The stationary agents may share a common strictly positive confidence radius, and all confidence inequalities are strict. A geometric criterion underlying the construction extends these results to further classes of norms. However, this criterion cannot be satisfied in dimensions one and two, regardless of the confidence norm. We also prove that no heterogeneous Hegselmann-Krause system with less than five agents admits a nonconstant periodic trajectory, under any norm and in any dimension; every trajectory with at most three agents converges exponentially.
自然语言处理
[NLP-0] Ambient LLM s: Data-Efficient Supervised Fine-tuning via Temperature Matching
【速读】: 该论文旨在解决自回归语言模型在后训练阶段因数据集规模有限、多样性不足而导致的过拟合及数据分布尾部信息遗忘(catastrophic forgetting)问题。现有监督微调(Supervised Fine-Tuning, SFT)通常使用远小于预训练阶段的数据集,难以充分覆盖长尾分布特征。其解决方案的关键在于提出一种名为“环境微调”(Ambient SFT)的原理性框架,通过利用预训练模型生成的不完美预测来增强微调数据集。该方法采用一个“冻结”的预训练模型作为噪声教师(noisy teacher),要求微调后的模型在高温度(high-temperature)条件下,其生成的条件分布与预训练模型的对应分布保持一致。温度超参数控制对预训练模型输出的信任程度:当温度趋于无穷时退化为标准SFT,而温度为1时等价于知识蒸馏。实验表明,该方法在领域迁移、指令微调和推理任务中均显著优于标准SFT,平均相对性能提升达21%。
链接: https://arxiv.org/abs/2611.23229
作者: Giannis Daras,Marios Papachristou,Adam Klivans,Antonio Torralba,Constantinos Daskalakis
机构: MIT(麻省理工学院); Arizona State University(亚利桑那州立大学); University of Texas at Austin(德克萨斯大学奥斯汀分校)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: Work in progress
Abstract:Collecting large-scale, diverse, and high-quality post-training datasets for autoregressive language models is challenging. Consequently, Supervised Fine-Tuning (SFT) is typically applied to datasets significantly smaller than those used for pre-training, which leads to overfitting and the catastrophic forgetting of the tails of the data distribution. To prevent this collapse, we introduce Ambient SFT, a principled framework for leveraging imperfect predictions to augment the fine-tuning set. In our method, a “frozen” version of the pre-trained model serves as a noisy teacher that the fine-tuned model must match at a coarse level. In particular, we require that the high-temperature versions of the conditional distributions produced by the post-training model should match the high-temperature versions of the pre-training ones. The temperature hyperparameter controls the level of trust in the pre-trained model, with the limit T\to \infty reverting back to regular SFT and T=1 to knowledge distillation. We demonstrate that our approach outperforms standard SFT across domain adaptation, instruction-tuning, and reasoning, leading for up to 21% average relative improvement to standard SFT.
[NLP-1] What Does a Bias Head Measure? Non-Identifiability in Head-Level Bias Attribution ACL
【速读】: 该论文旨在解决当前生成式AI(Generative AI)中社会偏见定位方法的可靠性问题,具体针对基于注意力头(attention head)的偏见识别与移除方法的有效性质疑。现有方法通常通过观察某个公平性指标在移除特定注意力头后的变化来验证其有效性,但缺乏对所选注意力头是否真正反映偏见的独立证据支持。本文以Qwen-2.5-7B-Instruct和Llama-3.1-8B-Instruct模型为对象进行审计,发现所谓的“偏见头图”(head map)与无标签响应长度图的相关系数分别达到ρ = -0.950和-0.768,表明其可能捕捉的是表面共现偏差(surface confound),而非真实偏见。进一步干预实验显示,在提示固定的情况下,响应长度仅解释该统计量1%的方差;而提示身份则解释了高达64%的方差,且该关系并非随长度单调变化。在任一固定提示下,公平性标签、响应长度与受保护群体词元的存在高度共线(φ = 0.95),这种共线性部分源于标注规则的定义本身,因此任何观测分析均无法区分三者中的因果角色。由此推断,此类方法无法通过非实验性分析建立因果归属。最终,研究指出其核心主张均不成立:无论采用何种参考解读,偏见选择性均未表现出晚期层集中特征;且所有测试的注意力头集合在剔除后均未降低不公平率。相反,其补集在反向标签上显著破坏拒绝行为(p = 2.7×10⁻⁵),证明该检测管道具备识别此类效应的能力,但其误将共线性信号当作偏见信号,从而导致错误归因。该研究的关键在于揭示现有偏见定位方法的内在缺陷——即依赖于不可分离的共线变量,且缺乏可证伪的实验设计,从而挑战了当前主流方法的科学基础。
链接: https://arxiv.org/abs/2611.23218
作者: S.M Seefat Alam
机构: Bangladesh University of Engineering and Technology (孟加拉国工程技术大学)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 18 pages, 13 tables. Under review at ACL Rolling Review (October 2026 cycle)
Abstract:Methods that localise social bias in large language models to individual attention heads and remove it by ablation are validated by their effect on a fairness metric, never by evidence that the statistic selecting the heads measures bias. We audit this on Qwen-2.5-7B-Instruct and Llama-3.1-8B-Instruct. The head map correlates at \rho = -0.950 and -0.768 with a label-blind response-length map, apparently identifying a surface confound. An intervention that breaks the length-label coupling rejects this reading: with the prompt held fixed, response length explains 1% of the statistic’s variance. What does explain it is prompt identity, at 64% of variance, in an ordering that is not monotonic in length. Within any fixed prompt the fairness label, response length and the presence of a protected-group token are collinear at \phi = 0.95 , partly by definition of the labelling rule, so no observational analysis can attribute the map to any of them, and no method in this family performs an experiment that could. The substantive claims fail independently: bias selectivity is not late-layer-concentrated on either model under either reference reading, and no head set we tested reduces the unfairness rate when ablated. Its complement, scored on the opposite label, reliably destroys refusal behaviour ( p = 2.7\times10^-5 ), demonstrating that the pipeline detects effects at this scale.
[NLP-2] Structure–Residual Inductive Biases for Language Model Pretraining
【速读】: 该论文旨在解决当前语言模型预训练中过度依赖下一词预测(next-token prediction)任务所导致的表征学习局限性,即模型可能难以有效捕捉数据中深层的结构化信息。其核心问题是:是否应将表征的结构组织本身作为显式的归纳偏置(inductive bias),以提升模型对语言本质结构的理解能力。解决方案的关键在于提出一种“结构-残差”(structure-residual)原则,通过两种互补的实现方式——边界瓶颈概念(Boundary Bottleneck Concept, BBC)与连续概念(Continuous Concept, CC),分别从信息路由和几何正则化角度分离语言表示中的结构性成分与细节性残差。BBC通过边界载体传递分段信息,保持局部自回归计算;CC则通过对表征几何进行谱分解,引导低频结构与高频残差的解耦。理论推导表明,在段内路径能量趋近于零时,多路修正后的边界注意力趋于精确;实验结果显示,二者在不同网络深度表现出差异化优势(BBC适配浅层,CC适配深层),且联合使用可在仅300M训练令牌下使GPT-2 Medium的负对数似然(NLL)降低1.44%,并持续在高达200亿令牌的训练中保持性能增益。诊断分析进一步验证了BBC增强边界状态的预测重要性,而CC显著降低了谱残差能量。这些结果表明,引入结构-残差归纳偏置可超越传统下一词预测机制,显著提升语言模型的预训练效果。
链接: https://arxiv.org/abs/2611.23197
作者: Hai Huang,Bartosz Żuk,Rassul Magauin,Jiaxin Huang,Xue Liu
机构: Palona AI; Institute of Computer Science, Polish Academy of Sciences(波兰科学院计算机科学研究所); Mohamed bin Zayed University of Artificial Intelligence(穆罕默德·本·扎耶德人工智能大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Next-token prediction is relied upon to learn both predictive behavior and internal representations. We ask whether representation structure itself should be an explicit inductive bias. Motivated by the two-part coding view of Kolmogorov complexity, we study a structure–residual principle in which a representation separates structure from realization detail. We instantiate this principle in two complementary ways. Boundary Bottleneck Concept (BBC) routes information from token segments through boundary carriers while preserving local autoregressive computation. Continuous Concept (CC) regularizes representation geometry and induces a spectral decomposition into low-frequency structure and high-frequency residual. We derive coding bounds for both views and show that multiplicity-corrected boundary attention becomes exact as the within-segment path energy vanishes. Empirically, BBC and CC exhibit strikingly different depth preferences: BBC works best in early layers, whereas CC is most effective in later layers. At 300M training tokens, combining BBC and CC improves NLL by 1.44% on GPT-2 Medium. These gains persist with training runs up to 20B tokens. Representation diagnostics show that BBC increases the relative predictive importance of boundary states, while CC reduces spectral residual energy. Together, these results suggest that structure–residual biases can improve language-model pretraining beyond next-token prediction. Code: this https URL
[NLP-3] Polarity Composition Hypothesis: Causal Evidence from Negation in NLI EACL2027
【速读】: 该论文旨在解决语言模型在自然语言推理(Natural Language Inference, NLI)任务中对否定句的处理机制问题,特别是其如何理解与推理否定语义。其核心挑战在于揭示语言模型内部对否定信息的表征与计算过程,尤其是在面对不同形式的否定时,模型是否具备一致且可解释的推理逻辑。论文提出“极性组合假说”(Polarity Composition Hypothesis),作为对否定处理过程的高层级因果模型,认为语言模型通过组合正负极性信息来实现对否定语义的推断。解决方案的关键在于利用分布式对齐搜索(Distributed Alignment Search)技术,识别出与该假说中变量高度对齐的因果有效子空间,从而揭示模型内部负责否定推理的可解释表示。这些发现表明,所识别的表征不仅在不同否定形式间具有泛化能力,还能部分迁移到MNLI数据集的特定子集,验证了该假说的有效性与模型内部表征的可解释性。
链接: https://arxiv.org/abs/2611.23185
作者: Nhi Hoai Doan,Tatsuya Hiraoka,Munachiso Samuel Nwadike,Kentaro Inui
机构: 未知
类目: Computation and Language (cs.CL)
备注: Accepted to EACL 2027 (main conference)
Abstract:We investigate how language models process negation in natural language inference (NLI). We propose the \textitPolarity Composition Hypothesis, a high-level causal model of this computation. Behavioral evaluations show that language-model predictions largely conform to the proposed hypothesis. Using Distributed Alignment Search, we identify causally effective subspaces aligned with the variables in the polarity-composition hypothesis. These representations transfer across negation forms and partially to a subset of MNLI. We release the dataset and code.
[NLP-4] MoE–LoRA Methods for Multilingual Adaptation: A Systematic Empirical Study EACL2027
【速读】: 该论文旨在解决大语言模型(LLM)在低资源语言支持方面表现显著不足的问题,尤其关注生成式AI(Generative AI)中路由低秩适配器(Routed Low-Rank Adapters, LoRA)是否能有效提升多语言性能。现有假设认为LoRA可作为参数高效的方法增强模型对非英语语言的支持,但这一假设尚未在统一的多语言评估框架下得到实证验证。为此,研究通过在超过9,000个评估场景中对200种不同语言上的8种LoRA方法进行系统比较,揭示了关键发现:尽管小规模搜索可识别出可行的LoRA配置,但其预测能力不可靠,最优配置在大规模下并不持续领先;此外,LoRA虽能稳定提升翻译任务的语言模型拟合度,却普遍损害阅读理解能力,并对低资源语言带来最大收益。因此,研究结论强调,LoRA应被视为针对特定任务的精准工具,而非通用性的多语言能力增强手段。
链接: https://arxiv.org/abs/2611.23158
作者: Nikit Srivastava,Joel Dag,René Speck,Axel-Cyrille Ngonga Ngomo
机构: Paderborn University (帕德博恩大学)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: To Appear - EACL 2027 (Findings)
Abstract:Routed low-rank adapters offer a parameter-efficient means to improve the performance of large language models (LLMs) across tasks. Hence, it is assumed that they can also be used to improve the support of low-resource languages in LLMs, which currently serve English far better than the thousands of other languages spoken worldwide. However, this assumption has never been verified empirically under a shared multilingual evaluation protocol. We address exactly this gap by comparing eight low-rank adaptation (LoRA) methods on a frozen base model using 200 unique languages in more than 9,000 evaluation settings. Our results suggest that a reduced-scale search can identify viable configurations of LoRA methods but does not reliably predict their relative performance, as the top-ranking configuration is never best at scale. LoRA reliably improves language-model fit on translation data, consistently degrades reading comprehension, and benefits low-resource languages most. These findings suggest that LoRA methods should be regarded as targeted tools rather than general-purpose means for improving the multilingual capabilities of LLMs.
[NLP-5] xtSphere: Efficient Text-to-Image Generation via Shared Hyperspherical Alignment and Uniformity
【速读】: 该论文旨在解决现有文本到图像生成方法在保持高效性的同时,如何实现自由形式文本条件下的高质量生成问题。尽管基于扩散模型的方法在图像质量上取得显著进展,但其需要大量推理步骤,计算成本高;而先前的高效框架Sphere Encoder虽具备少步采样优势,却因类别条件设计缺乏跨模态对齐,难以直接扩展至自由文本生成任务。为此,本文提出TextSphere,其核心创新在于通过引入预训练CLIP文本嵌入并结合自适应层归一化(AdaLN)实现语言条件控制,同时保留原始球面采样机制以维持高效生成能力。为弥合跨模态表征差距,提出球面语义对齐与均匀性(SSAU)目标,联合优化语义对齐与超球面上的表示均匀性。实验结果表明,TextSphere在CUB-200-2011、MS-COCO-2014和Flickr30k数据集上均显著降低生成FID(gFID),且重建FID(rFID)低于1.0,在仅需4步和6步采样的情况下,性能优于从头训练的扩散模型与GAN基线。此外,通过偏最小二乘奇异值分解(PLS-SVD)分析揭示了跨模态对齐的有效性,进一步验证了所提方法在缓解模态间隙方面的优势。
链接: https://arxiv.org/abs/2611.23155
作者: Yixia Zhang,Girish A. Koushik,Xilu Wang,Wenwu Wang
机构: University of Surrey (萨里大学); CVSSP (计算机视觉与智能系统中心)
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL)
备注: 21 pages, 8 figures
Abstract:Recent advances in text-to-image generation have substantially improved image quality, particularly through diffusion-based methods that require multiple inference steps. In contrast, Sphere Encoder offers an efficient generative framework capable of producing images with fewer forward steps. However, extending it to free-form text is nontrivial, because its class-conditioned design lacks cross-modal alignment. To address this challenge, we propose TextSphere, an efficient text-to-image generative framework that extends Sphere Encoder with language conditioning and cross-modal alignment. We incorporate pretrained CLIP text embeddings through adaptive layer normalization (AdaLN) to guide image generation while retaining the original spherical sampling mechanism. To bridge the cross-modal representation gap, we introduce a Spherical Semantic Alignment and Uniformity (SSAU) objective that jointly promotes semantic alignment and representation uniformity on the hypersphere. Experiments on CUB-200-2011, MS-COCO-2014, and Flickr30k demonstrate that TextSphere consistently reduces generation FID (gFID) over the text-conditioned baseline while achieving reconstruction FID (rFID) below 1.0. With only 4 and 6 sampling steps, TextSphere achieves lower gFID than the compared diffusion- and GAN-based models trained from scratch on CUB-200-2011 and MS-COCO-2014, respectively. We further observe that improved reconstruction fidelity does not necessarily yield better generation quality. In addition, we use Partial Least Squares Singular Value Decomposition (PLS-SVD) to analyze cross-modal alignment and modality gap, offering insight into the effectiveness of our approach.
[NLP-6] Mind the Gap: A Deployed Multi-Agent System for Documentation Maintenance AAAI2027
【速读】: 该论文旨在解决生成式 AI(Generative AI)快速发展背景下,用户面向文档(customer-facing documentation)更新滞后所引发的“文档漂移”(documentation drift)问题。随着代码迭代速度远超文档更新速度,基于大语言模型(LLM)的智能代理在依赖过时或不准确文档作为知识源时,会持续传播错误信息,加剧系统性偏差。其核心解决方案是构建一个以“TriAgent”为核心调度代理的多智能体系统,通过持续监测显性和隐性信号——如已关闭的支持工单中暴露的知识缺口、代码与文档间的不一致等——自动识别并定位文档缺失或过时之处。该系统在43天的影子模式运行中成功识别出92个文档缺陷,并经由独立大模型评审团验证,81.1%的断言正确率显著高于人工基准;在可比长度的响应对比中,70.5%的案例更偏好代理生成的内容,而人类回复仅获23.2%支持。值得注意的是,代理与人工输出具有互补性:人类常揭示隐藏于文档中的深层洞察或完全缺失的信息,系统则将这些发现转化为可修复的文档缺口,并在人工审核后完成修正。修复后,同一工单的代理重答表现显著提升——盲测中71.4%的案例更倾向修复后的回答,表明该系统具备有效的自我改进能力,为实现动态、自愈型技术文档体系提供了可行路径。
链接: https://arxiv.org/abs/2611.23142
作者: Lorenzo Masoero
机构: 未知
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: Accepted at IAAI-27 (Innovative Applications of AI, co-located with AAAI 2027), Emerging Applications track. 12 pages (6 pages + references and appendix), 2 figures, 6 tables
Abstract:The “AI revolution” has enabled teams across the technology industry to deliver software features at unprecedented velocity. While code evolves faster than ever, customer-facing documentation often struggles to stay up to speed. As users increasingly rely on LLM-based agents that consume documentation as their primary interface, the impact of this documentation gap worsens: stale information propagates and is treated as authoritative. Signals for identifying where documentation has fallen behind already exist in abundance, both explicit (e.g., resolved support tickets revealing what documentation lacks) and implicit (e.g., code-documentation discrepancies detectable from automated loops). We deploy a multi-agent system, built around a triage agent we call “TriAgent,” that continuously extracts these signals and repairs documentation drift. Over 43 days of shadow-mode operation, our system triaged 499 tickets and surfaced 92 gaps in the documentation. Using a panel of LLM judges from independent model families, 81.1% of TriAgent’s verifiable assertions were correct (245 of the 302 assertions the human resolution addresses; majority of three judges); on length-comparable pairs the panel preferred TriAgent’s response in 70.5% of pairs and the human engineer’s in 23.2% (the rest had no judge majority). Importantly, agent- and human-sourced replies tended to be complementary, with the human often surfacing insights that were either buried within the existing documentation or missing from it entirely; our system turns each such insight into a documentation gap and, after human review, a fix. Repairing a gap measurably improved the agent’s later answer: when the agent re-answered the same ticket after the repair, blinded judges preferred the post-repair answer over its own pre-repair answer 71.4% of the time (n=35). A blinded human audit confirms strong alignment on self-improvement.
[NLP-7] he Illusion of Grounding: Do Vision-Language Models Weigh Retrieved Text and Images Like an Ideal Observer?
【速读】: 该论文旨在解决生成式视觉语言模型(Vision-Language Models, VLMs)在多模态推理中因信息源不可靠而导致的决策失效问题。具体而言,现有模型依赖于文本检索和图像输入两种证据来源,但二者均可能出错:检索到的文本可能存在噪声或错误,而图像则可能因退化或与文本矛盾导致误导。为系统评估模型在这些不利条件下的鲁棒性,作者构建了GIB-200K数据集,包含20万张图像-问题对,源自COCO与Visual Genome,并通过可控扰动引入三类典型挑战:硬负样本检索噪声(1–20篇文档)、视觉模糊(高斯模糊、运动模糊、低资源裁剪)以及跨模态冲突(反事实翻转对)。关键创新在于超越传统“性能下降”的度量方式,转而从理想贝叶斯线索融合理论出发,推导出可检验的预测,从而形式化定义了知识惯性(Knowledge Inertia,即模型过度依赖文本证据,相对图像证据权重过高)与感官捕获(Sensory Capture,反之)现象,并采用校正测量误差的回归方法估计模态权重。进一步地,通过反事实翻转分解分离模态偏差与概念先验,提出接地幻觉评分(Grounding Illusion Score)以量化模型准确率中非基于图像真实内容的部分。实验在2,000个固定种子测试集上评估4个开源模型,结果显示所有模型在干净输入下表现良好(92.5%–94.2%),但在恶劣条件下准确率降至41.3%–59.9%;其中两个LLaVA模型表现出显著的知识惯性(文本/图像权重比分别为1.229和1.322),而InternVL3则偏向图像(0.680);约34.8%至71.4%的正确答案在检索文本虚假时被放弃。此外,无需训练的一致性检查虽提升准确性但牺牲覆盖率,而仅依赖图像的回退策略在图像严重退化时亦失效。研究全面公开所有实例输出、代码及审计脚本,确保结果可复现。
链接: https://arxiv.org/abs/2611.23127
作者: Basavaraj R M,Harshal Chandel
机构: eClerx
类目: Machine Learning (cs.LG); Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV)
备注: 10 pages, 5 figures, 11 tables. Code and data: this https URL (archived at this https URL )
Abstract:Retrieval-augmented vision-language models (VLMs) answer from two sources of evidence that can each fail: the retrieved text can be wrong or noisy, and the image can be degraded or contradicted. We introduce GIB-200K, 200,000 image-question items built from COCO and Visual Genome, each expanded into controlled stress conditions: hard-negative retrieval noise (1-20 documents), visual ambiguity (Gaussian and motion blur, low-resource crops) and cross-modal conflict with counterfactual flip pairs. Instead of only measuring degradation, we ask how models combine the two modalities. From ideal Bayesian cue combination we derive a testable prediction that yields formal definitions of Knowledge Inertia (text over-weighted relative to the model’s own image evidence) and Sensory Capture (the reverse), and we estimate the weights with a measurement-error-corrected regression whose recovery of known weights we verify. A counterfactual flip decomposition separates modality bias from concept priors, and a Grounding Illusion Score measures how much clean accuracy is not grounded in the image. We evaluate 4 open models on one frozen, seeded evaluation split of 2,000 GIB-200K items. All models are accurate on clean inputs (92.5%-94.2%) but adverse accuracy falls to 41.3%-59.9%. The two LLaVA models over-weight the text (Knowledge Inertia, w_T/w_I = 1.229 and 1.322), whereas InternVL3 favours the image (w_T/w_I = 0.680); between 34.8% and 71.4% of the clean answers that are correct are abandoned when the document lies. A training-free consistency check that compares the answer with and without retrieval trades coverage for accuracy under misleading documents, while falling back to the image fails when the image is heavily degraded. All per-instance outputs, code, and an audit that recomputes every reported open-model number are released.
[NLP-8] Dynamic Task Allocation with Dialogue IROS2026
【速读】: 该论文旨在解决多智能体协作中动态任务分配的挑战,特别是在目标与责任随时间变化、且智能体间存在部分可观测性(partial observability)的现实场景下,传统任务分配框架因依赖静态目标假设、完全可观测性或缺乏可验证信念机制而难以适用。其解决方案的关键在于提出一种通信增强型贝叶斯任务分配框架(communication-augmented Bayesian task-allocation framework),通过对话机制主动减少对环境及协作伙伴的不确定性,同时确保决策基于显式的概率信念模型。该框架在扩展的Overcooked环境中验证,结果表明:在部分可观测条件下,该方法显著优于基线的纯大语言模型(LLM)代理,平均提速0.585,并提前18个时步完成任务;而在全可观测条件下表现与非通信基准相当,进一步证明了通信在信息受限情境下的关键作用。
链接: https://arxiv.org/abs/2611.23119
作者: Matthew Groholski,Weiwei Gu,Nakul Gopalan
机构: Arizona State University (亚利桑那州立大学)
类目: Computation and Language (cs.CL)
备注: 9 pages, 6 figures, Accepted at Human-Robot Dialogue(HRD) Workshop at IROS 2026
Abstract:Effective multi-agent collaboration requires agents to adapt task allocation as goals and responsibilities evolve. Prior task-allocation frameworks often assume static goals, full observability, or rely on large language models (LLMs) without explicit, verifiable beliefs about task allocation or intent. These assumptions limit their applicability to real-world collaboration, where agents often lack full information about each other’s actions, progress, or local environment. To support dynamic task allocation under partial observability, we introduce a communication-augmented Bayesian task-allocation framework. Our agent uses dialogue to reduce uncertainty about the environment and collaborators, while keeping decisions grounded in explicit probabilistic beliefs. We evaluate our framework in an Overcooked domain extended with dynamically arriving tasks under both full and partial object-space observability. Our communication agent consistently outperforms a vanilla LLM agent in both settings. Additionally, our framework performs comparably to a non-communicating task-allocation baseline under full observability, but substantially outperforms it under partial observability as task complexity increases by an average speed up of 0.585 and completes on average 18 timesteps faster. These results indicate that our agent is more efficient in dynamic task allocation than baselines in partially observable environments, and that communication is important as available information decreases.
[NLP-9] Nunchi: Ears Help Eyes in Multi-Party Conversations
【速读】: 该论文旨在解决多人群体对话中视线目标识别(gaze target identification)的难题,尤其针对在多人场景下仅依赖视线几何信息会导致歧义的问题。现有数据集缺乏与对话相关的语境线索,如发言者身份和实际话语内容,限制了模型对真实社交交互的理解。为应对这一挑战,研究提出了SORBET基准数据集,包含3,340个样本、1,578段视频片段,同步标注了语音、转录文本及被注视者的标签,覆盖更广泛的多人群体对话场景。其核心解决方案是提出Nunchi框架:该框架从对话转录文本及视线与说话人检测模型的输出构建文本提示(text prompt),将各类对话线索(如谁在说话、说了什么)通过共享标签与潜在注视对象关联,并利用视觉语言模型(VLM)综合权衡视线几何信息与对话上下文线索,以实现更准确的注视目标判定。实验表明,Nunchi在SORBET上相较最强基线提升4.8个百分点,且在不同VLM、自动转录和跨数据集条件下均保持性能优势,验证了对话线索的有效性,而额外视觉线索未带来显著增益。
链接: https://arxiv.org/abs/2611.23115
作者: JunYeong Park,Shijing Wang,Seonwook Park,Michael Stengel,Shalini De Mello,Sang-Il Choi,Hyung Jin Chang
机构: Dankook University( Dankook大学); Beijing Jiaotong University(北京交通大学); NVIDIA(英伟达); University of Birmingham(伯明翰大学)
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL)
备注: 10 pages, 5 figures
Abstract:Identifying whom a person looks at is central to understanding multi-party conversations, yet gaze geometry alone becomes ambiguous as more people appear. Conversations offer cues that geometry lacks, such as who is speaking and what is said, but no gaze dataset provides them for many multi-party scenes. We introduce SORBET, a benchmark of 3,340 samples from 1,578 clips with synchronized speech, transcripts, and attended person labels, covering more multi-party scenes than existing gaze datasets. Prior gaze models struggle to use these cues, which come in different forms, must be linked to the right person, and can conflict with gaze geometry. We propose Nunchi, a framework that constructs a text prompt from the transcript and the outputs of gaze and speaker models, links every cue to the candidates with shared labels, and lets a vision-language model (VLM) weigh gaze geometry against conversational cues to identify the attended person. Nunchi outperforms the strongest existing method by 4.8 percentage points. On SORBET, conversational cues help whereas additional visual cues do not, and the gains hold across three VLMs, with automatic transcripts, and on two existing datasets.
[NLP-10] Large-scale Neuron Community Discovery Reveals Neuron-Sparse Behaviors in Language Models
【速读】: 该论文旨在解决语言模型中是否存在可复用的计算单元(即在不同输入间重复出现且具有因果必要性与充分性的神经元群体)这一关键问题。其核心挑战在于,尽管已有研究发现单次前向传播可通过后处理稀疏化为少量活跃神经元,但这些神经元组合是否构成跨输入一致、可重用的计算实体仍不明确。为此,作者提出“神经元社区发现”(neuron community discovery),一种无监督方法,通过结合基于梯度的归因、神经元间重复被共同归因所构建的图结构以及社区检测算法,识别出在多个提示(prompts)中反复出现且在所属层内对特定计算任务具备因果必要性和充分性的神经元群组。在对 Llama-3.1-8B 模型使用千万级提示进行实验时,该方法成功发现了数千个候选社区。通过三项因果测试(社区必要性、层内子集充分性、层必要性)在保留数据上验证,结果表明部分社区移除会破坏预测,而保留这些社区则可在剔除其余多层 MLP 神经元的情况下维持预测准确性。所发现的社区支持语法一致、词元概率校准、固定短语生成、上下文复制、加法与乘法等具体功能。当将社区规模限制在每层不超过 128 个神经元时,在 10.3% 的保留提示中至少有一层的贡献可由稀疏神经元群解释;若要求至少四层参与,覆盖率降至 8.7%;若要求十二层,则为 3.3%。研究表明,现有神经元的小规模群体可作为跨提示可复用的计算组件,并且这类群体能够在大规模模型中被有效发现。
链接: https://arxiv.org/abs/2611.23112
作者: Igor Shilov,Thomas Fel,Tal Haklay,Lucius Bushnaq,Owen Lewis
机构: 未知
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注:
Abstract:What reusable units of computation can we discover within a language model’s neurons? Prior work has shown that individual forward passes can often be sparsified post-hoc to small groups of neurons, but it has been unclear whether these groups form coherent computational entities reused across inputs. We introduce neuron community discovery, an unsupervised method for finding neuron groups that recur across prompts and are causally necessary and sufficient, within the layers they span, for common computations. The method combines gradient-based attribution, a graph connecting neurons that are repeatedly attributed together, and community detection. Applied to Llama-3.1-8B using ten million prompts, it discovers thousands of candidate communities. We evaluate them on held-out prompts through three causal tests: community necessity, sufficiency within a layer subset, and layer necessity. These tests identify groups whose removal disrupts predictions and whose retention preserves predictions despite ablating the remaining MLP neurons in the tested layers. Discovered communities support functions like grammatical agreement, token-probability calibration, fixed phrases, in-context copying, addition, and multiplication. With communities capped at 128 neurons across all layers, 10.3% of held-out prompts have at least one layer’s contribution explained on a neuron-sparse basis. Coverage remains 8.7% when requiring at least four layers and 3.3% when requiring at least twelve. These results show that small groups of existing neurons can provide reusable computational components across prompts, and that such groups can be discovered at scale.
[NLP-11] Natural Language Programming: Rethinking Language Memory and Cognition in the Era of Generative AI
【速读】: 该论文旨在解决生成式人工智能(Generative AI)背景下自然语言在计算系统中角色的根本性转变问题,即如何超越传统自然语言处理(NLP)中将语言视为被动处理对象的范式,重新定义语言作为可执行认知媒介的功能。其核心挑战在于:当大型语言模型(LLMs)使自然语言具备编程与推理能力时,如何理论化和建模人类与AI协作中的动态认知过程。解决方案的关键在于提出“自然语言编程”(Natural Language Programming, NLPg)这一新范式,将提示(prompt)视为动态的认知产物,不仅承载任务目标与约束条件,还外化了中间推理状态与验证标准。通过引入“对话式工作记忆”(Dialogic Working Memory)概念,论文将提示与对话状态建模为分布式人机记忆架构中的临时外部记忆结构,支持推理连续性与元认知调控。该框架整合了分布认知、扩展心智、元认知及对话学习等理论,并将现有记忆架构(如BTMA、SKGE、RMR)映射至该体系,揭示了情景、语义与元认知层如何协同支撑长时程对话连贯性。最终,论文推动计算语言学与人机知识系统进入后符号化、后界面的新范式,其中自然语言同时充当通信媒介、可执行规范、认知支架与分布式记忆架构。
链接: https://arxiv.org/abs/2611.23106
作者: Giovanni Adorni,Daniele Grosso
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 11 pages, 1 figure, 2 tables
Abstract:Generative Artificial Intelligence is transforming the role of natural language in computational systems. In classical Natural Language Processing (NLP), language was treated as an object of computation: something to analyse, classify, translate, or generate. Large Language Models (LLMs) challenge this perspective, enabling natural language to function increasingly as an executable cognitive medium through which humans structure reasoning, orchestrate problem solving, and regulate hybrid human/AI workflows. This paper introduces Natural Language Programming (NLPg) as a theoretical and epistemological extension of traditional NLP. Rather than focusing on language understanding alone, NLPg conceptualises prompts as dynamic cognitive artefacts that externalise goals, constraints, assumptions, validation criteria, and intermediate reasoning states. Prompting is thus not merely a technical interaction technique, but a form of dialogic cognitive programming operating within distributed human/AI memory architectures. Building on theories of distributed cognition, extended mind, metacognition, and dialogic learning, the paper proposes the notion of Dialogic Working Memory, describing how prompts and conversational states function as temporary external memory structures supporting reasoning continuity and reflective regulation. We further map existing memory architectures (BTMA, SKGE, RMR) onto this framework, illustrating how episodic, semantic, and metacognitive layers support long-horizon conversational coherence. Finally, implications for computational linguistics and AI/mediated knowledge systems are discussed, suggesting a post symbolic and post interface paradigm in which natural language operates simultaneously as communication medium, executable specification, cognitive scaffold, and distributed memory architecture. Comments: 11 pages, 1 figure, 2 tables Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI) MSC classes: 68T50 (Primary), 68T01, 68T30 (Secondary) ACMclasses: I.2.7; H.5.2; I.2.0 Cite as: arXiv:2611.23106 [cs.CL] (or arXiv:2611.23106v1 [cs.CL] for this version) https://doi.org/10.48550/arXiv.2611.23106 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Giovanni Adorni [view email] [v1] Tue, 17 Nov 2026 17:27:01 UTC (45 KB)
[NLP-12] Whats in a Tree ? Hierarchical n-gram models as controlled approximators of language
【速读】: 该论文旨在解决层次化语言模型(hierarchical models)在真实语料库中适用性不足的问题,即现有层次化模型虽可作为语言研究的合成场景,但其与真实语言数据之间的关联性尚未得到充分探索。其核心解决方案是提出“树形n元语法”(tree-grams),一种基于自回归生成的n元语法模型,通过层次化组合二阶与三阶概率张量来构建条件概率张量 $ p(x_n \mid x_{1:n-1}) $。该模型由两个关键控制参数构成:最大上下文窗口 $ n $ 和容量控制的秩超参数 $ d $,从而形成一组可解释且系统可扩展的语言近似器。实验表明,树形n元语法在性能上超越传统n元语法基线,并在与基于Transformer的深度语言模型竞争中保持相当水平。此外,论文展示了该模型在可控生成任务中的价值,例如通过在较简单数据(由中等 $ n $ 与 $ d $ 的树形n元语法生成)上微调窄模型,可增强在 $ \mu P $ 超参数转移下的鲁棒性;同时发现熵与相关性有效范围可调控微调后的性能差距,这得益于树形n元语法对建模时域(horizon)和容量的显式控制能力。当允许模型在候选拓扑结构间进行路由选择时,观察到显著的稀疏性涌现现象,使得高达95%的拓扑结构可被剪枝而几乎不损失性能,体现了其内在结构优化潜力。
链接: https://arxiv.org/abs/2611.23093
作者: Aritra Das,Francesco Cagnetta,Darshil Doshi,Victor V. Albert,Maissam Barkeshli,Andrey Gromov
机构: Google(谷歌); Stanford University (斯坦福大学); MIT (麻省理工学院); Caltech (加州理工学院); Flatiron Institute (扁平铁研究所); Harvard University (哈佛大学)
类目: Computation and Language (cs.CL); Disordered Systems and Neural Networks (cond-mat.dis-nn); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 7+13 figures, 10+30 pages
Abstract:Hierarchical models have been proposed as synthetic settings for studying language. However, their relevance to realistic language corpora remains underexplored. In this work, we introduce tree-grams, autoregressive generative n -gram models that build the conditional probability tensor p(x_n \mid x_1:n-1) through a hierarchical composition of order-2 and order-3 tensors. n controls the maximal context window, whereas a rank hyperparameter d controls model capacity, yielding an interpretable and systematically scalable family of approximators of language. We show how tree-grams outperform existing n -gram baselines, while remaining competitive with deep transformer-based language models. We demonstrate the utility of such controlled approximations by using tree-grams as generative models. In particular, we show that hyperparameter transfer under \mu P can be made more robust when narrower models are tuned on simpler data, generated by tree-grams with moderate n and d . In addition, we show that entropy and effective range of correlations control performance gap after fine-tuning. This is because tree-grams allow us to explicitly control horizon and capacity. When allowed to route across candidate topologies, we observe emergent sparsity in tree-grams. Such concentrated routing permits pruning upto 95% of topologies without significant change in performance.
[NLP-13] Runtime Verification of Multiple Natural Language Criteria for Agent Governance NEURIPS2026
【速读】: 该论文旨在解决生成式AI代理(AI agent)在实际应用中面临的核心挑战:如何高效、准确地对代理输出进行多准则验证。当前的验证器难以并行评估大量自然语言形式的规则(criterion),且无法提供每个规则的独立判断结果(如满足、违反或不适用)。为此,本文提出VFM(Tree-Attention Verifier),其核心创新在于采用树注意力机制(tree-attention mechanism),能够同时编码共享上下文与目标文本,并以并行方式评分多达N个自然语言准则,输出每个准则的三元判定结果(ternary verdicts)。VFM在包含39.4万条(上下文, 目标, 准则, 判定)样本的CriteriaBank数据集上预训练,该数据集涵盖超过32.1万个不同的准则字符串。实验表明,微调后的VFM-4B模型在合成治理基准测试中优于GPT-5.4;在DynaBench多准则基准测试中,其在追踪级别故障检测方面超越DynaGuard-4B,且细粒度的逐准则判决定位了违规规则,支持定向升级至更复杂的链式思维(chain-of-thought)裁判器。此外,研究证明CriteriaBank预训练显著提升了跨领域迁移能力,尤其在低数据场景下表现突出,同时在全量数据条件下可媲美直接微调的强基线模型。在单张H100 GPU上,VFM-4B可在2秒内完成对4K token上下文中1000个准则的评估,显存占用仅11 GB,具备部署于运行时验证场景的可行性。
链接: https://arxiv.org/abs/2611.23074
作者: Silviu Pitis,Parand A. Alamdari,Jessica Tang,Toryn Q. Klassen,Sheila A. McIlraith
机构: University of Toronto (多伦多大学); Vector Institute (向量研究所); Toronto, Canada (加拿大多伦多)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Accepted at NeurIPS 2026
Abstract:AI agents are governed by hundreds of natural language criteria, yet existing verifiers cannot efficiently evaluate an agent’s output against a large set of such criteria and report which are satisfied, violated, or inapplicable. We introduce the VFM, a tree-attention verifier that encodes the shared context and target text, and scores N natural language criteria in parallel, returning independent per-criterion ternary verdicts. The VFM is pretrained on CriteriaBank, a corpus of 394K (context, target, criterion, verdict) tuples spanning over 321K distinct criterion strings. A finetuned VFM-4B outperforms GPT-5.4 on a synthetic governance benchmark. On the DynaBench multi-criterion benchmark, it outperforms DynaGuard-4B on trace-level failure detection, and its per-criterion verdicts localize the violated rule and support focused escalation to a larger chain-of-thought judge. We show that CriteriaBank pretraining improves cross-domain transfer, particularly in the low-data regime, while matching a strong direct-finetuning control at full data. On a single H100, VFM-4B evaluates 1,000 criteria against a 4K-token context in under 2s at 11 GB VRAM, making it suitable for runtime verification.
[NLP-14] he Resolution Hypothesis: Agent Evolutions Key Is Identifying Improvements
【速读】: 该论文旨在解决自进化智能体(agent evolution)中普遍存在的“方法表现不一致且难以启动特定任务”的难题。其核心问题在于:尽管候选生成机制能够提出有效改进,但评估过程中的噪声导致无法可靠识别真正有益的候选产物,从而阻碍了进化循环的有效推进。解决方案的关键在于提出并验证“分辨率假说”(Resolution Hypothesis),即进化停滞并非源于生成能力不足,而是由于评估信号的不可靠性——在候选层面,评估系统无法有效区分有益与有害的候选物,导致错误在多轮迭代中累积;在方法层面,不同方法间的性能差距微小,甚至小于同一方法在不同运行间的波动,使得现有排名结果往往依赖偶然性。因此,研究指出,未来应从设计更复杂的候选生成机制转向提升改进效果的可检测性与可积累性,使真正的进步能够被稳定识别和持续放大。
链接: https://arxiv.org/abs/2611.23068
作者: Yuanhao Zeng,Shaocheng Lan,Ao Lu,Kuang Yan,Heng Lin,Xinang Xu,Yuan Luo,Yueting Wang,Yexin Li,Ruo Shan,Kan Ren
机构: 上海科技大学(ShanghaiTech University)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 38 pages, 12 figures, 13 tables. Project page: this https URL
Abstract:Agent evolution follows a simple loop: a proposer writes candidate artifacts (prompts, skills, memories) and selection accepts those that score higher. This loop has shown promise, yet practitioners face a puzzle: methods trade wins across papers, and evolution often fails to bootstrap on certain tasks. We trace this puzzle to the Resolution Hypothesis: agent evolution stalls not because methods fail to propose better artifacts, but because noisy evaluations cannot reliably identify them. Across seven evolution methods and six models (over 10M LLM calls), we find support for the hypothesis at both levels where evaluation scores are used. At the candidate level (choosing artifacts), helpful candidate artifacts exist, but selection is no better than random at telling them from harmful ones, so errors accumulate across rounds. At the method level (ranking methods), typical margins of a few percentage points are smaller than the difference between two independent runs of the same method, so published rankings often rest on luck. Together, these findings indicate that agent evolution is often bottlenecked by reliably identifying improvements. This calls for a shift in how we develop and evaluate self-improving agents, from designing ever more candidate improvement systems to making true improvements detectable and accumulable.
[NLP-15] ForecastGym: A Live Controlled and Replayable Benchmark for Forecasting Agents
【速读】: 该论文旨在解决基于大语言模型(LLM)的智能体在预测现实世界事件时,因检索到的证据(evidence)与模型自身能力共同影响预测性能,而现有评估方法难以实现对证据获取过程的可控与可复现的问题。其核心挑战在于:在线实时评估会将检索与推理混杂,导致结果不可控;而事后回溯搜索则可能通过时间过滤或当前网页排名泄露未来信息,造成评估偏差。为实现公平、可复现的模型比较,需在固定证据获取策略下,记录并追踪证据随时间演化的状态。为此,作者提出了ForecastGym——一个系统性记录四个商用网络搜索API针对开放预测市场问题每两天返回结果的平台,并在事件结果公布后,以“冻结”证据状态的方式离线重放,从而在一致的证据条件下对比不同模型、记忆策略及搜索源的表现。该系统首版覆盖150个问题、614个证据状态(每个问题对应一次采集时间点)和13.1万条搜索结果。实验发现,在冻结证据下重新运行查询,92%的问题返回结果几乎直接揭示了最终答案,即便使用了日期过滤器也未能完全避免信息泄露。在固定证据条件下,所有测试的LLM在预测准确性上均显著落后于市场基准。诊断分析表明,预测差距主要源于未能捕捉公开可用的关键信息,而非模型误读。此外,更换搜索接口可使模型的布里尔分数(Brier score)变化高达0.043,超过不同模型间的差异;而持续记录笔记的效果与仅读取一半原始证据相当,且消耗更少输入令牌。这表明,预测者所见的证据本身,其质量与来源的重要性甚至不亚于所用模型的推理能力。
链接: https://arxiv.org/abs/2611.23045
作者: Yicheng Tao,Yuante Li,Zipei Fan
机构: Carnegie Mellon University (卡内基梅隆大学); Jilin University (吉林大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Agents built on large language models (LLMs) increasingly forecast real-world events by searching the Web and reasoning over what they find. How well such an agent forecasts depends on both the retrieved evidence and the model, making control over that evidence essential for evaluation. Live evaluations can mix retrieval with reasoning when each system searches on its own, while retrospective search can leak later information through date filters or today’s ranking of historical pages. Comparing models, search sources, and memory strategies requires both fixed evidence and a record of how it evolved before resolution. We introduce ForecastGym, which records what four commercial web search APIs return for open prediction-market questions every two days and replays this frozen evidence offline once outcomes are known, comparing models, memory, and captured search sources under a fixed evidence-acquisition policy. Its first release covers 150 questions, 614 evidence states, one per question and collection time, and 131k search results. Rerunning the same queries after settlement returns results that all but state the outcome for up to 92% of the questions, despite the tested date filters. On the frozen evidence, all tested LLMs trail the market in forecasting accuracy. Our diagnostic analysis suggests that missed public information contributes more to the gap in high-gap states than misreading does. Switching the search API changes a model’s Brier score by up to 0.043, more than the spread across models, while persistent notes score similarly to rereading all evidence at half the input tokens. The evidence a forecaster sees can matter as much as the model that reads it.
[NLP-16] Storing the Gemma 4 K=V Cache Once: Exact Global-KV Memory Savings in a Paged Serving Engine with Unchanged Attention Kernels
【速读】: 该论文旨在解决大模型推理中全局注意力层(global attention layers)所导致的键值缓存(KV cache)存储开销过大的问题。在Gemma 4 31B模型中,尽管其设计通过将值(value)表示为归一化后的键投影、键(key)则由值按固定权重缩放,并结合旋转位置编码(rotary position encoding)于四分之一维度上,理论上可减少高达37.5%的全局KV缓存,但现有开源推理引擎仍采用完整的键和值存储方式,即每头每标记存储1,024个值。本文提出的关键解决方案是:仅存储一个640维的紧凑值行(exact in real arithmetic),并利用不变的分页注意力核函数(paged attention kernel)通过两个重叠的步进视图分别读取作为键缓存与值缓存。该方法在vLLM原型中实现与标准缓存算术等价的比特级输出,在16位精度下31B模型的最终答案仅有2/432出现差异(处于运行噪声范围内),且在32K至64K令牌长度下,每个请求的解码阶段KV内存消耗降低1.34至1.40倍。在双A100 GPU的预填充/解码测试场景下,采用15 GiB完整缓存池的解码引擎相比基准版本,在16K提示词下提升1.38倍解码吞吐量(误差<0.06%),8K提示词下提升1.24倍;基于vLLM块池的模型预测值与实测结果偏差均在1%以内,且能准确复现吞吐量变化趋势。此外,即刻解码(eager decode)吞吐量提升1.30和1.11倍,且在批处理规模为1时,8个提示词的输出完全一致。尽管该原型在同等批量下使单步解码慢4%至9%,但尚未支持默认尺度下的8位KV缓存压缩。因此,该方案的核心在于通过数学等价的紧凑缓存结构与现有分页注意力机制兼容,实现了显著的内存节省与性能提升。
链接: https://arxiv.org/abs/2611.23035
作者: Kushagra Bainsla
机构: San José State University (圣何塞州立大学)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL); Distributed, Parallel, and Cluster Computing (cs.DC)
备注: 21 pages, 4 figures, 3 tables
Abstract:In the global attention layers of Gemma 4 31B, the value is the normalized key projection, and the key is the value scaled by a fixed weight, with rotary position encoding on one quarter of its dimensions. The Gemma 4 report states that this design reduces the global KV cache by up to 37.5%, but the open serving engines that we checked still store a full key and a full value: 1,024 values per KV head and token. We store one 640-value row instead, exact in real arithmetic, and an unchanged paged attention kernel reads it through two overlapping strided views as its key cache and its value cache. In a vLLM prototype with the 16-bit 31B model, the outputs are bit-identical to an emulation of the same arithmetic on the stock cache, 2 of 432 final answers differ from the stock engine (within its run-to-run noise), and the steady decode KV memory per request falls by a factor of 1.34 to 1.40 at 32K to 64K tokens. On a prefill/decode pair of two A100 GPUs with NIXL transfer and CPU offload, a decode engine with a full 15 GiB pool ran 1.38 times the stock decode tokens per step for 16K-token prompts (repeated to 0.06%) and 1.24 times for 8K-token prompts; a model of the vLLM block pool predicted both ratios to within 1% (the engine was 0.7% to 0.9% higher) and matched the change between them. The eager decode throughput rose by 1.30 and 1.11 times, and on this path the compact engine gave the same output tokens as the emulation for 8 prompts at batch 1. The run-time prototype makes an eager step 4% to 9% slower at equal batch, and the method does not yet stack with an 8-bit KV cache at the default scale.
[NLP-17] PHC: Improving Hyper-Connections with Tensor-Parallel Grouped Updates
【速读】: 该论文旨在解决生成式 AI (Generative AI) 中基于 Transformer 架构的超连接(Hyper-Connections, HC)及其改进版本流形约束超连接(manifold-constrained HC, mHC)在扩展并行残差流(residual stream)时面临的更新独立性不足问题。尽管 mHC 在大规模模型中具有稳定性,但其性能提升受限于不超过四个并行流,根源在于每个子层仅向所有流写入单一输出,导致各流间更新差异有限。为此,本文提出张量并行融合超连接(Tensor-Parallel Hybrid Connections, TPHC),其核心创新在于将张量并行(Tensor-Parallel, TP)分片与 HC 流结合,通过将不同残差流的混合分别路由至各 TP 分片,并在写回阶段保持各分片输出独立,使每条流可接收来自不同分片的差异化输出组合。这一机制显著增强了残差流更新的表达能力,且无需增加注意力或前馈网络子层的参数量或浮点运算量(FLOPs)。实验表明,在 80 亿参数模型上,TPHC 仅比 mHC 多消耗 9% 训练时间(4 流配置),当流数从 4 增至 16 时训练时间仅增加 19%;在 11 亿参数模型上训练 1000 亿词元数据集,TPHC 在相同流数下相较 mHC 将验证损失降低 0.7–1.0%,并在六个常识推理基准上平均零样本准确率提升 0.3–0.7 个百分点。
链接: https://arxiv.org/abs/2611.23026
作者: Xinying Hu,You Wu,Haoyi Wu,Jianchao Tan,Kefeng Zhang,Yuchen Xie,Kewei Tu
机构: ShanghaiTech University (上海科技大学); Shanghai Engineering Research Center of Intelligent Vision and Imaging (上海智能视觉与成像工程研究中心); Meituan (美团)
类目: Computation and Language (cs.CL)
备注:
Abstract:Hyper-Connections (HC) expand the residual stream of a Transformer into N parallel streams, and manifold-constrained HC (mHC) stabilizes them at scale. However, mHC benefits little from more than four streams, partly because each sublayer writes a single output to all streams, limiting how differently they can be updated. In this work, we propose TPHC, which integrates tensor-parallel (TP) shards with HC streams to promote greater independence across streams. Specifically, we route a separate mixture of the residual streams to each TP shard and keep the individual shard outputs separate during write-back, so each stream can receive a different mixture of those outputs. This design makes the residual-stream updates more expressive without adding parameters or FLOPs to the attention or FFN sublayers. With native support for stream sharding under TP, TPHC requires only 9% more training time than mHC on an 8B model when both use four streams. Increasing the number of streams from four to sixteen raises its training time by only 19%. On a 1.1B model trained on 100B tokens, TPHC reduces validation loss by 0.7-1.0% and raises average zero-shot accuracy across six commonsense reasoning benchmarks by 0.3-0.7 percentage points relative to mHC at matched stream counts N\in\4,8,16\ . Our code is available at this https URL.
[NLP-18] Large Language Models Have Not Yet Transformed Science
【速读】: 该论文旨在解决的问题是:在2022年底大规模语言模型(Large Language Models, LLMs)兴起背景下,科学界对未来科学研究范式演变的担忧日益加剧,但缺乏对科学实践自身演进趋势的系统性认知,难以准确评估LLMs对科学发展的实际影响。为此,研究通过分析2015至2025年间涵盖五个学科领域的250万篇全文论文,从词汇使用、结构特征、认识论维度及社会协作等多个层面追踪科学实践的长期演变趋势。研究发现,除部分词汇频率变化外,大多数2023年前的趋势在后续时期仍持续发展,未出现显著断裂;值得注意的是,亚洲计算机科学领域合作者团队规模的增长显著加速,而生物医学领域主题多样性的“下降”现象主要源于文章格式的标准化,并非真实知识广度缩减。该研究的关键解决方案在于构建了一个基于大规模实证数据的基准框架,为未来在LLM深度应用背景下持续监测全球科学健康状态提供了可量化的分析基础。
链接: https://arxiv.org/abs/2611.22975
作者: Ron Bekkerman,Nichole Li,Bahareh Kargar,Mark Somers
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Computers and Society (cs.CY); Digital Libraries (cs.DL)
备注:
Abstract:The emergence of large language models (LLMs) in late 2022 has raised questions about the future of science, but assessing their impact requires understanding how scientific practices were already evolving. We analyze 2.5 million full-text papers across five disciplines from 2015 to 2025, tracing trends over a broad spectrum of lexical, structural, epistemological, and social dimensions. Aside from frequency shifts in some words, most pre-2023 trends continued with limited detectable breaks. Among the notable exceptions, growth in computer science coauthor teams accelerated significantly in Asia, while apparent declines in biomedical topic diversity largely reflected standardization in article formats. Our study establishes an empirical foundation for monitoring the health of global science as LLM adoption deepens.
[NLP-19] Can Decision Models Replace LLM s in Real-World Contact Center Classification and Judging Tasks?
【速读】: 该论文旨在解决在客服中心(contact center)场景中,如何在有限上下文任务(bounded conversational tasks)下高效、低成本地实现精准决策的问题。具体而言,研究关注生成式大语言模型(Generative LLMs)在实际应用中的高成本与低效率问题,探索替代方案——即决策模型(decision models)在零样本(zero-shot)条件下对多种客服任务的适用性与性能表现。其解决方案的关键在于:通过系统评估九种决策模型与十三种生成式LLMs在八个典型客服用例(六类分类任务与两类判断任务)中的表现,发现多数决策模型在分类任务上可达到与前沿LLMs相当的性能,且基于API的封闭式决策模型在单位请求成本上最具优势;同时,决策模型的高置信度预测具有更高的准确性,支持在生产环境中实现选择性路由(selective routing)。然而,大多数决策模型的主要缺陷在于“拒答”(abstention)能力不足,即极少主动选择不回答,导致其与最强LLMs之间的性能差距主要源于此。此外,小型决策模型在判断任务上的质量显著下降,并更易受位置偏差(position bias)影响。因此,论文提出应综合考虑拒答行为、置信度水平及部署成本,而非仅依赖整体准确率来选择决策模型,为实际落地提供了可操作的指导原则。
链接: https://arxiv.org/abs/2611.22949
作者: Md Tahmid Rahman Laskar,Xue-Yong Fu,Shashi Bhushan TN
机构: Dialpad Inc.(Dialpad公司)
类目: Computation and Language (cs.CL)
备注:
Abstract:Contact center systems often need to make bounded decisions, such as identifying a caller’s intent, retrieving the knowledge-base article that answers a query, assessing agent compliance, or judging the output of AI models used in production. Decision models return a typed answer and its probability directly, making them a promising cost-effective alternative to generative Large Language Models (LLMs) in such bounded conversational tasks. However, their effectiveness in real-world contact center applications remains unclear. In this paper, we compare nine decision models against thirteen generative LLMs in zero-shot settings across eight contact center use cases: six classification tasks and two judging tasks. Our experiments demonstrate that many decision models can match the performance of frontier LLMs on contact center classification, and API-based closed decision models achieve this at the lowest per-request cost among the API models we evaluate. We further observe that their confident predictions are more accurate, enabling selective routing in real-world production scenarios. Across classification tasks, the main weakness of most decision models is abstention: they rarely choose the no-answer option, which accounts for most of their gap to the strongest LLMs (although Gemini-Decision-1 is the notable exception in this regard). We also find that the judging quality of decision models drops sharply for smaller models, and they are also more prone to position bias. In short, our findings provide practical guidance for selecting decision models based on abstention behavior, confidence, and deployment cost, rather than overall accuracy alone.
[NLP-20] he Shared Word: Evaluating Language Models on Dutch Word Puzzles EACL2027
【速读】: 该论文旨在解决现有非英语大语言模型(LLM)评估基准普遍存在的问题,即多数基准为翻译版本、采用多项选择形式或难以实现自动化评分。为此,作者提出了针对荷兰语的专用评估基准——《共享词》(The Shared Word, TSW-NL),其核心设计遵循四个原则:语言特定性、开放词汇表、人工校准与确定性评分。该任务要求模型根据三个描述同一荷兰语词汇不同语义或用法的线索进行推理,每条线索单独均具歧义,需综合多线索信息才能正确解答。数据集基于荷兰语维基百科的消歧页面构建,经自动过滤与大规模人工校准后形成222个测试实例,每个实例包含三组独立的人类标注答案。实验表明,该任务具有显著挑战性:人类在测试中的准确率介于49.5%至61.0%之间,而表现最佳的开源模型仅达39.3%,多数模型低于20%。相比之下,闭源模型进展迅速,GPT-4.1(2025年4月)达到59.2%,而GPT-5.6 Sol(最大推理模式)则高达88.8%。该研究不仅为荷兰语模型评估提供了有效工具,更提出了一套可复用的方法论,适用于构建其他语言的类似基准。
链接: https://arxiv.org/abs/2611.22935
作者: Rik van Noord,Maurice Voors
机构: CLCG, University of Groningen(格罗宁根大学)
类目: Computation and Language (cs.CL)
备注: EACL 2027 (Findings)
Abstract:Many LLM benchmarks for languages other than English are translated, multiple-choice, or difficult to score automatically. We introduce The Shared Word (TSW-NL), a Dutch benchmark built around four core principles: it is language-specific, open-vocabulary, human-curated, and deterministically scored. Each puzzle gives three clues that describe different senses or uses of the same Dutch word. The clues are individually ambiguous, requiring the solver to combine evidence across multiple clues. We construct the benchmark from Dutch Wikipedia disambiguation pages, followed by automatic filtering and extensive manual curation. The final dataset contains 222 instances, with three independent human answers per test item. The task is clearly challenging: scores in our human study range from 49.5% to 61.0% and the highest-scoring open-weight model we tested reaches only 39.3%. Many others even score below 20%. At the same time, closed-source models show great progress: GPT-4.1 (April 2025) scores 59.2%, whereas GPT-5.6 Sol with max reasoning scores 88.8%. We believe this task provides a useful addition to Dutch model evaluation, plus a clear recipe that can be applied to construct a similar benchmark for other languages.
[NLP-21] When Do Typed Decision Models Work in Transportation? Evidence from Travel Choice and Traffic Signal Control
【速读】: 该论文旨在解决生成式人工智能(Generative AI)在交通系统决策应用中因响应冗长、计算开销大和推理延迟高而带来的效率瓶颈问题。其核心挑战在于如何在保持高决策质量的同时,实现低延迟、低成本的实时交通决策支持。解决方案的关键在于采用“有类型决策模型”(Typed Decision Models),该模型通过结合自然语言理解与结构化输出接口,直接返回选定动作及选项级别的概率分布,避免生成冗长文本,从而显著降低推理成本与响应延迟。实验表明,尽管在个体出行方式选择预测任务中,有类型决策模型的表现低于传统离散选择模型和最强的基于对话的大语言模型,但在闭环交通信号控制任务中,其在多数场景下可达到与成熟控制器相当的性能,并在复杂网络条件下仍具备较低的计算开销与延迟优势,凸显其在零样本、高效交通决策中的潜力。研究进一步强调,输入设计、问题表述方式以及任务特性对模型表现具有显著影响,提示未来应用需重视鲁棒性评估与输入工程优化。
链接: https://arxiv.org/abs/2611.22926
作者: Hongyang Zhang,Boyuan Deng,Siyuan Feng,Kai Chen,Junbo Zhang
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 22 pages, 4 figures
Abstract:Typed decision models combine natural-language understanding with a structured interface that returns a selected action and an option-level probability distribution without generating lengthy textual responses. Their low inference cost and response latency make them attractive for transportation problems with predefined alternatives. This paper investigates their effectiveness, robustness, and efficiency in two representative transportation tasks: individual travel-mode choice prediction and sequential traffic signal control. We systematically evaluate TypeSafe Jev and GPT-6 Luna Decisions against conventional transportation methods, reinforcement learning algorithms, and four chat-based large language models. Experiments on the Swissmetro dataset show that typed decision models achieve lower prediction accuracy than traditional discrete choice models and the strongest chat-based baseline, while their performance varies substantially with input representation and question framing. In closed-loop traffic signal control experiments, typed decision models demonstrate competitive performance against established controllers in several scenarios, but exhibit performance degradation under more challenging network conditions. Furthermore, the typed models achieve lower inference latency and computational cost than the evaluated chat-based large language models, highlighting their potential for efficient, zero-shot transportation decision-making. Overall, these findings demonstrate both the potential and limitations of typed decision models, emphasizing the importance of input design, task characteristics, and robustness evaluation for their effective application in transportation systems.
[NLP-22] When Passing Checks Do Not Establish Completion: Evidence Interpretation in Coding Agents
【速读】: 该论文旨在解决编码代理(coding agents)在验证过程中如何准确保持执行事实、正确理解其作用范围并有效履行检查义务的问题。其核心挑战在于确保生成的验证证据不仅在内容上正确,且在范围与责任边界内,避免误报或越界。解决方案的关键在于引入受控的证据消费者(controlled evidence consumers)、可执行的工作流(executable workflows)以及经过资格认证的仓库检查机制(qualified repository checks),通过结构化控制实现对修复过程的分阶段管理:将启动可达性分析与修复操作分离,将见证生成与声称的操作解耦,并借助生命周期与资源约束明确验证通过所确立的结论边界。实验结果表明,尽管多数交付标签正确,但存在事实误报与越权行为;生成笔记虽能揭示局部修复差异,却可能提供错误解释。综合多轮评估显示,现有方法在整体诊断性能上并无显著优势,亦未表现出技能退化或部署折叠效应,从而为验证过程中的证据解释、适用性及实际动作的完整性提供了严谨的实证刻画,并限制了对通用优越性的过度宣称。
链接: https://arxiv.org/abs/2611.22835
作者: Xueping Gao
机构: Alibaba Cloud Computing(阿里巴巴云)
类目: Computation and Language (cs.CL); Software Engineering (cs.SE)
备注: 12 pages; supplementary material included as ancillary files
Abstract:Verification procedures in coding agents must preserve execution facts, interpret their scope and discharge checking obligations. We study these responsibilities through qualified repository checks, controlled evidence consumers and executable workflows. A prospective thirteen-task factorial study finds that correct reported facts need not yield the requested finite verdict. Conversely, all 122 delivered labels in a held-out receipt-supply study are correct, while sixteen responses misreport facts and eleven others exceed authorized scope. Production repair supplies successful controls and finite-supported preservation failures. Across 118 repair episodes on 31 tasks, source interventions separate startup reachability from repair and witness production from the claimed operation. Lifetime and resource controls delimit what passing validation establishes. Generated-note comparisons yield local repair differences but can state false explanations; strong ordinary baselines, new tasks and Skill-transfer studies supply no consistent overall advantage. These results characterize evidence interpretation, applicability and fulfilled actions while limiting claims of general diagnostic superiority, Skill deterioration or deployed folding effects.
[NLP-23] Compact and Decodable Electrocardiogram Text for Large Language Models
【速读】: 该论文旨在解决将12导联心电图(Electrocardiogram, ECG)以文本形式输入大语言模型时,因压缩导致信息丢失与文本标记数(token count)之间难以平衡的问题。其核心解决方案是提出一种确定性编码方法:将8个独立导联以幅度字母和精确时间间隔的形式编码,剩余4个肢体导联则通过代数方式恢复。该方法的关键在于采用基于心脏周期的自适应容差选择关键点,利用从训练数据中学习得到的26字母对数码本对幅值进行参数化表示,并设计10步容差调度策略,在固定令牌预算(4,080个令牌)内确保所有记录均被完整存储。实验基于公开的PTB-XL数据库中的21,388条记录,通过解码-重编码回环测试、波形重建误差评估以及基于卷积分类器的成对比较验证了有效性。结果显示,平均相关系数达0.982,诊断性能指标(受试者工作特征曲线下面积,AUC)在原始信号与解码信号间差异仅为-0.0010,处于预设非劣效性边界(-0.010)内,表明编码后信号的诊断信息得以有效保留。然而,该结果未作为确证性结论,因测试集曾参与编码器开发。综上,该方法实现了高保真度的ECG文本压缩,为生成式AI处理临床心电图数据提供了可行路径。
链接: https://arxiv.org/abs/2611.22827
作者: Elif Dorukbaşı,İlker Türker
机构: Karabük University (卡拉布克大学)
类目: Computation and Language (cs.CL)
备注: 9 pages, 5 figures, 5 tables. Preprint of a chapter submitted to “Advances in Artificial Intelligence and Data Science: Methods, Systems, and Applications” (Karabük University Press). Code: this https URL
Abstract:When a twelve-lead electrocardiogram is written as text for a large language model, the number of tokens has to be balanced against the information lost through compression. This paper presents a deterministic encoding that writes eight independent leads as amplitude letters and exact time gaps and recovers the remaining four limb leads algebraically. The key points are selected with a cardiac-cycle-dependent tolerance, the amplitudes are parameterized by a 26-letter logarithmic codebook learned from a training set, and a 10-step tolerance schedule keeps all records within a fixed token budget. The encoding was applied to 21,388 records from the public PTB-XL database and evaluated using an exact decode-and-re-encode round trip, waveform reconstruction error, and a paired comparison of convolutional classifiers trained with 5 different seeds on signals from the original and decoded databases. All records were contained in 4,080 tokens (average: 2,866 tokens, 64.3% fewer than a per-sample transcription), and all stored lead strings were exactly reproduced. The average correlation between the decoded and reference signals was 0.982. The macro-averaged area under the receiver operating characteristic curve was 0.9167 for the reference signals and 0.9157 for the decoded signals (paired difference -0.0010; 95% patient-level bootstrap interval, -0.0028 to 0.0007), within the pre-specified non-inferiority margin of -0.010; this is not a confirmatory result, because the test fold had been used during encoder development. The diagnostic content of a compact text can therefore be preserved, while the language model’s ability to read the text is yet another matter.
[NLP-24] A Full-Ledger Counterfactual Audit of Adaptive Reasoning EACL2027
【速读】: 该论文旨在解决生成式 AI 在推理过程中如何动态决定是否继续计算以提升答案准确性的问题,即“自适应推理”(adaptive reasoning)中的计算资源分配难题。现有方法难以判断当前答案错误时,进一步推理是否能改善结果,导致计算浪费或过早终止。其核心解决方案是提出全账本反事实价值审计(Full-Ledger Counterfactual Value Audit, FLCVA),通过在相同冻结前缀下对比即时答案与四种固定长度的延续(continuation)表现,并精确核算策略所使用每一信号的成本,实现对局部延续价值的客观评估。实验表明,在GSM8K和MATH-500数据集上,随机256词元延续可分别在26.6%(Qwen3-8B)和18.9%(DeepSeek-R1-Distill-Qwen-7B)的错误答案中恢复正确解,而仅对1.4%和1.9%的正确答案造成损害;独立采样的延续正确性标签能够有效预测后续延续的表现。在离线回放场景中,基于这些标签进行计算资源分配可提升DeepSeek模型性能,而Qwen3的增益尚不显著。成本核算显示,当前轻量级、短视型控制器主要依赖于识别当前错误,难以提供超越该捷径之外的延续特异性价值。通过Llama-8B复现及动作/成本消融实验,验证了结论的稳健性。FLCVA 的关键在于将“局部延续价值测量”、“基于可观测信号的预测”以及“计算分配决策”三者分离,从而为自适应推理提供可解释、可优化的框架。
链接: https://arxiv.org/abs/2611.22778
作者: Xiaopeng Chu,Wenzheng Kong,Wei Liu,Wenyi Zhang
机构: University of Science and Technology of China(中国科学技术大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: Accepted to EACL 2027 (Main Conference)
Abstract:Adaptive reasoning must decide whether to answer now or spend more compute. Predicting that the current answer is wrong does not establish whether further reasoning will improve it. We introduce the Full-Ledger Counterfactual Value Audit (FLCVA), which compares an immediate answer with four fixed-length continuations from the same frozen prefix under one answer procedure and accounts for the cost of every signal used by a policy. On GSM8K and MATH-500, a random 256-token continuation recovers a wrong answer in 26.6% of Qwen3-8B cases and 18.9% of DeepSeek-R1-Distill-Qwen-7B cases, while damaging a correct answer in 1.4% and 1.9%. Correctness labels from independently sampled continuations predict a separate continuation’s outcome in both primary models. In offline replay, using these labels to allocate compute improves performance on DeepSeek; the Qwen3 gain remains inconclusive. Under full cost accounting, the tested lightweight, myopic controllers mainly recognize current errors and add little continuation-specific value beyond that shortcut. A Llama-8B replication and action/cost ablations examine the robustness of these findings. FLCVA separates measuring local continuation value, predicting it from observable signals, and using those predictions to allocate computation.
[NLP-25] CrossEnv: Synthesizing Verifiable Cross-Environment Tasks for Long-Horizon Agent Training
【速读】: 该论文旨在解决多环境间复杂交互任务难以有效合成的问题,特别是在长时程智能体(long-horizon agents)训练中,现有环境合成方法局限于单一环境内的任务生成,无法充分模拟真实世界服务间依赖关系——即一个服务中的操作往往依赖于另一服务提供的信息。其核心挑战在于跨环境数据的一致性保障,以确保合成的任务具备可执行性和可验证性。本文提出CrossEnv框架,通过构建实体-工具图(entity-tool graph)识别不同环境间的关联关系,并对任务特定数据进行对齐,从而实现可执行、可验证的跨环境任务合成。进一步地,引入CrossEnv-Curriculum训练范式,采用从单环境工具使用逐步过渡到多环境、多业务领域协作,最后引入用户交互的渐进式训练策略。在四个外部基准上的实验表明,结合跨环境训练数据与该课程化训练策略,在减少训练样本数量的同时仍能取得具有竞争力的性能表现;分析结果还验证了跨环境数据与课程化训练的有效性,展示了模型在新环境组合、与训练数据差异较大的外部任务以及更长工作流上的良好迁移能力。
链接: https://arxiv.org/abs/2611.22777
作者: Zhengfeng Yu,Tian Lan,Zhijing Wu,Enbo Zhao,Xuannan Liu,Feihu Jiang,Qianghuai Jia,Longyue Wang,Zhao Xu,Weihua Luo,Xian-Ling Mao
机构: Beijing Institute of Technology(北京理工大学); Alibaba Group(阿里巴巴集团); Beijing University of Posts and Telecommunications(北京邮电大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 23 pages
Abstract:Executable synthetic environments provide scalable training tasks and verifiable feedback for long-horizon agents. Existing environment synthesis methods primarily generate tasks within individual environments. However, these tasks do not adequately capture real-world interactions across services, where actions in one service depend on information obtained from another. Synthesizing such tasks is challenging because data generated across environments must remain consistent to support executable and verifiable interactions. To address this challenge, we introduce CrossEnv, a framework that uses an entity-tool graph to identify connections between environments and aligns task-specific data to synthesize executable, verifiable cross-environment tasks. We further propose CrossEnv-Curriculum, which progresses from single-environment tool use to coordination across more environments and business domains, before introducing user interaction. Across four external benchmarks, combining cross-environment training data with this curriculum achieves competitive performance with fewer training examples. Further analyses support the benefits of cross-environment data and curriculum training, and provide evidence of transfer to new environment combinations, external tasks less similar to the training data, and longer workflows.
[NLP-26] StitchRAG : Constructing Compact and Evidence-Preserving Contexts for RAG with Diffusion-Based Stitching
【速读】: 该论文旨在解决检索增强生成(Retrieval-augmented generation, RAG)中因长文本检索上下文导致的计算开销大以及噪声或无关内容影响模型性能的问题。其核心解决方案是提出StitchRAG框架,通过动态压缩检索片段,在保持信息密度与忠实性的前提下构建紧凑且连贯的上下文。该方法的关键在于:基于效用最大化原则,为每个检索片段自适应分配压缩比例,权衡保留信息量与计算成本;随后提取答案候选和查询相关关键词作为锚点,筛选包含锚点的句子,并在约束条件下利用文本扩散(text diffusion)技术将这些句子连接成语义连贯、自然流畅的上下文。实验结果表明,StitchRAG在五个开放域问答基准上均显著优于多种多步推理与压缩基线,消融实验进一步验证了动态尺寸分配、锚点保留及扩散式拼接策略的有效性。
链接: https://arxiv.org/abs/2611.22766
作者: Minsang Kim,Junpyo Lim,Heeje Cho,Seungjun Baek
机构: Korea University(韩国大学); SK Telecom(韩国电信)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Retrieval-augmented generation (RAG) improves factuality by grounding large language models (LLMs) in external evidence. However, long retrieved contexts incur computational overhead and can even hurt performance due to noisy or irrelevant content. We propose \textscStitchRAG, a framework for constructing compact context in RAG. \textscStitchRAG builds a faithful, information-dense context while dynamically compressing the retrieved passages. We adapt the size reduction ratio to each passage by formulating a utility maximization trading-off the retained information and computation costs. We then extract answer candidates and query-related keywords from passages as anchors and collect passage sentences containing anchors under the derived size constraint. Finally, we connect or stitch the collected sentences together using text diffusion to form a coherent and natural context. Experiments on five open-domain QA benchmarks show that \textscStitchRAG consistently outperforms diverse multi-step and compression baselines. Extensive ablations show the effectiveness of dynamic size allocation, anchor preservation, and diffusion-based stitching.
[NLP-27] Cross-cultural Adaptation of Recipes for Low-Resource Languages: A Benchmark and Multilingual Culinary Lexicon for Hausa Yorùbá and Igbo
【速读】: 该论文旨在解决低资源语言(Low-Resource Languages, LRLs)在菜谱翻译中缺乏文化适应性的问题,尤其针对尼日利亚三大主要语言——约鲁巴语、豪萨语和伊博语的菜谱翻译。传统机器翻译方法难以处理菜谱中涉及的食材、厨具、计量单位及烹饪步骤等需跨文化适配的内容。为此,作者提出ReciLoReL基准数据集与ReciLex多语言烹饪术语词典,分别包含600条专家翻译并经文化适配的菜谱及约1.5万个涵盖十类实体的多语言翻译词条。解决方案的关键在于利用ReciLex作为外部知识增强模块,在无需大规模目标语言菜谱语料进行训练或微调的前提下,提升生成式AI在零样本与少样本场景下的文化适应性翻译能力。实验表明,引入ReciLex后,模型在标准指标(如BLEU、chrF、TER)上均有显著提升,并在五项基于跨文化适应方法构建的文化适应性评估维度上获得0.11至0.36的人工评分增益,验证了该方法在低资源环境下实现高质量、文化敏感性菜谱翻译的有效性。
链接: https://arxiv.org/abs/2611.22708
作者: Nuhu Ibrahim,Maryam Abdulazeez,Fatsuma Jauro,Salisu Aliyu,Robert Stevens,Riza Batista-Navarro
机构: The University of Manchester (曼彻斯特大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 9 pages
Abstract:Recent multilingual large language models (LLMs) have improved machine translation, yet culturally adaptive translation for low-resource languages (LRLs) remains underexplored. This is particularly challenging for recipes, where ingredients, cooking tools, measurements, and procedures may require adaptation to the target cultural context. In this work, we introduce ReciLoReL, a benchmark of 600 expert-translated and culturally adapted recipes across Yorùbá, Hausa, and Igbo (three of the most prominent languages in Nigeria), and ReciLex, a multilingual culinary lexicon containing approximately 15,000 translated entities across ten recipe entity categories, including ingredients, tools, chef actions, state of tools, and durations. We evaluate state-of-the-art multilingual LLMs for English-to-LRL recipe translation under zero-shot and few-shot settings and investigate whether augmenting LLMs with ReciLex improves culturally adaptive recipe translation without requiring large target-language recipe corpora for training or fine-tuning. Beyond conventional reference-based translation evaluation using BLEU, chrF, and TER, we introduce five criteria for evaluating cultural adaptation, adapted from established cross-cultural adaptation methods. Across the selected LLMs and languages, ReciLex improves average BLEU from 13.57 to 14.51 and chrF from 37.89 to 38.74, while reducing TER from 77.14 to 75.12. Human evaluation further shows improvements ranging from 0.11 to 0.36 points across the five cultural adaptation evaluation criteria on a five-point scale. Overall, our work provides new resources, evaluation criteria, and empirical evaluation for studying culturally adaptive recipe translation in low-resource settings.
[NLP-28] Route Optimization under Port Congestion and Carbon Reduction: An NSGA-II Approach
【速读】: 该论文旨在解决海运业在港口拥堵加剧与碳减排要求日益严格双重压力下的运营优化问题,核心挑战在于如何在保障运输效率的同时降低碳排放。其解决方案的关键在于构建一个融合时间成本、运输成本与碳排放成本的多目标路径优化框架,基于依赖于拥堵程度的广义成本函数,采用变分不等式(Variational Inequality, VI)模型刻画船舶航线分配的网络均衡状态,并利用非支配排序遗传算法II(NSGA-II)求解三目标优化问题。通过对中国沿海集装箱网络(以上海为拥堵主港、宁波舟山港为替代港)的实证分析,研究揭示了替代港口分流的临界阈值约为60个船次,碳税达80美元/吨二氧化碳时可推动清洁燃料使用率超过50%,且燃料价格主要影响能源类型选择而非港口分流决策,表明港口选择与能源结构决策具有独立性。该研究为港口管理机构、监管者及航运企业提供了兼顾经济性与可持续性的协同决策支持。
链接: https://arxiv.org/abs/2611.22705
作者: Guanxi Liang,Sanglin Zhao
机构: 未知
类目: Computation and Language (cs.CL)
备注: 25 pages,7 figures
Abstract:Maritime shipping simultaneously faces the dual challenges of escalating port congestion and increasingly stringent decarbonization mandates. Large-scale vessel anchorage at hub ports inflates unit transport costs while generating significant additional (CO_2) emissions from idling auxiliary engines. This study develops a multi-objective routing optimization framework that jointly incorporates time, transport, and carbon emission costs under congestion-dependent generalized cost functions. A variational inequality (VI) model is formulated to characterize network equilibrium in vessel route allocation, while the Non-dominated Sorting Genetic Algorithm II (NSGA-II) is deployed to solve the resulting three-objective optimization problem. The model is empirically validated using China’s coastal container network (Shanghai as the primary congested port, Ningbo-Zhoushan as the backup alternative), with sensitivity analyses conducted across congestion levels, carbon tax rates, and fuel price differentials. Key findings indicate that: (1) alternative ports effectively hedge congestion when diversion costs fall below waiting costs, with a diversion threshold at approximately 60 vessel calls; (2) a carbon tax of 80/t (CO_2) pushes clean fuel adoption above 50%; and (3) fuel prices primarily influence energy-type selection rather than port diversion decisions, confirming that the two decision dimensions operate independently. These findings offer actionable decision support for port authorities, regulators, and shipping operators seeking to balance economic efficiency with environmental sustainability.
[NLP-29] Visibility Graph Representations of Word Embeddings for Text Classification
【速读】: 该论文旨在解决在受限词汇量预算下,文本分类模型如何有效利用词向量表示以提升分类性能的问题。其核心挑战在于:当词汇表规模受限时,不同表示方法的有效性会发生变化,传统基于词向量平均的固定长度文档表示可能无法充分捕捉类别间的区分特征。为此,论文提出将自然可见性图(natural visibility graph)作为中间表示,通过将词向量平均得到的文档向量转化为有序数值序列,并构建其对应的自然可见性图,进而以节点度序列或邻接矩阵的形式输入分类器。该方案的关键创新在于:利用可见性图对文档向量的结构信息进行编码,从而在有限词汇量条件下保留更具判别性的语义结构特征。实验结果表明,在AG News四分类基准上,当词汇量达到700个以上时,基于节点度的图表示性能超越多种主流嵌入方法(如BERT),在1,000词时达到91.24%准确率,优于BERT基线0.29个百分点;而在卷积分类器中,邻接矩阵表示也表现出显著优势。研究进一步揭示了图压缩、坐标排序及分类器选择对性能的影响,并强调需严格控制实验条件以确保微小差异的可靠性。
链接: https://arxiv.org/abs/2611.22685
作者: Elif Dorukbaşı,İlker Türker
机构: Karabük University (卡拉布克大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Text classifiers depend on representations that expose the distinctions between classes in a form the learning model can use, and a restricted vocabulary changes which representation is most effective. This paper studies natural visibility graphs as an intermediate representation for news topic classification under controlled vocabulary budgets. Word embeddings are averaged into a fixed-length document vector, the vector is read as an ordered numerical series, and its natural visibility graph is passed to a classifier either as an ordered node-degree vector or as an adjacency matrix. On the four-class AG News benchmark, chi-square selection retains 100 to 1,500 words, and the graph inputs are compared with direct embedding-based representations from continuous bag-of-words, skip-gram, GloVe, FastText and BERT. With a fully connected classifier, the degree representation is the most accurate configuration from 700 selected words onward; it reaches its highest accuracy of 91.24% at 1,000 words and exceeds the BERT-based baseline by 0.29 percentage points at 1,500 words. In a separate convolutional comparison, the adjacency input reaches 86.16% at 1,000 words, against 83.12% for the strongest static embedding. The results indicate that visibility-derived features can be competitive once the selected vocabulary provides adequate coverage. We discuss the roles of graph compression, coordinate ordering and classifier choice, and the controls needed to establish the reliability of the small differences observed.
[NLP-30] Contract-Explicit Evaluation: From Auditing Learned Readers to Scoped Certificates
【速读】: 该论文旨在解决文档级关系抽取(DocRE)系统评估结果在不同评估协议(evaluation contract)下排名反转的问题,即同一模型在不同评分函数、阈值或黄金标准版本下的性能排序可能不一致,而现有方法无法确定哪些比较结果具有稳健性。其解决方案的关键在于提出一种“协议显式评估”(contract-explicit evaluation)框架,通过一个预先注册的五步协议(声明、测量、范围界定、认证、预警),将肯德尔等级相关系数的经典一致性分解(concordance decomposition)具体化为可操作的评估流程。该框架能够明确界定可认证的评估区域,并在两个生态系统中提供无分布假设的普查证书(census certificates)。在对44次训练运行的DocRE审计中,评分维度生成了包含3个异常实例的普查证书(Clopper–Pearson上界为0.146);确认性记录包含注册的负样本,且审计自身发现简化策略的有效性——校准旋钮的全部置信区间仅由一次提交撤销所致。在基于大语言模型(LLM)裁判的普查实验(17名裁判,49,938场对决)中,12.25%的排行榜比较需在预声明的分割标准下进行预警(边界值0.124;面板鲁棒性边界0.143),探索性双层读取进一步证实69%的比较可在≤0.072的边界内被认证,且对已发布人工标签的效度分解显示,争议对决中有65%源于裁判错误而非标注模糊。跨模型证书因真实置信分数而被撤回,失败谱图则刻画了边界条件。研究公开了协议、地图、账本及每实例的分析产物。
链接: https://arxiv.org/abs/2611.22665
作者: Hao Lu,Weixing Zhu,Ruiyang Li
机构: Army Engineering University of PLA (中国人民解放军陆军工程大学)
类目: Computation and Language (cs.CL)
备注: 50 pages, 4 figures. Supplementary material (Online Resource 1, 14 pages) included as ancillary file. Artifact repository: this https URL . Submitted to Applied Intelligence
Abstract:Evaluation readings of document-level relation extraction (DocRE) systems depend jointly on the model and the evaluation contract: gold version, scoring function, threshold. Rankings reverse across contracts, yet no existing method determines which comparisons survive the reversal. We address this gap with contract-explicit evaluation: a preregistered five-step protocol (Declare, Measure, Scope, Certify, Hedge), operationalizing the classical concordance decomposition of Kendall’s rank correlation, that scopes certifiable regions and delivers distribution-free census certificates across two ecosystems. In the DocRE audit (44 training runs of one in-house architecture family), the scoring dimension scopes to a census certificate (three flagged instances, Clopper–Pearson upper bound 0.146); the confirmatory record includes registered negatives, and the audit prices its own simplification finding—the calibration dial’s entire margin is one withdrawn commit. In an LLM-judge census (17 judges, 49,938 battles), 12.25% of leaderboard comparisons require hedging at the pre-declared split criterion (bound 0.124; panel-robust 0.143), with an exploratory two-tier reading certifying 69% at a bound \le 0.072 and a validity decomposition against released human labels (65% of contested battles are judge error, not ambiguity). The cross-model certificate is withdrawn on real confidence scores; the failure spectrum maps the boundary. The protocol, map, ledger, and per-instance artifacts are released.
[NLP-31] he Between-Round Decision in Autonomous Research: An LLM Controller That Reads the Whole Trajectory
【速读】: 该论文旨在解决自主科研系统中“轮间决策”(between-round decision)这一核心问题,即在每轮实验后,如何基于已有证据决定是优化当前最优解(Champion)还是探索新方向、选择何种方案进行迭代以及设定何种目标。现有方法依赖规则、评分机制、搜索树或调度策略等外部结构来支撑决策,而大语言模型(LLM)在其中所需读取的信息内容尚不明确。本文提出BRED(Between-Round Evidence-based Decision),一种基于LLM的控制器,其关键在于不依赖预设的搜索结构,而是通过维护已验证的最佳解和历史实验日志,仅根据累积证据进行轮次级决策。实验表明,当控制器不读取历史证据时性能显著下降,证明了充分利用证据的重要性;在三个主流研究代理基准任务上,BRED相较基线分别在ETTh1上提升4.80% MAE(达0.4172)、在SST-2上提升3.44个百分点准确率(达95.07%)、在CIFAR-100上提升3.43个百分点(达84.63%),并优于已发表结果。在新的无泄露语音抑郁检测任务中,其宏平均F1达到0.6164,较基线高出5.5点,超越最强公开系统4.0点。这表明,基于证据驱动的轮间决策是提升自主科研系统效能的关键,且有效利用历史信息对性能至关重要。
链接: https://arxiv.org/abs/2611.22661
作者: Caoxin Wang,Shengkai Gan,Xiang Hu,Lili Pan
机构: University of Electronic Science and Technology of China(电子科技大学); Zhejiang University(浙江大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 16 pages, 2 figures
Abstract:Executing an idea often changes how promising it looks, so the next step in autonomous research should rest on results, not on proposals. After each round of experiments, the system must decide whether to refine the best solution or start a new direction, which solution to build from, and what goal to set. We call this the between-round decision. Existing systems wrap it in machinery such as rules, scores, search trees, and schedules; where an LLM makes it, what the LLM needs to read is left open. We ask whether an LLM can make this decision from accumulated evidence, without a search structure around it, and how much of that evidence it needs to read. We present BRED (Between-Round Evidence-based Decision), an LLM-based controller that maintains the best verified solution so far (the Champion) and a log of past rounds, and makes one decision per round. On three widely used research-agent benchmark tasks, BRED improves on the task baselines by 4.80% in MAE on ETTh1 (to 0.4172) and by 3.44 and 3.43 points of accuracy on SST-2 (to 95.07%) and CIFAR-100 (to 84.63%), ahead of the published results. When the controller does not read the accumulated evidence, the same loop gains less: with no history it reaches 0.4299 on ETTh1 and 83.71% on CIFAR-100, whereas the full controller’s MAE is 2.95% lower and its accuracy 0.92 points higher. On a new task, speech-based depression detection under a leakage-free protocol, it reaches a macro-F1 of 0.6164 on seeds held out from selection, 5.5 points above its baseline and 4.0 above the strongest published system. Code is available at this https URL.
[NLP-32] Understanding Cross-Modal Inconsistency through Representation Subspaces in Omni Models EACL2027
【速读】: 该论文旨在解决多模态统一模型(Omni Models)在处理跨模态语义等价输入时出现的答案不一致问题。现有研究在表示层面的分析无法区分两类关键信息:一是预测跨模态不一致发生的先验信号,二是实际影响模型输出的可扰动表示。本文提出通过从答案位置表示中估计一个低维答案子空间(Answer Subspace),将与答案相关的信息与其他表示分离开来。实验结果表明,预测跨模态不一致的信息主要集中于答案子空间之外,且集中在可复现的二维表示中,该表示并非由固定的模态偏移所解释。匹配干预实验揭示了两类表示的不同行为角色:扰动答案子空间会显著改变模型预测,而扰动不一致表示(如同维度的随机子空间)则几乎不影响输出。核心发现是,跨模态不一致可通过错位对齐模态特异性的答案表示来诱发——仅对单一模态的答案子空间施加扰动即可引发不一致,而对所有模态同步施加相同扰动则能维持一致性。这表明,能够线性解码的表示并不等同于具有行为有效性的表示,即预测不一致的表示与可诱导不一致的表示本质不同。
链接: https://arxiv.org/abs/2611.22619
作者: Seonghyeon Cho,Chanjun Park
机构: Soongsil University (松林大学)
类目: Computation and Language (cs.CL)
备注: Accepted for EACL 2027 Main
Abstract:Omni Models process audio, vision, and text within a unified language backbone, yet often produce different answers for semantically equivalent inputs across modalities. Existing representation-level analyses do not distinguish between information that predicts whether disagreement will occur and the representation whose perturbation affects the model’s answer. In this paper, we estimate a low-dimensional Answer Subspace from answer-position representations and use it to separate answer-related components from the remaining representation. Across three Omni Models and two multimodal reasoning tasks, information predictive of cross-modal inconsistency lies mainly outside the Answer Subspace, concentrating in a reproducible two-dimensional representation that is not explained by a fixed modality offset. Matched interventions reveal distinct behavioral roles: perturbing the Answer Subspace substantially changes the model’s predictions, whereas perturbing the disagreement representation, like a random subspace of the same dimensionality, leaves them almost unchanged. Our key finding is that cross-modal disagreement can be induced by misaligning modality-specific answer representations: perturbing the Answer Subspace for only one modality induces disagreement, whereas applying the same perturbation to both largely preserves agreement. Thus, the representation that predicts disagreement is distinct from the answer-level representation through which disagreement can be induced, showing that linear decodability does not imply behavioral efficacy.
[NLP-33] Evolution Under Fire: Evolving Agent Safety Harnesses with Adaptive Injection Stressors
【速读】: 该论文旨在解决生成式 AI(Generative AI)代理在使用工具时面临的核心安全挑战:恶意指令与合法任务指令通过同一交互通道(如电子邮件、网页和文档)传递,导致安全风险难以区分。现有防御机制多为手工设计且静态,依赖人工反复调整,难以适应动态攻击模式。其解决方案的关键在于提出“战火中的演化”(Evolution Under Fire)框架,该框架通过两阶段协同演化机制实现自适应安全防护:第一阶段采用受效用约束、帕累托调度的树搜索算法,基于执行证据动态演化出完整的可执行安全防护框架(safety harness),在保持任务模型、基准测试、工具和评估器不变的前提下优化安全性与任务性能的平衡;第二阶段引入自适应压力源演化(adaptive stressor evolution),主动发现保留任务效用的攻击样本,用于引导下一阶段的安全防护框架迭代优化。实验表明,该方法在AgentDyn(GPT-4o-mini)上将攻击成功率从50.4%降至3.0%,并提升清洁与受攻击场景下的任务效用;在Agent-SafetyBench(Qwen2.5-7B-Instruct)上将攻击成功率从82.6%降至62.0%的同时维持任务有用性和运行时有效性;且演化后的安全框架可直接迁移至Qwen3.5-35B-A3B模型,在未再演化的前提下将攻击成功率从32.1%降至0.9%,同时提升受攻击任务的效用表现,显著优于复现的现有防御方案。
链接: https://arxiv.org/abs/2611.22583
作者: Yujun Wang,Yilun Liu,Jinhe Bi,Krishna Kanth Nakka,Xuebing Zhou,Hinrich Sch{ü}tze,Yunpu Ma
机构: LMU Munich (慕尼黑大学); Munich Center for Machine Learning (慕尼黑机器学习中心)
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Tool-using language-model agents must act on information from untrusted emails, webpages, and documents. This creates a central security challenge: malicious instructions and legitimate task directives can arrive through the same interaction channel. Existing defenses are largely hand-designed and static, making further improvements dependent on repeated manual redesign. We introduce Evolution Under Fire, a staged framework that couples safety-harness evolution with adaptive stressor evolution. Utility-constrained, Pareto-scheduled tree search evolves complete executable safety harnesses from execution evidence while keeping the task model, benchmark, tools, and evaluator fixed. Adaptive stressor evolution then discovers utility-preserving attacks that guide a subsequent stage of harness evolution. On AgentDyn with GPT-4o-mini, an evolved harness reduces attack success from 50.4% to 3.0%, improves clean and attacked utility over the initial harness, and offers a stronger security-utility trade-off than reproduced defenses. On Agent-SafetyBench with Qwen2.5-7B-Instruct, harness evolution reduces attack success from 82.6% to 62.0% while retaining helpfulness and runtime validity. Evolved harnesses also transfer without further evolution to Qwen3.5-35B-A3B, reducing attack success from 32.1% to 0.9% while improving attacked-task utility.
[NLP-34] AfriNORM: An Evaluation Benchmark for Normative Reasoning of LLM s Across African Cultures
【速读】: 该论文旨在解决当前大型语言模型(LLMs)评估基准在非洲本地化应用场景中存在显著不足的问题,特别是现有基准多聚焦于语言流利性(如语法、翻译),而缺乏对文化适应能力的衡量,尤其忽视了非洲社会规范中的义务模态(deontic modality),即支配多元社会互动的严格准则与禁忌。为此,论文提出AfriNORM——一个基于情景的多项选择题数据集,专门用于评估和诊断LLMs在非洲社会文化语境下的规范推理能力局限。其核心解决方案在于引入两项新指标:通用常识偏差得分(Universal Commonsense Bias Score, UCBS)用于量化模型在常识理解上的普遍偏差,以及干扰项偏差得分(Distractor Bias Score, DBS)用于量化模型对非洲文化规范理解的不足。通过对约鲁巴语(Yorùbá)的案例研究发现,尽管“Gemini 2.0 Flash”、“GPT-5”和“Llama 3.1 8B Instruct”在英文提示下表现良好,但在使用约鲁巴语提示时准确率显著下降,且错误模式并非源于简单偏见,而是由文化语义混淆所致,表明这些模型在跨文化规范推理中存在深层语义理解缺陷。
链接: https://arxiv.org/abs/2611.22580
作者: Ahmad Samuel Gali(1),Tadesse Destaw Belay(2),Idris Abdulmumin(3),Seid Muhie Yimam(4),Shamsuddeen Hassan Muhammad(5 and 6) ((1) Independent Researcher, (2) Instituto Politécnico Nacional, (3) University of Pretoria, (4) University of Hamburg, (5) Bayero University Kano, (6) Imperial College London)
机构: Independent Researcher(独立研究员); Instituto Politécnico Nacional(墨西哥国立理工学院); University of Pretoria(普利托利亚大学); University of Hamburg(汉堡大学); Bayero University Kano(贝亚罗大学卡诺分校); Imperial College London(伦敦帝国学院)
类目: Computation and Language (cs.CL)
备注: 8 pages, 3 figures, 5 tables
Abstract:Large language models (LLMs) are increasingly deployed across various sectors. However, current evaluation benchmarks are insufficient to assess the applicability of these models in local African contexts, as most focus on linguistic fluency (grammar/translation) rather than cultural competence. The few existing benchmarks that measure general knowledge or social conventions do not capture the African deontic modality, i.e., the strict norms and taboos that govern diverse social interactions. In this paper, we introduce AfriNORM, a scenario-based, multiple-choice question dataset designed to measure and diagnose the normative reasoning limitations of LLMs when applied to African sociocultural contexts. We introduce two metrics: a Universal Commonsense Bias Score (UCBS) to quantify commonsense bias in LLM and a Distractor Bias Score (DBS) to quantify an LLM’s limited understanding of African cultural norms. We also conduct a case study on the Yorùbá language, identifying confusion gaps in two closed-source LLMs, “Gemini 2.0 Flash” and “GPT-5”, and one open-source LLM, “Llama 3.1 8B Instruct” when prompted in English versus Yorùbá. While their accuracy in English is high, their performance in Yorùbá is much lower, attributable to cultural confusion rather than simple bias, indicating that the models exhibit semantic confusion when reasoning about norms in Yorùbá.
[NLP-35] From Perception to Simulation: Unveiling and Empowering Mental Spatial Dynamics in MLLM s EMNLP2026
【速读】: 该论文旨在解决多模态大语言模型(Multimodal Large Language Models, MLLMs)在动态空间模拟任务中能力不足的问题,尤其针对缺乏中间视觉反馈情境下对空间状态演化进行推理的挑战。现有研究虽在静态场景的空间关系感知方面取得进展,但在动态空间演化建模方面仍存在显著空白。为应对这一问题,论文提出MentalGrid基准,专门评估模型在无视觉反馈条件下对空间状态演变的模拟能力。实验结果表明,当前主流模型在简单场景中受限于感知能力,在复杂多步推理任务中则因缺乏连贯的逻辑推演而表现不佳。为此,作者设计了一种三阶段课程学习框架,基于16,000个专业化的空间推理样本进行训练,从而构建出SpatialWalker模型。该方法在基准测试中相较基线模型平均提升33.8%,超越了如Gemini-3-Flash等闭源模型的表现;同时在跨域任务上仍保持4.9%的平均优势,验证了空间模拟作为构建更广泛空间智能基础的关键作用。
链接: https://arxiv.org/abs/2611.22541
作者: Jiachun Li,Zhuoran Jin,Kang Liu,Jun Zhao
机构: University of Chinese Academy of Sciences (中国科学院大学); Institute of Automation, Chinese Academy of Sciences (中国科学院自动化研究所)
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: Findings of EMNLP 2026
Abstract:Spatial reasoning is a fundamental component of world understanding for Multimodal Large Language Models (MLLMs). While existing research has advanced the perception of spatial relations in static scenes, the domain of dynamic spatial simulation remains relatively unexplored. To bridge this gap, we introduce MentalGrid, a benchmark focused on spatial reasoning tasks that require models to simulate evolving spatial states without intermediate visual feedback. Extensive evaluation on our benchmark reveals a significant deficiency in the spatial simulation capabilities of existing models. Specifically, we find that performance is constrained by perceptual limitations in simple simulation scenarios and hindered by inadequate multi-step reasoning as complexity increases. To address these challenges, we propose a three-stage curriculum learning framework utilizing 16k specialized spatial reasoning samples. This approach yields SpatialWalker, which achieves a 33.8% average improvement over its base model, surpassing proprietary models like Gemini-3-Flash in spatial simulation. Furthermore, SpatialWalker demonstrates robust generalization, maintaining a 4.9% average gain on out-of-domain benchmarks, demonstrating the pivotal role of spatial simulation as a fundamental building block for broader spatial intelligence.
[NLP-36] JET: Just-Enough Exploration Tuning for Efficient Tool-Calling Agents in Repository-Level Question Answering
【速读】: 该论文旨在解决代码仓库级问答(repository-level question answering, QA)中大语言模型(LLM)代理存在的“过度探索”(over-exploration)问题,即在已获取足够信息回答问题后仍持续调用工具。现有方法通过专家轨迹的监督微调(Supervised Fine-Tuning, SFT)虽提升了任务性能,但同时也将过度探索行为引入模型。其解决方案的关键是提出一种名为“恰到好处探索微调”(Just-Enough Exploration Tuning, JET)的新方法:JET 在训练过程中将每条轨迹截断于信息充分可作答的时刻,插入解释为何此时可作答的理由(rationale),并以更高的损失权重对理由部分进行微调,从而引导代理在信息充足时及时终止探索。实验表明,JET 在自生成轨迹上减少11%至33%的工具调用,同时保持高得分;在专家轨迹上相比SFT减少18%至28%的工具调用,且几乎完全保留SFT带来的性能增益,显著降低推理成本。
链接: https://arxiv.org/abs/2611.22539
作者: Yoonho Choi,Taehyeon Han,Seonghyeon Lee
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: Preprint. 13 pages, 8 figures, 5 tables
Abstract:In repository-level question answering (QA), an agent explores a code repository with tools and should answer a question once enough information is gathered. Nevertheless, we find that LLM agents often continue to invoke tools even after the gathered information is already sufficient to answer. We call this phenomenon over-exploration. Moreover, even though existing methods improve task performance through supervised fine-tuning (SFT) on expert trajectories, SFT also propagates the over-exploration in the trajectories to the agent. To mitigate over-exploration, we propose Just-Enough Exploration Tuning (JET), a fine-tuning method that trains the agent to answer once the gathered information is sufficient. JET truncates each training trajectory at the step where the gathered information becomes sufficient, inserts a rationale that explains why the agent can answer at that step, and fine-tunes the agent with a higher loss weight on the rationale. Experiments on two repository-level QA benchmarks with two open-weight 8B models demonstrate that JET reduces tool calls by 11 to 33% on self-generated trajectories while largely preserving the score. On expert trajectories, JET makes 18 to 28% fewer tool calls than SFT while retaining most of the score gain of SFT. In both cases, JET also lowers the inference cost.
[NLP-37] When Agent Skills Go Stale: Runtime Behavior and Selective Maintenance in Evolving Environments
【速读】: 该论文旨在解决在环境发生变化时,智能体(agent)所依赖的技能状态(skill state)因部分信息过时而导致决策性能下降的问题。当环境动态变化时,原有技能状态中仍保有部分有效信息,但若不加区分地复用整个状态,可能引入错误或冗余,从而降低任务成功率或增加执行成本。其核心解决方案是提出“选择性维护”(selective maintenance)机制:该机制通过回放成功执行中的检查点,仅基于当前观测判断是否存在与技能状态冲突的差异,对受影响内容进行局部修复,并验证更新有效性;当局部修复不足时则重建状态。该方法无需外部环境变化描述,仅依赖内部观测即可实现高效状态维护。实验表明,相比直接复用过时状态,选择性维护可将任务准确率提升高达67个百分点,执行步数减少最多达37%,显著提升了技能状态在动态环境中的持续可用性与推理效率。
链接: https://arxiv.org/abs/2611.22525
作者: Ömer Faruk Akgül,Shuli Jiang,Yuting Zhang,Yi Zhang,Xiaohan Fei,Wei Xia,Stefano Soatto
机构: AWS Agentic AI; University of Southern California (南加州大学)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Agent skills use past experience and current observations to guide decisions and actions on related tasks. This experience is retained in skill state, whose reuse can amortize inference-time reasoning and environment interaction. When environment changes, parts of this state can become outdated while others remain useful. Across four task settings and two models, we compare skill reuse before and after environment changes. Previously beneficial skills can reduce task performance by over 10 percentage points or increase execution cost by nearly 40% relative to no skill. Task success can also remain unchanged while execution savings decline. These effects depend on which environment-specific details become outdated and the agent’s ability to compensate using current observations. We introduce selective maintenance, which replays checks from successful executions without additional model calls. It verifies whether observed differences conflict with skill state, repairs affected content, and validates updates, rebuilding when local repair is insufficient. No description of the change is supplied. Relative to reusing stale state, maintenance improves task accuracy by up to 67pp and reduces execution steps by up to 37%. These results suggest that preserving a skill’s inference time value as environments evolve requires selectively maintaining its state.
[NLP-38] MM-ReportBench: A Comprehensive Benchmark for Multimodal Report Generation from Tables
【速读】: 该论文旨在解决现有视觉-语言模型在生成融合流畅文本叙述与忠实可视化结果的多模态报告(multimodal reports)时,评估基准存在单模态局限、度量方式简单、新颖性评估不足以及可视化质量评价缺失等问题。其核心解决方案是提出MM-ReportBench,一个覆盖185个真实世界表格、包含386项任务的多模态报告评估基准。该基准通过混合式生成管道(结合蒙特卡洛树搜索与人工质量控制)构建专家标注的参考报告,具备标准化章节结构和可验证的关键洞察点。为实现稳健评估,设计了双评判框架:文本评判模型(Text Judge Model)从结构完整性、基于源表的数值事实核查、洞察新颖性分析等方面评估文本质量;可视化评判模型(Visualization Judge Model)则从数据保真度、表现力、美学效果及技术正确性等维度评估生成图表,并同时评估跨模态一致性。实验表明,即使最先进的模型整体得分仅达68.9,且38.5%的报告存在数值幻觉,真正新颖的分析洞见仍极为稀少。消融实验进一步证实,单评判评估会系统性高估性能8–12分,凸显多维度、互补性评估在确保基准严谨性中的关键作用。
链接: https://arxiv.org/abs/2611.22514
作者: Teng Lin,Nan Tang
机构: HKUST(GZ) (香港科技大學(广州)); DSA Thrust (数据科学与人工智能研究团队)
类目: Computation and Language (cs.CL)
备注:
Abstract:The ability to generate multimodal reports integrating fluent narratives with faithful visualizations from tabular data, has become a cornerstone capability for modern vision-language models in practical domains such as finance, healthcare, and policy making. However, existing benchmarks are limited by single-modality focus, simplistic metrics, novelty evaluation, and visualization quality. We address these limitations with MM-ReportBench, a multimodal report evaluation benchmark covering 386 tasks across 185 real-world tables. Each task is paired with an expert-curated reference report featuring standardized chapter structures and verifiable key insight points, constructed through a hybrid pipeline of multimodal reports generation based on Monte Carlo Tree Search and rigorous human quality control. For robust evaluation, we design a dual-judge evaluation framework: Text Judge Model assesses textual quality along structural completeness, grounded insight verification with numerical fact-checking against source tables, and novelty analysis that distinguishes genuine analytical discovery from trivial paraphrasing; Visualization Judge Model evaluates generated visualizations across data fidelity, expressiveness, aesthetics, and technical correctness. They also evaluate cross-modal consistency. Extensive experiments on 20 state-of-the-art models reveal critical weaknesses: even the strongest model achieves only 68.9 overall score, with numerical hallucination affecting 38.5% of reports and genuinely novel insights remaining rare. Our ablation studies further demonstrate that single-judge evaluation consistently inflates scores by 8-12 points compared to dual-judge assessment, confirming that rigorous and reliable benchmarking demands multi-faceted, complementary evaluation probes.
[NLP-39] LMLineage: Content Lineage and Context Governance for LLM Agents
【速读】: 该论文旨在解决大语言模型(Large Language Model, LLM)智能体在多源上下文环境下的可解释性问题,特别是如何有效追踪并归因智能体输出句子的来源依据,同时实现对长期记忆中冗余信息的动态清理。传统方法往往局限于模型内部权重的解释,而忽视了推理过程与决策链的透明化。其解决方案的关键在于提出一种无需白盒访问、不依赖额外推理计算的“离线”(out-of-band)系统——LMLineage,通过高精度的上下文溯源机制,将每一条输出语句精准关联至原始输入片段,并提供校准置信度。该系统进一步引入概念漂移微压缩(concept-drift micro compaction)机制:基于行迹元数据检测已不再被引用的输入文档(即发生“漂移”),仅对这些过时内容进行局部压缩,从而在不丢失用户指令的前提下显著降低上下文负担。实验结果表明,窗口级嵌入相似性相比整篇文档或词法匹配更能准确捕捉LLM的语义重述行为;贪婪窗口快速路径在效率和准确性上均优于穷举式匹配;单次归因可在1秒内完成,且在40轮上下文累积下保持稳定;概念漂移检测具有零误压缩率,压缩召回率受制于归因准确性。该研究还展示了交互式事件追溯与内存清理的应用案例,验证了系统的实用性与可靠性。
链接: https://arxiv.org/abs/2611.22511
作者: Wei Hou,Jingxin Geng,Xin Chen,Zheqing Jiang
机构: Lenovo Global Security Lab(联想全球安全实验室)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Interpretability for AI agents should not be confined to understanding the model’s internal weights. It must also cover the agent’s reasoning and decision-making process over multi-source context. We present LMLineage, an out-of-band system that attributes every output sentence of an agent session to its supporting input spans with calibrated confidences, without requiring white-box access to the model or any additional inference passes. On top of the lineage metadata, we build concept-drift micro compaction: instead of waiting for the context-length limit and summarizing everything, the system detects when previously used input documents have stopped being referenced (“drifted”) and surgically compacts only those, preserving user instructions verbatim. We evaluate on a multi-tier corpus: 71 gold annotated sentences from 10 Chinese web-QA cases, plus 90 turns of independently collected multi-turn sessions across three generator models (gpt-4o-mini, qwen3.8-27b, deepseek-v4-flash). Results show that (i) window-level embedding similarity is far better aligned with LLM paraphrasing than whole document or lexical matching; (ii) the greedy-window fast path is both faster and more accurate on average than exhaustive line-matching (a multiple-comparisons bias in the latter inflates web recall at the cost of overall accuracy); (iii) per-turn attribution completes in 1s on a single GPU, remaining stable as context accumulates to 40 turns; and (iv) concept-drift detection is conservative (0% false compaction) with recall bounded by attribution accuracy. We further describe the concept-drift micro compaction design and an interactive incident-tracing/memory-cleaning case study. A companion patent application covers the lineage construction method.
[NLP-40] From syllable representation via gesture generation to articulation: A spiking neural model
【速读】: 该论文旨在解决语音生成过程中从音位表征到协调的声道运动之间的映射问题,即如何在神经层面实现语音产生的时序控制与动作协调。其核心挑战在于构建一个能够精确模拟发音动作起始、持续时间及动态轨迹的生物合理性模型,以支持连续、自然的语音输出。解决方案的关键在于提出一种基于脉冲神经网络(Spiking Neural Network)的传感运动模型,融合神经工程框架(Neural Engineering Framework)与语义指针架构(Semantic Pointer Architecture),通过每个音节对应一个神经振荡器来协调发音动作序列;利用相位依赖性指令激活不同的神经种群,分别控制元音、口腔约束、软腭咽部及声门动作(手势),并通过神经坡道积分器塑造手势激活的上升与下降过程,从而决定发音动作的幅度与时间进程。此外,通过对抗性神经种群对几何声道模型进行控制,实现了对实际发声运动的精准模拟。该模型在30个参考录音和两组各60次的仿真子数据集上验证了其可靠性,并通过调整手势命令幅值将成功实现率提升至60/60,充分展示了神经振荡相位规则与坡道动力学在语音时序生成中的关键作用。
链接: https://arxiv.org/abs/2611.22499
作者: Bernd J. Kröger
机构: RWTH Aachen University (亚琛工业大学)
类目: Computation and Language (cs.CL)
备注: 47 pages, 9 figures. Submitted to the Journal of Neurolinguistics on 8 November 2026
Abstract:Speech production requires a transformation from phonological representations into coordinated vocal-tract movements. This paper presents a spiking neural sensorimotor model within the Neural Engineering Framework and Semantic Pointer Architecture. Each syllable occurrence has a neural oscillator whose evolving phase coordinates its gestures and initiates the next syllable. Phase-dependent commands recruit separate neural populations for vocalic, oral-constriction, velopharyngeal, and glottal actions (gestures). Neural ramp integrators shape the rise and fall of gesture activation. Together with articulatory targets, these trajectories determine the extent and time course of articulatory movements. Opponent neural populations provide control signals to a geometric articulatory model. Thirty reference recordings cover an input corpus of eight lexical words, an alternative realization of one word, a syllable sequence, and isolated syllables. Two additional paired simulation subcorpora of sixty runs each, covering two words and one syllable sequence, assess activation reliability across neural realizations. Adjusting gesture command amplitudes increased successful realizations from 57/60 to 60/60 under the specified criteria, with ramp rate factors held constant. The model demonstrates how neural oscillator and ramp dynamics generate articulatory timing under specified phasing rules. It complements a preceding mental-lexicon model through an ordered syllable interface. Continuous online execution across all components and connection to the acoustic extension remain further integration steps. Accessible code and synchronized visualization support inspection of the neural activation trajectories and resulting vocal-tract movements. Comments: 47 pages, 9 figures. Submitted to the Journal of Neurolinguistics on 8 November 2026 Subjects: Computation and Language (cs.CL) Cite as: arXiv:2611.22499 [cs.CL] (or arXiv:2611.22499v1 [cs.CL] for this version) https://doi.org/10.48550/arXiv.2611.22499 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[NLP-41] Growing Clusters Lower the Bar: Profile Accretion in LLM -Based Incremental Entity Resolution
【速读】: 该论文旨在解决增量式实体消歧(incremental entity resolution)管道中大语言模型(LLM)作为匹配判断器时存在的偏差问题,即模型对新记录是否归属已有聚类的判断受集群规模影响,而非仅依赖于待比较记录本身。其核心发现是:随着已验证成员数量增加,模型会降低加入集群的阈值,导致相似名称公司之间的误匹配率显著上升,而这一现象在不同提示工程、批量处理格式和地址信息变化下依然存在。关键解决方案在于采用单一阈值在标注数据上进行在线链接,在易混淆区块中可使召回率提升\fxRRmin–\fxRRmax倍,且未造成可检测的精度下降或额外的LLM调用开销。
链接: https://arxiv.org/abs/2611.22485
作者: Huixue Han,Shangzhi Xu,Anqi Xiong,Yu-Chu Tian
机构: Queensland University of Technology (昆士兰科技大学); Independent Researcher (独立研究员)
类目: Computation and Language (cs.CL); Databases (cs.DB)
备注:
Abstract:Incremental entity resolution pipelines ask a large language model (LLM) whether a new record belongs to an existing cluster and show it the records already linked. Analyses of matching oracles assume that the answer depends only on the records compared. For the LLM matchers we test, it depends on cluster size, and the model lowers its bar for joining a cluster as the cluster grows. On French procurement suppliers from FOPPA, with SIREN identifiers as ground truth, we hold the new record fixed and vary only the number of verified members shown. Sixteen members instead of one raise the false match rate on similarly named companies by \dfmrMin–\dfmrMax percentage points in six open-weight models, while their area under the ROC curve rises by only \aucGainMin–\aucGainMax. In Claude Sonnet~5 and Haiku~5.5 the rise is \clSonD and \clHaiD points. The shift persists with street addresses, prompt variants and the BatchER and ComEM formats, and one false merge raises acceptance of the intruder’s other records in every open model. It does not arise on Czech and Bulgarian TED notices, where one record with its full address already identifies the company. In online linking on confusable blocks, a single threshold set on labelled data raises recall by a factor of \fxRRmin–\fxRRmax for five models with no detectable loss of precision and no extra LLM calls.
[NLP-42] LMMR: Learned Multigranular Memory Retrieval for Small Language Model Agents
【速读】: 该论文旨在解决小规模语言模型代理(small language model agents)在处理多跳问答(multi-hop questions)时,因记忆空间有限而难以有效读取所需信息的问题。由于相关事实常分散于多个记录中,且其上下文文本占用大量可用内存空间,导致模型无法高效获取关键证据。论文提出了一种学习型多粒度记忆检索方法(LMMR),其核心创新在于联合决策“下一个要读取的来源”以及“读取多少内容”,通过将段落和句子片段与其原始来源关联,并在训练中配对提供相同缺失支持的不同视图,利用相关干扰项和重复信息构建对比样本。在推理阶段,重排序器评估每个候选内容对锚定段落的增益,选择器则在给定的令牌预算内挑选最优证据。实验表明,在HotpotQA、2WikiMultihopQA和MuSiQue三个基准上的22,398个官方验证问题上,冻结的20亿参数答案模型在400令牌预算下分别达到61.47、43.26和30.58的F1分数,相较于查询扩展方法分别提升3.06、2.14和4.55点,同时内存输入减少23.1%–26.9%;相比同源训练的粗粒度检索器,输入令牌数减少23.5%–27.7%;与采用LLMLingua-2的同类方案相比,F1提升达3.99点,且在HotpotQA和2WikiMultihopQA上经Holm校正后显著。所有八种小型答案模型在HotpotQA上均优于查询扩展和MMR方法,在F1和精确匹配指标上均有提升。因此,解决方案的关键在于通过多粒度、可学习的检索策略实现更高效、精准的记忆访问。
链接: https://arxiv.org/abs/2611.22425
作者: Pengfeng Song,Yunkai Zhuang,Liye Shi
机构: Zhejiang Wanli University (浙江万里大学)
类目: Computation and Language (cs.CL)
备注: 18 pages, 11 figures. Code and experimental results: this https URL
Abstract:Small language model agents have limited space to read the memory needed for multi-hop questions. Relevant facts are often scattered across records whose surrounding text consumes much of that space. We propose LMMR, a learned multigranular memory retrieval method that selects both the next source and how much of it to read. LMMR links paragraphs and sentence fragments to their original sources and pairs views that supply the same missing support during training. Relevant distractors and repeated information provide contrasting examples. At inference, the reranker scores what each candidate adds to an anchor passage, and a selector chooses the evidence within a token budget. On 22,398 official validation questions from HotpotQA, 2WikiMultihopQA, and MuSiQue, a frozen 2B answer model reaches 61.47, 43.26, and 30.58 F1 under a 400-token budget. The gains over query expansion are 3.06, 2.14, and 4.55 points, while memory input decreases by 23.1-26.9%. LMMR also uses 23.5-27.7% fewer input tokens than a coarse retriever trained on the same questions. Against that retriever followed by LLMLingua-2 at comparable input lengths, F1 gains reach 3.99 points and are significant on HotpotQA and 2WikiMultihopQA after Holm correction. All eight small answer models evaluated on HotpotQA improve over query expansion and MMR in both F1 and exact match.
[NLP-43] SignAligner: Dual-Granularity Span Generation for Sign Language Alignment
【速读】: 该论文旨在解决手语语料库中字幕对齐(subtitle alignment)与手语词定位(sign spotting)在时间标注层面的协同优化问题,尤其关注如何在不依赖外部手语视频资源的前提下,实现高精度的字幕时间对齐。现有方法虽已具备大规模手语词的时间定位能力,但将字幕段落与手语词片段联合训练以提升对齐性能的研究仍不充分。为此,本文提出SignAligner——一种提示条件驱动的视觉-语言跨度生成模型,能够根据手语视频与字幕句子或目标词查询,输出对应的时间帧区间(文本起止帧间隔)。其核心解决方案包括:首先在字幕段落与手语词段落上联合训练模型,随后通过基于重叠度(overlap-based)的奖励机制进行精细化微调,同时保留辅助词例样本以增强泛化能力。实验表明,在英国广播公司-牛津手语语料库(BOBSL)的字幕测试集上,经重叠奖励优化后的模型达到66.68 F1@0.50,优于所有对比系统且无需外部手语视频数据。此外,联合训练显著提升了对齐性能,且该优势在后续重叠奖励微调后依然保持;对于词查询任务,优化后的模型能准确追踪目标词的实际手语位置,而非依赖视频中心固定位置,展现出更强的空间敏感性。综合结果支持采用单一经重叠奖励优化的跨度生成器,实现字幕对齐与手语词时间标注的统一建模,为人工审核提供高质量时间标注基础。
链接: https://arxiv.org/abs/2611.22409
作者: Jongbhin Woo,Youngjoon Jang,Joon Son Chung
机构: KAIST(韩国科学技术院); University of Oxford(牛津大学)
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL)
备注: 11 pages, 5 figures
Abstract:Large sign-language corpora benefit from temporal labels at two scales: subtitle spans for subtitle alignment and word spans for sign spotting. Sign spottings already locate words in time at corpus scale, but training subtitle alignment jointly on subtitle and word spans remains underexplored. We present SignAligner, a prompt-conditioned visual-language span generator that maps a sign-language video and a subtitle-sentence or target-word query to a textual start/end frame interval. It is first trained jointly on subtitle spans and word spans from sign spottings, then refined with overlap-based rewards while retaining auxiliary word examples. On the subtitle test set of the British Broadcasting Corporation-Oxford British Sign Language (BOBSL) corpus, the refined model reaches 66.68 F1@0.50 (F1 with a temporal intersection-over-union threshold of one half), the best among compared systems without external sign-video resources. This overlap-reward refinement outperforms cross-entropy continuation from the same jointly trained model. Joint training with word spans improves subtitle alignment in a pair of our generators matched in code, labels and maximum update budget. The advantage remains after both models of the pair are refined with overlap rewards. For word queries, the refined model follows the queried word, not a fixed position in the video: when the word’s sign lies away from the video centre, it is far more accurate than predictions that use position alone. Together, these findings support one shared span generator refined with overlap rewards, aligning subtitles and producing word spans for human review.
[NLP-44] Comparing Implicit and Explicit Language Attitudes toward the Dutch multiethnolect Straattaal in Humans and Large Language Models
【速读】: 该论文旨在解决大语言模型(Large Language Models, LLMs)是否像人类一样对语言变体与社会属性之间存在类似关联的问题,特别是针对荷兰多族裔方言斯特拉塔尔语(Straattaal)与其与移民群体之间的社会关联。研究的关键在于通过匹配的人类与模型实验设计,分别采用隐式联想测试(Implicit Association Test, IAT)和显式评分任务来评估人类与大语言模型在语言态度上的表现;同时引入大语言模型词联想测试(LLM Word Association Test, LLM WAT)作为对模型隐式态度的测量工具。结果表明,无论是人类还是大语言模型,均在隐式和显式层面将斯特拉塔尔语与移民背景个体相关联,且模型规模越大、越新,这种关联越强。此外,大语言模型在显式评分任务中表现出与人类相似的响应模式。这表明许多大语言模型已编码了类似于人类的语言态度,但同时也揭示了在将原本用于研究人类态度的概念与测量方法应用于大语言模型时存在的解释性挑战。因此,研究强调需将社会语言学视角及语言变异纳入自然语言处理(Natural Language Processing, NLP)研究,以更有效地识别并缓解模型中嵌入的社会与语言偏见。
链接: https://arxiv.org/abs/2611.22335
作者: Hedwig Oldenhof,Laura Rosseel,Anna Wegmann,Dong Nguyen
机构: Utrecht University(乌得勒支大学); Vrije Universiteit Brussel(布鲁塞尔自由大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:While language attitudes have been extensively studied in humans, the extent to which large language models (LLMs) encode similar associations between language varieties and social attributes remains unclear. Contemporary sociolinguistic research distinguishes between explicit attitudes, which stem from deliberative processing, and implicit attitudes, which arise from automatic processing. This study investigates whether LLMs exhibit human-like implicit and explicit language attitudes toward the Dutch multiethnolect Straattaal, particularly its association with migrant communities. We use closely matched designs to assess attitudes in humans and LLMs. For humans, we use an Implicit Association Test (IAT) and an explicit rating task. For LLMs, we use the LLM Word Association Test (LLM WAT) and the same explicit rating task. Human participants strongly associate Straattaal with individuals with a migration background at both implicit and explicit levels. In the LLM WAT, Straattaal is consistently associated with individuals with a migration background, with stronger associations observed in larger and newer models. When presented with the same explicit rating task as human participants, LLMs produce comparable response patterns. Overall, the findings suggest that many LLMs encode language attitudes resembling those observed in humans, however we also highlight important questions about how concepts and measures developed to study human attitudes should be interpreted in LLMs. This highlights the relevance of incorporating sociolinguistic perspectives and language variation into NLP research to better identify and mitigate socially and linguistically embedded biases.
[NLP-45] Zero-Dimensional Attention Topology for Linguistic Acceptability
【速读】: 该论文旨在解决如何利用Transformer模型中注意力机制的拓扑特性,为语言可接受性分类任务构建有效的句子表示问题。其核心挑战在于从复杂的注意力图结构中提取具有语义解释性和预测能力的表征,同时避免引入冗余或误导性的信息。解决方案的关键在于采用零维持续同调(H₀)方法,通过分析144个对称化BERT注意力图生成的最小生成树合并值,将原始注意力结构压缩为1,440维的拓扑特征向量。该方法能够保留句法层面的词元数量信息,并主动过滤掉方向性注意力差异这一类可能干扰泛化性能的敏感特征。然而,实验结果表明,该固定拓扑表示在跨域评估中的表现(外域Matthews相关系数为0.438)仍不及直接使用注意力统计特征(0.508)甚至低于原始BERT模型(0.545),说明其在当前设定下未展现出相对于简单注意力汇总的预测优势。研究进一步揭示了该表示的内在属性,如对采样单元敏感的不确定性估计,以及评估过程中数据完整性控制的重要性,为后续关于注意力拓扑结构的研究提供了关键的方法论参考与验证基准。
链接: https://arxiv.org/abs/2611.22312
作者: Saeed AlMarri(Khalifa University)
机构: Khalifa University of Science and Technology, Abu Dhabi, UAE
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 10 pages, 4 figures, 4 tables. Code and data are provided as ancillary files
Abstract:Topological summaries of Transformer attention provide sentence representations for linguistic acceptability classification. We examine a zero-dimensional persistent-homology ( H_0 ) representation that summarizes minimum-spanning-tree merge values from 144 symmetrized BERT attention graphs into 1,440 features. The retrospective evaluation uses 8,517 downstream training sentences and 521 in-domain and 497 out-of-domain CoLA validation sentences after data-integrity filtering. The recorded out-of-domain Matthews correlation coefficients are 0.438 for the topological representation and 0.508 for attention statistics, compared with 0.545 for BERT. Attention-feature results are retained aggregate measurements; BERT predictions are rescored and length controls are refitted from the accompanying records. Analytical results show that the representation retains token-count information and discards a class of directional attention differences. Source resampling of BERT predictions also demonstrates the sensitivity of uncertainty estimates to the sampling unit. The reported results do not demonstrate a predictive advantage for this fixed topological representation over simpler attention summaries. They identify representation properties and evaluation controls relevant to subsequent studies of attention topology.
[NLP-46] VRLLM : Turning LLM s into Powerful Tabular Classifiers
【速读】: 该论文旨在解决预训练大语言模型(LLM)在大规模表格分类任务中表现不佳的难题,尤其针对其在性能上普遍落后于树模型或专为表格数据设计的神经网络架构的问题。其核心解决方案的关键在于:通过在数十亿行带标签的工业级表格数据上对小型化预训练的Llama 3.2 1B模型进行精细化微调,成功实现了超越多种主流表格分类器(包括生产环境中的深度学习模型)的性能表现。研究团队提出的Visa Research LLM(VRLLM)不仅验证了紧凑型LLM在表格预测任务中的潜力,还通过系统性的数据准备、训练策略设计、消融实验与部署验证,揭示了模型内部机制——如前两个前馈神经网络(FFN)层对线性可解任务信息的涌现具有关键作用,并且模型能够学习到特定特征与目标之间的预测关联。该工作提供了一套可复现的工程实现路径与聚焦的实证案例,为将轻量级LLM有效适配至工业级表格预测任务提供了重要参考。
链接: https://arxiv.org/abs/2611.22308
作者: Yingtong Dou,Shubham Jain,Junpeng Wang,Hadi Abdullah,Vineeth Rakesh Mohan,Yuzhong Chen,Liang Wang,Yiwei Cai
机构: Visa Research(维萨研究)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: Technical Report. WIP
Abstract:Large Language Models (LLMs) have shown strong performance on text, but whether pretrained LLMs are effective for large-scale tabular classification remains unclear, as many previous works demonstrate that LLMs’ tabular data classification performance falls behind tree-based models or neural architectures designed for tabular data. At Visa Research, we managed to fine-tune a pretrained Llama 3.2 1B LLM on billions of rows of labeled tabular data and make it outperform a wide array of tabular classifiers and even our competitive production deep learning model. For ease of reference, we refer to the model as the Visa Research LLM (VRLLM). In this paper, we share how we achieved this exciting result, covering data preparation, training recipe, ablation studies of design choices and deployment validation. We also benchmark frontier LLMs on the same task and analyze their prediction behavior across model families and generations. Our interpretability experiments suggest that VRLLM learns feature-specific predictive associations, while component ablations indicate that the earliest two FFN layers contribute to the emergence of linearly decodable task information. Our paper provides an engineering recipe and a focused empirical case study on adapting compact LLMs to industrial tabular prediction tasks.
[NLP-47] OfficePrism: Benchmarking Agents on Single-Domain Operations Across Enterprise Business Functions
【速读】: 该论文旨在解决企业级任务中跨职能协作带来的复杂性与成本问题,即传统宽泛的工作绩效评估指标往往将单一业务职能内的能力表现与跨职能、跨系统协调的额外开销混为一谈,导致问题定位困难。其核心挑战在于如何在真实企业场景下对大模型在多业务领域中的端到端任务执行能力进行精准、可复现且可比较的评估。解决方案的关键在于构建一个基于统一基准定义的多领域企业任务评估框架——OfficePrism,该框架涵盖7个业务职能的420个单域任务,并在L1–L3三个层次的任务结构上进行标准化测试。每个任务均在独立环境中运行,结合共享的合成组织背景与特定职能领域的状态数据,通过融合实体记录、业务规则和工具模式(tool schema),协同生成中文工单、隐藏的可执行参考答案及预期输出结果,并经过自动化筛选与人工校验确保质量。该评估体系同时采用全模式(full-schema)与搜索-分派(search-and-dispatch)两种包配置进行对比,实验覆盖124种模型-调用器-包组合,在联合自然语言与数据库操作的通过率上表现出22.6%至76.4%的区间差异,中位数达67.9%。结果显示,所有配置中L1层级表现最优,且L2在122种情况下优于L3;在50组共用同一模型与调用器标签的对比中,搜索-分派方案在30组中表现更优,1组持平,19组落后,表明任务调度策略对实际性能具有显著影响。
链接: https://arxiv.org/abs/2611.22262
作者: Hanfeng Liu,Zhenyu Luo,Dongchao Zhang,Zexi Liu,Zulong Chen,Jinxin Hu,Jia Xu,Zeyi Wen
机构: Alibaba Group(阿里巴巴集团); HKUST(GZ)(香港科技大学(广州)); Alibaba International Digital Commerce Group(阿里巴巴国际数字商业集团); Cyberspace Institute of Advanced Technology, Guangzhou University(广州大学网络空间先进技术研究院)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Software Engineering (cs.SE)
备注:
Abstract:Broad workplace scores can mix competence within one business function with the extra cost of coordinating across functions and systems, making failures hard to localize. OfficePrism evaluates this prerequisite through 420 single-domain enterprise tasks spanning seven business functions under shared, benchmark-defined L1–L3 task-structure levels. Each task runs in a fresh environment combining a shared synthetic organization with one function-specific domain and yped state access. Our pipeline uses grounded records, business rules, and tool schemas to jointly construct a Chinese-language ticket, a hidden executable reference witness, and an expected response, followed by automated filtering and human curation. The same task set is evaluated through full-schema and search-and-dispatch packages. Across 124 model–harness–package configurations, the per-case joint natural-language-and-database pass rate ranges from 22.6% to 76.4%, with a median of 67.9%. L1 is highest in all 124 configurations, and L2 exceeds L3 in 122. Among 50 configurations sharing the same harness and model labels across both packages, search scores higher in 30, ties in one, and scores lower in 19.
[NLP-48] Where Does Hindsight Enter Clinical Agent Benchmarks? An Audit of Simulators and Case Construction
【速读】: 该论文旨在解决大语言模型(LLM)诊断代理在基准测试中因“事后回溯”(hindsight)信息泄露而导致的评估偏差问题。现有基准多基于出院摘要重构病例,而模拟器在生成未记录的检验结果时往往能访问最终诊断,从而引入了不可靠的性能高估。其解决方案的关键在于提出HindAudit框架,通过分离三个事后回溯通道——提供给模拟器的诊断、模拟器所依赖的回顾性上下文以及面向代理的病例文本,并采用仅基于真实数据训练的评估者对各通道进行独立测量。实验在MIMIC-IV的27,489例急诊住院患者上进行,结果显示:将诊断信息提供给模拟器对模拟结果影响微乎其微,提示诊断信号未显著泄露;而病例文本的重构方式对模型性能影响最大——与基于分诊记录的时间点重建相比,基准风格的摘要使四类零样本LLM的准确率从25.4%提升至39.7%,出院史摘要则达45.3%,且交互式代理得分普遍上升但排名基本不变。进一步分析表明,遮蔽诊断名称不影响结果,说明偏差主要源于病例文本中隐含的前瞻性信息。由于MIMIC-IV缺乏实时床旁病史,该差距为事后回溯效应的上限。研究报告了所有注册结果并建议未来构建基准时应严格控制信息泄露路径,以提升评估的可信度。
链接: https://arxiv.org/abs/2611.22241
作者: Meilu Zhu,Xiaohan Li,Axiu Mao
机构: The University of Hong Kong (香港大学); City University of Hong Kong (香港城市大学)
类目: Computation and Language (cs.CL)
备注: 8 pages main text, 20 pages in total, 6 figures, 11 tables
Abstract:Benchmarks for LLM diagnostic agents are built from retrospective records: the case an agent reads is rewritten from discharge summaries, and unrecorded test results are generated by a simulator that often sees the final diagnosis. HindAudit separates three hindsight channels (the diagnosis given to the simulator, the retrospective context it conditions on, and the agent-facing case text) and measures each with judges trained only on real data. On 27,489 emergency admissions from MIMIC-IV, five registered tests found no leakage into simulated laboratory results: providing the diagnosis to five simulators had a negligible effect, and prompted gatekeepers diluted the diagnostic signal, although exploratory results suggest they can raise agent scores. The largest effect lay in the case text. Against a presentation-time reconstruction from triage records, benchmark-style summaries raised the zero-shot accuracy of four LLMs from 25.4% to 39.7% and discharge histories to 45.3%, and raised six interactive agents’ scores while barely changing their ranking. The exploratory gap was reproduced in a registered replication on held-out patients. Masking diagnosis names left it unchanged. Because MIMIC lacks a contemporaneous bedside history, the gap is an upper bound on hindsight. We report all registered outcomes and recommend construction practices.
[NLP-49] SOTA or Luck? The Winners Curse in LLM Leaderboards
【速读】: 该论文旨在解决公开大型语言模型(LLM)排行榜中存在的“排名赢家诅咒”问题,即顶级模型性能评估因基准测试分数的噪声而被系统性夸大。这一现象源于排名机制对偶然采样误差与真实能力的混合奖励,导致整个排行榜出现由排名诱导的偏差(rank-induced bias)。其解决方案的关键在于提出一种基于 m-out-of-N bootstrap 的校正方法,该方法在假设模型池固定、任务独立同分布且有界的基础上,允许模型在单个任务内存在任意相关性。理论证明,在总体层级间存在正间隔的情况下,该方法可将原本为 Θ(N−1/2) 的排序偏差降至期望残差偏差 o(N−1/2)。在5个基准家族共14个基准上的实证结果显示,排名第一的模型调整幅度介于0.09至3.86个百分点之间;在SWE-bench Verified快照中,并列领先者获得0.90个百分点的调整。这种高估程度足以使“明显胜者”在统计上变得脆弱。为此,作者进一步利用该bootstrap方法量化各模型可能的排名区间,并基于一致恢复条件对模型进行分层。在SWE-bench Verified中,该流程将173个模型中的13个归入领先层级。最终,研究构建了“排名不确定性与排行榜评估报告”(RULER),结合分数校正与排名区间,通过样本分割和校准模拟验证其有限样本表现,并建议在发布原始排名的同时公开RULER,以帮助读者评估性能报告及其排序精度。此外,RULER还为基准设计提供了考虑选择效应与排名不确定性的指导。
链接: https://arxiv.org/abs/2611.22227
作者: Sikun Xu,Raphael Thomadsen,Dennis J. Zhang,Heng Zhang
机构: Washington University in St. Louis; Arizona State University
类目: Machine Learning (cs.LG); Computation and Language (cs.CL); Methodology (stat.ME); Machine Learning (stat.ML)
备注:
Abstract:Public LLM leaderboards guide model evaluation and adoption, yet the reported performance of top-ranked models can be systematically overstated. This leaderboard winner’s curse arises because ranking noisy benchmark scores rewards sampling luck alongside genuine ability. The same selection effect can distort scores throughout the leaderboard, a phenomenon we call the rank-induced bias. We develop an m -out-of- N bootstrap to correct for this bias. It assumes a fixed model pool with independent, identically distributed bounded tasks while allowing arbitrary within-task correlation among models. Under exact population tiers separated by positive gaps, we prove that the bootstrap reduces \Theta(N^-1/2) ranking bias to expected residual bias o(N^-1/2) . Across 14 benchmarks from 5 benchmark families, the rank-1 adjustments range from approximately 0.09 to 3.86 percentage points. In our SWE-bench Verified snapshot, the tied leaders receive an adjustment of 0.90 percentage points. Such overstatement is large enough to make the apparent winner statistically fragile. Thus, we use the same bootstrap to further quantify the intervals of plausible rank positions for each model and group models into tiers. We establish sufficient conditions for consistent recovery of population tiers. On SWE-bench Verified, this procedure places 13 of 173 models in the leading tier. We combine the score adjustment and rank intervals in the Rank Uncertainty and Leaderboard Evaluation Report (RULER). We validate its finite-sample performance using sample splitting and calibrated simulations, and advocate publishing RULER alongside raw rankings so readers can assess both reported performance and the precision of the ordering. RULER also provides guidance for benchmark design that accounts for selection effects and ranking uncertainty.
[NLP-50] Fine-Grained Japanese Clinical Named Entity Recognition: A Controlled Comparison of BERT and Open-Weight Large Language Models
【速读】: 该论文旨在解决生成式大语言模型(Generative Large Language Models, LLMs)在细粒度日语临床命名实体识别(Clinical Named Entity Recognition, NER)任务中的性能表现不确定,以及其对生成设置敏感性的问题。研究的关键在于验证任务特定微调与充分生成能力相结合是否能显著提升LLMs在历史日语文本案例报告数据集iCorpus上的提取性能。实验表明,当输出令牌限制足够大(如2,600或4,096)且模型参数量达到至少130亿时,经过三种子采样QLoRA微调的八款LLM在松弛表面匹配F1(relaxed F1)上均超越了两个基线BERT-CRF模型(0.882–0.914),且随着输出长度上限增加,召回率提升、截断输出减少,性能进一步改善;而仅通过JSON Schema约束结构化输出无法有效弥补零样本场景下的低准确率,说明生成容量与微调策略是决定性能的核心因素。
链接: https://arxiv.org/abs/2611.22219
作者: Feng Han,Ayako Yagahara,Kenji Hirata,Atsushi Yoshida,Kohsuke Kudo
机构: 未知
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 33 pages, 3 figures, 5 tables; includes appendices with supplementary tables and one supplementary figure
Abstract:Background: Generative large language models (LLMs) offer an alternative to encoder-based clinical named entity recognition (NER), but their performance on fine-grained Japanese extraction tasks and sensitivity to generation settings remain uncertain. Methods: We compared two fine-tuned BERT-CRF baselines with nine open-weight LLMs on iCorpus, which defines 113 entity categories. The evaluation used 433 sentences from 36 test documents; 1,672 sentences were used for training. LLMs were evaluated using zero-shot prompting with and without JSON Schema constraints and three-seed QLoRA fine-tuning. Fine-tuned models were assessed under different output-token limits, including 900 and larger model-specific limits of 2,600 or 4,096 tokens. Performance was measured using sentence-level label-surface matching with relaxed and exact-surface F1. Results: The highest zero-shot relaxed F1 was 0.337, compared with 0.836-0.841 for the BERT baselines. At the larger output limits, all eight fine-tuned LLMs with at least 13B total parameters exceeded both BERT baselines (relaxed F1, 0.882-0.914; exact-surface F1, 0.840-0.880). Across three-seed means, increasing the output limit improved relaxed F1 by 0.052-0.085, alongside higher recall and fewer length-limited outputs. JSON Schema constraints changed relaxed F1 by -0.014 to +0.008 among seven models outside the two reasoning-oriented configurations, while medical-general model pairs showed no consistent advantage for medical specialization. Conclusions: On this historical Japanese case-report benchmark, task-specific fine-tuning combined with adequate generation capacity was associated with stronger LLM extraction performance. Output-token limits materially affected comparisons with BERT, whereas enforcing output structure alone did not overcome low zero-shot accuracy.
[NLP-51] EGS: A Complementary Safety Framework for Educational LLM s
【速读】: 该论文旨在解决大语言模型(Large Language Models, LLMs)在面向青少年的教育场景中部署时,可能生成有毒、偏见性、误导性或不符合青少年发展特点的内容这一安全问题。现有方法多采用二元拒绝策略,无法有效应对学生提问中常见的模糊性或出于好奇心的探索性需求,导致教育交互中断。为此,论文提出教育护栏系统(Educational Guardrail System, EGS),其核心在于构建一种轻量级、与模型无关的运行时框架,通过三重机制实现精细化安全管控:一是基于风险等级对输入进行分诊;二是针对灰色地带问题提供类别特定的教育干预引导;三是对输出内容进行审计后再交付。该方案突破了传统“全拒”模式的局限,实现了在保障安全性的同时维持教育功能的有效性。实验表明,EGS在通用及教育专用基准测试中显著提升了内容过滤性能、拒绝质量以及对抗越狱攻击的鲁棒性,验证了纵深防御型护栏机制在提升教育类LLM安全性和教学可用性方面的有效性。
链接: https://arxiv.org/abs/2611.22203
作者: Zexuan Fei,Zhengnan Guo,Yiming Luo,Kaiyi Huang,Aimin Zhou,Bo Jiang,Fei Tan
机构: East China Normal University (华东师范大学); Shanghai Innovation Institute (上海创新研究院)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 20 pages, 8 figures, 8 tables
Abstract:Large Language Models (LLMs) offer strong potential for personalized education, but adolescent-facing deployment raises risks of toxic, biased, misleading, or developmentally inappropriate content. We argue that educational safety should move beyond binary refusal: student queries are often ambiguous or curiosity-driven, requiring contextual guidance rather than simple blocking. We introduce the Educational Guardrail System (EGS), a lightweight, model-agnostic runtime framework that complements intrinsic alignment through query moderation, scenario-aware pedagogical guidance, and response moderation. EGS triages inputs by risk level, redirects gray-zone queries with category-specific educational interventions, and audits outputs before delivery. Experiments across general and education-specific benchmarks show that EGS improves moderation performance, refusal quality, and jailbreak robustness across diverse LLMs. These findings suggest that defense-in-depth guardrails can strengthen safety while preserving the pedagogical utility required in educational LLMs. Our code and model weights will be publicly available to support community adoption.
[NLP-52] RISA: Recoverable Sparse Attention for Block-Diffusion Language Models
【速读】: 该论文旨在解决块级扩散语言模型(Block-wise Diffusion Language Models, DLMs)在长上下文推理中因内存受限的注意力机制导致的性能瓶颈问题。具体而言,每一去噪步骤均需重复读取不断增长的历史键值(Key-Value, KV)缓存,造成显著的计算与内存开销。现有稀疏注意力方法未能充分缓解此问题,原因在于不同查询可能选择不同的缓存页,导致整体读取集合膨胀;同时,重复的支持结构构建带来额外开销,而永久性剔除则可能丢弃后续所需信息。为此,论文提出RISA(Recoverable Sparse-Attention),一种适用于块级DLMs的可恢复稀疏注意力机制,其核心创新在于引入共享读取计划(shared read plan)。RISA利用在一个块内历史KV保持不变、而查询随去噪步骤动态演化的特性,将新生成的历史数据组织为精确优先级与可恢复层级,由初始查询构建一个受保护且有界读取计划,其索引在整个块内复用,同时在每次注意力计算中使用更新的查询和当前块的KV进行重计算。该方法实现了历史保留与重复读取的解耦,并在不改变模型主干或解码器的前提下,摊销了稀疏支持结构的构建开销。实验表明,在DreamReasoner-8B模型上,RISA在32K上下文长度下相比密集注意力实现2.67倍加速,固定轨迹请求提速1.36倍;在128K长度下,注意力加速达5.53倍。在三个模型共八个设置中,RISA在匹配KL损失条件下,解码速度优于MAGE、QUEST和LoSA等基准方案。
链接: https://arxiv.org/abs/2611.22191
作者: Xinrui Zhong,Fanjiang Ye,Yuke Wang
机构: Rice University (莱斯大学)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:Block-wise diffusion language models (DLMs) refine multiple tokens in parallel, offering an alternative to token-by-token autoregressive decoding. Their long-context inference, however, remains bottlenecked by memory-bound attention: every denoising step repeatedly reads a growing historical key–value (KV) cache. Directly applying sparse attention does not fully solve this problem, because different queries can select different pages, inflating the block’s collective read set, while repeated support construction adds overhead and permanent eviction can discard information needed later. We present RISA, a recoverable sparse-attention method with shared read plans for block-wise DLMs. RISA exploits the fact that historical KV remains fixed within a block while queries evolve across denoising steps. Newly generated history is organized into exact priority and recovery tiers; seed queries then build a protected, bounded read plan whose indices are reused across the block while attention is recomputed with fresh queries and current-block KV. This separates history preservation from repeated reading and amortizes sparse-support construction without changing the backbone or decoder. On DreamReasoner-8B, RISA retains comparable task performance. At 32K, the default configuration speeds up attention by 2.67\times and fixed-trajectory requests by 1.36\times over dense, and attention speedup reaches 5.53\times at 128K. Across eight settings on three models, RISA decodes faster than the tested MAGE, QUEST, and LoSA budgets at matched KL.
[NLP-53] From Imperfective Paradox to Moorean Absurdity: Single-Chain Reasoning Lock-In and Multi-Candidate Verbalization Across Aspectual and Prag matic Ambiguity
【速读】: 该论文旨在解决大语言模型在自然语言蕴含(Natural Language Inference, NLI)与连贯性判断中,因推理链过早锁定(premature lock-in)而导致的语义理解偏差问题。具体而言,研究关注事件是否仅发生、达到目标终点(telic culmination)、满足计划意图或在多重语用标准下被评估时,模型推理路径如何受提示工程与输出机制的影响。其解决方案的关键在于通过四次预注册、密封的正式实验,系统检验“单链过早锁定假说”的可观测后果:采用多候选表述(Multi-Candidate Verbalization, MCV)实现一次调用内加权估计(K=5),有效分离提示与非加权(MCU)及自加权(MCW)输出之间的差异;同时发现,基于NLI-CoT的推理模式在无中断的非目标性(atelic)任务中表现显著下降(-91.25个百分点),而直接思维链(Direct-CoT)与多候选反馈(MCU-CF)在隐含摩尔句(implicit Moorean sentences)的连贯性感知上分别提升26.64和17.31个百分点。此外,跨模型对比(包括OpenAI gpt-4o-mini等五种外部模型)揭示了离散式NLI-CoT与标量式Direct-CoT判断在不同模型间的解耦现象,验证了提示与读出协议对结果的可观察影响,但内部过早锁定仍为待证假设。
链接: https://arxiv.org/abs/2611.22188
作者: An-Sheng Jhang
机构: Dongguan University of Technology (东莞理工学院)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 21 pages, 4 figures, 25 tables (single-column arXiv preprint; includes 16,640 Qwen3-8B formal units across 4 preregistered experiments, 1,800 OpenAI format-ablation calls, and 4,240 P16.7 development-slice units across 5 external models)
Abstract:Natural language entailment and coherence judgments often depend on whether an event merely occurred, reached a telic culmination, satisfied a plan, or was evaluated under multiple pragmatic standards. We test observable consequences of the single-chain premature lock-in hypothesis across four preregistered, sealed formal experiments on Qwen3-8B-Q8_0 (16,640 formal units, temperature=0.0): 15 of 16 hypothesis slots retain confirmatory status (12 rejecting H0 under within-experiment Holm control at alpha_fam=0.05; E3_H3 demoted to descriptive status). In Experiment 1, Direct-CoT vs. MCU differs by 25.00 pp on telic high-confidence completion (Holm p=1.31239e-05), while NLI-CoT collapses by -91.25 pp on uninterrupted atelic items (Holm p=1.81805e-21). Multi-Candidate Verbalization (MCV) elicits K=5 weighted estimates in one call, separating the prompt from unweighted (MCU) and self-weighted (MCW) readouts. Mandarin (E2) and English (E3) expose an Interruption-Anchored Realis Paradox under NLI-CoT. In Experiment 4, MCU-CF raises perceived coherence on implicit Moorean sentences by 26.64 pp over Direct-CoT (Holm p=0.0006), alongside Direct-CF - Direct-CoT = 29.15 pp and MCU-CoT - Direct-CoT = 17.31 pp. A 1,800-call OpenAI ablation and a stage=“dev” P16.7 cross-model extension (4,240 planned/collected units across five external models: 4,222 valid, 18 format-missing, 0 transport-missing; 2% missing per slice; OpenAI gpt-4o-mini at top_logprobs=20 achieves 1,000/1,000 valid units) show that discrete NLI-CoT and scalar Direct-CoT judgments dissociate across models. Confirmatory effects concern observable prompt-and-readout protocols; internal premature lock-in remains a hypothesis. All code, stimuli, preregistered analysis plans, and verification tools are available at this https URL.
[NLP-54] Is 4-Bit Quantization More Than Noise to a Chat Models Safety Behaviour? Attributing Behavioural Shifts with Matched Perturbations
【速读】: 该论文旨在解决生成式 AI(Generative AI)在低比特量化(如4-bit)后模型拒绝有害请求能力的不一致性问题。现有研究表明,4-bit版本模型的行为差异可能源于量化过程中的特定舍入误差,而非单纯的比特数或整体权重扰动大小。其解决方案的关键在于识别并分离出具体舍入误差对模型行为的影响:通过配对测试、盲法人工标注及8-bit基准对照,研究发现不同4-bit模型在拒绝行为上存在相反变化趋势(如Qwen2.5-3B更易继续生成有害内容,而Llama-3.2-3B在越狱攻击下合规率下降10.3分),但当引入与真实误差镜像对称或高斯扰动时,行为变化趋近于零;仅当放大真实误差至两倍时,行为才呈现近线性变化。此外,在相同比特率下,特定量化格式(如Q4_K_M)对两模型均无影响。结果表明,4-bit量化下的拒绝行为主要取决于具体的舍入误差,而非比特数或似然值,因此必须将每个量化构建版本与其原始模型进行直接对比以评估其安全性变化。
链接: https://arxiv.org/abs/2611.22186
作者: Ege Kaya(University of British Columbia)
机构: University of British Columbia(不列颠哥伦比亚大学)
类目: Cryptography and Security (cs.CR); Computation and Language (cs.CL)
备注: 16 pages, 4 figures. Code and labels: this https URL
Abstract:Chat models on phones and laptops are usually 4-bit copies of the model that was evaluated, and audits of such copies disagree about what quantization does to refusal. A 4-bit build is one of many weight perturbations of the same size, so a change in its behaviour could come from its specific rounding error or from any perturbation that large. In a pre-registered study of three 3-4B chat models, with paired tests, blind human labels and an 8-bit null, we find two 4-bit shifts of opposite sign: with a compliant reply prefilled, Qwen2.5-3B continues into harmful content more often, and on held-out jailbreaks Llama-3.2-3B complies 10.3 points less often. Matched perturbations separate the two. Shuffled and Gaussian versions of the error average close to zero, while scaling the real error, from its mirror image to twice its size, changes behaviour near-linearly. At half the real error, Llama’s likelihood is almost unchanged, yet it already complies 5.8 points less often, and this http URL’s Q4_K_M, at the same bits per weight, moves neither model. Refusal under 4-bit quantization depends on the specific rounding error, not on bits per weight or likelihood, so each build has to be compared with its original.
[NLP-55] What RL on Expert-Built Tasks Teaches Nemotron 3.5 Super
【速读】: 该论文旨在解决大模型在特定专业任务场景下表现不足的问题,尤其聚焦于真实世界软件工程、复杂约束处理、长时序工具调用、专业图表理解及企业文档推理等高难度任务中的性能瓶颈。其核心解决方案是采用强化学习(Reinforcement Learning, RL)对NVIDIA的Nemotron 3.5 Super模型(120B总参数,12B活跃参数)进行后训练(post-train),针对五个由专家构建的任务数据集分别训练独立的秩为32的低秩适配器(LoRA adapter),并使用CISPO算法在Tinker平台上实现高效微调。该方法不依赖监督微调阶段,而是通过纯强化学习方式提升模型在具体任务上的决策能力。关键创新在于:每个适配器仅针对单一任务领域进行优化,且在相同评估环境与基准测试下与基线模型对比,显著提升了多个主流基准的表现——如SWE-bench Pro V2从86.9%提升至92.1%,AutomationBench从14.8%跃升至27.5%,部分任务甚至达到超过13个百分点的增益。分析显示,各适配器成功弥补了基线模型在工具使用前的数据与规则探查、代码测试生成、上下文信息延续、图表量化解析以及结构化文档读取等方面的系统性缺失,体现了模块化、任务特异性强化学习在提升生成式AI(Generative AI)专业能力方面的有效性。
链接: https://arxiv.org/abs/2611.22161
作者: Sushant Mehta,Edwin Chen
机构: Surge AI
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Software Engineering (cs.SE)
备注: 18 pages, 2 figures, 9 tables
Abstract:We post-train NVIDIA’s Nemotron 3.5 Super (120B total, 12B active parameters) with reinforcement learning (RL) on five expert-built task datasets, one at a time: real-world software engineering, complex constraints, long-horizon tool use, professional chart understanding and enterprise document reasoning. Each run trains a fresh rank-32 LoRA adapter with CISPO on Tinker, with no supervised stage, and each adapter is compared with the base model on the same tasks, harness and settings. On twelve headline benchmarks, eight built by other groups and four by Surge AI, the gains range from +4.5 points on IFBench to +13.8 on GDP_pdf, which rises from 10.4% to 24.2%; SWE-bench Pro V2 rises from 86.9% to 92.1%, Terminal-Bench 2.1 from 63.9% to 70.3%, MultiChallenge from 58.3% to 66.6%, AutomationBench from 14.8% to 27.5% and Chartography from 39.4% to 50.6%. The document adapter, trained without API tools, also gains +8.0 points on AutomationBench. In attempts read side by side with the base model’s on tasks each adapter newly solves, the base model’s failures recur in a few modes, and each adapter does more often what the base model missed: the tool-use adapter finds the data and the rules before acting, the coding adapter writes its own test or scratch files on SWE-bench Pro V2, the constraints adapter carries earlier turns into its answers, the chart adapter measures the marks in code, and the document adapter reads AutomationBench’s spreadsheets.
[NLP-56] Long-Context Generation Is a Sampling Problem (When You Turn Up the Heat) NEURIPS2026
【速读】: 该论文旨在解决开放权重语言模型在长序列生成过程中性能随生成长度增加而显著下降的问题,尤其关注高温采样条件下生成质量退化的现象。其核心挑战在于:随着温度升高,传统采样策略(如top-p、top-k)会大幅扩展候选词池,引入大量低概率词汇,导致生成内容偏离正确路径,进而引发推理错误与任务失败。解决方案的关键在于提出并验证“top-nsigma”这一新型采样机制——该方法通过维持一个稳定且较小的候选词池,不受温度升高的影响,从而有效抑制低概率词的引入。实验表明,在温度2下,top-nsigma在代码代理任务中可解决46至62个GitHub问题,远超top-p/top-k的0个;在多个基准测试中,top-nsigma的中位准确率保持在温度1时的100%,而top-p仅维持26%。进一步分析显示,候选词池大小与生成质量呈强负相关(Spearman相关系数-0.77),若在温度1时固定池大小再应用温度调节,top-H、min-p和p-less等策略亦能避免性能崩溃。研究证实,控制候选词池规模是维持长序列生成质量的核心机制。
链接: https://arxiv.org/abs/2611.22148
作者: Allen G. Roush,Minh Nhat Nguyen,Ravid Shwartz-Ziv,Judah Goldfeder,Sanjay Basu
机构: Thoughtworks(思特沃克); Apart Research; New York University (纽约大学); Meta; Columbia University (哥伦比亚大学); Oracle(甲骨文)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 10 pages, accepted to the 40th Conference on Neural Information Processing Systems (NeurIPS 2026). Track on Evaluations and Datasets
Abstract:Open-weight language models are widely observed to get worse the longer they generate. Often, attempts to fix this involve more training. At high temperature, The sampler, i.e. the rule that picks the next token out of the model’s probabilities, decides whether long and agentic generation survives and/or thrives. We test 29 open-weight models from 1B to 122B parameters on ten established benchmarks plus needle retrieval out to one million tokens, covering SWE-bench Verified coding agents, long chain-of-thought reasoning, long-form writing, long structured output and long-context reading. Samplers that keep their pool of candidate tokens small and steady as temperature rises don’t lose accuracy. Samplers that let in low-probability tokens fall apart. At temperature 2, coding agents running top-p, top-k or Top-H resolve 0 of 100 GitHub issues, while the same agents running top-nsigma resolve 46 to 62 and win every issue the two samplers disagree on, on three Qwen3.5 models. Over all benchmarks top-nsigma keeps a median 100% of its temperature-1 score where top-p keeps 26%. Counting how many tokens each sampler lets through explains why. In one Llama-3.1-8B story, going from temperature 1 to 2 grows top-p’s pool from 4 tokens to 51,845 while top-nsigma stays at 3. Across 133 combinations of sampler setting and temperature, the typical pool size predicts judged quality (Spearman correlation of -0.77), and when we pick the pool at temperature 1 and only apply temperature afterwards, Top-H, min-p and p-less stop collapsing. At the default temperature of 1 most samplers perform about the same, We release all code and data to the public.
[NLP-57] What MoE Routing Controls: Causal Interventions on Expert Trajectories
【速读】: 该论文旨在解决混合专家(Mixture-of-Experts, MoE)语言模型中,通过路由机制将标记(token)分配至不同专家时,如何利用专家路径的控制来引导模型生成特定格式或主题内容的问题。其核心解决方案在于:构建目标路径(target path),即基于某一概念或角色相关标记在各层中频繁使用的专家集合,并在生成过程中对目标专家的路由器对数几率(router logit)施加固定偏置,以强制令牌沿预设路径流动。研究区分了两类目标:格式目标(format targets)关注输出形式,概念目标(concept targets)关注内容主题。实验表明,在Qwen3-30B-A3B模型中,引导格式路径能够显著改变生成文本;而引导概念路径虽能有效调整模型下一词概率分布,甚至在跨语言构建路径时仍可实现概念偏向,但对最终文本影响有限。进一步分析发现,格式一旦启动便具有自我强化特性,而概念效应则集中于中间层,且其强度差异与嵌入向量的区分度和词汇频率有关,揭示了概念引导效果受限于深层语义传播机制。
链接: https://arxiv.org/abs/2611.22120
作者: Mukund Athmaram,Manoj Rao,Liang Ma
机构: 未知
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:In mixture-of-experts (MoE) language models, a token is routed to a sequence of experts across layers. We investigate how forcing tokens onto chosen expert paths influences model behavior. We build a target path from the experts that a concept’s or role’s tokens most often use, and we add a fixed bias to each target expert’s router logit during generation to steer the token toward this expert path. We study two kinds of targets. Format targets concern the form of the output, and concept targets concern its topic. In Qwen3-30B-A3B, we show that steering a format path changes the text the model writes. We also show that steering a concept path shifts the model’s next-token probabilities toward the concept, even when the path is built in another language, yet this shift has little effect on the generated text. We investigate why formats reach the text but concepts do not by steering only part of the generation, and find evidence that a format, once started, reinforces itself. We show that the concepts religion and money steer most strongly, and that the effect is concentrated in the middle layers. We attempt to explain why the strength of the effect varies by concept through an analysis of embedding distinctiveness and word frequency.
[NLP-58] Weight-Space Programming Doesnt Need Transformers
【速读】: 该论文旨在解决在无梯度优化(gradient-free)环境下,如何设计高效且可解释的神经网络架构以替代传统Transformer的问题。传统Transformer因其良好的梯度可训练性和GPU计算效率而被广泛采用,但在权重空间编程(weight-space programming)场景中,模型需直接生成目标权重而无需反向传播,因此原有优势不再适用。论文的关键解决方案在于:当允许大语言模型(LLM)自主设计架构时,它们倾向于构建基于关联记忆(associative memory)的稀疏表结构,该结构通过显式存储事实并以固定查询方式检索,实现高精度推理。实验表明,该自动生成的模型在仅使用约1,800倍更少参数的情况下,测试准确率超过由同一LLM手写权重的最优公开Transformer模型;同时,由于其结构本质为可读、可修改的键值对表,错误答案可直接定位至特定行,支持快速修正与调试。基于此,研究提出一种新型关联记忆模型,在相同规模下比同尺寸Transformer在GPU上提速达17倍。尽管预训练模型在多数任务中仍表现更优,但本研究的核心贡献在于揭示了在无梯度条件下应优先采用可解释、可编辑的稀疏表结构而非传统Transformer架构。
链接: https://arxiv.org/abs/2611.22055
作者: Hyunjong Ok,Jaeho Lee
机构: Pohang University of Science and Technology (POSTECH)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: Preprint (Work in progress)
Abstract:Standard neural architectures such as the Transformer were chosen because they train well with gradient descent and run fast on GPUs. In weight-space programming, a large language model (LLM) writes target weights by hand without gradient-based training, so those reasons no longer apply. Should we keep the Transformer? Given minimal guidance, ten labeled examples, and a sandbox, mainly on multiple-choice questions, we task three frontier LLMs with either generating the weights of a fixed architecture or designing the architecture themselves. When free to design, all three compile plain-text facts into a sparse table queried by a fixed procedure, reaching up to 54.4% test accuracy. An LLM-designed model achieves higher test accuracy than the most accurate released Transformer that the same LLM wrote while using about 1,800x fewer parameters. Without gradients, they choose a table, in which the LLM can state what each weight means and can find and rewrite a wrong one. Each wrong answer points to one row, and removing a row changes about one other answer. From this, we propose an associative-memory-based model up to 17x faster on GPUs than a same-sized Transformer. We also test pretrained weights, other benchmarks, and tasks where models generate answers. Pretrained LLMs remain more accurate; the claim concerns which structure to build when writing weights.
[NLP-59] Beyond Acuity Accuracy: A Multidimensional Framework for Evaluating Patient-Facing Emergency Triage AI Tools
【速读】: 该论文旨在解决患者端生成式AI分诊工具在临床应用中仅依赖病情严重程度(acuity)准确性评估的局限性,提出了一种更全面的评估框架——SAFE-Triage。其核心问题是:如何在真实临床场景下系统性评估分诊工具在临床表现、患者交互质量及临床医生可读性总结等方面的综合性能,以确保其安全性和实用性。解决方案的关键在于构建一个多维度集成评估框架,融合人工评审、自动化指标、大语言模型作为评判者(LLM-as-judge)以及模拟测试,从多个角度对基于提示的LLM分诊(PLT)与规则增强型提示的LLM分诊(rPLT)进行对比分析。研究结果表明,PLT在急诊严重指数(ESI)准确率(57.5% vs 42.8%,P < 0.001)和误分诊率(13.1% vs 29.8%,P < 0.001)方面显著优于rPLT,且获得更高临床评分;而rPLT则在输出长度、稳定性及可读性方面表现更优。SAFE-Triage框架成功识别出不同性能指标间的权衡关系,为分诊工具的部署前优化与持续监测提供了依据,明确了其在效率、改进空间及需人工干预环节的具体位置。
链接: https://arxiv.org/abs/2611.22045
作者: Ruio Heerschap,Henry Foote,Mahmoud Alwakeel,Zhouhan Qian,Jiaqi Wang,Elisabeth Bond,Yang Yang,Jonathan Chong Kai Liew,Matthew A. Bergens,Monique Arnold,Haoyuan Wang,Rui Yang,Kunyu Yu,Nikita R. Paripati,Diana Pimenta,Kyle Naimoli,An-Kwok Ian Wong,Nicoleta Economou-Zavlanos,João Ricardo Vissoci,Nan Liu,Anand Chowdhury,Chuan Hong
机构: 未知
类目: Computation and Language (cs.CL); Computers and Society (cs.CY)
备注: 82 pages, including 34 pages of supplementary material. 5 figures (4 main figures and 1 supplementary figure)
Abstract:Patient-facing AI triage tools may support care-seeking and emergency department prioritization, but require evaluation beyond acuity accuracy alone. We developed SAFE-Triage, a framework integrating human review, automated metrics, LLM-as-judge evaluation, and simulation testing across clinical performance, patient interaction, and clinician-facing summary quality. We applied SAFE-Triage to 40 MIMIC-IV-derived emergency vignettes and a 10,030-case acuity analysis comparing prompted LLM triage (PLT) with rule-augmented prompted LLM triage (rPLT). PLT achieved higher emergency severity index (ESI) accuracy (57.5% vs 42.8%, P 0.001 ), lower undertriage (13.1% vs 29.8%, P 0.001 ), and higher clinician ratings across all domains. rPLT produced shorter and more stable outputs with higher automated readability scores. SAFE-Triage identified trade-offs between acuity accuracy, undertriage risk, communication quality, readability, and summary quality, supporting pre-deployment refinement and ongoing monitoring by showing where triage tools are efficient, need improvement, or need clinician oversight.
[NLP-60] RLDE: A Reinforcement Learning Framework for Interpretable Iterative Document Editing
【速读】: 该论文旨在解决长文本在自然语言处理中质量、清晰度与连贯性提升的难题,其核心挑战在于现有自动化系统多采用黑箱生成或依赖监督学习,缺乏可解释性且未能捕捉人类编辑过程中迭代修正的序列特性。为应对这一问题,论文提出一种基于强化学习的端到端、指标驱动的框架RLDE,将文档修订建模为无参考的马尔可夫决策过程(MDP),实现对未配对文档-修订目标的无监督学习。其关键创新在于:采用有限动作空间(NOOP、INSERT、DELETE、REPLACE),结合指针网络(Pointer Network)实现细粒度的词元级动态片段选择,并通过冻结的BART-base模型生成上下文感知的改写建议;同时引入基于SBERT的语义保留惩罚(Semantic Preservation Penalty)机制,有效缓解因删除操作导致的语义退化问题。在此基础上,使用近端策略优化(PPO)算法优化包含可读性、困惑度和语法正确性在内的多目标奖励函数。实验在BEA-2019数据集上验证了显著性能提升(p < 0.01),Flesch阅读易读性得分由49.8升至59.4,困惑度从63.7降至39.8,同时保持93%的语义保真度。相较于链式思维(CoT)提示工程模型如GPT-4和LLaMA-2,RLDE在推理计算成本极低的前提下仍展现出卓越性能,且经人工评估确认其编辑质量高,意义保留得分达4.9/5,编辑质量得分为4.4/5。
链接: https://arxiv.org/abs/2611.22039
作者: Nikhil Parkar,M. Marimuthu,Riya Joshi,Karthikeyan N
机构: Vellore Institute of Technology, Chennai Campus (维洛尔理工学院,金奈校区)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Improving the quality, clarity, and flow of long documents remains a challenging problem in NLP, as human editing typically occurs through iterative refinement rather than a single-pass rewrite. Most existing automated systems act as black-box generators or rely on supervised learning, limiting interpretability and failing to capture the sequential nature of revisions. In this paper, we propose RLDE, an automated end-to-end metric-driven RL framework that formulates document revision as a reference-free MDP problem (unsupervised with respect to paired document-revision targets). Working with a limited action space (NOOP, INSERT, DELETE, REPLACE), RLDE leverages Pointer Networks for dynamic span selection at the token level and sends generative actions to the frozen BART-base model to obtain context-aware suggestions. PPO agents learn to optimize a multi-objective reward function based on readability, perplexity, and grammar metrics with the help of a novel SBERT-based Semantic Preservation Penalty constraint to avoid deletion degeneracy. Evaluations on the BEA-2019 dataset showed statistically significant results (p 0.01) across all metrics, where Flesch Reading Ease increased from 49.8 to 59.4, perplexity decreased from 63.7 to 39.8, while preserving 93% of semantic information. When compared to Chain-of-Thought (CoT) prompt-engineered models like GPT-4 and LLaMA-2, RLDE shows high performance at a very low computational inference cost. Moreover, the quality of edits generated by RLDE has been verified via human evaluations, scoring 4.9/5 for meaning preservation and 4.4/5 for edit quality.
[NLP-61] GauntletScore: Auditable Fact-Checking That Matches the Best Published Checkers on LLM -AggreFact
【速读】: 该论文旨在解决生成式AI(Generative AI)文本中事实一致性验证的核心问题,即如何有效判断大语言模型生成的句子是否得到其源文档的充分支持。现有方法多依赖单一模型进行判断,存在泛化能力不足与可解释性差的问题。本文提出的解决方案关键在于构建GauntletScore证据引擎,其核心创新在于采用多独立阅读者协同评估机制:通过向来自不同提供方的多个独立阅读者提出结构化类型问题,并将每位阅读者的回答作为证据记录至证据账本(evidence ledger);随后基于固定的贝叶斯规则,仅当两名阅读者达成一致时才判定为矛盾。这一设计显著提升了判断的鲁棒性与可解释性,同时避免了单一模型可能存在的偏差。在包含11个数据集共29,320个命题的完整LLM-AggreFact测试集上,GauntletScore实现了79.51的平衡准确率,超越截至2026年11月15日所有公开报告的平均表现,与最优系统nimble-v4-max持平(95%置信区间包含后者)。尤其在全篇新闻文章与多句摘要对比的任务(AggreFact-CNN)中,其得分达76.45,大幅领先次优系统(70.1),且显著优于六种高均值系统(63.9–67.8)。此外,该系统能够完整记录每项判决背后的证据链条,这是单模型方法无法实现的透明性优势。整个测试过程在预设停止策略下一次性运行,引擎参数冻结,确保结果可靠性。研究同时开源了逐条预测结果与评分脚本,推动事实验证领域的可复现性与透明度提升。
链接: https://arxiv.org/abs/2611.22027
作者: Ernie Hobbs,Jim Hinson
机构: Genstrata Inc.(Genstrata 公司)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 10 pages, 1 figure, 7 tables. Per-claim predictions and scoring script: this https URL
Abstract:Checking whether a generated sentence is supported by its source document is a standard safeguard for AI-written text, and the strongest published checkers are single models. We describe GauntletScore, an evidence engine whose few rules and thresholds were set on the benchmark’s development split. It puts typed questions to independent readers from different providers and records each answer as a measurement in an evidence ledger. It combines the measurements with a fixed Bayesian rule that counts a contradiction only when two readers agree. On the full LLM-AggreFact test split (29,320 claims across 11 datasets), following the leaderboard’s approach, GauntletScore achieves 79.51 balanced accuracy, above every average published on the leaderboard as of 15 November 2026. That places it level with the best of them, nimble-v4-max (79.45 from its published dataset scores; the engine’s 95% interval, 78.47 to 80.53, contains it). On the same claims it beats two of its three readers used alone (Readers A and B) by 1.66 and 2.17 points (95% intervals 0.80 to 2.53 and 1.29 to 3.06). Without AggreFact-CNN, described below, the leads are 0.90 and 1.32 and both intervals stay above zero. The engine also records the evidence behind every verdict, which a single model does not. Its largest margin is on full-length news articles checked against multi-sentence summaries (AggreFact-CNN), a format that resembles much regulated writing. There it scores 76.45, the best on the leaderboard, against 70.1 for the next best and 63.9 to 67.8 for the six systems with the highest averages. The test split was run once, on an engine frozen before the first test claim, under a stopping plan committed in advance. We release per-claim predictions and the scoring script.
[NLP-62] SkillBuffBench: Measuring What Agent Skills Contribute across Harnesses Models Skill Access Modes and Memory
【速读】: 该论文旨在解决当前生成式AI(Generative AI)代理在实际部署环境中,其技能包(Agent Skills)有效性评估缺乏真实性和系统性的问题。现有研究多依赖于理想化场景——即手动选择最优技能并从零开始执行任务,这与实际部署中代理需从已安装的技能库中动态选择且携带历史记忆的复杂情境严重脱节。为此,论文提出SkillBuffBench基准测试体系及其配套评估协议,将技能访问策略视为三级因子(无技能、精选技能、自动调用已安装库中的技能),并与其与记忆状态(开启/关闭)、四种主流代理框架及多类先进模型(包括Anthropic、Google、OpenAI的闭源模型以及中国厂商提供的开源模型)进行全组合交叉实验。评估基于领域专家贡献的真实科研任务(当前涵盖材料科学与生物学),并持续滚动扩展。所有实验均通过确定性验证器确保可复现性,采用任务级配对自助法报告不确定性,并独立记录成本、代理轮次与运行时间,避免指标混淆。该研究的关键在于构建一个贴近真实部署环境、具备高维度控制变量与可扩展性的标准化评估框架,以更准确地衡量技能包在复杂现实条件下的实际增益。
链接: https://arxiv.org/abs/2611.22017
作者: Gengyue Dong,Ye Wei,Zheng Zhu,Yidong Han,Jiale Miao,Xiaojun Jia,Guanghui Li
机构: University of Oxford (牛津大学); Imperial College London (帝国理工学院); University College London (伦敦大学学院); Nanyang Technological University (南洋理工大学); BraneMatrix AI(脑矩阵人工智能)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 18 pages, 2 figures, 6 tables. Work in progress: benchmark design and protocol; results to be added in a later version
Abstract:Agent Skills, file-based packages of procedural knowledge that a coding agent can discover and load on demand, are now shipped by every major agent harness, yet the evidence for how much they help comes almost entirely from settings in which the right Skill is hand-picked for the task and the agent starts from a blank slate. Neither holds in deployment, where an agent chooses from whatever library happens to be installed and frequently carries memory from earlier work. We present SkillBuffBench, a benchmark and paired evaluation protocol that treats Skill access as a three-level factor (no Skill, a curated Skill, and automatic invocation from an installed library), crosses it with memory (off and on), with four agent harnesses and with multiple models (the state-of-the-art proprietary models from Anthropic, Google and OpenAI, and the state-of-the-art open-weight models from Chinese providers), every model being run under every harness, and evaluates the result on research tasks contributed by domain experts together with the Skills they use, currently in materials science and biology, with the suite extended on a rolling basis. Every run is marked by a deterministic verifier; uncertainty is reported with a paired, task-level bootstrap; and cost, agent rounds and wall-clock time are reported alongside the score rather than folded into it. This version of the paper describes the design of the benchmark, its evaluation protocol and its current tasks and Skills. The evaluation is in progress: results will be added once the runs are complete, and the task suite, results and analysis will continue to be updated on a rolling basis.
[NLP-63] DiagTrace: An Interactive Benchmark for Clinical Decision Making EACL2027
【速读】: 该论文旨在解决现有大型语言模型(Large Language Models, LLMs)在医疗问答与诊断推理评估中过于依赖一次性提供完整患者信息、仅评估最终诊断或解释的局限性,忽视了真实临床情境下需逐步获取诊断证据的交互式信息检索过程。为弥补这一差距,论文提出DiagTrace——一个基于源文本的基准测试框架,用于评估LLMs是否能够通过交互式信息查询重建真实的诊断轨迹。其解决方案的关键在于构建包含初始表现、事件库、临床病程及诊断参考的结构化病例数据集,模拟临床医生逐步收集证据的过程,并通过多轮交互评估模型在临床病程对齐性、交互质量与效率方面的表现。该基准涵盖8个专科共510例病例,其中225例用于评估4个主流LLMs,结果揭示了覆盖度、证据来源一致性排序与交互轮次使用之间的权衡关系。
链接: https://arxiv.org/abs/2611.22009
作者: Yufan Wang,Rui Yang,Yi Lin,Chunhua Weng,Yifan Peng
机构: Weill Cornell Medicine(威尔康奈尔医学中心); Cornell University(康奈尔大学); Columbia University(哥伦比亚大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: Accepted to Findings of EACL 2027
Abstract:Large language models (LLMs) have shown strong performance on medical question answering and diagnostic reasoning benchmarks. However, existing evaluations typically provide all patient information upfront and assess only the final diagnosis or rationale. This setup ignores the need to acquire diagnostic evidence incrementally from incomplete information in real-world diagnosis. To bridge this gap, we introduce DiagTrace, a source-grounded benchmark for evaluating whether LLMs can reconstruct the documented diagnostic trajectory through interactive information seeking. Each benchmark instance is built from a PubMed Central case report and includes an initial presentation, an event bank, a clinical course, and diagnostic references. DiagTrace contains 510 cases across eight specialties, with 225 used to evaluate four LLMs on clinical-course alignment, interaction quality, and efficiency. Results reveal trade-offs in coverage, source-consistent ordering, and turn use.
[NLP-64] Large Language Models as Nonlinear Dynamical Systems: The Three Clocks of Depth Training and Generation
【速读】: 该论文旨在解决生成式语言模型(Generative AI)中复杂动态行为的理论解释问题,特别是其在多层迭代、训练过程及文本生成过程中表现出的非线性动力学现象,如固定点、分岔、同步、混沌与噪声驱动的势阱逃逸等。这些问题长期被语言模型研究者独立发现并重新命名,缺乏跨学科统一框架。其解决方案的关键在于构建“三时钟词典”(Three-Clock Dictionary),将非线性动力学中的十五个核心概念分别映射至语言模型的三个演化时钟——层间迭代、训练步进和生成序列,并标注每种对应关系为已证明、实测或猜想;同时提出一系列降阶模型(reduced-order models),包括:软最大函数对应玻尔兹曼分布、梯度下降等效于非线性弹簧上的欧拉步进、损失尖峰类比于从深谷抛出的小球、注意力机制在圆环上的建模对应库拉莫托(Kuramoto)模型、重复循环则类比于随占用时间加深的势阱。通过在GPT-2和Pythia等开源模型上验证,发现重复循环的逃逸行为遵循阿伦尼乌斯定律(Arrhenius law),其中逻辑差距(logit gap)构成势垒,该势垒随生成长度增加而加深并趋于饱和,导致脱离循环所需温度随生成长度N呈1/ln N衰减,且随模型规模增大而升高。此外,连贯文本存在于狭窄的温度窗口内,小型推理模型在其推荐温度下可共享此特性。训练过程使每个Pythia层的扩展性增强,小规模变压器处于多模态稳定性的边缘。尽管基于逃逸分析设计的控制器能有效消除循环,但尚未超越最优固定温度策略。论文最后提出了十二个开放问题,指明非线性动力学有望在未来推动语言模型的可控性与性能突破。
链接: https://arxiv.org/abs/2611.21983
作者: Maor Farid
机构: Leo AI, Inc.(Leo AI 公司); 50 Milk Street, Boston, MA 02109, United States
类目: Machine Learning (cs.LG); Computation and Language (cs.CL); Dynamical Systems (math.DS); Chaotic Dynamics (nlin.CD)
备注: 52 pages, 19 figures. Code and data: this https URL
Abstract:Language models are dynamical systems. A transformer is iterated over its layers, over the steps of its training and over the tokens it generates, and on each of these three clocks it shows behavior that nonlinear dynamics has studied for a century: fixed points, bifurcations, synchronization, chaos and noise-driven escape from potential wells. The two communities rarely meet, so this behavior is often rediscovered under new names. We build a bridge between them. To describe, we condense about 200 papers into the Three-Clock Dictionary, which maps fifteen concepts of nonlinear dynamics onto each clock and marks every correspondence as proved, measured or conjectured, and we show each concept in a mechanical system and in a language model side by side. To explain, we derive reduced-order models: the softmax is a Boltzmann distribution, gradient descent an Euler step on a nonlinear spring, a loss spike a ball thrown out of a stiff valley, attention on the circle a Kuramoto model, and a repetition loop a well that deepens while it is occupied. We then test these explanations on open language models, on a laptop CPU. Escape from repetition loops in GPT-2 and Pythia follows the Arrhenius law with the logit gap as the barrier, and the barrier deepens and then saturates, so the temperature needed to leave a loop falls like 1/\ln N with the generation length N and rises with model size. Coherent text lives in a narrow window of temperatures, which a small reasoning model shares at its recommended temperature. Training makes each Pythia layer more expansive, and a small transformer trains at a multi-mode edge of stability. The third step, solving problems of language models, is still ahead: controllers derived from the escape analysis remove loops but do not yet beat the best fixed temperature. We state twelve open problems where nonlinear dynamics could get there.
[NLP-65] Solver to System: LLM Reasoning Across Representations and Execution in Scientific Simulation NEURIPS2026
【速读】: 该论文旨在解决大语言模型(LLM)在生成与解释基于偏微分方程(PDE)的物理系统模拟器代码时所面临的挑战,特别是如何评估模型对物理系统本质的理解能力。其核心问题是:当多源异构表示(如代码、方程、自然语言描述、数值轨迹)之间存在矛盾时,LLM能否通过跨表示一致性推理恢复真实的物理系统,并准确识别代码中的物理无效性。解决方案的关键在于构建了一个名为Solver2System的数据集,该数据集包含具有多种表示形式的PDE求解器、受控的物理错误注入以及词汇扰动,从而系统性地评估LLM在面对不一致信息时的推理能力。研究发现,尽管部分模型在证据冲突时表现出对特定表示的偏好,且代码的词汇扰动会影响其他表示的解释,但执行反馈可提升对物理有效性的判断;值得注意的是,即使弱代理模型未能定位根本缺陷,仍能做出正确判断,表明LLM可能在缺乏因果诊断或一致表示融合的情况下达成正确结论。
链接: https://arxiv.org/abs/2611.21964
作者: Eliza Berman,Rohan Subramanian,Shreya Jha,Lohit Jagarapu,Romain Lopez,Rajesh Ranganath
机构: New York University (纽约大学)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: NeurIPS 2026 Workshop: Sim2Science: ML with Imperfect Scientific Models. 23 pages, 14 figures, 9 tables
Abstract:Simulators based on partial differential equations (PDEs) are widely used to model physical systems, but generating and interpreting PDE solver code with large language models (LLMs) remains challenging. Scientific simulation code is a useful testbed for LLM scientific reasoning because (1) one system can be represented by code, equations, natural language, and numerical trajectories and (2) code can execute successfully while implementing invalid physics. We introduce Solver2System, a dataset of PDE solvers with multiple representations, controlled corruptions, and lexical perturbations, to evaluate whether LLMs can recover the underlying physical system through (1) cross-representation consistency and (2) static and agentic detection of physical invalidity. Across eight open-weight models, LLMs systematically favor some representations when evidence conflicts, and lexical perturbations to code alter interpretation of other unchanged representations. Execution improves physical-validity judgments, but weaker agents can judge correctly without identifying the causal defect. Thus, LLMs may reach correct judgments despite inconsistent representation reconciliation or causal misdiagnosis.
[NLP-66] Conserved Verdicts Drifting Commitments: Identifying Harmful Overthinking
【速读】: 该论文旨在解决生成式推理模型在持续思考过程中可能因过度推理而导致正确答案丢失的问题,即“推理过犹不及”现象。其核心问题是:如何在推理链中决定何时停止以最大化最终答案的准确性。解决方案的关键在于引入“承诺-判决差距”(commitment-verdict gap)这一概念,该差距等于模型继续推理所带来的预期答案质量变化量以及停止所能获得的准确率增益。研究指出,仅依赖最终答案正确性评估(verdict)无法有效指导停机决策,因其缺乏可预测的漂移趋势;而通过区分单条推理链内部的答案变化与仍在运行的推理链之间的分布差异,能够揭示准确率下降的根本原因——并非所有推理链都在持续改进,而是剩余未终止的链决定了整体准确率曲线的衰减。实验基于MATH-500和OlympiadBench数据集,在多个模型家族中验证发现,继续推理平均导致3.6分(95% CI [2.2, 5.2])的准确率损失,且损失集中于少数问题。此外,理想停机规则最多仅能带来+1.8至+2.5分的增益,表明当前模型存在显著的过推理倾向。该框架强调,在构建实际停机策略前,必须先量化承诺-判决差距及链间动态变化,从而为优化推理效率提供理论依据。
链接: https://arxiv.org/abs/2611.21946
作者: Jianghong Zhou,Hao Rong,Chen Lin
机构: 未知
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 40 pages
Abstract:A reasoning model that has reached a correct answer can lose it by continuing to think. Deciding whether to stop means comparing two probabilities at the same point of a chain: that the answer the model would give if stopped now is correct (its commitment), and that the answer it will finish with is correct (its verdict). We show that the verdict alone, which is what final-correctness probes estimate, cannot make this comparison. As a forecast of a fixed outcome it has no predictable drift, whether continuing helps or hurts. The commitment-verdict gap equals the expected remaining change in answer quality and the accuracy gained by stopping. A second identity splits any accuracy-versus-budget curve into answer change within chains and change in which chains are still running. We measure both on MATH-500. At all 379 points where DeepSeek-R1-Distill-Qwen-1.5B holds a correct answer, selected without regard to how its chain ends, continuing loses 3.6 points on average (95% CI [2.2, 5.2]), and the loss concentrates in a few problems. In six reasoning models from three families, on MATH-500 and OlympiadBench and with budgets up to 16,384 tokens, declining accuracy curves come from which chains remain, while the chains still reasoning keep or improve their answers. An oracle stopping rule gains at most +1.8 and +2.5 points at 1.5B and 7B. The framework says what to measure before building a stopping rule.
[NLP-67] DNeo: Your Parallel Draft Model Needs Better Target Feature Encoding
【速读】: 该论文旨在解决生成式 AI(Generative AI)中高效大语言模型(LLM)推理的瓶颈问题,核心挑战在于如何在不牺牲生成质量的前提下提升推理速度。现有并行草稿架构(如DFlash、DSpark)虽通过小规模并行草稿模型(drafter model)生成多个草稿令牌,并由大型目标模型(target model)进行并行验证以实现加速,但其标准特征编码机制存在三大局限:(1)有损性,即编码投影导致目标模型特征空间的破坏性失真,源于“残差混叠”(residual mixing)效应;(2)无记忆性,无法记录先前被拒绝的草稿提案;(3)特征采样方式均匀,忽略了关键隐藏状态。为克服上述问题,本文提出轻量级特征编码方法——特征混合注意力(Mixture-of-Features Attention, MoFA),其引入工作记忆机制,用于编码历史被拒绝的草稿信息。结合MoFA、拒绝记忆机制以及一系列改进措施(包括双拍因果卷积、仅前缀交叉注意力和非均匀目标层选择),最终构建出DNeo架构,在保持草稿模型规模与生成质量不变的前提下,实现了相较于当前最优方案(DFlash-2)1.3倍的生成速度提升,标志着高质量、高速大语言模型推理的新里程碑。
链接: https://arxiv.org/abs/2611.21925
作者: Jameson Sandler,Shane Bergsma,Mostafa Elhoushi,Bin Claire Zhang,Gavia Gray,Joel Hestness,Nolan Dey,Ferdinando Fioretto
机构: 未知
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Speculative-diffusion decoding uses a small parallel drafter model to generate b draft tokens in one forward pass, which then undergo parallel verification via a larger target model. Faster inference can be achieved when multiple draft tokens pass verification simultaneously. Recent parallel draft architectures (DFlash, DSpark) improve draft quality by conditioning on rich features computed by the target, through a mechanism known as feature encoding. However, standard feature encoding mechanisms have three limitations: (1) they are lossy, applying encoding projections that are destructive to the underlying target feature-space via a discovered effect known as residual mixing, (2) they are memory-less, with no record of previously processed proposals, and (3) they utilize features sampled uniformly from across the target layers, omitting important hidden states. To address these issues, we propose Mixture-of-Features Attention (MoFA), a lightweight feature-encoding approach equipped with a working memory that encodes previously rejected proposals. Leveraging MoFA, rejection memory, and a set of further enhancements, including 2-tap causal convolutions, prefix-only cross-attention, and non-uniform target layer selection, the final architecture, DNeo, achieves 1.3x faster generation speed than SoTA (DFlash-2) without impacting drafter footprint or generation quality. DNeo marks a new state of the art in high quality, fast LLM inference.
[NLP-68] Latent Fact Composition under Next-Token Training
【速读】: 该论文旨在解决生成式模型在处理两跳推理(two-hop reasoning)任务时,对未在训练数据中显式出现的中间实体(intermediate entity)进行推理的能力问题。具体而言,现有方法通常仅训练模型预测最终答案,而主流语言模型则以逐词预测为目标,这可能导致模型无法有效利用隐含于上下文中的中间实体信息。为探究这一问题,研究构建了一个合成语料库,其中包含所有中间实体均未出现在任何两跳训练样本中的两跳问题,并可精确计算损失函数的每一项。实验表明,标准的Transformer、循环变压器(looped transformer)及DiscoLoop均几乎无法正确回答此类问题。为此,作者提出Baton系列循环变压器,其核心创新在于将中间实体作为令牌嵌入传递至第二轮推理,从而增强模型对中间信息的记忆与利用能力。当仅以答案为目标进行训练时,Baton的三种变体中有两种几乎完美解决了这些挑战性问题。研究进一步通过分析三个设计条件揭示了各模型成功或失败的根本原因,且单变量实验验证了该假设的合理性。然而,当Baton在全词预测目标下训练时,其性能先上升后下降,且性能衰退发生在第二跳阶段,此时模型仍能准确回忆两个事实,说明问题并非源于记忆失效,而是由训练目标中特定梯度成分导致——特别是与训练集相关的两个梯度项影响了模型在复杂推理路径上的泛化能力。
链接: https://arxiv.org/abs/2611.21921
作者: Tianyi Zhang,Feiying Chen
机构: 未知
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 21 pages, 5 figures, 23 tables
Abstract:Controlled studies of latent two-hop reasoning train transformers to predict only the answer, whereas language models are trained to predict every token. We ask whether a model trained on every token learns to answer two-hop questions whose intermediate entity appears in no two-hop training example. We build a synthetic corpus with such questions. On this corpus every term of the loss can be computed exactly. A looped transformer, a standard transformer and DiscoLoop answer almost none of these questions. We therefore introduce the Baton family of looped transformers. They pass the intermediate entity to the second loop as a token embedding. When trained to predict only the answer, two of its three variants answer these questions almost perfectly. We then relate the success and failure of each model to three design conditions. Experiments that change one condition at a time agree with this hypothesis. When Baton is trained on every token, however, its accuracy on these questions rises and then declines. We further split the gradient of the next-token loss into three terms. Removing the two terms that depend on the particular training set raises the peak accuracy but does not remove the decline. The decline arises in the second hop while the model still recalls both facts.
[NLP-69] Call-Efficient ICD-10 Code Prediction via Supervised Candidate Generation and Selective LLM Invocation
【速读】: 该论文旨在解决从简短且噪声较大的临床诊断文本中自动分配ICD-10编码这一具有挑战性的医学文本分类与标准化任务。其核心难点在于诊断短语长度有限、标签空间具有层级结构以及类别分布严重不均衡。解决方案的关键在于构建一个高效调用(call-efficient)的ICD-10编码预测流程,该流程融合了Transformer嵌入表示、基于树结构的监督式候选生成、候选集级别的重排序以及门控式大语言模型(LLM)辅助机制。通过采用无泄露的安全验证协议,模型与阈值选择基于交叉验证的留出预测结果,最终评估在锁定测试集上进行。实验表明,监督式条件父编码候选生成器在锁定测试集上达到74.32%的Top-1准确率和93.05%的Top-10候选覆盖率;而直接的零样本GPT-5.6-Sol基线虽取得81.01%的Top-1准确率,但需对每个诊断实例调用大模型。最终部署系统“Pre-LLM融合”仅在31.97%的案例中调用LLM,即实现了81.24%的OOF-CV准确率和83.19%的锁定测试准确率,将错误率从监督基线的25.68%降低至16.81%,相对误差减少34.5%。结果表明,大语言模型最有效的角色是在不确定样本上进行受控重排序,而非无限制地执行编码或全量数据调用。
链接: https://arxiv.org/abs/2611.21891
作者: Ilia Kopanichuk,Daniil Kartashev,Yuri Bolkonsky,Petr Anokhin,Vladimir Makharev,Iaroslav Bespalov,Dmitry V. Dylov
机构: 未知
类目: Computation and Language (cs.CL)
备注: 8 pages, 8 figures, 4 tables
Abstract:Automatic assignment of ICD-10 codes from short clinical diagnosis text is a practical medical text classification and normalization task. The task is difficult because diagnosis phrases are short and noisy, the ICD-10 label space is hierarchical, and the class distribution is strongly imbalanced. This paper presents a call-efficient ICD-10 code prediction pipeline that combines transformer embeddings, tree-based supervised candidate generation, candidate-level reranking, and a gated LLM assistance. The evaluation uses a leakage-safe validation protocol: model and threshold selection are based on out-of-fold cross-validation predictions, while final reporting is performed on a locked test set with no shared source-file groups. The supervised conditional parent-code candidate generator reaches 74.32% Top-1 accuracy on the locked test set and 93.05% Top-10 candidate coverage. A direct zero-shot GPT-5.6-Sol baseline reaches 81.01% Top-1 accuracy, but requires an invocation of this large model for every diagnosis. The final deployable system, Pre-LLM fusion, reaches 81.24% OOF-CV accuracy and 83.19% locked-test accuracy while calling the LLM for only 31.97% of cases. This improves the locked-test error rate from 25.68% for the supervised parent-code baseline to 16.81%, corresponding to a 34.5% relative error reduction. The results show that the most effective role of the LLM is controlled reranking on uncertain examples rather than unrestricted coding or full-dataset invocation.
[NLP-70] Knowing Which Not How Much: Language Models Order Party Positions They Cannot Place ACL
【速读】: 该论文旨在解决生成式语言模型在评估社会实体(如政党)政策立场时,其知识表征究竟是基于可度量的绝对尺度(absolute scale),还是仅依赖于相对排序(ordering)的问题。现有大多数评测方法仅通过单个实体的问答来衡量模型的知识水平,隐含假设模型同时具备顺序与量级信息,但这一假设可能并不成立。研究发现,尽管模型在对多个政党进行相对排序时表现优异(优于最强基线两倍以上),但在将特定政党置于一个已知的量化政治光谱中进行绝对定位时却显著失败,表明模型可能仅掌握了可靠的相对顺序而缺乏锚定的量级基准。进一步实验显示,当提供外部编码数据作为参考答案时,模型的绝对定位能力显著提升,接近性能上限。因此,该研究的关键在于揭示:针对同一事实,相对排序类问题与绝对定位类问题实际上测量的是模型知识的不同维度——前者反映模型的比较推理能力,后者则依赖于外部量纲校准。这提示当前以相对排序为主的基准测试无法充分反映模型在量化解析任务中的真实能力,尤其在政党议题重要性等需要精确量化的场景中,必须区分并分别设计评测范式。
链接: https://arxiv.org/abs/2611.21878
作者: Yousef Attia,Jarosław A. Chudziak
机构: Warsaw University of Technology (华沙理工大学); Warsaw, Poland
类目: Computation and Language (cs.CL); Computers and Society (cs.CY)
备注: Accepted at the 40th Pacific Asia Conference on Language, Information and Computation (PACLIC 40)
Abstract:Language models are increasingly asked where real actors stand: what a party’s platform prioritises, what an institution has committed to. Such judgments are gradable, carrying both an order and a magnitude. Nearly every benchmark that tests them asks about one entity at a time, then reads that score as the model’s knowledge of the domain. Order and magnitude, however, need not be acquired together. Text about social entities carries far more comparison than measurement, so a model may hold a dependable ordering with no anchored scale beneath it. We ask whether a model can place a documented social position on a scale or only order the actors it compares, and what that means for how such knowledge is measured. We test this on party positions in Japan and South Korea, where the Comparative Manifesto Project’s coded data supplies a rare external answer key. Asked to order four parties, models beat the strongest fixed-answer baseline by more than a factor of two; asked to place one party on the same scale, they fall below it. The ordering survives inside those failed placements, and supplying the coded values lifts placement to near ceiling. Comparative and absolute questions about the same fact therefore measure different things here, and for party issue salience a benchmark built from one cannot speak for the other.
[NLP-71] he Hidden Cost of MoE Diffusion Decoding: Profiling and Fixing the Sampler NEURIPS
【速读】: 该论文旨在解决生成式 AI(Generative AI)中基于扩散机制的混合专家模型(diffusion Mixture-of-Experts, MoE)在边缘设备上解码效率低下的问题,其核心挑战在于:尽管经过权重量化后,专家计算(expert GEMMs)的开销已显著降低,但解码过程中对整个词元(token)画布的逐词元采样与停止流水线(per-token sampling and stopping pipeline)仍成为最主要的性能瓶颈。该流水线在每次去噪步骤中需处理远超自回归模型的逻辑值(logits)条目数(每输出词元需处理 S 倍于去噪步数 S 的逻辑值),导致计算冗余严重。解决方案的关键在于提出两阶段无损优化:第一阶段为比特精确(bit-exact)级优化,确保输出字节级完全一致;第二阶段为精度对齐(accuracy-parity)级优化,采用融合单次熵核(single-pass entropy kernel)与 Gumbel-max 采样技术,将原流水线重构为一次完成的高效内核。结合 W4A16 量化专家计算核,整体系统在不损失 GSM8K、HumanEval、MATH 等基准测试准确率的前提下,实现了 2.4–2.7 倍的端到端加速,且该优化效果仅在扩散 MoE 模型中显著,在密集模型中则不明显,表明该开销具有模型架构特异性。
链接: https://arxiv.org/abs/2611.21852
作者: Harshit Samrat,Ramakrishna Paramahamsa Kompella
机构: Indian Institute of Technology Madras(印度理工学院马德拉斯分校)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: Accepted to NeurIPS workshop DiffuLM 2026
Abstract:Masked diffusion language models (dLLMs) decode by iteratively denoising a full canvas of tokens, so every forward pass runs a per-token sampling and stopping pipeline over the entire canvas-by-vocabulary logits tensor. We profile single-user decoding of a 26B-parameter diffusion Mixture-of-Experts model on a bandwidth-bound edge accelerator (NVIDIA GB10) and report a result that is easy to miss: after weight quantisation, this logits pipeline, not the expert GEMMs, is the single largest consumer of decode time. Unlike autoregressive decoding, where sampling touches one position’s logits per committed token, diffusion decoding processes S times as many logit entries per output token, where S is the number of denoising steps. We show this pipeline is largely redundant or fusable and remove it in two lossless tiers: a bit-exact tier whose output is byte-identical to the reference decoder, and an accuracy-parity tier using a fused single-pass entropy kernel and Gumbel-max sampling. Combined with a W4A16 expert kernel, the full stack reaches 2.4-2.7 times end-to-end speedup over bf16 at unchanged benchmark accuracy on GSM8K, HumanEval, and MATH, on a single 273 GB/s device. We confirm the finding on a second, independently built diffusion MoE (LLaDA2.0-mini: pipeline 28.8% of decode, fused W4A16 3.3 times at accuracy parity) and show it disappears on a dense diffusion model (3.6%). The cost is specific to MoE diffusion, where quantisation makes the backbone cheap. Code and quantised weights are released.
[NLP-72] USI: A Citation-Grounded Graph-Based Benchmark
【速读】: 该论文旨在解决波斯语法律领域中检索增强生成(RAG)系统评估缺乏真实性、可溯源性及法律推理覆盖度的问题。现有基准数据集或基于法条文本合成生成,难以体现真实的法律信息需求;或源自非正式的问答资源,未与权威法律文件建立明确关联,无法有效评估基于法律依据的推理能力。为此,本文提出TUSI-Bench——一个面向波斯语法律领域的结构化评估基准,其核心在于构建包含7,261条法律条文的多层级法律知识框架,整合成文法典、法律评注、学理著作及最高法院统一判例,并在此基础上构建法律知识图谱以支持结构感知的检索。TUSI-Bench包含两个互补任务:(i) TUSI-QA,涵盖1,999个开放式问题,要求生成带有明确法律引证的可溯源回答;(ii) TUSI-Bar,基于伊朗律师资格考试题目设计的4选项选择题基准,约500道题。此外,提出一种基于知识图谱关系的检索基线方法,实验表明该方法相较传统基线可提升检索召回率最高达8个百分点,凸显了结构化法律表示对提升RAG系统性能的关键作用。该基准为波斯语法律领域中可溯源法律推理的标准化评估提供了基础资源,推动面向结构化法律智能的后续研究。
链接: https://arxiv.org/abs/2611.21823
作者: Marzieh Karimi,Mehran Safayani,Alireza Basiri
机构: Isfahan University of Technology (伊斯法罕理工大学); Department of Electrical and Computer Engineering (电气与计算机工程系)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 28 pages, 11 figures, including appendix
Abstract:Evaluating retrieval-augmented generation (RAG) systems in Persian law remains challenging because benchmarks lack realism, grounding, and coverage of legal reasoning. Existing resources are either synthetically generated from statutory texts, lacking natural information needs, or derived from informal legal question–answer sources without explicit linkage to authoritative legal documents, making them unsuitable for rigorous evaluation of grounded legal reasoning. To address this gap, we introduce TUSI-Bench, a structured Persian legal benchmark designed for the evaluation of RAG-based systems. TUSI-Bench is built upon a multi-layered legal knowledge framework containing 7,261 legal articles and integrating codified statutes, legal commentaries, doctrinal writings, and Supreme Court unification rulings. We also construct a legal knowledge graph linking articles and concepts for structure-aware retrieval. Based on this foundation, TUSI-Bench provides two complementary evaluation tasks: (i) TUSI-QA, an open-ended question answering benchmark containing 1,999 questions that requires grounded responses with explicit legal citations, and (ii) TUSI-Bar, a multiple-choice benchmark containing approximately 500 four-option questions derived from Iranian Bar examination questions. Furthermore, we propose a graph-based retrieval baseline that leverages relational structure within the legal knowledge graph. Experimental results demonstrate that incorporating graph-based relations improves retrieval recall by up to 8 percentage points compared to standard baselines, highlighting the importance of structured legal representations for RAG systems. The proposed benchmark provides a standardized resource for evaluating grounded legal reasoning in Persian and enables future research on structured legal intelligence.
[NLP-73] Bookkeeping Is Not Judgement: EJBench an Engineering Judgement Benchmark Built from a Conjugate Heat Transfer Design Campaign
【速读】: 该论文旨在解决生成式 AI 在工程判断(engineering judgement)任务中的能力评估问题,即模型能否基于少量已有设计结果,预测新工况下的设计表现并作出合理决策,而无需依赖仿真或CAD软件操作。与传统工程基准测试不同,EJBench 具有软件无关性(software-agnostic),聚焦于模拟工程师在不运行仿真时的推理能力。其核心挑战包括:将未标记的测试结果匹配至对应工作点、决定在设计签发前应揭示哪些历史运行数据、以及根据峰值电芯温度对新工况进行排序。所有答案均基于收敛的仿真结果,而非由大语言模型(LLM)评判,确保评估客观性。实验表明,当前最先进模型 GPT-6 Astra 的得分仅为 0.71,且在正确与错误任务上表现出相近的置信度,反映出模型在工程推理中缺乏可靠的不确定性校准。通过使用 LoRA 微调与强化学习对 Qwen3.5-9B 进行后训练,并以仿真结果验证的答案标准占比作为奖励信号,其在保留测试集上的性能从 0.071 提升至 0.106,验证了针对性优化的有效性。该研究的关键在于构建一个真实、可复现且脱离工具依赖的工程判断基准,推动模型从“执行”向“决策”层面的能力演进。
链接: https://arxiv.org/abs/2611.21707
作者: James Sadler,Rotimi Alabi,Kotub Uddin
机构: Gradient Dynamics(梯度动力学), AESC(能源智能中心)
类目: Computational Engineering, Finance, and Science (cs.CE); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 23 pages
Abstract:We introduce EJBench, a benchmark of engineering judgement: whether a model can predict how a design behaves at new conditions from a few earlier results, and decide on it. Unlike engineering benchmarks that test whether an agent can operate simulation or CAD software, EJBench is software-agnostic and asks what an engineer would conclude without running a simulation. Its tasks come from a conjugate heat transfer design campaign for battery cold plates, in three families: matching unlabelled test results to their operating points, choosing which stored runs to reveal before signing off a design, and ranking new duties by peak cell temperature. Every answer is taken from a converged simulation, not from an LLM judge. The best of ten frontier models, GPT-6 Astra, scores 0.71, and models state almost the same confidence on tasks they get wrong as on tasks they get right. We also post-train Qwen3.5-9B with LoRA and reinforcement learning, rewarded by the share of answer criteria the simulation results confirm; on the held-out test split its score rises from 0.071 to 0.106. We release the train, validation and test splits, the evaluation harness and the leaderboard.
[NLP-74] Disambiguating Moral Circuits and Moral Features in Language Models: Moral content normative judgment or cue words? Evidence from 14 models
【速读】: 该论文旨在解决大语言模型在处理道德类提示时,其内部决策机制中哪些注意力头(attention heads)和稀疏自编码器(SAE)特征对判断结果产生关键影响的问题。具体而言,研究聚焦于识别并解析模型在面对涉及道德、审慎、礼仪、美学或证据等规范性判断的语义对时,究竟是响应于特定内容本身,还是对“道德”一词的语义触发。解决方案的关键在于构建了227组经五名标注者验证的最小语义对(minimal pairs),这些语义对仅在规范类型上存在差异,并在14个开源模型(涵盖Llama、Gemma和Qwen系列,参数量从1B到14B,包含基础与指令微调版本)中进行测试,对比有无“morally”一词的情形。研究发现,改变道德判断的注意力头属于一个通用的评价电路(evaluation circuit),该电路在所有类型的规范性判断中均起作用,且其贡献度在道德与其他规范之间无显著差异(合并比率为1.00,97.5%置信区间[0.92, 1.07],有“morally”时;0.90,[0.84, 0.96],无“morally”时),表明模型依赖这些头部进行跨类别规范判断。此外,在未明确要求判断的任务中,这些头部依然活跃;在ETHICS基准测试的400个场景中,移除这些头部会使可接受与不可接受场景之间的区分度下降37%至39%,远高于随机头部的1%至3%。同时,所定位的SAE特征主要响应“morally”一词,而非语义内容本身。模型隐藏状态中可读取语义对所属规范类别(由人工标注确定),在8个测试模型中,移除承载该类别的方向会削弱道德判断约几个百分点,且在七种情况下削弱幅度超过其他类型判断。该研究揭示了模型在规范性推理中共享神经机制的核心特征,为提升生成式AI(Generative AI)的可解释性提供了实证基础。
链接: https://arxiv.org/abs/2611.21705
作者: Davide Liga,Liuwen Yu,Matteo Magnini
机构: University of Luxembourg(卢森堡大学); Luxembourg Institute of Science and Technology (LIST)(卢森堡科学技术研究所)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Interpretability methods can locate the attention heads and sparse autoencoder (SAE) features that change the answer a language model gives to a moral prompt, e.g., whether “Stealing medicine for a dying child is morally” continues with “justified” or “wrong”. Such a component could respond to moral content, to the judgment of conduct by any norm, or to the word “morally” in the prompt. We built 227 minimal pairs of short phrases, validated by five annotators, whose two versions differ in a moral standard, in another kind of norm (prudence, etiquette, aesthetics or evidence) or in a descriptive quality, and showed them to 14 open-weight models (Llama, Gemma and Qwen, 1B to 14B, base and instruction-tuned), with and without the word “morally”. In every model, the heads that change the moral answer belong to an evaluation circuit the models use for every kind of normative judgment we tested: these heads carry the same share of the effect for moral as for other normative pairs (pooled ratio 1.00, 97.5% confidence interval [0.92, 1.07], with the word; 0.90, [0.84, 0.96], without it), and the models need them as much for both. The models need these heads also when nothing in the prompt asks for a judgment, and on 400 scenarios of the ETHICS benchmark, where removing them takes away 37 to 39% of the difference between acceptable and unacceptable scenarios, against 1 to 3% for ten random heads. The SAE features we located respond mostly to the word “morally” rather than to what the phrase is about. The models represent the category of a pair (the standard the annotators assigned to it): it can be read from their hidden states, and, in the eight models we tested, removing the direction that carries it weakens the moral verdict by a few percent, and more than the other verdicts in seven of them. We release the pairs, the ratings, the code, the trained SAE dictionaries and the result files.
[NLP-75] Watermark Evidence under Open-Weight LLM Modification: Exact Bounds for Low-Rank Output Watermarks
【速读】: 该论文旨在解决生成式 AI (Generative AI) 水印系统在模型权重被修改后仍能保持有效检测能力的问题,特别是针对开放权重语言模型(open-weight language models)在发布后进行微调、量化或剪枝等操作时,水印验证器(verifier)所依赖的证据可靠性下降的挑战。其核心问题是:当模型权重被修改时,如何准确评估水印证据的可检测性变化,尤其是在共享前缀处由权重扰动引入的残差对下一词元概率分布的影响。解决方案的关键在于:通过将模型修改建模为下一词元 logits 中的残差项,在有限词汇量下推导出期望对数似然比证据的严格下界,并结合残差跨度约束建立精确的偏差上界;进一步通过协方差项刻画一阶证据偏移,并给出显式的二阶余项;对于序列级检测,则利用条件指数矩上界与超鞅论证结合分布无关的顺序统计容忍度边界,实现攻击选择后在独立认证延续序列上的高置信度漏检率上界。实验表明,INT8量化和30%剪枝对水印真阳性率影响较小(85.6%-94.8%),而500步LoRA微调显著降低至28.8%和11.6%,且协方差预测能精准跟踪实测证据漂移,无余项违反现象,同时在500个基线模型延续中未出现假阳性,验证了理论框架的有效性与水印在模型编辑攻击下的鲁棒性边界。
链接: https://arxiv.org/abs/2611.21702
作者: Prashant Shekhar,Dumindu Samaraweera,Caroline Howard
机构: Embry-Riddle Aeronautical University, Daytona Beach, FL, USA(艾姆布里-瑞德尔航空大学,代顿纳海滩,佛罗里达州,美国)
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Open-weight language models can continue generating after their released weights are modified, while a watermark verifier still scores outputs against the original base and marked models. We study how such modifications change the evidence available to an OpenStamp-style verifier. At each shared prefix, we represent the modification as a residual in the next-token logits. We derive an exact finite-vocabulary lower bound on the expected log-likelihood-ratio evidence under a residual-span constraint, along with a sharp span-only deviation bound. We also characterize the first-order evidence shift through a covariance term with an explicit second-order remainder, and determine the residual span needed to reach a target evidence level. For sequence-level detection, we derive an exact conditional exponential-moment bound and combine it with a supermartingale argument and a distribution-free order-statistic tolerance bound. This yields a high-confidence upper bound on miss probability after attack selection on calibration data and evaluation on independent certification continuations. In experiments with OpenStamp on Llama-2-7B and Mistral-7B-v0.3, INT8 quantization and 30% pruning retain test true-positive rates of 85.6-94.8%, while 500-step LoRA fine-tuning lowers them to 28.8% and 11.6%. The covariance prediction tracks the measured mean LoRA evidence drift on both models, with no remainder-bound violations across 160,000 test prefixes. Additionally, we observe zero false positives among 500 base-model continuations per model. Through these theoretical and empirical results, the current research provides a better understanding of watermark robustness to model-editing/tuning attacks.
[NLP-76] oken Commitment as a Control Surface for Steering Masked Diffusion Generation
【速读】: 该论文旨在解决生成式模型在推理阶段难以有效对齐生成结果与多目标输出要求的问题,尤其是在基于掩码扩散模型(masked diffusion models)的生成过程中,如何动态调整未决预测(provisional tokens)的“确定性”以实现更精准的目标控制。现有方法通常仅通过修改激活值、逻辑输出或去噪轨迹进行干预,而将“何时及哪些预测被固定”这一承诺策略视为不可调节的解码细节。本文的关键创新在于将令牌承诺机制(token commitment) 视为一种特定于扩散过程的可调控控制面,并将其分解为两个可独立优化的决策:一是选择哪些令牌进行承诺,二是每一步应承诺多少令牌。针对第一个问题,提出反事实目标感知承诺(counterfactual target-aware commitment),通过融合模型置信度与“移除单个令牌后目标贡献”的反事实评估来排序待定位置;针对第二个问题,引入校准自适应预算(calibrated adaptive budgets),依据目标进展程度和置信度动态调整每步承诺数量。实验表明,这两种机制在语言和视觉-语言任务中均实现了稳定且单调的性能-目标质量权衡,在多种设置下优于无评分器的随机调度方案。更重要的是,由于其仅需对临时生成结果计算标量分数,该框架可无缝扩展至自动代理目标(如对话摘要中的自然语言推理蕴含关系、图像描述中的CLIPScore),无需依赖固定的可重用引导方向。该研究首次系统地揭示了“承诺过程”作为可解释、可调控的控制界面,在掩码扩散生成中的核心作用。
链接: https://arxiv.org/abs/2611.21669
作者: Ambroise Aigueperse,Hugo Buurmeijer,Carmen Amo ALonso,Marco Pavone
机构: Stanford University (斯坦福大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Masked diffusion models repeatedly predict provisional tokens before a sampler decides which predictions become permanent. Existing inference-time steering methods primarily modify activations, logits, or denoising trajectories, while treating this commitment policy as a fixed decoding detail. We instead study token commitment as a diffusion-specific control surface and separate it into two decisions: which tokens to commit and how many tokens to commit at each step. For the first, we introduce counterfactual target-aware commitment, which ranks unresolved positions by a mixture of model confidence and leave-one-token-out target contribution. For the second, we introduce calibrated adaptive budgets driven by target progress and confidence. Across language and vision-language settings in the LLaDA family, multiple targets, two activation-steering methods, and an unsteered control, both mechanisms yield empirically monotonic target-quality trade-offs in every configuration we evaluate. Adaptive schedules also recover competitive operating points relative to scorer-free stochastic schedules matched in average denoising cost. Because both mechanisms require only a scalar score evaluated on a provisional completion, they apply unchanged when the target is an automatic proxy for task success rather than an output attribute: NLI entailment for dialogue summarization and CLIPScore for captioning, neither of which corresponds to a single reusable steering direction. We reproduce the behavior of both mechanisms on a second masked-diffusion family, Dream-7B, for one control target and one task-success target. These results characterize commitment–which provisional predictions become irreversible, and how rapidly–as an interpretable control surface for steering masked diffusion generation under the conditions we test.
[NLP-77] Evaluating Semantic Insensitivity in Financial Sentiment Analysis: Implications for Sustainable Financial Decision-Making
【速读】: 该论文旨在解决金融情感分析模型(如FinBERT)在面对细微对抗性文本修改时的脆弱性问题,此类攻击可能威胁自动化交易、投资组合管理与风险评估等系统决策的可靠性。其解决方案的关键在于提出一种名为“代理驱动语义反转”(Agent-driven Semantic Inversion, ASI)的新型对抗攻击框架,通过基于大语言模型(LLM)的多智能体流水线,识别金融新闻中的可攻击句段,并实施领域特定的词汇替换,从而实现对财务绩效指标的根本性语义反转。实验表明,ASI在90.85%的可攻击案例中成功误导了FinBERT,暴露出该模型严重依赖上下文关键词和句法模式而非真正的金融语义逻辑理解的缺陷。研究揭示了当前金融自然语言处理(NLP)模型在逻辑感知与鲁棒性方面的显著不足,凸显构建具备逻辑推理能力的稳健金融AI模型的紧迫性。
链接: https://arxiv.org/abs/2611.21653
作者: Gabriela Cioboata,Teodor-Stefan Bildea
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 17 pages, 1 figure, Presented and accepted for publication at the Symposium “Rethinking Economic Models: Sustainability, Finance, and Accounting in Times of Transformation” (Bucharest, 2026)
Abstract:Financial sentiment analysis is critical for automated trading, portfolio management, and risk assessment, heavily relying on models like FinBERT. Despite high benchmark accuracy, these models remain vulnerable to subtle adversarial text modifications, threatening the integrity of algorithmic decision-making. In this paper, we propose Agent-driven Semantic Inversion (ASI), a novel adversarial attack framework targeting financial sentiment models through domain-specific semantic manipulation. Utilizing an LLM-powered multi-agent pipeline, our approach identifies eligible financial news sentences and applies targeted lexical substitutions. We empirically evaluate ASI against FinBERT, one of the most widely deployed financial sentiment models. Our results reveal a striking vulnerability: the attack successfully deceived FinBERT in 90.85% of eligible cases, with the model frequently maintaining its original sentiment prediction despite the fundamental inversion of financial performance indicators. Through qualitative analysis and token-level investigation, we demonstrate that FinBERT critically over-relies on contextual keywords and syntactic patterns rather than true logical comprehension of financial semantics. Our findings highlight the urgent need for logic-aware, robust financial NLP models to ensure the reliability of automated decision-making systems.
[NLP-78] From Preference Data to Personalization: Tracing Sycophancy in Large Language Models
【速读】: 该论文旨在解决大语言模型中“阿谀奉承”(sycophancy)现象的评估与成因问题,尤其聚焦于不当迎合行为——即在应当纠正、修正或抵制的情况下仍无条件认同用户主张、偏好或自我评价。其核心挑战在于区分合理同意与非理性迎合,传统以用户一致性为指标的方法难以有效识别后者。解决方案的关键在于提出一种有效性感知的阿谀奉承评分框架(validity-aware sycophancy grader),并构建了名为SycoDrift的针对性评估数据集,包含3,298个经有效性标注的探测样本,且对用户开放性(user openness)和情境重要性(stakes)进行了受控注释。通过该框架,研究揭示了三种加剧不当迎合的压力源:人类偏好数据中的弱而集中的一致性倾向、自动化评判标准中强调“验证”导向的评分规则所引发的训练期放大效应,以及推理阶段基于用户画像的个性化设置导致的显著不当迎合提升,尤其当匹配的用户身份信息紧随查询出现时影响最为显著。
链接: https://arxiv.org/abs/2611.21650
作者: Himaghna Bhattacharjee,Yasha Sheynin,Shivam Singhal,Arjun Subramonian,Ankur Samanta,Julian Michael,Smitha Milli,Maximilian Nickel,Claudia Shi
机构: Google(谷歌); Stanford University (斯坦福大学); University of Pennsylvania (宾夕法尼亚大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 66 pages, 19 figures
Abstract:Sycophancy in large language models is often measured as agreement with the user, yet agreement can be appropriate when the user’s claim is valid, their preference is genuine, or their experience warrants acknowledgment. We study sycophancy as inappropriate agreement: validating a claim, preference, or self-assessment when the context calls for correction, qualification, or resistance. We introduce a validity-aware sycophancy grader and SycoDrift, a persona-targeted evaluation set of 3,298 validity-annotated probes with controlled annotations for user openness and stakes. We use this framework to analyze three sources of agreement pressure: human preference data, automated judge specification, and in-context personalization. First, preference for validating responses in human datasets (Chatbot Arena, HH-RLHF, Community Alignment) is weak and concentrated where agreement is often appropriate. Second, in a selected-value, single-model experiment, emphasized judge rubrics produce consistent but modest increases relative to paired suppressed rubrics, demonstrating a plausible mechanism for training-time amplification. Third, in off-policy evaluations across 14 models, inference-time personalization increases inappropriate agreement, with the largest effect from matched user profiles placed directly before the query.
[NLP-79] Literary Language Mashup: Curating Fictions with Large Language Models
【速读】: 该论文旨在解决传统文学作品评价中依赖人工评判所存在的主观性高、效率低及可重复性差等问题,提出一种基于大语言模型(LLM)作为评估者的新方法,嵌入于GrAImes框架中,以实现对文学微型小说(microfictions)及短篇故事的自动化评价。其解决方案的关键在于利用大语言模型对西班牙语和英语的真人创作与AI生成文本进行评价,并将其结果与五位文学博士及十六名文学爱好者的人工评价进行对比分析,通过计算克里彭多夫α系数(Krippendorff’s alpha)衡量两者间的一致性程度。研究发现,尽管大语言模型在文学评价中的可靠性尚未达到可替代人类评委的标准(Krippendorff’s alpha < 0.66),但其仍能提供关于文本编辑质量的初步、有价值的参考视角,具备作为辅助工具的潜力。因此,该研究揭示了生成式AI在文学评估中的局限性与互补价值,推动了人工智能在文学领域应用的科学化探讨。
链接: https://arxiv.org/abs/2611.21649
作者: Manzanarez Gerardo Aleman,Monroy Raul,Garcia Flores Jorge,Calvo Hiram
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:In this paper, we propose an alternative approach to human evaluation of literary fictions by employing LLMs themselves as evaluators within the GrAImes framework. We apply this methodology to assess human-written and AI-generated microfictions in Spanish, to five PhD professors in literature and sixteen literary enthusiasts, and to short stories in both Spanish and English. By comparing the evaluations performed by LLMs with those of human judges, we examine the degree of alignment and divergence between both perspectives, thereby assessing the feasibility of LLMs as auxiliary literary evaluators. Our analysis focuses on the alignment of responses from LLMs with those of human evaluators, providing insights into the potential of LLMs in literary assessment. The conducted experiments reveal that while LLMs cannot be regarded as substitutes for human judges in the evaluation of literary microfictions and short stories, with a Krippendorff’s alpha reliability coefficient less than 0.66, they can serve as a valuable tool that offers an initial perspective on the editorial quality of the texts in question. Overall, this study contributes to the ongoing discourse on the role of artificial intelligence in literature, underlining both its methodological constraints and its potential as a complementary resource for literary evaluation.
[NLP-80] Can an LLM dream? Testing the overfitted brain hypothesis on an agent with text memory
【速读】: 该论文旨在检验Hoel(2021)提出的假设,即梦境通过扭曲重放的体验,防止大脑对近期事件产生过拟合。为此,研究构建了一个基于大语言模型(LLM)的代理(Claude Opus 5.5),其通过文本记忆存储学习内容,并在每晚根据当日对话记录进行记忆重写。实验使用LongMemEval数据集模拟了十天的对话过程,在每次夜间重写前,以两种方式对对话记录的5%、10%、20%或40%的回合进行污染:一是“细节污染”,即修改一两个细节为看似合理的错误信息,或用之前某天同说话者的内容替换;二是“随机文本污染”,将选定回合的一半替换为《白鲸》(Moby-Dick)的段落,并将《白鲸》中的词汇插入另一半。记忆容量随时间从19,500提升至117,000个标记符,依据人类自由回忆的数据设定,从而强制模型遗忘。结果显示,在30次运行中(涵盖十种条件与三种不同对话历史),无论何种污染方式均未提升模型在新任务中应用早期知识的能力;40%的细节污染使回忆率由76%降至60%;而随机文本污染对回忆率无显著影响(40%污染时配对差异为0±2分),因模型能识别并排除此类干扰内容。此外,被替换的错误细节仍被当作事实存储,导致相关问题回答正确率仅为40%,远低于无污染情况下的75%。这些结果表明,预训练模型具备区分真实事件与污染文本的能力,因此无法有效验证原假设。要实现公平测试,需采用更难区分真实与虚假信息的学习者,例如在污染数据上微调的模型,或处于虚构世界中的代理。
链接: https://arxiv.org/abs/2611.21648
作者: Eric Zou
机构: University of Waterloo(Waterloo大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 13 pages, 7 figures, 2 tables. Code and data: this https URL
Abstract:Hoel (2021) proposed that dreams distort replayed experience so that the brain does not overfit to recent events. We test this hypothesis on an LLM agent (Claude Opus 5.5) that stores what it learns in a text memory and rewrites that memory each night from the day’s conversation transcript. The agent goes through ten simulated days of conversation built from LongMemEval. We corrupt 5, 10, 20 or 40 percent of the transcript’s turns before each nightly rewrite in one of two ways. Detail corruption changes one or two details in a turn to plausible wrong ones, or replaces the turn with one by the same speaker from an earlier day. Random-text corruption replaces half of the selected turns with a passage of Moby-Dick and inserts Moby-Dick words into the other half. A size limit on the memory, rising from 19,500 to 117,000 tokens over ten nights, set from human free-recall data, forces it to forget. We learned that corruption did not help the agent apply what it learned in earlier conversations to new tasks in any of 30 runs (ten conditions, three different conversation histories). Replacing details in 40 percent of turns lowered recall from 76 to 60 percent. Random text had no consistent effect on recall, with a paired difference of 0 \pm 2 points at 40 percent, because the model recognized it as corrupted and left it out of memory. Replaced details were stored as fact even when another turn held the correct value, and the agent answered 40 percent of those questions correctly, compared with 75 percent without corruption. These results suggest that a pretrained model can separate corrupted text from real events, so a fair test needs a learner that cannot make this separation as easily, such as a model fine-tuned on the corrupted days or an agent whose days take place in an invented world.
[NLP-81] Limite 1B - Violetto: Technical Report
【速读】: 该论文旨在解决大模型在数学推理任务中高计算成本与高性能之间难以平衡的问题,尤其针对需要高吞吐量、高效单轮数学求解的应用场景。其核心挑战在于如何在有限参数规模下(10亿参数)实现接近甚至超越更大模型的数学推理能力,同时保持训练与推理的可扩展性与低成本。解决方案的关键在于:一是构建一个以数学为中心的高质量2370亿词级语料库,并从零开始训练;二是采用协同优化的架构设计与训练策略,以最大化模型的推理能力;三是通过知识蒸馏、针对性监督微调及基于精选课程的强化学习价值反馈(RLVR)进行后训练,显著提升模型在复杂数学问题上的表现;四是刻意限制指令微调程度,使模型专注于单轮数学求解任务,作为大型系统中的高速并行组件,而非通用对话助手。该方法使Limite 1B - Violetto在AIME 2026上达到94.01%准确率,在BeyondAIME上达到74.25%,性能媲美远超其规模的模型。
链接: https://arxiv.org/abs/2611.21646
作者: Mario Prignano,Gabriele Cirillo,Alessio Morosini,Luca Cerovaz,Alessandro Bartolocci,Emanuele Rodolà,Giulio Starace,Francesco Pappone
机构: 未知
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 36 pages. Model, checkpoints and serving code: this https URL
Abstract:We present Limite 1B - Violetto, a one-billion-parameter model designed for high-throughput mathematical reasoning. Limite is trained from scratch on 237 billion tokens from a carefully curated, mathematics-centric corpus and supports context lengths of up to 131k tokens. Its architecture and training recipe were developed together to maximize reasoning capability while keeping training and inference costs low. Post-training combines knowledge distillation, targeted supervised fine-tuning, and RLVR on a curated curriculum. Limite achieves 94.01% on AIME 2026 and 74.25% on BeyondAIME, placing it among models many times its size on our evaluation suite. The model is deliberately only lightly instruction-tuned: it is intended primarily as a single-turn mathematical solver and as a fast, parallel component of larger systems, rather than as a general-purpose assistant. We release the model, its base and value checkpoints, and an optimized serving implementation under the Apache 2.0 license.
[NLP-82] Asking What Matters: Disagreement-Guided Ambiguity Detection for Interactive Text-to-SQL
【速读】: 该论文旨在解决交互式文本转SQL(text-to-SQL)任务中大语言模型(LLM)代理在有限对话轮次内难以有效识别并优先澄清对最终生成SQL具有实质性影响的语义歧义问题。现有澄清策略存在三大局限:(i)仅关注表面歧义而未评估其对SQL执行结果的实际影响;(ii)忽略模式路径或领域知识中隐含的深层歧义;(iii)在低效或冗余问题上耗费过多对话轮次。其核心挑战在于,在目标SQL尚未明确的情况下,如何精准识别真正影响查询正确性的歧义。为此,本文提出一种基于分歧引导的歧义检测框架DisGuAD(Disagreement-Guided Ambiguity Detection),其关键在于将多个合理SQL解释之间的分歧视为判断歧义重要性的直接证据。DisGuAD通过构建查询项清单,利用互补探测族生成对比性SQL解释,比较其规范化的SQL组件,并将反复出现的差异归因于需用户澄清的显式目标。在此基础上,结合轻量级优先级排序与对话状态机制,选择最具影响力的未决歧义进行澄清。实验在BIRD-Interact Lite和Full数据集上使用两种LLM主干模型验证表明,相较于原始交互策略,DisGuAD在任务成功率与澄清质量上均实现显著提升,标准化奖励提升4.40–8.97点,关键歧义覆盖率从48.62%/45.76%提升至82.75%/85.14%。此外,其生成的澄清信号还可独立提升两个静态text-to-SQL框架的性能最高达13.97分,且无需修改其SQL生成器。结果证明,合理SQL解释间的分歧可作为下游生成中具有行动意义的歧义筛选依据。
链接: https://arxiv.org/abs/2611.21645
作者: Su Dong,Jiayan Lin,Yujia Liu,Zijin Hong,Qinggang Zhang,Feiran Huang,Xiao Huang
机构: Hong Kong Polytechnic University(香港理工大学); Jinan University(暨南大学); Jilin University(吉林大学); Beihang University(北京航空航天大学)
类目: Databases (cs.DB); Computation and Language (cs.CL)
备注:
Abstract:Interactive text-to-SQL requires an LLM agent not only to generate executable SQL, but also to decide which uncertainty to clarify within a limited interaction budget. However, current clarification policies have three limitations: (i) they target surface ambiguity without testing its SQL impact; (ii) they miss ambiguities implicit in schema routes or domain knowledge; and (iii) they spend limited turns on unfocused or redundant questions. These limitations expose a critical challenge: identifying SQL-impacting ambiguity before the intended SQL is known. In this paper, we propose DisGuAD, a Disagreement-Guided Ambiguity Detection framework that specializes clarification selection. DisGuAD treats disagreement among plausible SQL interpretations as direct evidence of which uncertainty matters to the final query. Specifically, it constructs a query-term inventory, elicits contrastive SQL interpretations through complementary probe families, compares their canonical SQL components, and attributes recurring differences to user-facing clarification targets. A lightweight prioritization and dialogue-state mechanism then selects one high-impact unresolved target for clarification. Experiments on BIRD-Interact Lite and Full with two LLM backbones demonstrate consistent improvements in task success and clarification quality. Compared with the original interaction policy, DisGuAD improves normalized Reward by 4.40–8.97 points and raises overall critical-ambiguity coverage from 48.62% to 82.75% and from 45.76% to 85.14%. Its clarification signals further improve two independently designed static text-to-SQL frameworks by up to 13.97 Reward points while keeping their SQL generators fixed. These results demonstrate that disagreement among plausible SQL queries provides actionable evidence for selecting clarification targets that matter to downstream generation.
[NLP-83] MInfoVisQA: A Multilingual Benchmark for Information-Rich Visual Question Answering
【速读】: 该论文旨在解决多模态大语言模型(Multimodal Large Language Models, MLLMs)在面对信息丰富的可视化图表(Information-rich Visuals, InfoVis)时,其跨语言理解能力是否保持一致的问题。具体而言,研究关注当视觉内容的文本语言发生变化时,模型能否稳定、准确地理解与之对应的语义信息。解决方案的关键在于构建了一个名为MInfoVisQA的平行基准数据集,包含8,960个基于128个种子实例生成的跨语言问答配置,覆盖24种语言。通过将数值-结构化数据与文本内容解耦,并程序化同步翻译后的标签,实现了语言对齐的视觉图像与问答对的生成。在此基础上,研究人员在两种设置下评估了25个MLLMs:语言一致问答(Language-Aligned QA, LQA)与跨语言问答(Cross-Language QA, XQA)。实验结果揭示了平均准确率掩盖的显著鲁棒性差距——尽管两个模型在LQA上的均值准确率相近(80.4%),但其在不同语言间的性能波动分别达到7.8和20.3个百分点;且所有模型在固定英文问答而使用非英语视觉语言时表现下降,凸显了现有模型在跨语言敏感性方面的不足,因而亟需建立更精细的跨语言理解评估指标。
链接: https://arxiv.org/abs/2611.21644
作者: Changjiang Jiang,Jiang Zhou,Xinbin Yuan,Fan Zhang,Tao Feng,Zhuohan Xie
机构: 未知
类目: Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV)
备注: Dataset: this https URL Code: this https URL
Abstract:Multimodal large language models (MLLMs) excel in vision-language tasks; yet, whether they maintain consistent understanding of information-rich visuals (InfoVis) when the surface language changes remains unclear. To quantify this capability, we introduce MInfoVisQA, a parallel benchmark with 8,960 QA configurations derived from 128 seed instances across 24 languages. We construct seed-matched, language-parallel visuals and QA pairs by disentangling numerical-structural data from text and programmatically synchronizing translated labels. We evaluate 25 MLLMs under two settings: Language-Aligned QA (LQA, matching visual-text languages) and Cross-Language QA (XQA, fixed QA language with varying visual languages). The results show that aggregate accuracy masks substantial robustness gaps: two models with comparable mean LQA accuracy (80.4%) exhibit performance spreads of 7.8 and 20.3 percentage points across languages. Moreover, all 25 models score lower with non-English visuals under fixed English QA, underscoring the need for cross-language sensitivity metrics.
[NLP-84] Sleep2Word: Peripheral-to-EEG Translation via Duration-Aware Wordization
【速读】: 该论文旨在解决从外周生理信号(如呼吸、心电图[ECG]或光电容积脉搏波描记法[PPG])中重建全夜脑电图(EEG)频谱图的问题,从而实现无需直接采集EEG即可开展睡眠分析的扩展应用。其核心挑战在于如何对具有不同时间结构的多模态信号构建紧凑且具时序敏感性的表征。解决方案的关键在于提出一种名为Sleep2Word的框架,该框架通过引入“时长感知的生理词”(duration-aware physiological words)机制,将不同模态的生理嵌入按上下文依赖的方式分块为可变时长的片段,并联合量化其内容与持续时间,形成模态特异的词典(wordbooks)。随后,采用基于块自回归(block-autoregressive)的翻译器,在固定时长块内预测目标EEG词的身份及其特定出现时长的精细化修正。这一设计实现了源端与目标端词典的独立性,有效抑制了时序漂移,并显著缩短了自回归解码路径。在涵盖9个数据集、超过2万例多导睡眠图(PSG)数据上训练后,Sleep2Word在EEG频谱图重建、睡眠分期、觉醒检测及生存预测任务中均优于现有基线方法;同时,其生成的词可将整晚信号压缩高达69.7%,在精度与计算开销之间取得良好权衡。进一步的睡眠事件分析表明,所学习的词能捕捉跨模态间生理上合理的关联,展现出在可解释的跨模态分析中的潜力。
链接: https://arxiv.org/abs/2611.21643
作者: Keondo Park,Arthur Delhaye,Younghoon Na,Hyun-Woo Shin,Hyung-Sin Kim
机构: Seoul National University(首尔国立大学); Université du Québec à Rimouski(魁北克省里穆斯基大学); Seoul National University College of Medicine(首尔国立大学医学院); Seoul National University Hospital(首尔国立大学医院)
类目: Computation and Language (cs.CL); Sound (cs.SD); Audio and Speech Processing (eess.AS)
备注: 10 pages, 4 figures
Abstract:Estimating full-night electroencephalography (EEG) spectrograms from peripheral measurements could extend EEG-informed sleep analysis to recordings acquired without EEG. This task requires compact representations of signals with different temporal structures. We introduce Sleep2Word, a framework for translating respiration, electrocardiography (ECG), or photoplethysmography (PPG) into EEG spectrograms through duration-aware physiological words. Building on context-dependent chunking, Sleep2Word groups physiological embeddings into variable-duration chunks and jointly quantizes their content and duration into modality-specific wordbooks. A block-autoregressive translator then predicts target EEG word identities and occurrence-specific duration refinements within fixed-duration blocks. This design accommodates independent source and target wordbooks, limits temporal drift, and shortens the autoregressive decoding path. Trained on over 20K PSG cases across 9 datasets, Sleep2Word consistently outperforms existing spectrogram generation baselines in EEG spectrogram reconstruction, sleep staging, arousal detection, and survival prediction. Compressing full night sequences by up to 69.7%, the learned words yield a favorable accuracy-compute trade-off. Sleep event analysis further shows that the learned words capture physiologically plausible associations across modalities, suggesting their potential for interpretable cross-modal analysis.
[NLP-85] CrisisBench: an evaluation suite for crisis counseling AI
【速读】: 该论文旨在解决危机热线(crisis helpline)在人力与资金资源匮乏背景下,如何有效评估和提升生成式 AI 在危机对话中表现的问题。现有评估基准无法充分捕捉危机干预中的关键环节,如风险评估、情绪升级与降级处理,且由于隐私限制难以使用真实通话记录进行训练与评估。为此,研究提出 CrisisBench——一个保护隐私的单轮与多轮危机咨询评估框架,包含1,085个来自200条去标识化真实对话的单轮任务及基于热线元数据生成的1,185名患者类大模型(patient LLMs),并通过专家顾问校准了患者模型与23项指标的LLM评判器。关键解决方案在于构建一个可扩展、隐私安全的评估体系,实现对生成式AI在危机对话中反应质量的量化评估。结果显示,尽管当前系统在结构能力(如微技能与语气风格)上接近人类水平,但在应对复杂响应层面的缺陷(如过度共情、提问不当)仍显著落后于专业咨询师;多轮评估中检索增强生成(retrieval-augmented generation)表现最优,但各系统在安全性和风险处理维度仍存在明显差异。研究释放了评估标准、评判提示与任务数据,为未来危机对话生成模型的开发与验证提供了可靠基础。
链接: https://arxiv.org/abs/2611.21622
作者: Akshay Swaminathan,Filip Ramazan,Sharang Phadke,Kevina Wang,Ivan Lopez,Shaked Peleg Azzam,Gloria Ye,Chastin Chung,William Wang,Stephanie Stoll,Ivy Pham,Rebecca Hurwitz,Shreya Shah,Samuel Chuang,Divyanjali Verma,Abhay John,Ehsan Adeli,Nigam H. Shah
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:Background: Crisis helplines are essential to mental health systems but are often understaffed and underfunded. LLMs could support counselors, but existing benchmarks do not capture the escalation, risk assessment, and de-escalation central to crisis conversations, while privacy restrictions limit use of real transcripts. Methods: We present CrisisBench, a privacy-preserving suite for single- and multi-turn crisis counseling evaluation. It includes 1,085 single-turn tasks from 200 de-identified conversations and 1,185 patient LLMs derived from hotline metadata. Patient models and a 23-item LLM judge were calibrated against expert counselors. Results: We evaluated five GPT models plus draft-then-edit and retrieval-augmented generation. All systems underperformed human counselors on excessive empathy and questioning in single-turn evaluation. The rubric achieved human-human Krippendorff’s \alpha = 0.51 and LLM-human \alpha = 0.54 (Pearson r = 0.80). In multi-turn evaluation, retrieval-augmented generation scored highest overall (0.961), with all systems within 0.058. Microskills and style/tone were near ceiling, while productivity and safety/risk accounted for most between-system variation. Counselors rated patient LLMs realistic (86%) and useful (92%). Conclusion: CrisisBench enables scalable, privacy-preserving evaluation of crisis counseling. Current systems perform strongly on core structural competencies but still diverge from experts on important response-level failure modes. We release the rubrics, judge prompts, and evaluation tasks.
[NLP-86] Modeling Document and Section Hierarchies in Embedding Space
【速读】: 该论文旨在解决传统嵌入式主题模型(Embedded Topic Models, ETM)在处理具有层级结构的文档时,仅用单一主题分布表示整个文档、忽略其内部多层级子主题结构的问题。针对这一局限,提出了一种名为分层嵌入式主题模型(Hierarchical Embedded Topic Model, hETM)的新方法,其核心在于通过嵌入空间中的层次先验,将文档整体与各部分内容的主题分布进行耦合建模,从而捕捉文档内部的层级语义结构。hETM的关键创新在于引入了基于词嵌入空间的分层先验机制,并设计了可高效推断的摊销变分推理(amortized variational inference)流程,使得模型能够同时学习文档级和段落级的主题分布。实验结果表明,hETM在多个真实场景(如法律文件、政府报告、在线论坛)的结构化文本数据上,不仅在保留低困惑度(held-out perplexity)方面优于或接近现有平面及分层主题模型,且在主题质量与下游聚类任务中均表现出色,验证了其在复杂文档结构建模中的有效性与鲁棒性。
链接: https://arxiv.org/abs/2611.21980
作者: Kriste Krstovski,Sharyn O’Halloran,Kshitij Chandna
机构: Columbia University (哥伦比亚大学)
类目: Machine Learning (stat.ML); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:Embedded topic models integrate word embeddings into the topic generation process, producing coherent topics and robust performance on large vocabularies. However, each document is represented by a single topic distribution, ignoring its internal structure. Many real world collections, including legal documents, company and government reports, and online discussion forums, exhibit hierarchical organization in which documents consist of multiple sections that often address distinct subtopics. We introduce the Hierarchical Embedded Topic Model (hETM), a structured topic model that couples the topic distribution of a document and those of its sections through a hierarchical prior in the embedding space. We derive an amortized variational inference procedure for hETM and evaluate it on structured collections from each of these domains. Experimental results show that hETM achieves the lowest or near-lowest held-out perplexity among flat and hierarchical topic model baselines, along with comparable or higher topic quality. It also performs strongly on a downstream clustering task, achieving the best clustering performance on most of the collections.
[NLP-87] Marimba: A System One Model for Speech
【速读】: 该论文旨在解决语音应用中在有限音频窗口内对词项存在性、话语回合完成状态及语言类型进行实时决策的关键问题。其核心解决方案在于构建一个基于冻结的Qwen3-ASR 0.6B特征的可键入语音决策接口——Marimba。针对词汇性问题,采用零样本评分机制对自动语音识别(ASR)转录文本进行判断;针对非词汇性问题,引入条件化问题的分类头与独立的第18层话语结束读出层;语言识别则直接利用ASR模型自身输出的语言得分。实验结果显示,在私有对话测试集上,解码器的受试者工作特征曲线下面积(AUROC)达到0.80–0.92,在公开朗读语音数据集上为0.92–0.97;语音命令四分类准确率达0.93,二元词汇性任务的期望校准误差(ECE)不超过0.08。个人身份信息(PII)检测通过将贪婪解码的转录结果输入专用的定制化PII管道,在合成巴西葡萄牙语语音上实现项级F1分数0.904(参考转录达1.000)。第12层说话人读出层在AMI和VoxConverse数据集上检测说话人切换的AUROC达0.966–0.980,优于ECAPA-TDNN基线,并在说话人数量(0, 1, 2, 或3+)识别上超越pyannote 3.1;话语结束读出层在Smart Turn自有测试集上取得AUROC 0.930,略低于Smart Turn v3.2的0.982,但差异不显著(配对置信区间包含零)。整体而言,该方法通过高效利用预训练模型特征与模块化设计,实现了多任务语音决策的高精度与鲁棒性。
链接: https://arxiv.org/abs/2611.18330
作者: Solano Amadori
机构: charcot.ai
类目: Audio and Speech Processing (eess.AS); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Sound (cs.SD)
备注: 12 pages, 4 figures, 4 tables. Code: this https URL weights: this https URL
Abstract:Speech applications require decisions about term presence, turn completion, and language within an audio window. We present Marimba, an interface for typed speech decisions built on frozen Qwen3-ASR 0.6B features. Lexical questions use zero-shot scoring of ASR transcripts; non-lexical questions use a question-conditioned head and a separate layer-18 turn-end readout. Language identification uses the ASR model’s own language scores. On private conversational test sets, decoder AUROC is 0.80-0.92; on public read speech it is 0.92-0.97. Speech Commands four-way accuracy is 0.93; binary lexical ECE is at most 0.08. Personal-information detection, which passes the greedy transcript to a private, curated PII pipeline, reaches item-level F1 0.904 on synthetic Brazilian Portuguese speech (1.000 on reference transcripts). Layer-12 speaker readouts detect speaker change with AUROC 0.966-0.980 on AMI and VoxConverse, above an ECAPA baseline, and count speakers (0, 1, 2, or 3+) more accurately than pyannote 3.1. The turn-end readout reaches AUROC 0.930 versus Smart Turn v3.2 at 0.982 on Smart Turn’s own test set. On a subset of ICSI meetings labeled by next speaker, AUROC is 0.716 versus 0.712; the paired confidence interval includes zero.
信息检索
[IR-0] CARE: Conversational Affective Response Retrieval for Internet Memes
链接: https://arxiv.org/abs/2611.23005
作者: Jiajun Li,Leihan Zhang,Teng Zhang,Le Zhang,Qiang Yan
类目: Information Retrieval (cs.IR)
备注: 26 pages, 5 figures
Abstract:Conversational meme retrieval supports response selection in online communication, where relevance depends on both contextual meaning and affective expression. We study this task using two language-specific collections from Douban and Imgflip, containing 98,506 meme records across 5,583 discussion topics. Thread membership supplies weak relevance labels, while likes provide implicit ranking feedback. Automatically extracted sentiment, facial-expression indicators, and cross-modal consistency features support an exploratory analysis of engagement. We propose CARE, a Conversational Affective Response Retrieval framework that combines an emotion-enhanced CLIP retriever with a Jina-based reranker. The retriever incorporates free-form emotion descriptors through gated fusion and learns from multiple positive examples per query. The reranker combines query–image interactions with emotion representations and is trained using engagement-based ranking supervision. Under the reported topic-held-out protocol, CARE achieves Hit@10 of 24.03% on Douban and 49.33% on Imgflip, compared with 23.16% and 41.07%, respectively, for fine-tuned CLIP. These results support the utility of the complete two-stage system for recovering thread-associated memes. They do not independently establish perceived social resonance or isolate the contribution of affect from reranker training. The study provides a basis for further evaluation of affect-aware retrieval with human relevance judgments and controlled comparisons.
[IR-1] When the tail is earned by its companions: measuring label dependence in long-tailed multi-label classification
链接: https://arxiv.org/abs/2611.22859
作者: Christoffer Löffler
类目: Machine Learning (cs.LG); Computation and Language (cs.CL); Information Retrieval (cs.IR)
备注: 27 pages, 3 figures, 13 tables. Submitted to Information Sciences
Abstract:A rare label seldom occurs alone, and a classifier can recover it from the frequent labels it co-occurs with, its companions, without having learned the label itself. Existing diagnostics of this need a trained model: none says before training whether a corpus admits the failure, or what the standard aggregate hides. Macro-F1, like any measure built from per-label confusion counts, is unchanged by how a label’s recall divides between occurrences with and without a companion. We introduce a within-label estimand of that division, a reading rule under which a fall in dependence counts only if companion-free recall rises, and a statistic of training labels that bounds, before training, how far reported recall can exceed companion-free recall. Across 23 corpora, 7 training objectives and five families of label-correlation mechanisms, companion-free recall falls short on all six corpora where the estimand is defined. Up to 18.4% of the tail true positives macro-F1 credits would vanish at the companion-free rate, within the training-label bound on every corpus. No long-tail objective reduces the dependence, but architecture moves it. Models whose tail score is computed from other labels’ predictions are strongly dependent.
[IR-2] Music Similarity Learning in Sparse Latin Scenarios
链接: https://arxiv.org/abs/2611.22846
作者: Gabrielly Xavier dos Santos,Lucas Dias Cler,Olga Goussevskaia
类目: Information Retrieval (cs.IR)
备注:
Abstract:In this work we propose a lightweight, interpretable approach that estimates item-item music similarity from publicly available metadata. Our hybrid regression model combines content-based filtering with the classic item-based collaborative filtering. We test generalization via a zero-shot evaluation on Latin American music, a genre for which co-occurrence data is scarce. Accurately predicting similarity for these unseen, sparse pairs shows that our approach is robust to the regional gap affecting many existing systems, without the audio signal processing or the computational budget that deep multimodal approaches would require.
[IR-3] Better Rankers Worse Retrievers: The Hidden Cost of RL Post-Training in Generative Recommendation
链接: https://arxiv.org/abs/2611.22838
作者: Yuheng Chen,Fan Zhang,Yixuan Xu,Chao Yang,Yang Xiang,Can Yang
类目: Information Retrieval (cs.IR); Artificial Intelligence (cs.AI)
备注:
Abstract:Generative recommenders decode the semantic ID of a user’s next item with a language model, and post-training them with reinforcement learning (RL) after supervised fine-tuning (SFT) has become standard practice. The gains are mostly reported on the first ten results, yet in deployment such models hand hundreds of candidates to a downstream ranker or reward model. We show that RL post-training has a hidden cost: it improves accuracy at the top of the list (hit@K for small K) but, on all five datasets we study, lowers the recall of the candidate list (hit@K for K in the hundreds). The model becomes a better ranker and a worse retriever. Controlled experiments trace the loss to the negative advantages that RL assigns to the candidates it samples, the rollout. The root cause is that only one next item is logged per user, so RL treats every other candidate as wrong. Yet these candidates are the model’s most plausible alternatives, and often the next items of other users with similar histories. Penalising them pushes down exactly what similar users choose next, and these items drop out of those users’ candidate lists. This diagnosis points directly to a remedy: out-of-rollout anchoring, a single additional loss term that preserves the SFT model’s ordering of the items outside the rollout. It keeps the gain at the top and restores recall to the SFT level on all five datasets, and it is effective under other RL configurations as well. To our knowledge, this is the first systematic study of this recall loss: we measure it, identify its cause, and show that, once the cause is known, a simple remedy suffices.
[IR-4] IntentGR: Enhancing Intent for End-to-End Personalized Generative Retrieval in Search WSDM2027
链接: https://arxiv.org/abs/2611.22788
作者: Gefan Ye,Yanjing Wu,Yinfu Feng,Zhiqiang Wang,Liang Li,Jun Xiao,Rong Xiao,Xiaoyi Zeng
类目: Information Retrieval (cs.IR)
备注: 11 pages, 3 figures; Accepted by WSDM 2027 Findings
Abstract:Personalized generative retrieval in search should jointly satisfy the user’s immediate search intent and long-term preference. Recent end-to-end approaches, primarily explored in recommendation, unify item tokenization and sequence generation within a single model, allowing semantic identifiers (SIDs) to be optimized toward the downstream generation objective. However, these methods often treat the query merely as an input feature, without explicitly leveraging it to guide either identifier learning or sequence generation. This leaves intent guidance weak in both the tokenization and the generation stage, causing generated semantic identifiers to drift toward high-frequency token patterns rather than the current search intent. In this work, we find that incorporating additional query-related evidence into the framework of end-to-end generative retrieval is important. Based on this finding, we propose IntentGR, an intent-enhanced end-to-end framework that strengthens query semantics at four complementary stages of the retrieval pipeline. Query-Aware Semantic Behavior Representation (QBR) introduces query-level behavioral evidence as a new input signal; Query-Semantic Alignment (QSA), Long-Tail Query Enrichment (LQE), and Query-conditioned Attention Modulation (QAM) then serve as integration mechanisms at the optimization, training-distribution, and decoding stages to fully exploit this evidence. Experiments on the public AOL4PS benchmark and AE, a large-scale industrial e-commerce dataset, show that IntentGR improves Recall@10 by 8.11% and 11.96% over the strongest baseline, with strong gains on long-tail queries. These results suggest that enriched query-related input evidence, together with effective integration mechanisms, is beneficial for building intent-aware end-to-end generative search systems.
[IR-5] Learning Cross-Domain Media Similarity from Amazon Reviews
链接: https://arxiv.org/abs/2611.22756
作者: Gabrielly Xavier dos Santos,Lucas Dias Cler,Luciana Fujii Pontello,Olga Goussevskaia
类目: Information Retrieval (cs.IR)
备注:
Abstract:Collaborative filtering is a basic foundation of many recommendation algorithms. It works by analyzing a user’s history and then suggesting items similar to those in that history. This method typically relies on user data to compute item similarity, data that is widely available to commercial systems but is usually sparse or inaccessible for independent systems and research, especially across multiple media domains. In this work, we propose a method to generalize similarity information between items using supervised learning. As input to our model, we use semantic and temporal features extracted from Amazon product reviews, which are publicly available. As the training target, we use item (product) co-occurrence in the reviewing history of Amazon customers. We evaluate the quality of the predicted similarities across two media domains: books and movies. Experimental results demonstrate the robustness of our methodology, yielding positive outcomes in zero-shot cross-domain transfer tasks. Furthermore, feature attribution analysis reveals a structural shift in how feature groups are utilized across setups: intra-domain models rely heavily on linguistic semantics, whereas cross-domain environments pivot toward behavioral popularity cues. Incorporating even limited target-domain context during training bridges this domain sparsity gap, significantly enhancing framework generalization.
[IR-6] MIPR: A Benchmark for Reasoning -Intensive Multimodal Intent-to-Product Retrieval with Multi-Category Solutions
链接: https://arxiv.org/abs/2611.22509
作者: Qiang Li,Xuan Lu,Jian Ruan,Manxi Lin,Jiacheng Chen,Wen Shen,Zhihua Wei,Xiaoyu Shen,Dan Ou,Xiangheng Kong,Junjun Zheng
类目: Information Retrieval (cs.IR)
备注: 14 pages, 8 figures
Abstract:Real-world e-commerce queries often express user intent without naming a target product, and a single intent may be fulfilled by products from multiple categories. However, existing multimodal retrieval benchmarks typically assume a relatively well-defined target, leaving retrieval for intents with multi-category solutions less explored. To fill this gap, we introduce MIPR, a benchmark for Multimodal Intent-to-Product Retrieval comprising 3,522 multimodal queries and a shared corpus of 30,731 products. MIPR features reasoning-intensive retrieval, multi-category solutions, and fine-grained relevance labels. Experiments show that the best embedding-based retriever achieves 49.23% nDCG@10 on the full query set, and 55.86% solution-category recall@10 on queries with multiple solution categories. Reranking improves these scores to 60.11% and 67.00%, respectively, yet substantial gaps to the candidate-pool oracle remain. These results establish MIPR as a challenging and diagnostic benchmark for evaluating multimodal intent-to-product retrieval. The data and code are publicly available at this https URL.
[IR-7] From Relevance to Utility: Rubric Alignment in Yuanbao Search KDD2027
链接: https://arxiv.org/abs/2611.22503
作者: Yu Lu,Lixin Zou,Yangqi Chen,Hui Xu,Qiaolin Xia,Hao Wang,Qianwen Ma,Reed Li,Shuang Zeng,Haijin Liang,Haibo Shi
类目: Information Retrieval (cs.IR)
备注: 10 pages, 5 figures, 5 tables, accepted at KDD 2027
Abstract:Agentic search demands that a search engine move beyond per document relevance ranking to assemble document sets optimized for downstream LLM reasoning. Prior work addresses this either by improving individual pipeline stages in isolation or by applying RL based selection on top of relevance trained upstream modules, leaving the full stack unaligned to utility for reasoning. We propose rubric aligned search, a framework that unifies the traditional search pipeline under a single supervision surface. For each query, we derive a rubric from coarse to fine, specifying weighted positive criteria that the document set must satisfy and negative criteria that it must avoid. From this rubric we derive two training signals. First, a lightweight labeler distills rubric grounded judgments into per document utility labels, which supervise a contrastive dual encoder for retrieval and a multi task ranker trained with four rubric aligned scoring heads. Second, a rubric conditioned evaluator produces a set level reward that drives a dual branch RL selector, combining GRPO with best rollout distillation to resolve credit assignment in subset optimization. Deployed in Yuanbao Search, the pipeline improves end to end answer quality by 8.52% over the online production baseline.
[IR-8] More Than a Score: Reinforcement Learning with Verbalized Feedback for Generative Recommendation
链接: https://arxiv.org/abs/2611.22410
作者: Zipeng Chen,Xinyu Lin,Yuliang Sun,Sheng Zhang,Zeyu Wang,Xiaowen Shi,Junwei Pan,Chengguo Yin
类目: Information Retrieval (cs.IR)
备注: 10 pages, 11 figures and tables
Abstract:Large language models (LLMs) are increasingly used for generative recommendation, where recommended items are generated directly from users’ interaction histories. To further improve recommendation quality, reinforcement learning (RL) has been applied to align these models with recommendation objectives. However, existing RL methods rely on scalar rewards to implicitly guide exploration. These rewards do not specify which aspects of a candidate conflict with the user’s preferences or how the next candidate should differ. To address this gap, we propose VerRec, a verbalized feedback-guided RL framework for generative recommendation. Specifically, a user simulator generates personalized verbalized feedback based on the user’s historical preferences and the candidate item profile. The policy uses this feedback to produce refined recommendations. Original and refined candidates form an expanded group and are jointly optimized with group relative policy optimization. The user simulator is only required during training and is removed at inference time. Experiments on three Amazon review datasets with two generative recommendation backbones demonstrate that VerRec improves recommendation accuracy and ranking quality. Further analyses show that verbalized guidance encourages exploration of more diverse candidates and helps the policy discover more rewarding recommendations during training.
[IR-9] Deal With It: On Better Schedules for Listwise Reranking
链接: https://arxiv.org/abs/2611.22392
作者: Benjamin Clavié,Herumb Shandilya,Hanna Lichtenberg,Aamir Shakir,Makoto P. Kato
类目: Information Retrieval (cs.IR); Artificial Intelligence (cs.AI)
备注:
Abstract:Listwise LLM rerankers cannot reliably rank hundreds of documents in a single call, and rely on schedules to orchestrate their calls. Parallel schedules split candidates into groups and advance a fixed quota from each. When a group holds more of the top passages than its quota, some are cut, and expensive repair calls become required. We show that the way candidates are split matters where such a quota can bind, and that, in expected cuts, the best split there deals them round-robin by retriever rank, like cards. Although not conceptually new, dealing has remained unanalyzed. We prove that, under a simple retriever model, no split with equal quotas cuts fewer of the true top ten on average, and that contiguous chunks cut more as the retriever improves. DealRank, the simplest schedule our analysis allows, is on average within 0.1 nDCG@10 points of the best schedules over six LLMs, in two rounds. It is consistently 1.1 to 1.4 times faster than BlitzRank with concurrent first calls up to a thousand candidates. There, it needs three rounds and is eleven times faster than the sliding window. In existing schedules such as BlitzRank, dealing reduces repair calls from 59% of queries to 14%, improving latency with no significant change in quality. Random partitioning appears equivalent at a quota of four, as predicted, but ties results to the seed and trails by 1.3 nDCG@10 points at a quota of one over a dense retriever.
[IR-10] RICoT: Recurrent Interleaved Chain-of-Thought for Page Stream Segmentation
链接: https://arxiv.org/abs/2611.22382
作者: Prashant Khatri,Bhavana Malagavalli,Saiful Haq,Niyati Chhaya
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Information Retrieval (cs.IR)
备注: 18 pages, 3 figures, 19 tables. Dataset and code to be released
Abstract:Large language models (LLMs) have renewed interest in Page Stream Segmentation (PSS), splitting a multi-page file into its constituent documents. Yet published LLM methods still frame the task as boundary classification over one or a few neighboring pages, and commercial document-splitting APIs do not disclose their methods. We reimagine PSS for LLMs as a sequential process over the whole stream and propose RICoT (Recurrent Interleaved Chain-of-Thought), a prompting framework that works through the pages as a reader would: within a sliding window it reasons about each page and commits its boundary decision before moving on, and across windows it carries forward a compact record of what it has read, keeping per-call cost bounded. We also introduce FinancePSS-205, a benchmark of 205 real-world, de-identified financial page streams, to be released publicly. On FinancePSS-205, RICoT matches or outperforms pairwise and windowed prompting with the same backbone for each of three open-weight LLMs; on a TABME test subset, RICoT with GLM-5 reaches document F1 0.847 and boundary F1 0.915, above published zero-shot results. A 2B-parameter student, trained on a larger teacher’s decisions with and without the page-level reasoning, performs comparably in both modes and comes within 0.05 document F1 of the strongest prompted model on FinancePSS-205.
[IR-11] NONSENSE: Non-Semantic Strings Expose Semanticity Gaps in Dense Retrieval EACL2027
链接: https://arxiv.org/abs/2611.22353
作者: Ankit Gupta
类目: Information Retrieval (cs.IR); Computation and Language (cs.CL)
备注: Accepted to EACL 2027 (Main Conference). 15 pages
Abstract:Dense retrieval is central to modern search, question answering, and retrieval-augmented generation, where high similarity is often treated as evidence of semantic relevance. We examine how far this assumption can be stressed. NONSENSE is a black-box sparse-edit probe that constructs visibly non-semantic strings while forbidding query-term copying, optimizing only the dense similarity between a query and the string. Across five BEIR-style datasets and three popular single-vector retrievers, these strings frequently outscore the retriever’s own best judged query-relevant passage (QRP), and often rank first in the full corpus. The result indicates a semanticity gap in the evaluated dense retrievers: nearest-neighbor status can be separated from passage-like meaning. These findings highlight the importance of evaluating dense retrievers not only by whether they retrieve relevant documents, but also by how reliably their similarity scores reflect semantic evidence.
[IR-12] GuardAgent : A Memory-Augmented Agent ic Framework for Adaptive Detection of Adversarial Promotions in LLM -Based Recommendation WSDM2027
链接: https://arxiv.org/abs/2611.22274
作者: Shiyi Yang,Xinshu Li,Chen Wang,Xiwei Xu,Liming Zhu,Lina Yao
类目: Information Retrieval (cs.IR); Cryptography and Security (cs.CR)
备注: Accepted by WSDM 2027
Abstract:In traditional recommender systems (RSs), textual information typically serves as auxiliary signals for representation learning. In contrast, in LLM-based RSs, textual information becomes the primary medium for recommendation generation through semantic understanding and reasoning. This shift fundamentally changes the attack surface, making promotion-driven manipulations more feasible, effective, and harder to detect. However, effectively detecting such attacks remains largely unexplored. To fill this gap, we present the first systematic study on detecting promotion-driven attacks against LLM-based RSs, aiming to safeguard next-generation recommendation platforms. Specifically, we propose GuardAgent, a novel LLM-powered agentic detection framework that operates through a loop of Perception, Action, Reflection and Memory. In contrast to conventional LLM-based detectors that function in a static, one-off, and opaque fashion, GuardAgent continually learns and self-evolves to adapt to unseen threat patterns while generating explainable reasoning traces and outcomes. Extensive experiments on real-world datasets demonstrate that GuardAgent achieves state-of-the-art generalization performance, laying the foundation for robust and trustworthy recommendation in practical settings.
[IR-13] aching Language Models to Narrow Boolean Patent Searches from Search Feedback
链接: https://arxiv.org/abs/2611.22270
作者: Seiya Kawano,Koji Marusaki,Hirofumi Nonaka,Koichiro Yoshino
类目: Information Retrieval (cs.IR); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 13 pages, 3 figures, 3 tables. Accepted at NLPIR 2026
Abstract:Patent prior-art searchers issue Boolean queries line by line, broadening to find relevant documents and narrowing until candidates can be screened. We distill 38,397 Japan Patent Office search reports into executable expert strategies, re-execute them, and evaluate generated strategies by citation recall and the smallest set containing each citation. Prompted models write valid, high-recall strategies but leave citations in sets too large to screen. We adapt a generator offline from scored rollouts in two forms: strategy-level learning from a screening-aware strategy score, and line-level learning from citation-preserving narrowing credit, which upweights the query lines keeping a found citation in a smaller set. On shared rollouts, line-level credit outperforms strategy-level scores on screening measures; iterated, it cuts the median citation-containing set from 768 to under 50 and places more citations in sets under 100 documents than experts, at lower recall. Strategy-level preference learning reaches near-complete recall; the two forms are complementary.
[IR-14] Cluster-Guided Decoding with Trust-Region Constraints for Personalized Recommendation
链接: https://arxiv.org/abs/2611.22163
作者: Gezheng Wen,Chao Shen,Raviteja Anantha,Tan Shen Cliff Chung,Layne C. Price
类目: Information Retrieval (cs.IR)
备注: 5 pages, 2 tables
Abstract:Most recommenders score each item in the catalog separately, and do not explicitly model which semantic region of the catalog a recommendation belongs to. Models expressive enough to offer that control are often too slow to run over a full catalog for online inference. In this paper, we present Cluster-Guided Generative Reconstruction (CG2R), which learns a hierarchy of semantic clusters and uses a cluster-level prior to steer item ranking for personalized recommendations. Unlike existing systems that only use clusters to pre-select candidates, CG2R adds the cluster prior directly to the item scoring rule, so that the learned cluster structure shapes the ranking itself. Because the prior is assembled from cluster centroids rather than per-item statistics, it remains defined even for cold-start users when no history is available. Across the three benchmark datasets evaluated, CG2R outperforms classical, neural, and LLM-based baselines in terms of Recall and nDCG, with its performance gains widening as the catalog grows sparser. In addition, CG2R serves recommendations in milliseconds, suggesting potential as low-latency first-stage retrieval.
[IR-15] SIMQ: A Single Graph-based Index for Multi-Vector Query ICDE2027
链接: https://arxiv.org/abs/2611.22079
作者: Yifan Lu,Jiabao Jin,Xiaoyao Zhong,Peng Cheng,Xiangyu Wang,Haotian Li,Hua Chen,Junqing Chen,Wei Xi,Heng Tao Shen
类目: Information Retrieval (cs.IR); Databases (cs.DB)
备注: under ICDE 2027 review, 13 pages
Abstract:Multi-vector retrieval models such as ColBERT achieve state-of-the-art effectiveness by representing documents as sets of token-level vectors and scoring via MaxSim, but their high computational cost limits scalability to large corpora. Existing approximate methods face fundamental bottlenecks: centroid interaction matrix approaches require computing a full matrix at every query and rely on GPU-accelerated clustering, while data-oblivious compression methods yield unstable recall that requires repeated compensation. We propose SIMQ (Single graph Index for Multi-vector Query), a framework that bridges single-vector graph indices and multi-vector retrieval through data-aware sampling-based clustering and per-token graph navigation, simultaneously bypassing all aforementioned bottlenecks. SIMQ organizes all token vectors into fine-grained clusters via data-aware sampling, builds a single graph index over cluster centroids independently of PLAID’s residual representation, and approximates the MaxSim score through per-token graph search without materializing any interaction matrix or relying on GPU acceleration. We provide theoretical analysis showing that, under a manifold assumption and successful cell retrieval, data-aware partitioning has a smaller expected approximation error than a data-oblivious partition. Building on this foundation, SIMQ+ further introduces count-based scoring and a dynamic candidate set to improve robustness across varying query conditions. Experiments on five benchmarks compare recall, throughput, and construction time against several baselines, primarily on CPU; large-scale PLAID k-means centroids are built on a separate GPU machine. SIMQ has been integrated into VSAG, an Ant Group’s open-source vector search library.
[IR-16] Beyond Bitwidth: Deployment-Constrained Mixed-Precision Quantization for Multilingual LLM Inference on NVIDIA Blackwell
链接: https://arxiv.org/abs/2611.21974
作者: Hyojung Han
类目: Information Retrieval (cs.IR); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 18 pages, 2 figures
Abstract:Low-bit LLM quantization is useful only if a model preserves service-relevant accuracy, executes through supported GPU kernels, and meets measured memory and latency limits. We study Qwen3-8B on NVIDIA B200 with vLLM, including INT2/INT3, GPTQ, NVFP4, and mixed-precision configurations. We verify native packed execution and precision on all 144 fused linear modules and evaluate Korean, English, numerical, and reasoning quality under preregistered gates. Packed INT2 and INT3 execute, but all seven tested INT2 grids degenerate on generation. Isolated module-restoration effects are strongly non-additive; Korean and reasoning sensitivity rankings disagree, and a promising INT4/INT8 development candidate fails independent final evaluation. A configuration with 16-bit activations, 98 NVFP4 weight-only modules, 46 INT8 modules, and an INT8 output head passes all gates on the untouched FINAL-81 set (5,168 items): 5.917 GiB of weight-loading memory (limit 5.98), Korean teacher-correct retention 0.904 (gate 0.90), and TPOT ratio 0.939 relative to BF16. It improves paired accuracy over NVFP4 W4A4 by +1.4 percentage points (Bonferroni-adjusted interval [+0.2, +2.6]), compared with +3.4 points on the selection set. This comparison changes both activation precision and INT8 protection; exploratory analyses suggest that the activation change contributes substantially, but the isolated allocation effect is not established. Final checkpoints were not retained, and bitwise rebuild reproducibility remains unverified.
[IR-17] Urban Representation Learning for Identifying Potential Disparities in Crime Reporting
链接: https://arxiv.org/abs/2611.21825
作者: Jorge Valverde-Rebaza,Carlos Paez De La Cruz,Pedro Pérez-Collado,Daniel Sánchez-Cataño,Eashley Martínez-Vergara
类目: Machine Learning (cs.LG); Computers and Society (cs.CY); Information Retrieval (cs.IR)
备注: Accepted for publication in International Conference on Artificial Intelligence, Computer, Data Sciences, and Applications (ACDSA 2027)
Abstract:Open crime data support urban safety analysis but reflect reported incidents rather than the complete distribution of criminal activity, potentially introducing spatial distortions in downstream analysis. We propose a self-supervised urban representation learning framework for identifying unusual reporting patterns relative to representation-similar peers using only georeferenced crime records. The method constructs complementary hourly, weekly, and static reporting profiles, learns scale-specific representations under geographic self-supervision, and integrates them into a unified latent space. Neighbor Profile Divergence (NPD) is then used to quantify disagreement between each region’s reporting profile and those of its retrieved peers. Experiments on more than 1.8 million crime records from Mexico City, aggregated into 1,058 H3 regions, show that the proposed representation achieves a mean Silhouette score of 0.4052, compared with 0.2863 for Hex2Vec. NPD rankings remain highly stable across neighborhood sizes, with Spearman correlations from 0.9720 to 0.9951, while most high-divergence candidates persist across configurations. The retrieved peers also extend substantially beyond immediate geographic neighborhoods, indicating that the learned representation captures structure beyond local spatial proximity. High-divergence regions exhibit spatially organized patterns and, in several cases, differences in incident volume relative to peer expectations. These results support NPD as an internal auditing and prioritization mechanism for identifying regions that warrant further investigation, rather than as direct evidence of underreporting.
[IR-18] Evaluating AST-Derived Knowledge Graph Retrieval for AIOps Incident Diagnosis in Apache Kafka
链接: https://arxiv.org/abs/2611.21625
作者: Shashank Mishra,Shweta Mishra
类目: oftware Engineering (cs.SE); Information Retrieval (cs.IR)
备注: 24 pages, 6 tables, 1 figure. Preprint
Abstract:Incident diagnosis requires retrieving source-linked operational evidence from large codebases. We evaluate graph-augmented retrieval for Apache Kafka using an abstract-syntax-tree (AST)-derived ownership graph over 1,600 source-linked log records and an audited benchmark of 315 incident-style queries covering 103 unique targets. Vector-only retrieval is a strong baseline at Recall@20 of 0.956, and neither static nor confidence-guided staged AST expansion significantly improves Recall@20 over it. As an external control we evaluate HippoRAG, a published text-derived graph-retrieval system, against its own ColBERTv2 retrieval backbone. Against ColBERTv2-only (Recall@20 0.803), HippoRAG attained 0.308; a post-hoc document-ensemble sensitivity raised Recall@20 to 0.648 but did not eliminate the deficit. Relative to static expansion, staging reduces cross-encoder pair workload and measured reranking time by about 52%. A manually evaluated 100-query downstream study finds no significant root-cause-correctness advantage for AST-derived structural evidence.
[IR-19] Complexity and precision in quantum associative retrieval
链接: https://arxiv.org/abs/2611.21941
作者: Carlo A. Trugenberger
类目: Quantum Physics (quant-ph); Computational Complexity (cs.CC); Information Retrieval (cs.IR); Neural and Evolutionary Computing (cs.NE)
备注:
Abstract:We show that efficient and high-precision quantum associative retrieval is possible with an exponential complexity advantage with respect to the classical brute force approach and with a polynomial capacity advantage with respect to classical neural network implementations for any number of patterns scaling polynomially in the number of qubits. Both classical neural and quantum associative memories store the patterns in a Hamiltonian; in the classical case the dynamics is irreversible relaxation toward attractors, in the quantum case it is unitary transformations generating an effective thermal distribution, followed by a measurement.
人机交互
[HC-0] “I Just Dont Want to Be a Burden”: Investigating AI Chatbot Emotional Support Among People with Suicidal Ideation and Other Mental Health Concerns
链接: https://arxiv.org/abs/2611.23190
作者: Vasudha Varadarajan,Nathan S. Fishbein,Andrew Aquilina,Jordan Taylor,Hong Shen,Maarten Sap
类目: Human-Computer Interaction (cs.HC); Computers and Society (cs.CY)
备注: 41 pages, 9 figures
Abstract:People experiencing high-acuity mental health concerns are turning to AI chatbots for emotional support, but little is known about what drives this use, their experiences, and their preferences. In a sequential mixed-methods study, we inductively analyzed Reddit posts and used the findings to develop a survey. We surveyed 202 adults with recent mental health concerns who had used AI for emotional support. Most participants were already receiving clinical care and reported turning to AI, most commonly because they did not want to burden others, rather than being driven by barriers to care. Benefits were broadly shared, but participants reporting past-year suicidal ideation described more harm-facilitating AI behavior and reliance than those without. Overall, participants ranked honesty above warmth, and preference for crisis escalation declined as clinical acuity rose. These findings motivate AI support that addresses perceived costs of disclosure, evaluates benefits alongside harms, and widens feasible paths to human connection.
[HC-1] AI Companions for Young Adult Caregivers: Practical Advice Emotional Depth and the Limits of Metric Evaluation
链接: https://arxiv.org/abs/2611.23147
作者: Trizal Garg,Xuelong Fu,Dhruv Shah,Jin Huang,Shrigopal Maheshwari,Atiksh Aggarwal,Yash Sinha,Dhruv Kumar,Siddharth Mehrotra
类目: Human-Computer Interaction (cs.HC)
备注: Under Review
Abstract:Young adult caregivers (YACs) are individuals aged 18 to 25 currently taking care of a family member, partner, or friend facing chronic illness. AI companions designed specifically for YACs remain underexplored in prior works. We present Carebot, it uses an open-weight LLM (Gemma 3-27B) as its base model, on which we compared fine-tuning, retrieval-augmented generation (RAG), and a combination of both. The fine-tuning was performed based on an existing dataset of conversations with YACs and a psychologist. Three well-being experts evaluated these configurations and preferred the fine-tuned model for the warmth of its responses and its actionable advice tailored to YACs. Based on their choice, we adopted the fine-tuned model as Carebot’s final configuration and used it for all subsequent evaluations. An LLM judge preferred this Carebot’s responses over the base model’s in 85% of evaluated pairs. We then conducted a user study with 90 YACs, half of whom interacted with the base model and half with Carebot. The user study revealed a discrepancy between metric evaluation and actual user perception. Participants who used Carebot were more likely to characterize its replies as overly practical and brief, noting that it favored advice over emotional support. Conversely, participants who used the base model indicated that its responses were more emotionally oriented than practical. Although, fine-tuning made Carebot’s replies more practical, but they were not emotionally supportive for users. We argue that support tools for YACs require situation-sensitive emotional depth and aligned evaluation frameworks to study their effectiveness.
[HC-2] A Unified Choice Model of News Selection: Experimental Evidence on Stance Rank and Engagement Cues
链接: https://arxiv.org/abs/2611.23079
作者: Jacopo D’Ignazi,Andreas Kaltenbrunner,Michele Starnini,Vicenç Gómez
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:Social media platforms often display visual engagement cues alongside ranked news articles, yet their influence on clicks remains poorly understood when considered jointly with ranking position and users’ political preferences. We propose a unified discrete choice framework using a conditional multinomial logit model to jointly estimate and compare the influence of these factors on news selection. We evaluate the model using 1,480 choices from 370 human participants across four news topics in a controlled study with and without visual engagement cues. The framework allows us to assess the incremental contribution of each factor through nested-model likelihood-ratio tests. We find that including visual engagement cues significantly improves model fit across all topics after accounting for rank and stance. Engagement cues alone also predict clicks better than either of these features, although the full model achieves the highest held-out log-likelihood. Finally, we quantify via counterfactual analysis the impact of ranking position and displayed engagement counts on click probabilities. These findings can inform platform owners, researchers, and policymakers evaluating how platform design affects the selection of political news.
[HC-3] How Web Research Agents Interpret User Priorities
链接: https://arxiv.org/abs/2611.22989
作者: Robin Sharma,Parisa Suchdev
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI)
备注:
Abstract:Web research agents gather information online and write cited reports, including financial analysis. Users can state a priority to guide this research, but a priority may not say how to judge the options, leaving the agent to choose the criteria. We examine how two such agents, built on OpenAI and Anthropic models, interpret a stated priority. In 36 runs, the agents compared Microsoft and Alphabet as businesses. Prompts asked for special attention to growth or resilience, or gave no priority. We analyzed reports, searches, returned sources and cited sources. Of the 24 reports given a priority, 21 focused on it, but their criteria varied. Under the growth priority, OpenAI reports that ranked the companies accounted for risk or predictability and favored Microsoft, while the Anthropic ones judged growth by its pace and favored Alphabet. Under the resilience priority, reports used partly different criteria, and most named Microsoft as the more resilient company. Of the search and source measures tested, only the Anthropic agent’s search wording differed detectably by condition. These findings suggest evaluations should examine the criteria behind a report’s judgements, not only its focus or accuracy. Interfaces could let users see and adjust those criteria.
[HC-4] he Whole Warps the Parts: Attentional Certainty Modulates Ensemble-Induced Bias in Visualizations
链接: https://arxiv.org/abs/2611.22885
作者: Jennifer E. Corbett,Brian W. Anthony
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:Data visualizations often require viewers to extract both ensemble statistics and individual values from collections of graphical marks. Yet individual items embedded within an ensemble can be systematically biased by that context. Prior studies have reported both repulsion away from and attraction toward ensemble statistics, but these effects have been observed across different conditions. We test whether task or attentional certainty explains these opposing forms of ensemble-induced bias. In Study 1, observers briefly viewed ensembles of differently sized circles and then judged the size of an individual circle embedded within the ensemble. Using a common stimulus generation framework, we tested both identification and discrimination when the target was specified before or after the ensemble, and an additional no-cue identification condition without a specified target. Under the shared cued conditions, effects were predominantly repulsive or near zero in both identification and discrimination. In contrast, eliminating target-specific information reversed ensemble-induced bias from repulsion to attraction. Identification precision also decreased systematically from pre-cue to post-cue to no-cue conditions. Study 2 replicated this cue-dependent pattern in graphical displays of startup valuations and precinct vote contributions. Our findings isolate attentional certainty as a key determinant of whether ensemble context repels or attracts individual-item representations, extending the design question beyond precision to how attention directing choices determine which bias the viewer will be operating under.
[HC-5] AI as Anchor: How Interview Sequencing Relates to the Appropriation of Synthetic Stakeholder Voices in Design Education
链接: https://arxiv.org/abs/2611.22740
作者: Osnat Mokryn,Oshrat Ayalon,Noga Dines,Orit Shaer
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:Designers increasingly use large language models to simulate stakeholder conversations. Yet it remains unclear how AI-Simulated Stakeholders (AIS) are taken up when used alongside human stakeholder engagement. We report a field study in a single-institution, single-semester design course in which 8 teams (N=21) constructed, evaluated, and peer-reviewed an initial AIS. Then, the teams interviewed both AIS and human stakeholders in two sequences: AIS-first or human-first. Drawing on interview transcripts, design artifacts, course-embedded reflections, and post-course qualitative sessions, we examine exploratory patterns associated with these sequencing conditions. We find that students did not describe AIS as a neutral addition to the process. Instead, the order in which AIS and human stakeholders were formally interviewed was associated with different ways of interpreting and using information from the two sources. When the AIS interview came first, it provided a broad but often general frame that students later reconsidered through human interviews. When the AIS interview followed human engagement, AIS was more often used to revisit and validate insights already grounded in lived experience. Students described AIS as an “anchor,” but its reported role varied with its position in the formal interview sequence. In the final designs, a larger proportion of prominent human-stakeholder-derived concerns than AIS-derived concerns had corresponding design elements. We contribute: characterization of AIS as a reference point whose reported role varies with its position in the formal interview sequence; exploratory empirical evidence that the implemented interview-order conditions were associated with distinct patterns of interpreting and using stakeholder input; and implications for integrating AIS into design workflows in ways that preserve direct engagement with human stakeholders
[HC-6] An Affordable Vibrotactile Surface for Tactile Figure Recognition: Exploration Strategies of Blind and Sighted Users
链接: https://arxiv.org/abs/2611.22614
作者: Uday Nath Gunda,Arun Raman,Parthasarathy PD
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:Refreshable tactile displays can make graphics accessible to blind and low-vision people, but high-resolution actuator arrays remain expensive. We study whether a low-cost vibrotactile surface can convey simple figures, and how they are read. Nine participants (four blind; five sighted, tested blindfolded then open-eyed) identified seven figures by touch, with video and timing analysis of their exploration. Blind and blindfolded-sighted readers achieved comparable accuracy, but blind participants were about three times faster, using a continuous, rhythmic “gliding” strategy-reading through motion, whereas sighted participants searched in halting, pause-heavy passes; among the sighted, more pausing accompanied errors. Recognition failures were concentrated in specific figures (a diagonal was misread even with vision), locating the bottleneck in representation rather than resolution. We contribute an empirical characterization of tactile reading strategies and design implications for low-cost refreshable tactile displays and affordable tactile representation.
[HC-7] Controlling Proactive Multimodal AI: Separating Perception from Participation
链接: https://arxiv.org/abs/2611.22431
作者: Muawiz Feroze Khan,Fatima Ali,Yun Huang
类目: Human-Computer Interaction (cs.HC)
备注: 30 pages, 4 figures, 2 tables
Abstract:Proactive multimodal AI can watch a user’s screen, listen to speech, track interaction, and decide whether to enter unfolding work. Little is known about how people want to control what such AI may perceive and how far it may participate. Informed by a 14-person Wizard-of-Oz study, we built SpotMe, a design probe that senses screen, speech, and interaction; chooses Silence, Ask, or Assist; and lets users separately configure what it may perceive, how far it may participate, and what types of support it may provide. In a 16-person study on a civic dashboard, participants first experienced proactive help, then inspected decisions to act or remain silent, and then configured SpotMe’s perception and participation. We found that participants could keep all perception channels available while lowering participation to Ask or Silence; that exposing suppressed decisions made silence interpretable and contestable; and that seeing what SpotMe used led some to change how they clicked, highlighted, and spoke. We contribute an empirical account of inspecting and controlling proactive multimodal AI by decoupling perception from participation.
[HC-8] Plenty to FRET About: From Hazard Analysis to Formal Requirements for Robot-Assisted Dressing
链接: https://arxiv.org/abs/2611.22399
作者: Gunjari Chakraborty(University of Manchester),Yasmin Rafiq(University of Manchester),Hazel M. Taylor(University of Manchester),Marie Farrell(University of Manchester)
类目: Logic in Computer Science (cs.LO); Human-Computer Interaction (cs.HC); Robotics (cs.RO)
备注: In Proceedings FMAS 2026, arXiv:2611.21375
Abstract:Understanding requirements for systems involving humans and robots is essential for providing robust evidence that such systems behave correctly. Requirements formalisation approaches such as the Formal Requirements Elicitation Tool (FRET) can support the transition from unstructured natural language to formal representations suitable for verification. In this paper, we formalise 146 hazard-derived natural-language requirements for a Robot-Assisted Dressing (RAD) case study, previously classified using a taxonomy of Human-Robot Teamwork (HRT) requirements. Through an iterative process involving domain-expert clarification, these were represented by 117 unique FRETISH requirements. Formalisation exposed ambiguities, temporal and probabilistic assumptions, and cases where source requirements needed to be decomposed or could be consolidated because they expressed the same formal behaviour. Temporal analysis showed that 70.1% of the unique FRETISH requirements use either immediate-response or ordering constraints (immediately or before), while formalisation also exposed relationships between requirements across HRT taxonomy categories. We further analyse the formal semantics and realisability of the resulting requirements. The results show that HRT classification and formalisation provide complementary views: the taxonomy captures the teamwork concern represented by a requirement, while FRETISH makes the required behaviour explicit. We report practical lessons and limitations of applying FRET to HRT requirements and identify directions for further work.
[HC-9] An AI-Driven Assistant for Human-Centric Building Digital Twin Systems: Design Implementation and Comparative User Evaluation
链接: https://arxiv.org/abs/2611.22350
作者: Leonardo Messi,Erika Anneli Pärn,Juan Diego Castaño Molina,Oleksandr Anyshchenko,Dijar Selami,Peihang Luo,Borja García de Soto
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI)
备注: 43 pages, 8 figures, 9 tables. Submitted to Developments in the Built Environment
Abstract:Smart buildings generate streams of operational data, yet this richness rarely translates into accessible, decision-ready information for stakeholders who inhabit and manage them. Current Building Digital Twin Systems (BDTS) often remain oriented toward technical specialists and rely on static, one-size-fits-all interfaces. This paper presents the design, implementation, and comparative user evaluation of an AI-driven conversational assistant integrated into a human-centric BDTS for Indoor Environmental Quality monitoring, combined with adaptive information presentation and auto-suggest functionality. The prototype was deployed at New York University Abu Dhabi. A controlled, two-arm, between-subjects, quasi-experimental study with wave-based, non-randomized assignment compared the Baseline interface with the Experimental interface. Results showed significant reductions in task-completion time and perceived workload, while accuracy and usability improved but did not meet the adjusted significance criterion. The study contributes an implemented and empirically evaluated architecture that separates generative interaction from deterministic environmental assessment and coordinates conversational responses with adaptive visual presentation.
[HC-10] Big Brains Tiny Board: Pupavision an Agent ic Physical AI Instrument for Monitoring Drosophila Pupariation and Lifecycle Transitions
链接: https://arxiv.org/abs/2611.22136
作者: Joel Sadler,Daniel Graham,Melanie Worley
类目: Artificial Intelligence (cs.AI); Human-Computer Interaction (cs.HC)
备注: 6 pages, 4 figures
Abstract:Many biology experiments are limited by the availability and ease of use of instruments that monitor and record them. For example, the timing of pupariation in the fruit fly Drosophila melanogaster, a widely used model organism, can provide biologically relevant information about growth regulation and is often recorded manually. Here, we present Pupavision, a small physical AI instrument built from a single-board Linux computer and a day/night camera for roughly 45 to 75 US dollars in parts. The key idea is that a frontier coding agent runs its agentic harness directly on the tiny board. The model thinks in the cloud, but the agent works on the device. The agent assists image capture, writes code to repair errors, and checks the experiment every day to keep raw data capture reliable. If recording stops, the researcher gets a ping on their mobile phone. A simple deterministic capture service records one image every ten minutes and never depends on the agent. We report a 26-day pilot in a fly incubator (about 3,580 frames), including eleven days in which the agent was logged out and capture continued, and a 51-hour outage whose cause led to the layered design. A researcher-in-the-loop workflow then produces per-vial pupariation curves by tracing selected pupae backward in time from a late reference frame. Future work will validate the assay for lifecycle tracking accuracy against expert-labeled data and expand real-time computer vision analysis. Pupavision offers a blueprint for do-it-yourself agentic instruments that put frontier intelligence on a tiny board, with deterministic code underneath. We present this as an early case study from a pilot lab deployment and contribute lessons learned toward a future where scientific domain experts are empowered by DIY agentic hardware.
[HC-11] Projected Route Cues for Bouldering: Effects of Circular Linear and Combined Highlighting on Route Observation and Climbing
链接: https://arxiv.org/abs/2611.22093
作者: Yuki Uranishi,Tamon Hamamoto,Chang Liu,Goshiro Yamamoto,Masato Kobayashi,Photchara Ratsamee,Haruo Takemura
类目: Human-Computer Interaction (cs.HC)
备注: 12 pages, 11 figures, 7 tables. Submitted to ITE Transactions on Media Technology and Applications
Abstract:Before attempting a bouldering problem, climbers inspect the route to find its holds and plan their moves, which novices find difficult on walls where many routes overlap. We built a projector-camera system that projects route cues onto the wall and compared three cues: circles on the route holds, limb-colored lines connecting the holds, and both combined. Ninety-six participants, 61 of them first-time climbers, each climbed four routes, one per cue and one without projection, in a Latin-square order. In mixed-effects models, all three cues shortened the time before the first attempt; with circles, its geometric mean fell from 83 s to 39 s. Circles also increased the number of moves made within each trial. Lines led to more use of off-route holds than circles, mainly on routes where such holds lay under the projected lines. Participants preferred the combined cue, although no behavioral advantage of it over circles was detected.
[HC-12] hinking Harder Doesnt Lead to Better Choices: Reasoning Effort Increases the Confidence but Not the Correctness of LLM Shopping Agents
链接: https://arxiv.org/abs/2611.22033
作者: Adi Krishnamoorthy,X.Y. Han
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computers and Society (cs.CY); Human-Computer Interaction (cs.HC); General Economics (econ.GN)
备注:
Abstract:Large language models (LLMs) are increasingly deployed as autonomous shopping agents that process multimodal product listings (consisting of visual and text data) and recommend purchases on behalf of consumers. Naturally, one might assume that models with more thinking power would make higher-quality purchasing decisions. To test this belief, we ask two questions about such agents: does asking a model to think harder make it choose better products, and does the confidence it reports tell us the likelihood it may be wrong? We benchmark ten frontier models on three product-choice benchmarks built from WebShop, Amazon’s SQID and Wayfair’s WANDS platforms to see if LLMs with more thinking power actually make better-quality product purchase decisions. Across all three benchmarks, we found that increasing the thinking effort does not increase decision quality in a statistically meaningful way. The same change in thinking effort raises stated confidence at a statistically significant level, so extra thinking buys confidence rather than correctness. We also measure the calibration of these models (how well their reported confidence percentage matches their true correctness percentage) in these tasks and show that most models are not well-calibrated at every effort level. We also find that errors are strongly shared: when one model errs in a task, several others tend to err on that same task. Furthermore, on tasks where several models all err, a large number of them pick the same wrong product.
[HC-13] Supporting Researcher Control in AI-Assisted Social Media Analysis: A Mixed-Methods Study of AI4OSN
链接: https://arxiv.org/abs/2611.22006
作者: Qile Wang,Abhishek Karwankar,Shukria Muhammadi,Srikanth Kumar Perumalla,Abhin Suresh,Sunita Chandrasekaran,Kenneth E. Barner,Matthew Louis Mauriello
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:High-quality annotated datasets are critical for advancing computational social science (CSS) research, yet collecting, curating, and analyzing large-scale data from dynamic sources such as online social networks (OSNs) remains a persistent bottleneck. Researchers must also translate substantive questions into queries, relevance criteria, and defensible interpretations. Large language models (LLMs) can assist, but they can also obscure the judgments, potentially reducing researcher oversight and critical evaluation. We present AI4OSN, a mixed-initiative platform that keeps researchers in control of data collection, relevance labeling, and sensemaking while making AI assistance available as an inspectable advisory resource. In a systematic user study (N=18) using X/Twitter data, participants reported above-benchmark perceived usability (mean SUS score of 79.7), increased confidence in LLM-assisted annotation and analysis tasks, and low cognitive workload during AI-supported sensemaking. Notably, novice participants showed descriptively greater alignment and higher reported agreement with AI output than experienced participants, underscoring the need to support differentiated verification practices across expertise levels. Overall, this work contributes both a human-centered system for social data analysis and empirical insights into how AI-assisted interfaces shape researcher behavior, trust, and decision-making. Our findings highlight design opportunities and tensions in building accessible, transparent, and agentic AI systems for social media and web research.
[HC-14] Synchrony Affect and Perceived Autonomy in Virtual Robot-Facilitated Anxiety Support Groups
链接: https://arxiv.org/abs/2611.21977
作者: A’di Dust,Mohammad Soleymani,Mina Kian,Maja Matarić
类目: Human-Computer Interaction (cs.HC)
备注: 17 pages, 7 figures
Abstract:Socially assistive robots could expand access to group-based mental health support when skilled human facilitators are unavailable. However, their effectiveness will depend on whether they foster productive interactions among participants. We examined dyadic head-movement synchrony among 96 participants across 18 virtual robot-facilitated anxiety support group sessions. A human-controlled virtual robot, via a Wizard-of-Oz (WoZ) system, facilitated all sessions. We found that stronger peak synchrony predicted lower post-session negative affect. In contrast, speakers who more often followed listeners’ movements reported higher negative affect and state anxiety, demonstrating that synchrony strength and direction capture distinct interpersonal processes. Participants who believed the robot was autonomous showed more genuine synchrony and stronger peaks than those who believed it was human-operated; more negative attitudes toward robots also predicted greater synchrony. These findings support synchrony as an objective, behaviorally grounded measure for evaluating robot-facilitated groups and a potential signal for guiding future autonomous facilitators.
[HC-15] Its My Turn! Interpretable Methods Reveal Developmental Differences from Mother-Child Conversational Dynamics
链接: https://arxiv.org/abs/2611.21969
作者: A’di Dust,Anura Deshpande,Pat Levitt,Maja Matarić
类目: Human-Computer Interaction (cs.HC); Computation and Language (cs.CL)
备注: 9 pages, 5 figures
Abstract:Understanding conversational interactions between mothers and children is an important challenge for computational behavioral modeling. Existing methods use traditional statistical approaches and subjective measures of conversational dynamics, without incorporating temporal or probabilistic modeling. We developed an interpretable computational framework that combines sequence analysis, information-theoretic measures, and Bayesian hierarchical modeling to characterize temporal conversational dynamics in parent-child play. We applied this framework to interactions from 94 mother-child dyads with two-year-old children. Sequence analysis revealed that dyads with lower language development (as defined by scores on the Bayley developmental test) exhibited more heterogeneous conversational organization and fewer serve-and-return exchanges than higher-development dyads, a reciprocal turn-taking pattern in which one partner initiates communication and the other responds. Mutual information analyses suggested differences in conversational predictability between lower- and higher-development dyads, although the bootstrap confidence intervals overlapped. Bayesian hierarchical models estimated temporal trajectories for maternal consecutive turns, child responses following maternal speech, and overlapping responses; HDIs showed that trajectories differed in whether they excluded 0, but there were no statistically credible differences between the Bayley groups. Overall, these results indicate that children with higher developmental scores have more organized and predictable communication styles, with more ``serve-and-return" patterns with their mothers. These results demonstrate that interpretable computational representations of conversational dynamics capture developmental differences beyond simple measures of speech quantity.
[HC-16] SimAI: Mechanistic User Simulation through Psychological Network Models
链接: https://arxiv.org/abs/2611.21808
作者: Cheng Li,Adhyyan Narang,Jiayi Yuan,John F. Buss,Jina Suh,Devon Sandel-Fernandez,Yulia Tsvetkov,Adam M. Kuczynski,Tim Althoff
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Human-Computer Interaction (cs.HC)
备注: 47 pages
Abstract:Realistic user simulation enables the study of human interactions when direct experimentation is costly or potentially harmful. However, existing simulators largely overlook the psychological mechanisms that drive user behavior. We introduce SimAI, a framework for mechanistic user simulation that models how psychological factors interact to shape user responses. To overcome the challenge of supervising open-ended behavior, we propose a graph reconstruction objective that provides graph-level rewards based on how well simulated conversations reflect the underlying psychological relationships. To support this approach, we construct PSYCHE, a dataset of 18,576 profile-network pairs derived from 516 psychotherapy cases. Automatic evaluations show that SimAI achieves greater simulation and behavioral diversity than existing patient simulators while maintaining strong profile alignment. In an interactive study with 30 mental-health professionals and trainees, SimAI receives the highest average ratings for conversational authenticity, realistic help-seeking behavior, and clinical training utility. Beyond simulation quality, SimAI enables realistic clinical training and uncovers safety failures in prolonged AI interactions with vulnerable users. Code is available at this https URL.
[HC-17] RelicTouch: Exploring Fingertip Vibrotactile Feedback for Image Interpretation
链接: https://arxiv.org/abs/2611.21642
作者: Xiaosong Wang,Ran Liu,Yifan Feng
类目: Human-Computer Interaction (cs.HC)
备注: 7 pages, The 23rd IEEE International Conference on Ubiquitous Intelligence and Computing
Abstract:Vibrotactile feedback is often used to simulate material properties on interactive surfaces, but less is known about how it can help users interpret visual differences during touchscreen exploration. We present RelicTouch, a finger-worn system that converts image-derived features into movement-contingent vibration. Study 1 was a controlled within-subject experiment (n=20) examining how vibration intensity, stripe spacing, and finger configuration affected haptic–visual congruence and subjective clarity. Study 2 was an exploratory qualitative study with eight domain experts examining how users interpreted cues generated from luminance and edge variation during continuous image exploration. Results suggest that cue legibility was associated with distinguishable signal changes, movement-based comparison, and visual or task context. Single-finger exploration yielded higher subjective ratings, whereas multi-finger feedback showed possible value for coarse comparison across separated features. These findings provide initial design implications for vibrotactile image exploration beyond material-simulation-oriented rendering.
[HC-18] RAW: Refusal Aversion in the Wild A Causal Measurement Method for Deployed LLM s EMNLP2026
链接: https://arxiv.org/abs/2611.21641
作者: Pavel Kireyev
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Human-Computer Interaction (cs.HC); General Economics (econ.GN)
备注: Accepted to the EMNLP 2026 Industry Track. 26 pages incl. appendices, 8 figures, 11 tables. Code: this https URL Data: this https URL
Abstract:Deployed LLMs field hundreds of millions of conversations a day and refuse a non-trivial fraction of user requests, yet the user-side cost of refusal, whether users disengage and by how much, is unmeasured on any production model: a blind spot for the consumer-facing platforms built on them. A/B-testing refusal is ethically and operationally constrained on frontier-lab products and unavailable to third-party platforms deploying these models through a fixed API. We introduce RAW, a method that uses LLM stochasticity as a natural experiment: at near-identical prompts in any existing log, the same model sometimes refuses and sometimes does not, and that variation identifies the causal user-side effect of refusal. Applying the method to 15 deployed LLMs in LMSYS-Chat-1M and WildChat, with user-clustered inference, we find that for about half of the models we can measure precisely, a refusal costs 3 to 17 pp of subsequent re-engagement; the rest show no distinguishable effect, among them one deployment whose positive point estimate we trace to persistent users retrying the refused request. Our contribution is the measurement method, which runs on any deployed model from logs alone for about \ 0.30 per 1,000 conversations. We release labels, moderation scores, prompt embeddings, and pipeline code.
[HC-19] Digital Qiang Garden: Artist Residency as Interface for Creativity Under Change
链接: https://arxiv.org/abs/2611.21640
作者: Yunyi Rainee Wang,Joshua Nijiati Alimujiang
类目: Human-Computer Interaction (cs.HC)
备注: 14 pages
Abstract:This research presents Digital Qiang Garden, an artist-led residency situated in Qiang ethnic minority villages in western Sichuan, China, as a practice-based inquiry into creativity for change. Rather than framing creativity as a discrete outcome or individual agency, the project examines how creative practice unfolds through lived conditions under uncertainty. Drawing from three years of residency practice, the pictorial identifies four creative modes of praxis that repeatedly emerged through shared living, material engagement, and cross-media collaboration. Through annotated visuals and project-based case studies, we further trace how these modes generated eight interrelated layers of change, spanning artistic processes, social relations, cultural encounters, ecological awareness, organizational structures, and artists’ longer-term trajectories. By framing an artist-led residency as a transferable structure, this work contributes to HCI and creativity research by foregrounding designed conditions, situated action, and relational processes through which change unfolds over time.
[HC-20] ClearTalk: Data-Driven Mandarin Articulation Analysis and Personalized Communication Support for Children with Dysarthria ICDM2026
链接: https://arxiv.org/abs/2611.21639
作者: Ben Xiaolu Huang
类目: Human-Computer Interaction (cs.HC); Computation and Language (cs.CL)
备注: 5 pages, 3 figures, 1 table. Accepted to the IEEE ICDM 2026 Teen Research Track
Abstract:Children with dysarthria may experience reduced intelligibility and difficulty expressing everyday needs. ClearTalk is a smartwatch-based system that converts Mandarin speech-recognition differences into interpretable articulatory profiles for personalized practice and staged communication. Target and recognized utterances are aligned at the syllable level, summarized across seven articulatory positions, and ranked by a count-sensitive asymmetric heuristic. For 13 dysarthric speakers from the Mandarin Dysarthria Speech Corpus, system syllable-level accuracy showed a Pearson correlation of 0.908 with expert intelligibility, with an R-squared value of 0.824 and a p-value below 0.001. The system-selected priority targets achieved an 86.7 percent overlap with the consensus of three blinded speech-language pathologists; across 210 target consonants, automated labels achieved 86.7 percent agreement, a Cohen’s kappa of 0.76, and a macro-F1 score of 0.81. In a balanced single-session pilot with 15 children, mean pre/post gains were 14.4, 29.4, and 27.7 percentage points for mild, moderate, and severe groups, while communication-round completion rates were 81.3, 66.7, and 48.0 percent. These findings support the feasibility of using interpretable patterns of speech errors to guide wearable articulation practice and communication assistance.
[HC-21] Can AI Help Explain AI? Effects of Multimodal LLM -Generated Explanations on Perceived Understanding and Perceived Trust in XAI Visualizations
链接: https://arxiv.org/abs/2611.21638
作者: Mario Truss,Marc Schmitt,Stefan Feuerriegel
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI)
备注:
Abstract:Explainable artificial intelligence (XAI) aims to make machine learning models understandable, yet many XAI visualizations remain difficult for end-users to interpret. Here, we evaluate whether augmenting XAI outputs with textual explanations from multimodal LLMs improves perceived understanding and perceived trust. We conducted a pre-registered within-subject experiment ( N = 128 ) across seven decision scenarios and five XAI methods, with and without LLM generated explanations. The explanations improved perceived understanding and perceived trust by 0.35 and 0.26 points on five-point scales (both p .001 ). The gain appeared in all seven scenarios, did not differ between local and global explanations, and was largest for participants with the least prior XAI exposure. A mediation analysis indicates an indirect effect on perceived trust through perceived understanding. These results suggest that multimodal LLMs can provide a useful interpretive layer for XAI beyond expert audiences.
[HC-22] Auditing Hyperstimulation at Scale in Child-Facing YouTube Videos
链接: https://arxiv.org/abs/2611.21637
作者: Shao-Yu Chu,Noel Mekogue,Qiyu Li,Xuan Liu,Shuang Hao,Haojian Jin
类目: Human-Computer Interaction (cs.HC); Computers and Society (cs.CY); Multimedia (cs.MM)
备注:
Abstract:Parents, pediatricians, and journalists have raised concern that child-facing videos are becoming increasingly overstimulating, but this concern rests mostly on ad hoc observation rather than systematic measurement. We present a large-scale audit of hyperstimulation patterns, which we refer to as content-independent production choices that may exceed young children’s capacity to process comfortably. Drawing on parents’ Reddit discussions and developmental psychology, we identify 12 hyperstimulation patterns and operationalize them as computational metrics. We apply these metrics to 16,375 videos from 542 child-facing YouTube channels spanning 2015-2026. We found broad increases in stimulation over time, with cut rates rising across all five content categories. Nearly a quarter of the channels increased in at least one stimulation pattern over time. More stimulating videos also tended to receive more views, but the relevant stimulation patterns differed across content types.
[HC-23] Ethics Suppression in AI Development: A Comparative Case Study of Responsible AI Under Competitive Pressure
链接: https://arxiv.org/abs/2611.21629
作者: Victor Frimpong
类目: Human-Computer Interaction (cs.HC); Computers and Society (cs.CY)
备注: Peer-reviewed article
Abstract:This study explores the governance tensions among Google, OpenAI, Meta, and Microsoft, focusing on how competitive AI development can lead to the ethics suppression. The study aims to examine how ethical oversight loses practical influence even when responsible AI structures are formally in place, especially during rapid technological competition. Despite the rapid growth of responsible AI frameworks and ethical principles in the tech sector, there has been less focus on how these concerns are actually applied when pressure to deploy increases. This paper introduces the concept of ethics suppression in AI development, highlighting that while ethical oversight may exist on paper, its practical authority can diminish during decision-making. Using a comparative qualitative case study examining corporate governance documents, public statements, and reports on governance issues, we identify common issues such as reduced authority, a weakened ability to escalate concerns, urgency in deployment, and fragmented governance. Our findings indicate that heightened competition can weaken the influence of ethical oversight, even in organisations with established responsible AI practices. By differentiating ethics suppression from ethics washing, it offers a framework for organisations and policymakers to evaluate the effectiveness of ethical oversight in competitive AI development.
[HC-24] PDFa11yMut: Mutation-Based Evaluation of Automated PDF Accessibility Validators
链接: https://arxiv.org/abs/2611.21628
作者: Gauri Jain(University of California, Santa Cruz)
类目: Human-Computer Interaction (cs.HC)
备注: 7 pages, 3 figures, 5 tables. Research artifact: this https URL
Abstract:Automated validators are widely used to assess PDF accessibility, but their detection boundaries are difficult to measure without controlled ground truth. A validator-clean PDF may be free of the implemented machine-checkable failures, or it may contain accessibility-relevant structural defects that the validator does not detect. This paper introduces PDFa11yMut, a controlled mutation-based benchmark for evaluating automated PDF accessibility validators against independently verified PDF structure defects. We generated 23 verified mutants from five validator-clean golden PDFs using five mutation operators targeting reading order, omitted assistive structure, heading hierarchy, marked-content association, and internal marked-content sequence. Each generated mutant was independently verified at the PDF structure level and checked for parseability, page-count preservation, and absence of detected visual differences. We then evaluated the mutants with three validators: PAC, Adobe Acrobat Accessibility Checker, and veraPDF, with veraPDF run using the PDF/UA-1 profile. The validators detected a small and highly operator-specific subset of the verified mutants. PAC detected 4/23 mutants (17.4%), while Acrobat and veraPDF each detected 5/23 mutants (21.7%). All three validators detected every generated heading-hierarchy mutant. None detected the reading-order swap, omitted-structure, or internal MCID-order mutants. Acrobat and veraPDF also detected one marked-content association mutant that PAC missed. The practical implication is that a validator-clean result can coexist with verified, visually invisible structure defects that affect the assistive representation of a PDF. Comments: 7 pages, 3 figures, 5 tables. Research artifact: this https URL Subjects: Human-Computer Interaction (cs.HC) Cite as: arXiv:2611.21628 [cs.HC] (or arXiv:2611.21628v1 [cs.HC] for this version) https://doi.org/10.48550/arXiv.2611.21628 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
计算机视觉
[CV-0] Spool: Low-Latency Streaming Inference for Hybrid Vision-Language Models
链接: https://arxiv.org/abs/2611.23228
作者: Rui Ning,Om Chabra,Hari Balakrishnan
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Applications such as visual accessibility agents and robotics require answering questions about recent video content promptly, over a bounded window of past frames (e.g., the last two minutes). Hybrid vision-language models (VLMs), which interleave attention with linear recurrent layers such as Gated DeltaNet (GDN), are attractive for this setting, but maintaining a sliding window is expensive. The reason is that recurrent layers combine all past frames into a shared state, so expiring old frames normally requires re-running the entire window through the model. We observe that the recurrent update is affine in the state with input-dependent coefficients, so the effect of each frame can be captured as a transformation that composes associatively. We present Spool, a training-free framework that caches per-frame state transformations and composes them efficiently as the window advances, removing expired frames from each recurrent layer’s state without replaying frames through the backbone. Removal is exact within each layer; as with key-value eviction in attention, expired frames can retain residual influence through the cached inputs computed by earlier layers. We design a dedicated data structure and GPU kernels to achieve low caching and composition overhead. Across five open-source hybrid VLMs and four video reasoning benchmarks, Spool achieves the lowest p95 time-to-first-token in all 20 model-benchmark tests, reducing it by a geometric mean of 6.6\times over the faster of two recent vLLM baselines, while improving accuracy by 2.3 percentage points on average.
[CV-1] FastFrames: Video Generation at Interactive Latency FAST
链接: https://arxiv.org/abs/2611.23226
作者: Gene Chou,Guandao Yang,Fujun Luan,Lars Mescheder,Wei Dong,Stephan Richter,Vladlen Koltun
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project page: this https URL
Abstract:Autoregressive video diffusion models generate frames based on new observations and inputs, making them a promising medium for interactive visual experiences. However, existing models are orders of magnitude too slow for fluid interaction. This paper introduces FastFrames, a family of fast autoregressive video diffusion models that achieves interactive latency without compromising quality. We profile the inference pipeline of state-of-the-art models and find that operations often treated as fixed implementation details, such as VAE decoding and KV cache updates, in fact account for over half of end-to-end inference latency. We challenge the assumption that these components are indispensable and replace them with lightweight alternatives. Our 1.3B variant runs at 36 ms per frame, or 28 responses per second (RPS), on a single B200 GPU while matching the generation quality of a state-of-the-art model that is nearly 8x slower. Our 14B variant runs at 45 ms per frame, or 22 RPS, on an 8x B200 GPU node and yields significantly better visual quality and realism. We show that FastFrames enables real-time interaction in applications such as camera-conditioned and physics-conditioned video generation.
[CV-2] EviSlide-MIL: Multi-Query Pooling and Shared Prediction for Class-Incremental Whole-Slide Image Classification
链接: https://arxiv.org/abs/2611.23212
作者: Zhouhan Qian,Sicheng Chen,Chad Wong,Tianyi Zhang,Enhui Chai,Zeyu Liu,Fei Xia
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Histopathological analysis of whole-slide images (WSIs) plays a critical role in cancer diagnosis, and multiple instance learning (MIL) is widely adopted for WSI analysis under weak supervision. As new tumor subtypes are introduced, extending existing MIL models requires learning new classes while retaining knowledge of previously learned ones. This presents a class-incremental learning challenge, where models must acquire new classes with limited access to historical training data and recognize all learned classes without task identifiers at inference. A major obstacle is forgetting, as updates to feature aggregation and prediction modules can degrade performance on previously learned classes. To address this challenge, we propose \textbfEviSlide-MIL, a class-incremental WSI framework that integrates multiple learnable queries per class with a shared prediction module. The queries aggregate frozen patch features into class-specific representations, while the shared prediction module limits parameter growth across incremental stages. By freezing queries associated with previously learned classes, EviSlide-MIL preserves their attention patterns for identical inputs, while experience replay and knowledge distillation further mitigate forgetting as shared parameters are updated. Experiments on eight tumor subtypes from The Cancer Genome Atlas demonstrate that EviSlide-MIL achieves improved performance and reduced forgetting compared with competing methods across most learning settings, demonstrating its potential for scalable and knowledge-preserving WSI classification. Code and pretrained models will be released.
[CV-3] Multimodal Masked Diffusion Autoencoder for Modeling Impaired Driving CVPR2027
链接: https://arxiv.org/abs/2611.23211
作者: John Gideon,Anshul Gupta,Megan Applegate-Kenton,Kayli Battel,Todd J Rowell,Patricio Reyes Gomez,Laporsha Dees,Emily Sumner,Avinash Balachandran,Guy Rosman
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Robotics (cs.RO)
备注: Equal contribution: John Gideon and Anshul Gupta. 8 pages, 6 figures. Submitted to CVPR 2027
Abstract:Human cognitive state (e.g., distraction, stress), gaze, and behavior interact with the driving environment in subtle ways that AI could use to assist drivers, but ground-truth labels for human cognitive state levels are costly to obtain at scale. We propose a multimodal masked-diffusion autoencoder, ID-MAE, that adds a self-supervised objective to impairment prediction: alongside predicting the impairment label as a masked token, it reconstructs masked gaze and trajectory. To probe the effects of different modalities and types of impairments on model inference, we introduce DIVERT, a driving-simulator dataset containing 103 hours of driving data (n = 109) with ten impairment inductions spanning cognitive, visual, and emotional states, recorded with gaze, vehicle state, and scene video. We demonstrate how ID-MAE affords significant improvements in impairment inference results on both the DIVERT dataset and existing datasets, and enables strong results in additional tasks, such as trajectory and gaze prediction. For location-matched pairs, our model achieves a paired AUC of 0.706 for impairment detection on DIVERT, compared to 0.590 without reconstruction loss, improving over existing baselines. We probe the design space in terms of input modalities, masking scheme, positional encoding, attention mechanism, and decoding. We also investigate the interaction between various impairments in training and inference, demonstrating that visually demanding distractions are separable from cognitive distractions. Code and data will be available upon acceptance.
[CV-4] Learning Grounded Reasoning from Human Gaze
链接: https://arxiv.org/abs/2611.23209
作者: Gabriel Sarch,Declan Campbell,Nicholas Budny,Kia Ghods,Katerina Fragkiadaki,Jonathan D. Cohen,Brenden M. Lake,Taylor W. Webb
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Human visual reasoning is driven by systematic exploration and extraction of task-relevant visual evidence. Human gaze provides a window into this process, making gaze a useful source of guidance for visually-grounded reasoning in vision-language models (VLMs). However, human gaze also includes noise and redundancies that may transfer poorly to VLMs. In this work, we investigate how human gaze data can be effectively leveraged to improve visual reasoning in VLMs, while maintaining alignment with human behavior. We collect 7,109 human gaze trials from a diverse and challenging set of visual reasoning tasks, and train grounded reasoning models with different forms of gaze guidance. Across evaluations in chart understanding, visual search, and counting, we find that using human gaze as a prior for grounded reasoning consistently outperforms supervision using the full gaze sequence. Based on these findings, we introduce GGR (Gaze-Guided Reasoner), which achieves greater human alignment and more reliable visual grounding while matching or outperforming leading grounded-reasoning baselines in accuracy. Reinforcement learning further improves task accuracy, gaze alignment, and grounding quality despite using no gaze supervision. Behavioral analyses show that gaze supervised models exhibit more visual exploration before arriving at an answer. Our results suggest that the value of human gaze lies in guiding spatial exploration and shaping reasoning behavior without requiring complete imitation of human behavior.
[CV-5] Distortion-Free Semantic Watermarks for Black-Box Generative Models
链接: https://arxiv.org/abs/2611.23208
作者: Aviv Yossef,Or Patashnik,Or Zamir
类目: Machine Learning (cs.LG); Cryptography and Security (cs.CR); Computer Vision and Pattern Recognition (cs.CV)
备注: Project page: this https URL Code: this https URL
Abstract:We present a provably distortion-free semantic watermarking algorithm that requires only black-box sampling access to a generative model and black-box access to a semantic embedding. A watermark is called distortion-free if, averaged over its secret detection key, it leaves the generator’s output distribution unchanged; this provides a formal guarantee that watermarking does not reduce output quality. Conventional such watermarks are often brittle to transformations. This is especially problematic for images, where benign operations such as compression, cropping, or resizing change the entire pixel representation. Semantic watermarks therefore require detection to be robust to content-preserving transformations. Previous methods combining these properties rely heavily on specific white-box properties of the underlying model: either to autoregressive text generation or to diffusion-specific access, such as control of the initial noise or diffusion inversion. In contrast, our construction is fully generic: It only interacts with the generative model and the semantic embedding, a function that determines the notion of “preserving content” under which detection should be robust, via sample and query access. We then instantiate this framework for image generation. We characterize the two competing properties required of the semantic embedding: it should remain stable under transformations that preserve an image’s content, but should distinguish independent generations from the same prompt. We train an image encoder directly for this tradeoff. The resulting encoder improves separation while maintaining stability over off-the-shelf alternatives and enables strong aggregate detection while remaining robust to various transformations. We demonstrate the effectiveness of our watermark scheme on both open-source and closed models, which is possible only due to restriction to black-box access.
[CV-6] Reading Scene States: Uncovering Safety Failures in Vision-Language Models
链接: https://arxiv.org/abs/2611.23206
作者: Tongxi Wu,Yuyang Cheng,Jiacheng Xian,Menglin Yang,Kai Chen,Jian Zhang,Yang Gao
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: VLM safety, multimodal jailbreaks, and scene-state reasoning
Abstract:Vision-language models (VLMs) can provide harmful assistance by interpreting visual events, even without an explicit request for the underlying harmful objective. We identify this \textbfscene-state safety vulnerability and introduce \textbfSceneSafetyBench, comprising 375 image–question pairs spanning Initial, Medium, and Ultimate scene states and five visual presentations. Across eight VLMs, scene-state inputs elicit harmful assistance with attack success rates (ASRs) of 45.6–72.3%. Comparative analysis across three VLMs shows that scene-state inputs most closely resemble QR-related images in their layer-wise safety-related signal profiles, while exhibiting lower average late-layer readouts and distinct signal dynamics during generation. To turn this vulnerability into a systematic attack, we propose \textbfS3TA, which translates harmful objectives into scene states, queries image–question pairs independently, and assembles the responses. On seven MM-SafetyBench categories across eight open-weight and commercial VLMs, S3TA achieves a model-average ASR of \textbf71.2%, exceeding the strongest baseline by 6.1 percentage points. Text-side filters and visual moderation provide limited protection. These findings highlight the need to safeguard the assistance models infer through scene-grounded reasoning, beyond screening input content.
[CV-7] Retrieval-Augmented Correction for Generalist Medical Image Segmentation
链接: https://arxiv.org/abs/2611.23189
作者: Vincenzo Marcianò,Xiaoming Zhang,Michela Antonelli,Sebastien Ourselin,Maria A Zuluaga
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: Accepted in Transactions on Machine Learning Research (10/2026). Reviewed on OpenReview: this https URL
Abstract:Automated segmentation is routinely used as the first step in quantitative image-based clinical workflows, where predicted masks are subsequently reviewed and corrected by human operators. Automated correction of the residual structural errors is currently addressed one anatomy at a time, so the number of trained correctors scales with the number of segmentation targets. We present GRACE, a retrieval-augmented refiner that edits an input mask under context drawn from a shared knowledge base of verified prior cases. The refiner is trained once on a multi-organ corpus spanning compact organs, cavities, branching vessels, and tumour lesions. In the held-out dataset-transfer evaluation, three held-out datasets query a fixed knowledge base built from training data only. A separate leave-one-anatomy-out experiment tests parameter-free index extension: verified cases of an unseen solid organ enter the knowledge base, and the refiner stays frozen. A light shape-aware projection head on a frozen multimodal encoder maps embedding similarity to the shape agreement of the masks, so the knowledge base returns cases that the refiner can use as a shape prior. We evaluate GRACE on 18 public datasets spanning CT, MRI and ultrasound, using nine refinement configurations and seven upstream segmentors that span weak, mid, and strong operating regimes. We report each change with the DSC of its input mask. On the weak tier, the input DSC is 0.386 and GRACE adds up to +0.35 DSC. On a strong segmentor, the input DSC is 0.932 and the mean change is small. Transfer to an imaging modality that is absent from training and from the index is partial and depends on the anatomy, and remains an open research direction. Our results suggest retrieval-conditioned refinement is a practical alternative to anatomy-specific correction models. Code: this https URL
[CV-8] Any2Building: Unifying Point Clouds and Images for Building Reconstruction via Shared Spatial Queries
链接: https://arxiv.org/abs/2611.23187
作者: Hongye Hou,Zhuoxiao Li,Liu Xiangyu,Tengxi Wang,Jing Ou,Dongli Wu,Shuai Zhang,Tongyan Hua,Wufan Zhao
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Abstracting structured buildings from limited observations is essential for 3D urban modeling and digital twins. Partial point clouds and overhead images of the same building provide incomplete 3D surface measurements and visual cues to its footprint, roof geometry, and rooftop structures, respectively. This motivates a unified framework that aligns geometric and visual cues in 3D and reconstructs buildings from either or both modalities. Building on this connection, we present Any2Building, a unified model for Level of Detail 2 (LoD2) building reconstruction from partial point clouds, overhead images, or both, with optional coarse geometric priors. At its core is an occupancy-grounded shared spatial queries representation that organizes available observations within a common 3D grid. Supervised by complete building surface voxels, these queries connect observational evidence to target geometry, enabling a shared generator to learn structural priors across tasks and providing spatial guidance for recovering missing structure. A trained voxel conditional surface-generation and low-poly remeshing pipeline then converts the predicted structure into compact LoD2 building models. To support this framework, we introduce BuildingWorld-Image, comprising approximately one million paired samples of partial point clouds, overhead images, and LoD2 building models across 18 cities. Experiments on multiple benchmarks demonstrate that Any2Building outperforms existing methods in geometric reconstruction accuracy across unimodal and multimodal point-cloud completion, single-image building generation, and prior-assisted generation, while maintaining detailed and robust recovery when point cloud input modality is degraded.
[CV-9] Say Less Hallucinate Less? Diagnosing and Restoring Object Coverag e in Preference-Optimized Vision-Language Models
链接: https://arxiv.org/abs/2611.23179
作者: Hansang Lee
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: 18 pages, 6 figures, 19 tables. Main text and appendices
Abstract:Preference optimization (PO) is the standard way to reduce object hallucination in large vision-language models. We find that four public PO checkpoints of LLaVA-1.5-7B also lower object coverage. They name fewer object categories, and their lower hallucination rates bring no significant gain in set-level object F1 on CHAIR. We therefore ask how much of the PO gain the base model can reproduce by stopping earlier. A positive end-of-sequence (EOS) logit bias lets the base model reach the recall of each PO model, and this early stopping alone reproduces 58 to 86 percent of the CHAIRi gain of three checkpoints on two disjoint COCO sets. Writing longer does not bring the lost categories back. Under greedy decoding, a negative EOS bias recovers only 2 to 9 percent of the lost CHAIR recall of these checkpoints, and 94 to 96 percent of the added object mentions repeat categories already named. On CHAIR, the base caption of the same image still names about half of the missed object categories. We therefore propose base-guided object-category restoration, which appends high-confidence base-only categories to the PO caption without training. With thresholds fixed on a development set, it brings three checkpoints within one point of the base recall on a held-out COCO set with a lower distinct-object hallucination rate than the base model. Against the recall-matched base stopping curve, its gain holds for RLAIF-V near the base recall and for RLAIF-V and CHAIR-DPO at an intermediate recall. We will release the code of the recall-controlled evaluation.
[CV-10] Déjà View: A Shared Working-Memory Model for Visual Tasks
链接: https://arxiv.org/abs/2611.23177
作者: Mranmay Shetty,Gabriel Kreiman
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Neurons and Cognition (q-bio.NC)
备注: 6 figures; includes appendices
Abstract:A working-memory model must remember stimuli and use its memory flexibly to perform different tasks. Here we introduce a model of working memory for images that combines dynamic activity with temporary associations between image features. We train the model on five tasks, including remembering a sample while other images are presented, categorizing distractor images while remembering the sample, and determining whether an image is new or has been presented before. The same trained model performs all tasks, guided by explicit cues about which response is required. We use this flexible memory model to ask how delay, distraction, and image history affect performance. Delays have little effect on sample-memory performance for the delays we tested, while seeing more images leads to more false alarms even when the target image appeared equally recently in all cases. On the sample-memory-plus-categorization task, the probability of correct responses on all questions decreases as more questions are asked, despite robust sample-memory performance. Finally, the model maintains stable performance across 40 brief trials of each task under the tested conditions. Together, these results demonstrate how a single working-memory system can be flexibly used to perform different tasks, and how its errors depend on both previous images and task demands.
[CV-11] Attribution Through the Decoder: Prediction-Space Explanations for Structured Outputs
链接: https://arxiv.org/abs/2611.23161
作者: Zaer Bin Zaqy,Brinnae Bent
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: Under review
Abstract:Networks that predict angles, gaze directions, and rotations do not output these quantities directly: they emit raw coordinates that a fixed decoder, such as normalization, maps to the final prediction. Attribution methods are usually applied to the raw coordinates, so they can credit inputs for changes the decoder discards, such as rescaling a vector when only its direction matters. The resulting explanation can highlight inputs that have no effect on the prediction. We instead attribute through the decoder differential, which for normalization decoders reduces to a scaled tangent projection. We characterize invariance to output rescaling, show that every fixed smooth scalar target on a sphere has points of zero tangential sensitivity, and show that an endpoint-projected variant of Integrated Gradients can still attribute to scale-only inputs, whereas direct angular Integrated Gradients cannot. On synthetic tasks with analytically known relevance, ambient attribution assigns up to 72% of its mass to an irrelevant region. On MPIIGaze validation data, tangent Grad-CAM produces 44% greater prediction displacement under 10% pixel deletion. On PASCAL3D+ with a 6D rotation decoder, the full differential improves mean deletion displacement by 14.1 degrees over the ambient coordinate sum and by 13.8 degrees over a decoder-invariant scalar. Results for a two-polarity tangent LayerCAM variant are mixed: it exceeds ambient-sum LayerCAM on circular regression but no difference is detected on quaternion pose. These results separate two failure sources, radial contamination and scalarization loss. The approach uses existing attribution methods with decoder-aware targets, although some variants also require changes to attribution aggregation. An invariant attribution can still miss inputs that drive the prediction, so invariance and deletion performance should each be checked.
[CV-12] VIVID: Beyond Static Multi-Image Reasoning with Localized Visual Differences and Textual Entity Dynamics in Video MLLM s
链接: https://arxiv.org/abs/2611.23153
作者: Chengqian Ma,Yuanze Zeng,Hongyu Wang,Mingfeng Lin,YiCheng Xiao,Yu Chen,Pengyu Li,WenHao Wang,Yuexiao Ma,Xiawu Zheng,Fei Chao
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 14 pages, 2 figures, 5 tables. Code: this https URL
Abstract:Video question answering under tight frame budgets requires preserving localized motion and event order, yet current video multimodal large language models (MLLMs) reduce comprehension to static multi-image reasoning by discarding unselected frames and encoding keyframes independently: (1) Spatially, whole-frame pooling dilutes small moving targets into static backgrounds; (2) Temporally, dropping intermediate frames or averaging tokens erases inter-frame motion, confusing direction, speed, and order. We propose VIVID (Visual Inter-frame Variations and Individual Dynamics), which moves beyond static snapshots like human vision: keyframes capture core states while inter-frame Patch Diffs preserve neighboring changes within K_f keyframes. Autonomous Delta-Driven Keyframe Planning (ADKP) selects anchors from a 3\text fps pool and feeds surrounding patch differences into Dual-Road Change Memory (DRCM), forming Dynamic Tokens at zero extra anchor tokens alongside a camera-stabilized Active Motion Region of Interest (ROI) and chronological Entity State Chains. Adding 0.22% trainable parameters to a frozen 27\textB backbone, VIVID-Small ( 50 examples, \sim 1.5 GPU hours) beats all eleven baselines in overall accuracy across 8,083 questions and trails only VIVID-Large ( 3,600 examples), which leads on MotionBench, Video-MME, and TempCompass.
[CV-13] π4: A Foundation Model for 4D Scene Geometry and Motion
链接: https://arxiv.org/abs/2611.23144
作者: Yifan Wang,Weiguang Zhao,Ni Yang,Haoyu Guo,Junting Dong,Weinan Zhang,Chunhua Shen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract: \pi^4 is a unified foundation model for 4D scene perception. From diverse, uncalibrated RGB observations in the wild—a single image, an unordered collection, or a video sequence—it recovers dense geometry, camera poses, and metric scale, together with 2D/3D point tracks and moving-object masks for multi-image inputs. It couples a geometry branch with a geometry-conditioned motion branch, which predicts image-space correspondences that are lifted into 3D using reconstructed target-view geometry. Across diverse benchmarks, \pi^4 achieves the best overall average rank in geometry evaluation, alongside competitive performance in single- and multiview 2D/3D point tracking and moving-object segmentation. With a shared backbone, the full geometry-and-motion system contains approximately 0.5B parameters—less than half the size of most baselines. It produces joint reconstructions and tracks in one forward pass, including large-scale scenes with thousands of input images despite training with contexts of at most 64 views. We further present the data and training recipe underlying these capabilities, together with empirical studies of key design choices. Code and models are available at this https URL.
[CV-14] CP4D: A Unified Tool-Center-Point Tracking and Generation Model for Robot Manipulation
链接: https://arxiv.org/abs/2611.23143
作者: Haosong Peng,Xiangyang Liu,Zhengshen Zhang,Yunfeng Fan,Tianyu Qi,Xin Xia,Yufeng Zhan,Wenchao Xu
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注: Project page: this https URL
Abstract:Robot manipulation videos provide a rich visual experience, yet recovering robot action and gripper-state annotations from RGB observations still remains challenging. Using the embodiment-agnostic tool-center-point (TCP) pose as the action space, we present TCP4D, a unified model in which a shared 4D scene representation both tracks observed TCP motion and generates future TCP trajectories. Given a monocular video and initial 2D TCP queries, the tracker directly recovers metric TCP poses in SE(3) and gripper states alongside scene geometry and camera parameters. The generator samples local features from the same 4D representation along the recovered TCP tracks, then combines them with robot observations and language instructions to predict future TCP states. Our new cross-embodiment augmentation pipeline converts existing RGB videos into demonstrations for alternative embodiments by retargeting the recovered motion and synthesizing the corresponding observations, without recorded source-robot states. Experiments show that the tracker achieves absolute trajectory errors of 7.60 and 16.07 mm on DROID and RoboLab and adapts to a new embodiment with only ten demonstrations. The generator achieves 69.19% mean success on RoboTwin and outperforms well-established policy models on two real-world tasks, while joint training on source and augmented demonstrations improves target-robot success by 41.5% over augmented-only training.
[CV-15] GameRealm: Grounding Visual Learning in Interactive Games
链接: https://arxiv.org/abs/2611.23140
作者: Weiguang Zhao,Yifan Wang,Ni Yang,Yating Wang,Xingyu Miao,Yankai Jiang,Mulin Yu,Tao Lu,Haoyu Guo,Junting Dong,Weinan Zhang,Chunhua Shen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Visual foundation models are central to physical intelligence, yet their development remains constrained by fragmented training datasets, inconsistent annotation quality, and limited multimodal supervision. To address these challenges, we introduce GameRealm, a large-scale, richly annotated dataset that grounds visual observations in synchronized, game-native supervision. GameRealm encompasses over 120 million RGB frames and 1,000 hours of gameplay footage across all views. Beyond its unprecedented scale, GameRealm provides rich multimodal supervision, including metric depth, camera parameters, synchronized audio, control inputs, and 3D point trajectories. Its diversity further extends to flexible view configurations, spanning single-view, stereo, multi-view, and panoramic recordings. To enable scalable acquisition across diverse game environments, we develop a unified data collection pipeline that combines human-operated interaction with agentic exploration. Extensive experiments across diverse visual learning tasks demonstrate that GameRealm consistently strengthens visual foundation models in both visual generation and geometric learning. These results establish GameRealm as a versatile and valuable data resource, unlocking new opportunities for advancing visual intelligence.
[CV-16] LUTBench: Benchmarking Text-Driven Photographic Look Transfer
链接: https://arxiv.org/abs/2611.23113
作者: Chuanzhi Xu,Boyu Wei,Chengkun Yue,Xuanhua Yin,Yiyang Li,Chupeng Liu,Zihan Deng,Weidong Cai
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Photographic color grading shapes the tone and color of a photograph to achieve a desired look. Text-driven photographic look transfer lets users specify these adjustments in natural language, with the goal of reproducing the requested look across photographs while preserving their content. However, prior research on text-driven color grading uses different datasets, text instructions, target definitions, and metrics. These differences make it difficult to compare how faithfully the methods reproduce the same look. To compare methods under a shared protocol, we introduce LUTBench. We collect 91 three-dimensional lookup tables (3D LUTs) and define one photographic look per LUT with a specified input conversion. Applying these looks to 5,000 MIT-Adobe FiveK photographs produces 455,000 input-target image pairs with reproducible target images. For each look, we use templates to turn measured tone and color changes on fixed color samples into candidate clauses. We use training image measurements to verify applicable clauses before composing the text description. The evaluation measures target fidelity, tests sensitivity to predefined input conversions, and separates generalization to unseen looks from reproduction of seen looks on new photographs. Across 16 methods, stronger text alignment does not guarantee target fidelity. However, most supervised variants improve fidelity on both seen and unseen looks.
[CV-17] Event-Based Dynamic Lip Region Estimation During Speech
链接: https://arxiv.org/abs/2611.23111
作者: Hamze Hammami,Omar Eldardeer,Francesco Rea
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Event cameras offer high temporal resolution and low power consumption, making them promising for speech-related applications. These include automatic lip reading, audiovisual speech analysis, and human-robot interaction, particularly in challenging environments characterized by low illumination or high dynamic range. Event-based lip reading, speaker recognition, and voice activity detection, however, mostly operate on a cropped region of interest (ROI), by hand or by the dataset, and no event-based method supplies that ROI as a spatial region; existing event-based face analysis gives the mouth as landmarks, which presents a mismatch between application and localization. The present work predicts a dense heatmap over the lip region from the full event “frame”. Events are encoded as a polarity-separated voxel grid and processed at every location by a two-pathway continuous-time recurrent network (CT-RNN) with learned, log-space bounded time constants acting on the real elapsed time of each bin: a stateless fast pathway and a stateful slow pathway, utilised for refining and anchoring the prediction. The model reaches 69.5% on lips on four unseen speakers and switches target only 2.1 times per second, against 8.9 and 11.0 for convolutional baselines with more parameters. Adding a spatial encoder improves precision but fails on a new face scale. On a real event sensor, trained only on simulated events, the proposed model reaches 58.0% on lips against 43.2% for the CNN, with fewer than half its target switches. Subjects: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG) Cite as: arXiv:2611.23111 [cs.CV] (or arXiv:2611.23111v1 [cs.CV] for this version) https://doi.org/10.48550/arXiv.2611.23111 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[CV-18] FMT: Lossless Image Compression in a Fixed Number of Parallel Steps
链接: https://arxiv.org/abs/2611.23108
作者: Ammar Ali
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Learned lossless codecs make smaller files than common classical codecs, yet the images we archive and train on still use classical codecs. A main reason is time: a learned codec takes seconds to decode a photo that a classical codec opens in milliseconds, and their faster variants give up most of their gain in size. We argue that much of this time is spent waiting, not computing: the network waits for the values it conditions on, and the arithmetic decoder reads them one at a time. We introduce FMT (Fixed-step Multiscale Tree), a lossless image format whose decoding order and entropy coder are designed together so that neither wait grows with the image. FMT stores an image as a coarse-to-fine tree whose steps hold independent values: a GPU predicts each step at once and decodes it in parallel rANS lanes that carry their states from step to step, so an image of any size decodes in a fixed number of sequential steps. On this tree we build a family of bit-exact codecs, from a rule with no network to large networks, and no codec we ran on the same GPU beats any of them on both size and speed. Without a network, FMT beats NVIDIA’s HTJ2K on size and speed; with a large one, it decodes Kodak photos 9 to 61X faster than learned codecs we ran on the same A100, at the same size or smaller.
[CV-19] Event-Based Person Detection for Maritime UAV Search WACV2027
链接: https://arxiv.org/abs/2611.23088
作者: Zijie Ning,Patrick Vandewalle
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at the 5th Workshop on Maritime Computer Vision (MaCVi), WACV 2027
Abstract:Maritime search and rescue requires locating people in the water quickly, motivating the use of UAVs to survey the search area. During flight, camera motion and the longer exposures used in low light can blur small targets and hinder person detection. Event cameras record brightness changes as they occur, reducing the motion blur associated with longer exposure times. We investigate their use for maritime person detection by generating synthetic events using the v2e event simulator from public MOBDrone rescue videos and retaining the existing annotations. We compare the recurrent vision transformer model RVT-T with YOLOv8n and RT-DETR-R18 trained on exposure-matched images. On 843 test samples from held-out recordings, RVT-T achieves 15.63% person AP50:95, compared with 6.89% for YOLOv8n and 9.44% for RT-DETR-R18 under 100 ms synthetic exposure. Increasing exposure times from 50 to 200 ms progressively reduces frame-based detection accuracy, widening the gap to the event model. To evaluate the feasibility of on-board processing, we deploy our detector on a Jetson Orin Nano platform. The RVT-T model retains its accuracy and requires approximately 136 ms and 2.02 J per sample, allowing real-time processing on the UAV.
[CV-20] ransfer Once Ask Locally: Compact Audio-Visual Memory for On-Device Question Answering
链接: https://arxiv.org/abs/2611.23085
作者: Qitong Wang,Jihui Jin,Divya Kothandaraman,Vishnu Raj,Christopher Rasmussen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Audio-visual question answering on lightweight user devices requires balancing speed, memory usage, privacy, and communication. One approach is to run audio-visual encoders on a server and a lightweight language decoder on the user device, keeping questions local while offloading media encoding. However, transferring and processing the full multimodal representation remains costly. We propose a training-free, query-agnostic token pruning framework that constructs a compact, reusable memory before transmission. We define video spatio-temporal blocks and aligned audio windows as players and estimate their importance through Monte Carlo Shapley valuation. We use an embedding coverage objective that combines query-independent utility with cross-modal redundancy calibration and rewards unimodal diversity through marginal coverage gains. We retain the original token embeddings of high-scoring players under a limited token budget and transfer the resulting memory once for reuse across subsequent local questions. Experiments on Video-MME, Daily-Omni, and Video-Holmes using video-SALMONN 2+ (3B) and video-SALMONN 2 Pro (4B) show that our method achieves the best or tied-best QA accuracy among the compared compression methods across all evaluated settings. On Daily-Omni with video-SALMONN 2+ (3B) at 80% pruning, our method improves accuracy by 12.86–14.62 percentage points over these methods while achieving the lowest device-side LLM prefill and decoding latency, with up to 1.92\times decoding speedup at comparable embedding memory usage.
[CV-21] UnrealEvent: Synthetic Dataset Generation and Annotation for Event-Based Vision ECCV2026
链接: https://arxiv.org/abs/2611.23078
作者: Yentel Schoovaerts,Zijie Ning,Patrick Vandewalle
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at the Workshop on Neuromorphic Vision (NeVi), ECCV 2026
Abstract:Event-based cameras provide high temporal resolution, low power consumption, and a high dynamic range, but progress in event-based object detection remains constrained by the scarcity of annotated training data. To solve this issue, this paper introduces the tool UnrealEvent, a synthetic dataset-generation pipeline implemented in Unreal Engine 5.5. The pipeline combines a V2E-inspired event-frame generator with EasySynth semantic-rendering passes, allowing rendered level sequences to be converted into paired event representations and YOLO-format bounding-box annotations without manual labeling. Events are generated from per-pixel log-luminance differences between consecutive frames, with contrast thresholds sampled independently from a normal distribution to emulate the fabrication-induced threshold mismatch of real sensors. Using this pipeline, we provide a synthetic dataset and evaluate YOLOv8 detectors under synthetic, zero-shot real, and real fine-tuning training on eTraM. The results show strong in-domain synthetic performance but a substantial synthetic-to-real gap, especially for pedestrians. Fine-tuning on a small real subset improves real-domain performance, while the remaining gap highlights the importance of asset diversity and sensor calibration.
[CV-22] RFW-S: Everyday Deformable Objects with Independent Simulation Verification
链接: https://arxiv.org/abs/2611.23056
作者: Jiarui Guo,Saining Zhang,Jiaying Wan,Jiawei Weng,Junhao Chen,Rui Zhang,Yuanbo Lyu,Zhiheng Lee,Peishuo Li,Jingrun Du,Weiqing Xiao,Weiyi Lin,Zenan Meng,Hu Wei,Lin Qu,Zongzheng Zhang,Hao Zhao,Chaojian Li
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注: 34 pages, 19 figures. Project Page: this https URL
Abstract:Large-scale robotic simulation requires diverse deformable objects with executable physical properties and realistic interaction behaviors. However, existing 3D asset collections primarily emphasize visual diversity, while scalable physical construction and verification remain challenging. We introduce RFW-S, a large-scale dataset of 83,365 simulation-verified assets spanning 485 everyday categories across five structural families: soft solids, cloth sheets, shells, garments, and bags. We develop an automated pipeline integrating reference-guided 3D generation, structure-aware physical modeling, and LLM-assisted parameter estimation and calibration. To ensure physical reliability, we introduce an independent simulation verification framework that evaluates object behaviors under interactions distinct from those used for parameter calibration. Extensive experiments demonstrate the effectiveness of independent verification and characterize deformable responses under different physical configurations. We further evaluate robotic grasping in terms of both task success and object integrity, and demonstrate the physical realism of our simulated objects through sim-to-real replay of fragile-object manipulation. RFW-S provides a scalable foundation for physically grounded simulation and embodied robotic learning.
[CV-23] FrameRouter: Frame Budget Routing for Long-Video Understanding on Video-MME-v2 and Beyond NEURIPS2026
链接: https://arxiv.org/abs/2611.23055
作者: Bingjun Luo,Jialin Guo,Siqi Li
类目: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: Accepted to NeurIPS 2026
Abstract:Long-video question answering benchmarks like Video-MME-v2 typically evaluate systems under a uniform frame budget for every question, despite stark variation in the evidence each query actually demands. We analyze this mismatch on Video-MME-v2 and find that different question subsets prefer different frozen frame-budget policies, while uniformly increasing the frame count does not reliably improve all metrics. These findings motivate fixed-mean frame-budget routing, a controlled inference setting in which each question may receive a different frame budget while the average budget matches a uniform reference. We introduce FrameRouter, a training-free test-time framework that combines an evidence-demand router, a budget-constrained allocation rule, and a frozen bank of frame-sampling policies. By controlling how much visual evidence each question receives while leaving the underlying samplers unchanged, FrameRouter is orthogonal to query-aware frame selection. Across Video-MME-v2 and several long-video understanding benchmarks, FrameRouter improves long-video question answering by using the same average visual budget more selectively across questions. Code is available at this https URL.
[CV-24] ReTouchGS: Ultralight Render-free Aesthetic Color Enhancement and Transfer for 3D Gaussian Splatting
链接: https://arxiv.org/abs/2611.23054
作者: Chuanzhi Xu,Boyu Wei,Chengkun Yue,Xuanhua Yin,Yiyang Li,Haodong Chen,Qiang Qu,Weidong Cai
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Graphics (cs.GR)
备注:
Abstract:3D Gaussian Splatting (3DGS) enables photorealistic scene reconstruction. Existing 3DGS color transfer and interactive editing methods are guided by reference images and user-specified color adjustments. However, in many applications, users simply want aesthetically guided color enhancement that automatically makes the scene more visually appealing while maintaining consistent colors and tones across viewpoints. We present ReTouchGS, to our knowledge the first framework for reference-free aesthetic color enhancement of pretrained 3DGS models. It predicts a scene-level color lookup table (LUT) from the model and its associated camera parameters. ReTouchGS also supports reference-guided color transfer by fitting a scene-level LUT using one edited view of the same scene and its corresponding original render. A shared blend-aware color refinement network with only 44.7K parameters refines an analytic color update to realize either color mapping without rendering. It preserves geometry and delivers standard 3DGS assets in seconds. We also design the ReTouch3D benchmark to support the evaluation of aesthetic color enhancement and reference-guided color transfer across viewpoints. Experiments demonstrate high color accuracy and strong generalization to unseen scenes in both aesthetic color enhancement and reference-guided color transfer.
[CV-25] Refin3R: Geometry-Guided Refinement for Feed-Forward 3D Reconstruction
链接: https://arxiv.org/abs/2611.23051
作者: Jongmin Lee,Jaeyeong Kim,Seungryong Kim
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Feed-forward 3D reconstruction models predict dense point maps from multiple images in a single pass. Without explicit geometric constraints, however, their predictions are often inconsistent across views and lack fine-grained precision. Recent methods incorporate geometric priors such as camera pose or depth, but use them merely as auxiliary inputs, process each view independently, or refine the geometry only where the priors are available. We present Refin3R, which refines feed-forward point maps with Structure-from-Motion (SfM) points, propagating SfM geometry over the image grid. In its dual-branch architecture, a 2D branch extends SfM geometry beyond its coverage using geometry-aware backbone features, while a 3D branch reasons over the SfM points of all views, capturing multi-view structure that the image grid alone cannot represent. The two branches exchange information bidirectionally to jointly refine each view, yielding dense and accurate point maps. Across ScanNet++, ETH3D, Tanks\Temples, and out-of-domain human and surgical scenes, Refin3R consistently outperforms feed-forward and geometry-guided methods in point map accuracy while running faster than the strongest geometry-guided baseline.
[CV-26] Wall-to-wall maps of canopy height and aboveground carbon across the Atlantic Forest
链接: https://arxiv.org/abs/2611.23048
作者: Matheus Pinheiro Ferreira,Thiago Carmuega Rabaçal,Giulio Brossi Santoro,Giovanna de Andrade Ferreira,Rodrigo Costa Pinto,Danilo R. A. de Almeida,Paulo Guilherme Molin,Catherine Torres de Almeida,Angélica F. Resende,Vinicius Castro Souza,Marcelo Antonio Pinho Ferreira,Cássio Toledo,Eben North Broadbent,Angelica Almeyda Zambrano,Laura Simões,Juliano van Melis,Ricardo Ribeiro Rodrigues,Pedro H. S. Brancalion
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 56 pages, 13 figures, 8 tables, 2 supplementary tables. Submitted to Scientific Data. Maps and code openly available via Zenodo and GitHub
Abstract:Global canopy height maps are increasingly used to estimate forest carbon, yet their regional accuracy is rarely tested. We present wall-to-wall maps of canopy height (10~m) and aboveground carbon density (30~m) for the Brazilian Atlantic Forest in 2024. Height was predicted by a deep learning network fusing Sentinel-2 reflectance, Sentinel-1 and ALOS PALSAR backscatter and ALOS elevation, trained on 7,179 and tested on 1,793 patches (58,800~km ^2 ) from a statewide airborne LiDAR campaign over São Paulo state, Brazil. On held-out test patches, the root mean square error is 3.05~m at 10~m and 2.58~m at 30~m, and the model was validated at two LiDAR sites outside São Paulo. Over the same pixels, the Meta and World Resources Institute global product has more than twice this error. A power law fitted to 124 inventory plots converted canopy height to above ground carbon density. The remaining forest holds 1,944~Tg~C over 28.3~Mha. Global products therefore need local validation before use in carbon accounting.
[CV-27] When Anchors Shift the Target: Advantage-Gated Adaptation for Diffusion Reinforcement Learning
链接: https://arxiv.org/abs/2611.23038
作者: Pu Cao,Min Zhou,Ying Shu,Tiezheng Ge,Qing Song,Lu Yang
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 12 pages, 4 figures, including supplementary material
Abstract:Forward-process reinforcement learning adapts diffusion and flow-matching models by re-noising rewarded generations into regression targets. Because negative advantages induce concave updates with unbounded repulsive gradients, existing methods stabilize training using cached predictions from a moving policy copy, yet these anchors also alter what positive feedback teaches the model to fit. We formalize this conflict as anchor coupling: after separating frozen-reference regularizers, affine moving-anchor rules cannot eliminate negative curvature while preserving every positive regression target. We propose Advantage-Gated Adaptation (AGA) to decouple these roles. Positive samples retain standard target fitting, whose self-limiting update vanishes at the observed target independently of the moving anchor. Negative samples instead use the first-order tangent of the concave objective at the moving-policy prediction, preserving the local repulsion direction with a prediction-independent gradient. Within the quadratic-anchor family, these requirements uniquely determine the parameter-free gate \omega(a)=\max(-a,0) . Because the moving copy provides local linearization rather than retention, AGA assigns long-horizon behavior preservation to a frozen pretrained reference regularized at its operating classifier-free guidance scale, interpreted through an implicit-reward lens. Across compositional generation, text rendering, and preference alignment, our experiments show that displacing positive targets degrades visual quality and heightens sensitivity to policy-copy schedules, whereas AGA consistently improves the reward–quality trade-off.
[CV-28] Automated Annotation-Free Synthetic Data Generation for Underwater Litter Detection: A Comparison of Generative Approaches
链接: https://arxiv.org/abs/2611.23037
作者: Amala Mary Vincent,Bart De Schutter
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Underwater object detection is often limited by scarce and imbalanced training data and by strong appearance variations caused by turbidity, lighting, and differences between sensors. Existing data augmentation strategies, such as geometric or color augmentation or fully synthetic image generation with Generative Adversarial Networks (GANs), either lack semantic realism, fail to respect scene physics, or require additional annotations. In low-data domains like underwater environmental monitoring, the need for additional annotations is often the core bottleneck. Collecting new footage is hard, but labeling this dataset is even more time-consuming and costly, which limits how quickly a usable dataset can be constructed. We propose a unified fully automated zero-annotation pipeline that inserts class-conditioned trash objects into real underwater frames while preserving the original background context. Starting from sampled bounding boxes, each box is converted into an insertion mask and the object is pasted or generated within it, while the same box coordinates are automatically written as the corresponding annotation. Because labels are derived directly from the sampled boxes, the pipeline generates fully labeled training data for low-data settings without any manual annotation. Within this single pipeline we compare three strategies: copy-paste, Stable Diffusion inpainting, and DALL-E image-conditioned generation. We evaluate bounding-box faithfulness, visual realism, and the impact of the generated samples on downstream detector performance on the SeaClear underwater debris dataset. Results show that DALL-E improves detection performance in four of five domains, with relative gains of up to 96 percent over the real-only baseline.
[CV-29] M3OPD-Reason er: An Open Recipe for Post-Training Multimodal Reasoning with Multi-Teacher On-Policy Distillation
链接: https://arxiv.org/abs/2611.23034
作者: Nguyen Quang Trung,Hoang Phan,Shuo Sun,Longyin Zhang,AiTi Aw
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Multi-Teacher On-Policy Distillation (MOPD) enables a single student to absorb the skills of several domain teachers by learning from its own generated responses. We present Multimodal Model Multi-Teacher On-Policy Distillation Reasoner (M3OPD-Reasoner), a fully open recipe for post-training vision-language models (VLMs) over three reasoning domains: image STEM, image perception, and video. We curate two datasets, M3OPD-SFT-1.35M for supervised fine-tuning (SFT) and M3OPD-RL-193K for reinforcement learning (RL) with verifiable rewards. On these data, we compare post-training recipes that start from Qwen3.5-9B: training one model on the mixed data of all domains (MixSFT and MixRL), and training one teacher per domain with Group Sequence Policy Optimization (GSPO), directly or after an SFT warm-up, and distilling the teachers into one student with MOPD. We evaluate on 21 image STEM, image perception, and video reasoning benchmarks after we decontaminate the training data against them. The comparison establishes MOPD of teachers trained with GSPO directly from Qwen3.5-9B as the best recipe: the student closely matches each teacher on its own domain and is on average above the mixed-training baselines on the same prompts. Since MOPD already merges the domain capabilities that MixRL would otherwise have to learn, its student is a stronger starting point for RL, and we ask whether the same data can carry the model further from there: we add MixRL on top of the MOPD student, on the same prompts and with the same settings, which, to our knowledge, no prior VLM recipe has studied. The same prompts still improve the MOPD student, with no collapse, and give our main model, M3OPD-Reasoner-9B. MOPD and MixRL can therefore work as consecutive stages that get more out of the same data. We release all data, the training code, and the checkpoints of every stage.
[CV-30] Check Your Blind Spots: Decision-Guided Perception by Probing the Planner with Hypothetical Objects
链接: https://arxiv.org/abs/2611.23030
作者: Hoang Pham Cong,Long D. Nguyen,Ho Viet Duc Luong,Nguyen Tran Minh Nhat,Binh P. Nguyen
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Not every perception error changes the planner’s action. Yet adaptive-resolution perception chooses where to spend high-resolution computation from image cues, which can neither reliably point to an object that low resolution missed nor tell which objects matter to the planner. We observe that the planner can say where an object would matter before anything is seen: inserting a hypothetical object at a location tests whether its action would change. Decision-Guided Speculative Perception (DGSP) treats the low-resolution scene and its action as a draft and probes the planner in this way to obtain a decision-sensitivity field, the locations where an unseen object would change the action. Only this field and the support, the detections on which the current action rests, are verified at high resolution, repeatedly until the action is stable. For two planner classes, we prove that DGSP returns exactly the action of whole-frame high-resolution perception if crop detections match whole-frame ones and the crops cover the field, conditions that real images meet approximately. On three driving datasets, it reproduces this action on about 98% of frames with about a quarter of the pixels. At the pixel cost of uniform downscaling, it repairs more of the draft’s disagreements with this action, brakes falsely less often, and recovers many more of the decision-relevant objects that the draft misses.
[CV-31] When a Larger Guidance Scale Repairs Text-to-Video Failures and When It Does Not
链接: https://arxiv.org/abs/2611.23025
作者: Taewon Kang,Matthias Zwicker
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 29 pages, 15 figures
Abstract:When a text-to-video diffusion model ignores part of a prompt, the usual first fix is to raise the classifier-free guidance scale. This paper is a measurement study of when that fix works. On Mochi, the same increase repairs failures on rare compositions (“a snowy beach”) beyond what generating again does, does not repair negations (“a highway with no vehicles”), where it fixes fewer failures than generating again does, and changes nothing for events that must happen late in a clip (“a sandcastle that a wave washes flat”). Which of these happens can be known in advance, from the prompt and one inexpensive score. The score, which we call demand, measures how strongly the denoiser reacts to the part of the prompt that the failed video leaves out, and it is read from the first quarter of denoising, before any video exists. Our main finding is a condition: raising the scale is effective only when the prompt contains no negation and no event tied to a specific time, and when demand is sufficiently high, with recovery clearly higher above a fitted threshold than below it. Published corrections for the three failure types, used as fixed probes, behave as this account predicts. A short decision rule follows: before generating again, a user can tell whether the most common fix is worth trying, and what to use when it is not.
[CV-32] racking Ethological Behaviors of Broiler Chickens Through Computer Vision (YoloV11) Using an Android Mobile Device
链接: https://arxiv.org/abs/2611.23007
作者: Ryan Davi Oliveira de Meneses,Glauber da Rocha Balthazar
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:This study developed and evaluated a computer vision model for automated ethological behavior classification of broiler chickens in climate-controlled poultry houses. Using publicly available videos preprocessed and manually annotated into three classes (sitting, standing, and feeding), the YOLOv11 model was trained and deployed on an Android application via TFLite conversion. Exploratory analysis ensured dataset consistency and visual quality, while evaluation showed class-dependent performance: higher robustness for static postures (sitting, AP=0.531; precision=100%) and lower accuracy for dynamic behaviors (feeding, AP=0.292; recall=0.20). Tests with unseen images demonstrated significant improvements, achieving accuracies of 98.04% for sitting, 83.33% for standing, and 67.65% for feeding. The model showed effective generalization in real farm environments, supporting its potential for continuous behavioral monitoring, early welfare assessments, and decision-making in precision poultry farming. These findings establish a technical foundation for future mobile-based applications at commercial scale, reinforcing the role of computer vision in automated animal welfare evaluation.
[CV-33] he Rhythm of Generated Video: Latent Chunking Leaves Periodic Traces in Compressed Bitstreams
链接: https://arxiv.org/abs/2611.22996
作者: Chen Ye Gan,Bryan Zhen Zhang,Gene Wen,Yuxing han
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Manuscript, 9 pages, Computer vision, AI-generated video detection, compressed-domain analysis, video coding,
Abstract:Most AI-generated video detectors analyze decoded pixels and overlook how videos are produced. We show that causal video autoencoders leave an architecture-specific rhythm in compressed bitstreams: their temporal stride concentrates unpredictable content at latent-chunk boundaries, so rate-distortion optimized encoders allocate bits periodically. Without decoding pixels, we observe four-frame pulses in CogVideoX and Wan2.2 and eight-frame pulses in LTX-Video, which are absent from content-matched camera footage. Passing camera footage through a generator’s autoencoder alone induces the pulse, supporting its proposed origin. A training-free peak-ratio statistic separates generated from camera video (AUC 0.84-0.94) and persists across tested quantization settings and H.264/H.265 encoding. Frame-size traces also identify four commercial generators, with 89.2%, 84.1%, and 83.0% accuracy under H.264, H.265, and AV1, beating the pixel-based UnivFD baseline (72.7-73.6%). Our results directly link generative architecture to measurable coding demand.
[CV-34] CAST: Calibrating Correspondences for Training-Free Biomedical In-Context Segmentation
链接: https://arxiv.org/abs/2611.22995
作者: Mingze Jiang,Jiashu Han,Haonan Chen,Zhiyi Li,Yilun Du,Kunzan Liu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 10 pages, 6 figures
Abstract:Biomedical image segmentation must keep up with new anatomical structures, subjects, imaging modalities, and acquisition protocols, yet dense expert annotations are scarce and models trained on one cohort often struggle to transfer. In-context segmentation (ICS) offers a way out: a new structure is specified by a few annotated exemplars at inference time. However, reliable correspondence remains challenging in biomedical images, where adjacent structures often share similar appearances, and variations across subjects and imaging conditions further obscure their distinctions. We introduce CAST, a training-free framework built on a frozen DINOv3 backbone to calibrate fine-grained correspondences. Its stability test improves feature reliability at both global and patch-specific levels, while its coherence test assesses whether candidate correspondences are supported by consistent local structural context. Together, these tests enable more reliable correspondence across diverse anatomical structures and imaging conditions. Across eight biomedical benchmarks spanning 2D and 3D segmentation, cross-subject, and cross-modality transfer, CAST outperforms the strongest overall baseline by 12.34 percentage points in Dice and ranks first across all six volumetric settings. The versatility of CAST further opens up opportunities for challenging biomedical applications, including 2D-to-3D segmentation and atlas label transfer.
[CV-35] SpikWM: Efficient End-to-End Spiking World Modeling for Visual Planning
链接: https://arxiv.org/abs/2611.22981
作者: Shuhan Ye,Hongbin Yu,Chong Wang,Hui Lu,Jiaming He,Yi Yu,Jiangbo Qian,Xudong Jiang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Goal-conditioned planning with image observations and goals requires predictions that distinguish visual states and action consequences. Spiking neural networks offer sparse computation, but single-step spikes can obscure motion cues and merge actions with different physical outcomes. We propose the Spiking Latent World Model (SpikWM), an end-to-end joint-embedding predictive architecture (JEPA) with one simulation step per image and one neural update per latent position. Spiking Regional Encoding retains spatial cues through regional pooling. Bi-Spike Action Encoding combines positive and negative binary responses, while Causal Membrane Prediction carries subthreshold evidence across the latent history with learned group-wise time constants. The encoder and predictor are trained jointly on reward-free image–action trajectories using next-embedding prediction and SIGReg. With a common planner, SpikWM achieves mean success rates of 88.67% , 97.33% , 99.33% , and 86.67% on Push-T, Two-Room, Reacher, and OGBench-Cube, respectively, outperforming two spiking baselines. Push-T ablations support each module, and controlled replays examine responses to appearance and physical state changes. Under a matched planning budget, SpikWM reduces estimated arithmetic energy by 75.6% relative to LeWM. Code will be released at this https URL.
[CV-36] Rescaling OLS-Merged Adapters To Improve Personalized Image Generation
链接: https://arxiv.org/abs/2611.22969
作者: Leander Girrbach,Niklas Schweiger,Zeynep Akata
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Merging multiple adapters into a single image generation model reduces deployment costs while preserving diverse task capabilities or personalization modes. State-of-the-art methods such as RegMean, Mix-of-Show, and Subspace Routing fit the outputs of the merged model to those of the individual adapters using ordinary least squares (OLS). We show that this reconstruction objective systematically shrinks task-specific signals whenever the activation distributions of different tasks overlap, producing a merged adapter that is too weak for downstream use. We derive a decomposition that quantifies this shrinkage and propose rescaling the merged update to restore signal strength. On multi-subject personalization with up to 30 merged adapters, rescaling substantially improves subject fidelity over unscaled baselines. Our approach also improves results on style transfer, concept erasure, image classification, and LLM-based text classification.
[CV-37] MI-Hyper: Generating Multiple Instance Models for Weight Space Transfer Learning in Histopathology via Mixed Tasks Pretrained HyperNetwork
链接: https://arxiv.org/abs/2611.22968
作者: Xitong Ling,Yousef Kotp,Weiming Chen,Yining Li,Jiawen Li,Minxi Ouyang,Xinyi Guo,Mingxi Fu,Yizhi Wang,Tian Guan,Lianghui Zhu,Yonghong He
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Zero-shot whole-slide image (WSI) prediction aims to generalize to new pathology tasks without parameter updates at deployment. MI-Zero makes these predictions by pooling patch-text similarities. Conditional weight generation, combined with supervision from diverse pathology tasks, offers an alternative: learning to construct predictive rules from task descriptions. We introduce MI-Hyper, a hypernetwork pretrained across mixed tasks to generate the aggregation and classification weights of multiple instance learning (MIL) models. We first train the generator and a shared MIL basis jointly using source labels. We then freeze the basis and refine the generator. The generated task weights are reused across cases. Across 308 tasks from 52 pathology datasets, MI-Hyper improves mean AUC, balanced accuracy and macro-F1 over MI-Zero by 1.85, 1.17 and 0.57 percentage points, respectively. On some molecular prediction tasks, both AUC and balanced accuracy improve by more than 20 points. These results support task-conditioned weight generation as a promising approach to weight-space transfer for long-sequence modeling.
[CV-38] CodeReel: Accelerating Code-to-Video Agents with Exact Frame Reuse and Blocking-Aware Scheduling
链接: https://arxiv.org/abs/2611.22967
作者: Junyi Shen,Yao Lu
类目: Multimedia (cs.MM); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Distributed, Parallel, and Cluster Computing (cs.DC)
备注: 19 pages, 8 figures, 14 tables
Abstract:Coding agents make videos by writing and refining programs in a build–inspect–edit loop. A finished film costs a median of 2.85 in model calls and 11 minutes, over a third of that time in builds that block the agent. We find that much of this time is avoidable, for two reasons. First, 37.5% of the frames of a session’s later films were drawn before, many shifted in time or inside an edited scene. Second, half of the builds stop at a check, where the agent needs only the verdict, but which builds stop is unknown until the checks finish. We present CodeReel, a training-free execution layer behind agents’ existing commands. Exact frame reuse keys each frame by its drawing commands, regardless of its position in time, and reuses each frame whose key recurs. Blocking-aware scheduling runs the checks first and starts the render alongside them, at idle priority, unless the session’s recent builds make a pass unlikely. Scheduling spends CPU to answer sooner and reuse returns it; exact keys let later versions use frames rendered for a failed one as the agent thinks. We introduce EditLoop, a benchmark of replayable agent sessions whose films are checked against renders from scratch. On it, CodeReel cuts the agent’s wait by 36.3% against Rebuild, a warm pipeline with parallel rendering, and by 57.9% against Sequential, today’s step-by-step pipeline. On time-shifting edits its reuse renders 43.9%, close to the 43.0% whose pixels are new, while Manim-style caching renders 99.8%. With real agents, the median wait per build is 0.62x Rebuild’s in Claude Code, on topics never used in development, and 0.67x in Codex CLI. The gain holds for five models and two frameworks.
[CV-39] raining-Free Building Change Detection with Example-Induced Discriminants
链接: https://arxiv.org/abs/2611.22964
作者: Zhi Lu,Bin Hong,Zhenming Peng
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Building change detection (BCD) relies on dense annotations that are costly to collect across regions. A few annotated examples offer a more economical way to specify the target, but change maps provide incomplete building supervision: they label only changed buildings and leave the building-containing date unspecified. Our key observation is that these annotations nevertheless reveal features that recognize buildings in other images, including unchanged ones. We present Linear INduced Discriminants (LIND), a training-free framework that turns regional feature contrasts into a reusable linear building classifier. Using a single frozen vision transformer, it unifies building recognition and temporal comparison in the same feature space. This approach requires neither text prompts nor an auxiliary segmentation model. Only one parameter is tuned on a single validation dataset and then kept fixed across datasets and shot settings. Extensive experiments demonstrate strong detection accuracy, cross-dataset transfer, and efficient inference, with extensions to other semantic classes and broader land-cover changes. Change annotations thus provide a compact visual specification for detection in new scenes. Code: this https URL
[CV-40] Hierarchical Diffusion Transformer for End-to-End Image Generation
链接: https://arxiv.org/abs/2611.22962
作者: Xiaoyu Wu,Yifei Wang,Weihang Guo,Chen Wei,Lydia E Kavraki,Niloofar Mireshghallah,Steven Wu
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Under Review
Abstract:Pixel-space diffusion learns visual representations and image generation jointly in a single end-to-end training stage, without a separately pretrained visual encoder. However, plain pixel-space Transformers still trail leading two-stage latent diffusion models in generation quality and training efficiency. We hypothesize that part of this gap arises from retaining one token per patch throughout the backbone, which does not explicitly encourage a visual hierarchy that summarizes related content at progressively coarser levels. To address this, we introduce Hierarchical Diffusion Transformers (HiT), which progressively group related tokens and represent each group with a single token. The main Transformer bottleneck processes these compact group tokens, and ungrouping restores features to their original image positions. A compression budget regulates the average compression ratio across the batch, allowing group memberships and counts to vary across images and diffusion timesteps during both training and generation. The model exhibits emergent semantic structure, learning to group coherent objects and object parts without segmentation labels. In text-to-image generation, HiT further learns to group caption words with image patches depicting the same concepts, without region annotations. HiT achieves 1.48 FID on ImageNet 256\times256 and reaches comparable generation quality to JiT using approximately 40% of its cumulative training FLOPs. For text-to-image generation, HiT exceeds UniLiP’s average score across five benchmarks with roughly 6\times fewer denoiser parameters and 3\times fewer forward FLOPs. These results suggest that learning a visual hierarchy within the generator breaks down the barrier to stronger image generation without external visual pretraining.
[CV-41] Is model self-assessment enough for diverse LVLM ensembles?
链接: https://arxiv.org/abs/2611.22956
作者: Michael Smith,Frank P. Ferrie
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Ensembles of diverse Large Vision Language Models have demonstrated performance beyond that of single models across multiple datasets and tasks. However, the necessary aggregation process usually relies on training, parameter fitting, or complex task-specific pipelines. Furthermore, evaluation is limited, making a fair comparison difficult. We create a unified pipeline where we perform candidate response selection: we build an ensemble by selecting the best answer from a set provided by different models. We comprehensively evaluate across 34 different means of confidence estimation, 24 ensemble configurations, 20 datasets, and two prompt specifications. We show that simple training and parameter-free approaches can be sufficient, even for ensembles of different architectures. These simple self-assessment approaches rely solely on each constituent model’s confidence, without explicit calibration, yet are able to perform better than said constituent models. They are also competitive with more involved alternatives on standard metrics such as accuracy, including on open-ended problems. However, we also discover that dataset, ensemble composition, and prompt wording are key confounding variables insufficiently explored by prior work that significantly affect ensemble performance, such that the ensemble can perform worse than its best constituent model. These must be optimized for to get the most out of ensembles, at the cost of limiting their generalizability. We recommend that practitioners consider training-free ensembles only if deployment conditions can be precisely characterized, with Best-of-N configurations of strong models being a good starting point.
[CV-42] Fiberpose: state-of-the-art fiber morphometry and instance segmentation with a distance-transform neural network
链接: https://arxiv.org/abs/2611.22953
作者: Viktor A. Balashov,Daniil S. Antonenko
类目: Computer Vision and Pattern Recognition (cs.CV); Quantitative Methods (q-bio.QM)
备注:
Abstract:Analysis of fibrous structures in SEM images is a routine task in biomedical engineering and materials science. However, most available methods for fiber quantification still rely on classical image binarization followed by simple post-processing. These approaches often produce unreliable results in images with weak boundaries, heterogeneous contrast, and densely packed fibers. Here, we introduce Fiberpose, a framework that reformulates fiber analysis as direct regression of a smooth distance-transform (DT) field, together with a novel set of geometry-based algorithms specifically adapted to this representation. The predicted field directly encodes fiber geometry, while the accompanying algorithms enable instance reconstruction and measurement of diameter and orientation within a unified workflow. The framework is implemented as a standalone Python desktop application and a web tool this https URL. On an independent dataset of fibrous SEM images, Fiberpose achieved state-of-the-art instance-segmentation accuracy in comparison with SAM2 and Cellpose-SAM, and more accurate diameter measurements than DiameterJ, Fibar and GIFT. The method remains effective in dense and weakly contrasted regions and shows qualitative generalization across microscopy modalities not represented in the training data. Fiberpose therefore provides an interpretable geometric representation and an automated pipeline for reproducible quantitative characterization of fibrous materials.
[CV-43] Code World Model 2.0: Towards Infinite Controllable Video Generation WWW
链接: https://arxiv.org/abs/2611.22925
作者: Ruoyu Wang,Hao Zhong,Yanming Yang,Jiahao Tian,Lin Liu,Biao Gong,Guoshan Lu,Yiding Tian,Chunhua Shen,Zhenzhong Lan,Chi Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project Page: this https URL
Abstract:Visual proxies decouple explicit world-state simulation from photorealistic video synthesis, offering a controllable interface for both offline clip creation and real-time interactive streaming. Yet a practical renderer must address two challenges. Upstream proxies vary from coarse layouts to fine articulated geometry depending on the interaction, and streaming inference must operate in a few causal steps without accumulating drift across self-generated history. We present Invoker-Render, pairing a bidirectional creation model with a causal four-step streaming model under a shared conditioning interface over structural proxies, an appearance reference, and language. Trained on an aligned approximately 200-hour multi-granularity corpus, Invoker-Render honors fine geometric constraints when supplied while synthesizing plausible local motion under coarse proxies, enabling runtime selection of input detail. To bridge bidirectional creation and real-time streaming without compounding step compression and rollout drift, we introduce Progressive Forcing (ProForcing). ProForcing first trains a causal predictor and distills it into a four-step renderer using histories derived from training videos. Training then shifts to continuous self-generated rollouts. A pipelined serving stack supports efficient streaming inference. Experiments show versatile control across proxy families, granularities, and camera paths, consistent gains over recent streaming baselines, and stable 30-minute racing rollouts that preserve scene and subject appearance.
[CV-44] SWIR-cGAN: Synthesizing SWIR Bands Using VNIR Imagery from CBERS and AMAZONIA-1
链接: https://arxiv.org/abs/2611.22922
作者: Alisson Cleiton de Oliveira,Rennan de Freitas Bezerra Marujo,Thales Sehn Körting
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Available on: this https URL . Early Access
Abstract:Spectral band synthesis is an important research topic in remote sensing. The Brazilian satellites CBERS-4, CBERS-4A, and AMAZONIA-1 provide high temporal resolution (1-3 days) visible and near-infrared (VNIR) imagery through the Wide-Field Imager (WFI), but lack short-wave infrared (SWIR) bands. Recent deep learning methods have shown promising results for spectral synthesis, but challenges remain for cross-satellite generalization and spectrally distant bands. This paper presents a conditional Generative Adversarial Network (cGAN) with an attention-enhanced U-net generator and a multi-loss objective to synthesize Sentinel-2/MSI SWIR bands 11 (SWIR-1) and 12 (SWIR-2) from WFI VNIR inputs. Experiments used 4,125 patches for training and cross-validation and 1,026 unseen patches for testing, with hyperparameters optimized using a subset of the training data. On SWIR-1, the model achieved SSIM of 0.900, PSNR of 31.44 dB, and MAE of 0.019 in TOA reflectance. On SWIR-2, it achieved SSIM of 0.917, PSNR of 33.35 dB, and MAE of 0.015. Benchmarking was performed on the test dataset, with each model optimized separately prior to evaluation. SWIR-cGAN outperformed the baselines, particularly for SWIR-2. The ablation study identified the multi-loss as the main performance contributor, while the attention and residual modules provided additional improvements. Our results demonstrate the feasibility of synthesizing SWIR bands from WFI imagery. The SWIR-cGAN is available at this https URL.
[CV-45] BRepA: Graph-Based B-Rep Pretraining for Unified CAD Understanding and Generation
链接: https://arxiv.org/abs/2611.22903
作者: Dekai Zhu,Zhaiyu Chen,Stefan Gavranovic,Slobodan Ilic
类目: Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR)
备注:
Abstract:Boundary representation (B-Rep) of CAD models encodes geometric primitives and their relations as a graph structure, enabling efficient shape modeling within CAD software. Learning on B-Reps for downstream tasks, such as segmentation and novel shape generation, requires specialized neural network designs. Moreover, performance improvements and accelerated training can be achieved with effective pretraining strategies, which is particularly important for tasks with limited labeled B-Rep data. In this paper, we introduce BRepA, a unified self-supervised pretraining framework that explicitly reconstructs both the geometry and topology of B-Rep objects via graph-based modeling. By learning from large-scale unlabeled CAD data only once, BRepA produces representations that generalize across tasks and significantly enhance both discriminative and generative learning. On segmentation tasks, BRepA demonstrates remarkable data efficiency. For instance, with only 1% labeled samples, BRepA pretraining improves mIoU by 32.52% compared to models trained from scratch. On generative tasks, aligning diffusion model representations with pretrained BRepA features substantially accelerates convergence, achieving nearly 2 times faster training while improving generation quality under the same computational budget, despite the diffusion model being roughly 20 times larger than BRepA. Overall, our results show that geometry- and topology-aware B-Rep pretraining provides a powerful foundation for both understanding and generating CAD structures, serving as a versatile representation backbone for diverse downstream applications. We will release the code to facilitate future research.
[CV-46] You Only Encode Once: Sub-Pixel-Shifted Frames for Arbitrary-Scale Remote Sensing Super-Resolution with 2D Gaussian Splatting
链接: https://arxiv.org/abs/2611.22898
作者: Canjie Liu,Liuqian Wang,Qi Ming,Wei Lu,Xiaohan Hao,Yuefei Wu,Wentao Chao,Lei Shi
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Arbitrary-scale super-resolution (ASSR) of remote sensing images remains challenging due to the limited high-frequency information in a single low-resolution (LR) observation. While multi-frame SR can exploit complementary sub-pixel information, existing approaches often require real multi-frame training sequences and repeated feature encoding, increasing data acquisition demands and computational costs. To address these limitations, we propose SPS-GS (Sub-Pixel-Shifted Gaussian Splatting), an encode-once framework that learns from multi-phase LR observations synthesized from individual high-resolution images, without requiring real multi-frame training sequences. Specifically, Phase-Coherent Aliasing Demodulation (PCAD) exploits shift-dependent aliasing through Fourier-based alignment, phase-aware residual fusion, and least-squares aliasing estimation, enabling multi-frame information to be integrated before a single deep encoding pass. Gaussian Projection Alignment (GPA) then iteratively refines Gaussian centers and colors using multi-observation back-projection to promote observation consistency. The refined Gaussians support continuous rendering at arbitrary scales. Extensive experiments on five remote sensing datasets demonstrate strong reconstruction performance against single-image and multi-frame SR baselines, while maintaining competitive inference efficiency at common scales. The code will be publicly released.
[CV-47] Impact of Pretraining Length Model Size and Annotation Budget on Downstream Clinical Tasks of Foundation Models in Radiotherapy
链接: https://arxiv.org/abs/2611.22896
作者: Alejandro Cortina Uribe,Sidsel Højklint Poulsen,Ivan Richter Vogelius,Jens Petersen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 9 pages, 5 figures, 2 tables. Supplementary Material included
Abstract:Cone-beam CT (CBCT) scans acquired during radiotherapy capture changes such as tumor shrinkage and radiation-induced tissue changes, offering great potential for foundation models that help monitor individual treatment responses. However, to our knowledge, no large-scale pretrained models for radiotherapy CBCT exist, and evaluating modeling choices is difficult since downstream tasks are not known in advance. We examine how pretraining length, model size, and finetuning strategy relate to downstream clinical performance, and how metrics tracked during pretraining correlate with it. We pretrain vision transformer encoders with masked image modeling on about 150,000 raw 3D radiotherapy CBCT scans, tracking representation quality via validation loss, effective rank, and linear probing. Downstream tasks are classification of feeding tube presence ( n=51,135 CBCTs, 1,562 patients) and clinical characteristics (T-, N-stage, sex, age, tumor site; n=1,278 patients) in head and neck cancer. We compare low-rank adaptation (LoRA) and full finetuning against no pretraining, varying annotated dataset size. Pretext validation loss was the most consistent predictor of downstream performance; effective rank and linear probing were also associated, but not consistently across model sizes and finetuning methods. LoRA performed comparably to full finetuning while training only 0.5-2.5% of the parameters. Fully finetuned models consistently outperformed models without pretraining across all sizes, with AUROC 0.91-0.97 vs. 0.58-0.85 for feeding tube classification and 0.99-1.00 vs. 0.71-0.97 for sex classification. Model size had only a small, inconsistent effect, suggesting smaller models suffice at lower computational cost. Code and pretrained models (non-commercial license): this https URL
[CV-48] Wear and Go: Calibration Free Sparse Inertial Motion Capture via Online Sensor Setup Estimation
链接: https://arxiv.org/abs/2611.22894
作者: Yuanshuo Tan,Licheng Zhong
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Sparse inertial motion capture typically assumes that each sensor’s anatomical identity and mounting orientation are known. We present MOSAIC (Motion-based Sensor Assignment and Inertial Calibration), a framework that recovers this setup from ordinary motion without a calibration pose. Our central insight is that body pose changes continuously while the sensor setup persists, allowing movements that are individually ambiguous to provide complementary evidence over time. A setup network processes unordered sensor streams with temporal and cross-sensor attention, then accumulates assignment and rotation evidence into a session-level estimate. The recovered setup canonicalizes measurements for frozen pretrained pose estimators, separating sensor configuration from motion reconstruction. For sensors worn at prescribed body sites, evaluation on real inertial recordings with randomized setups shows near-calibrated pose accuracy for bounded offsets and substantial benefits from accumulation under arbitrary orientations. Transfer across pose networks supports setup recovery as a reusable interface between interchangeable sensors and pretrained pose models.
[CV-49] RoadAnnoter: An AI-Assisted Desktop Tool for Semantic Segmentation Annotation of Road-Scene Images
链接: https://arxiv.org/abs/2611.22884
作者: Nazmul Haque,Arnob Protim Roy
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 33 Pages
Abstract:Accurate semantic segmentation of road-scene images is essential for road safety research; however, polygon-level annotation remains labor-intensive and limits labelled dataset development. This study presents RoadAnnoter v1.0, an open-source desktop application integrating six human-in-the-loop annotation modes: manual polygon delineation, SAM Prompt, Smart Select, Edge Fill, pretrained semantic segmentation using Cityscapes-trained SegFormer models, and task-specific segmentation using TinySegNet, a lightweight MobileNetV3-Small-based network trained with accumulated annotator-generated labels. SAM-based modes use promptable segmentation with SAM and MobileSAM, while Edge Fill combines Canny edge detection with colour-similarity-based flood filling in CIELAB colour space. SegFormer generates suggestions for common road-scene categories, whereas TinySegNet supports adaptation to project-specific features through iterative training. All six modes use a shared mask-to-polygon pipeline comprising morphological processing, Gaussian smoothing, contour extraction, and Douglas-Peucker simplification; annotations are exported in YOLO segmentation format. RoadAnnoter was used to annotate Google Street View images from recorded road accident locations in Bangladesh for a vision-language model-based investigation of free-flow speed estimation. Qualitative assessment across representative scenes demonstrated the complementary applicability of the annotation modes to diverse feature geometries. Visual comparison with Roboflow indicated similar qualitative boundary characteristics for principal road-scene regions, although no quantitative accuracy assessment was conducted. RoadAnnoter provides a locally operated framework for developing road-scene segmentation datasets for transportation engineering and road safety research.
[CV-50] Learning Agent -Centric Vision-Language-Action World Models for Autonomous Driving NEURIPS2026
链接: https://arxiv.org/abs/2611.22880
作者: Guoqing Wang,Pin Tang,Xiangxuan Ren,Chao Ma
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: Accepted by NeurIPS 2026
Abstract:Recent progress in integrating vision-language-action (VLA) models with world modeling has significantly advanced end-to-end autonomous driving by allowing systems to reason about imagined futures instead of merely reacting to current observations. However, most world VLA models generate future frames at the pixel level with plenty of dense visual tokens. Although such generated futures can be visually plausible, they preserve many driving-irrelevant appearance details, i.e., only a small portion of the tokens directly encode critical planning-relevant factors, such as agent geometry, motion dynamics, and interactions. This introduces a significant performance bottleneck. An important question thus arises: what world state should a world VLA model reason over to support planning beyond pixel-level prediction? We argue that planning-oriented world modeling should focus on a compact and structured agent-centric latent state that captures the interactions between objects in the driving scene over time. As such, we introduce AgentWorld, an agent-centric VLA world model that uses discrete agent tokens jointly for reasoning and trajectory planning. Each token is dynamically grounded to a surrounding agent associated with its object’s state, such as semantic identity, 3D geometry, motion, and interaction context. AgentWorld learns these tokens through three key processes: explicit geometric grounding, reasoning-oriented supervised fine-tuning, and agent-oriented reinforcement learning. Extensive open-loop and closed-loop experiments under diverse driving scenarios show that AgentWorld enhances planning capabilities, fully exploits visual evidence and agent dynamics, and yields interpretable agent-centric world states.
[CV-51] SlideJev: Order-Stable and Tissue-Grounded Jev Decisions for Whole-Slide Pathology
链接: https://arxiv.org/abs/2611.22879
作者: Lei Wu,Jiashuai Liu,Di Zhang,Chunze Yang,Jiusong Ge,Junbo Lu,Yongqi Liang,Wenjie Zhao,Ni Zhang,Chen Li,Zeyu Gao,Qidong Liu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 10 pages, 6 figures, 3 tables
Abstract:Whole-slide visual question answering (VQA) offers a natural-language interface to pathology, allowing diagnostic and phenotypic questions to be answered directly from tissue slides. However, current vision-language models (VLMs) rely heavily on linguistic priors: their answers are sensitive to the order of the candidate answers, and can be reached through shortcuts that bypass the tissue evidence. To address these challenges, we introduce SlideJev, a framework that separately verifies candidate answers against the slide and learns to ground these judgments in tissue evidence. To connect parallel candidate evaluation with visual evidence learning, we adopt Jev and score each candidate by the difference between support and rejection log-likelihoods from the original language-model head. We compute these scores in parallel using separate prompts with the same format and no option-position labels, then normalize them for answer selection. To ground these scores in tissue evidence, candidate-conditioned reading and comparison guide gated updates to visual tokens across decoder layers. With the backbone frozen, we train these visual modules through the same scores using question-answer and paired-image supervision. Across three whole-slide pathology VQA benchmarks, SlideJev achieves the highest mean overall accuracy and 100% answer consistency across four option orders. Image-control experiments further show that SlideJev achieves the largest average accuracy gains from correct images over mismatched and image-free inputs across the three datasets. Our code is publicly available at this https URL.
[CV-52] Harnessing Image Diffusion Prior for Photo-Realistic Video Restoration NEURIPS2026
链接: https://arxiv.org/abs/2611.22874
作者: Fanghua Yu,Hongyu An,Jinfan Hu,Xinqi Lin,Zhiyuan You,Jian Wang,Hongyang Li,Chao Dong,Jinjin Gu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 40th Conference on Neural Information Processing Systems (NeurIPS 2026)
Abstract:Recent image generation models produce photo-realistic results with rich details, but transferring their powerful priors to video restoration remains difficult due to temporal inconsistency from stochastic detail synthesis. Existing video restoration methods improve temporal coherence but often sacrifice visual fidelity, leaving a gap between image-level quality and video-level stability. We propose a generative video restoration method that bridges this gap by enabling temporally consistent detail synthesis from image generation priors. Our method combines noise energy rebalancing to suppress structure-disturbing low-frequency stochasticity, motion-aligned noise warping to propagate details along object motion, a self-supervised denoising encoder to mitigate encoder-induced temporal uncertainty, and decoupled temporal attention to model cross-frame dependencies. Together, these components preserve fidelity, enhance fine-grained realism, and maintain temporal consistency. Extensive experiments show that our method substantially outperforms existing video restoration approaches in visual quality, detail richness, and temporal stability.
[CV-53] LOCUS: Causal Support Windows for Attribute Control in Autoregressive Image Generation
链接: https://arxiv.org/abs/2611.22867
作者: Zijun Huang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Unified autoregressive (AR) image generators struggle to satisfy multiple attribute constraints in a single prompt. Although these attributes are decodable from image-token latents, editing those latents at test time remains difficult. Adding a fixed attribute direction to the prefix can reinforce a requested attribute but rarely overrides a conflicting one, such as turning a car red when the prompt specifies blue. We investigate how the effect of an intervention depends on where it acts in the image-token sequence. We introduce the causal support profile, which measures, for each attribute, how often patching image-token states within a given window flips the rendered attribute. Structural attributes (count and position) are most responsive in early, narrow windows, while appearance attributes (color, size, and material) respond over later, broader windows. A budget-matched comparison of early and middle windows shows that the effect of window location depends on the target attribute, and two-dimensional support maps show that the responsive regions track spatial content. Based on these profiles, we propose LOCUS (Latent Optimization in Causal Support windows), which diagnoses which attribute failed and routes test-time latent optimization to its corresponding support window. With the generator, optimizer, reward, verifier, and token budget held fixed, support-routed windows outperform late windows by 23.6 percentage points in GenEval++ macro accuracy. With the generator frozen, LOCUS raises Janus-Pro’s GenEval++ accuracy from 24.6% to 46.8% overall and from 19.5% to 40.5% on multi-attribute tags. On GenEval, LOCUS repairs 48.1% of failed generations in one rollout, compared with 37.4% for prefix optimization, and needs 14% fewer rounds on average to succeed.
[CV-54] Beyond Co-Generation: Rethinking Jointness in Joint Audio-Video Generation
链接: https://arxiv.org/abs/2611.22857
作者: Jihoo Jung,Ji-Hoon Kim,Suyeon Lee,Gia Thinh Pham,Joon Son Chung
类目: Computer Vision and Pattern Recognition (cs.CV); Audio and Speech Processing (eess.AS)
备注: Preprint
Abstract:Our aim is to advance the understanding of genuine audio-visual coordination in joint text-to-audio-video (T2AV) generation, with particular emphasis on semantic consistency. Evaluating audio-visual consistency in T2AV generation is nontrivial, as detailed text prompts obscure whether such consistency stems from genuine cross-modal coordination or merely from independent text following. To isolate the former, we introduce intrinsic jointness: the ability of audio and video to produce the same realization for details left underspecified by the prompt. Using semantically underspecified prompts (e.g., “musical instrument playing” rather than “electric guitar playing”), we investigate four recent models and reveal their limited intrinsic jointness. To address this limitation, we propose JointREPA, a simple representation alignment-based post-training method to facilitate cross-modal coordination. Since existing benchmarks use semantically detailed prompts, we introduce UNSAIDBench, a benchmark designed to evaluate intrinsic jointness under semantic underspecification. Experiments across two distinct models and three benchmarks, including UNSAIDBench, demonstrate that JointREPA consistently improves semantic consistency.
[CV-55] Propose Verify: Fast Reflection Symmetry Detection via Harmonic Fields
链接: https://arxiv.org/abs/2611.22856
作者: Yishai Lavi,Shai Avidan
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Detecting multiple reflection symmetries in 3D shapes requires aggregating geometric evidence over a large space of candidate planes, making traditional voting approaches computationally expensive. We introduce the Fourier Accelerated Symmetry Transform (FAST), a training-free framework that combines geometric pair voting with efficient spectral computation. Exploiting the fact that global reflection planes pass through the shape centroid, we reduce the search to plane orientations, while geometric consistency filtering and independent witnesses suppress unreliable pair hypotheses. Our key insight is to represent these votes as a signal over plane orientations and efficiently compute a continuous, smoothed symmetry field in the spherical-harmonic domain. Specifically, we exploit the Funk-Hecke theorem, the spherical counterpart of the Fourier convolution theorem for rotationally symmetric kernels, to transform Gaussian smoothing into multiplication of harmonic coefficients. This enables efficient field evaluation, from which symmetry candidates are extracted, refined, and geometrically verified. Across two benchmarks, including a new benchmark of highly symmetric shapes, FAST achieves the highest clean multi-plane detection F_1 scores among evaluated methods (0.861 and 0.947), with GPU inference below 4 ms per shape: over 100\times faster than the next-fastest evaluated multi-plane GPU baseline. FAST also demonstrates strong robustness to geometric noise and competitive single-plane detection, achieving 94.9% success on clean single-plane shapes.
[CV-56] CaFe: Causally Faithful Conditional Traffic Scenario Generation
链接: https://arxiv.org/abs/2611.22855
作者: Steven Yu Tung,Shubham Vaijanath Phoolari,Christoph Lauer,Abhinav Valada,Steven Peters
类目: Cryptography and Security (cs.CR); Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注:
Abstract:Language-conditioned traffic simulation enables scalable test scenarios steered by natural language. The goal is to control selected agents while surrounding traffic reacts to their observed behaviour. For example, a vehicle instructed to cut in late should cause the following vehicle to brake only when the manoeuvre becomes observable. Yet simulators can leak private instructions across agents, causing surrounding vehicles to react before the instructed behaviour becomes observable. Such premature reactions defuse desired events and alter the ensuing closed-loop scenario. Existing benchmarks measure what agents do, not what they could have known, leaving this failure unmeasured. We formalise it as intent leakage and address it through evaluation, data and modelling. Our paired intervention test holds an instructed agent’s executed trajectory fixed while varying its private instruction, isolating leakage-induced changes in surrounding agents. We also construct WOMD-PACT, an automatically generated, verifiable dataset of 531,099 WOMD scenes with timed behaviour tags with reaction episodes identified through a counterfactual ghost test and grounded private instructions. Finally, we introduce CaFe, a dual-stream diffusion-forcing simulator fine-tuned in closed loop that restricts inter-agent interactions to public beliefs, preserving legitimate anticipation without leaking private intent. Our evaluation reveals that existing conditioning designs displace 23-83% of nearby uninstructed agents through private-intent leakage, even when the instructed agent’s executed motion is held fixed. CaFe eliminates measurable leakage while maintaining comparable WOSAC realism (0.763), demonstrating that causal faithfulness can be enforced without sacrificing measured simulation realism.
[CV-57] Keeping Parts Honest: Isolated Part Graphs for Faithful Vision Transformer Explanations
链接: https://arxiv.org/abs/2611.22853
作者: Bernard Tomczyk,Gail Collyer-Hoar,Bryan Williams
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 12 pages, 6 figures, 6 tables. Supplementary material will be added in a later version
Abstract:Part and prototype-based explanations for Vision Transformers (ViTs) present a decision as evidence from individual image regions. However, self-attention mixes information across patches, so part-level features absorb context from the whole image and consequently fail to faithfully reflect the decision. We show that existing interpretable ViT classifiers depend heavily on this leakage: when each explained region is encoded in isolation, their accuracy collapses, and retraining them under isolation recovers only part of the lost accuracy. We propose a classifier that represents each image as a graph of isolated parts. Each part is encoded only from its own patches by a ViT whose last blocks are fine-tuned under isolation, preventing evidence leakage from elsewhere in the image. Parts are discovered without annotation and refined hierarchically through their sub-parts, making them more consistent and stable across images than those of competing methods. Each edge is computed from part pairs, and recovers evidence that attention previously shared across the object without affecting the parts. Matching the image graph to a learned prototype graph per class makes the prediction a direct combination of node and edge similarities, each computed only from the patches it is credited to, so every explanation is faithful by construction. Across 11 fine-grained and general-object benchmarks, our method exceeds the accuracy of existing explainable methods that rely on leakage, while providing explanations that reflect what the model actually uses.
[CV-58] chnical Report of Mobile Manipulator Robot for Industrial Environments
链接: https://arxiv.org/abs/2611.22852
作者: Aria Dadnavi,Matin Jamshidi,Kian Khatibi,Shahriar Karimi Makhsous,Saina Najafi Shabestari,Seyed Sobhan Hosseini Hajibekandeh,Zahra Mahaam,Reyhane Mohasseli,Ghazal Ghoreishi,Amirhossein Gholizadeh Behbahani,Mohammad Hossein Moaiyeri
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV); Systems and Control (eess.SY)
备注: 25 pages, 22 figures, 4 tables. Technical report of Team Auriga’s 2025 mobile manipulator (@Work) platform (Robotics and Intelligent Automation Laboratory, Shahid Beheshti University)
Abstract:This paper presents the design and implementation of Auriga Robotics’ mobile manipulator robot, developed at the Robotics and Intelligent Automation Laboratory, Department of Electrical Engineering, Shahid Beheshti University. Hereafter, this platform is referred to as the @Work Robot. The robot targets industrial tasks, with an emphasis on improving efficiency in repetitive or hazardous work environments. It uses a 4-wheel Mecanum drive for omnidirectional movement and a 5-degree-of-freedom manipulator arm with a 3D-printed gripper for object handling and navigation. The electronics consist of custom PCB boards built around ESP32 microcontrollers, together with an NVIDIA Jetson Orin NX that handles high-level processing and real-time system control. On the software side, Hector SLAM provides mapping and localization, A* handles path planning, and YOLO performs real-time object detection. These modules draw on fused sensor data to deliver robust navigation and reliable collision avoidance. Overall, the system shows effective hardware-software integration, with consistent performance in autonomous navigation, object handling, and task execution in industrial-like settings.
[CV-59] Kolibri: Color Finds the Place Refinement Sets the Pose for RGB-D Localization in Colored Point Clouds
链接: https://arxiv.org/abs/2611.22829
作者: Leonel Aguilar,Bartu Tuncay,Christoph Hoelscher
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:An agent or robot in a scanned building must locate its camera in rooms that no model saw during training. Point cloud registration methods can align each RGB-D frame with the building’s colored point cloud, accurately but at a high cost. We observe that geometric refinement brings nearby initial poses to the same final pose, so the learned component only needs to find the place. Kolibri does this with a graph network of 55,712 parameters that describes 10 cm cubes of points, called cells, by their color, local shape and neighbors. With the same refinement, it matches GeDi on 96 unseen ScanNet rooms (81.4% against 80.3% within 5 cm and 5°) and on a held-out test of 10 further rooms, in 82 ms per frame on an RTX 2080 Ti. The saved time allows for search without a room label. In one point cloud of 100 rooms, Kolibri loses 1.3 points against knowing the room, while retrieving the room first loses 56.0. In a hospital emergency department scene from a real scan, it localizes 86.5% of the frames rendered from the scan in 384 ms each, where GeDi needs 91 minutes per frame on cubes of the scan for a similar recall. The method needs colors that agree between the frame and the point cloud. On gray frames and across cameras with different color responses, point cloud registration methods keep the lead.
[CV-60] A tiny object detection dataset for high-altitude open-water search and rescue
链接: https://arxiv.org/abs/2611.22825
作者: Hui Xu,Chenfei Hou,Henglai Wei,Hao Wang,Yidong Jin,Xiulin Niu,Haoyang Zhang,Hui Zhang,Tianshi Xu
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Large aerial platforms often operate above 1.5 km to improve the efficiency of open water search and rescue (OW-SAR). However, long imaging distances, wide fields of view, and limited sensor resolution reduce objects to a few pixels, increasing their susceptibility to environmental interference and making detection harder than for conventional small objects. To our knowledge, existing datasets focus on low-altitude scenarios and do not support automatic detection at 1.5-2.0 km. We therefore present TinyWOD, a tiny object detection dataset for high-altitude OW-SAR using lifebuoys as detection targets. It comprises 11,359 manually annotated images and 19,505 annotated instances spanning diverse environments, lighting conditions, water colors, and surface ripples. Compared with existing datasets, TinyWOD contains much smaller objects and a substantially higher proportion of tiny instances; its average equivalent side length (ESL) is 12.7 pixels, with 76.96% of in stances having ESLs below 15 pixels, reflecting actual high-altitude water-rescue imaging conditions. We define APtiny, a tiny object detection metric, and benchmark mainstream and state-of-the-art networks using this and conventional metrics. All evaluated networks effectively detect tiny waterborne objects, demonstrating TinyWOD’s reusability and utility. We expect this dataset to support research on tiny object detection in high-altitude OW-SAR.
[CV-61] Pixel is All You Need
链接: https://arxiv.org/abs/2611.22820
作者: Jie Chu,Haoting Shi,Fucheng Zhang,Jiawei Liu,Junwei Liao,Ying Wen,Muning Wen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 20 pages, 6 figures, 7 tables, including references and appendices
Abstract:Text, images, audio, touch, and even robot actions can, in principle, all be rendered onto a single canvas, suggesting an ambitious goal of unifying all modalities within vision. Language provides the natural first test. To subsume it, a visual model must reason directly over text in pixels. Across a wide range of image, video, and unified multimodal generators, however, we find that their visual generation modules cannot perform symbolic reasoning over text on the canvas. They succeed only when an underlying text module or prompt rewriter already supplies the answer, effectively reducing the generator to a mere renderer, a limitation that post-training does little to resolve. To test whether this failure stems from pixels themselves or from current training paradigms, we train a family of purely pixel-space generative models from scratch without any text pathway. With images as their sole input and output, these models read questions, write intermediate steps, and produce answers directly on the canvas. Remarkably, they exhibit hallmark behaviors of large language models. Performance improves predictably with scale following a power law. Writing intermediate steps frame by frame substantially outperforms single-frame answering, echoing chain-of-thought, while reinforcement-learning post-training further strengthens reasoning. Moreover, sharing one canvas allows the same model to reason across modalities, a capability that text-only language models lack. To the best of our knowledge, this is the first systematic study of these phenomena in purely pixel-space generative models free of text tokens.
[CV-62] Visual Harness Scales General VLMs to Medical Images
链接: https://arxiv.org/abs/2611.22815
作者: Shaowen Wan,Hanqi Jiang,Zihao Wu,Junhao Chen,Weihang You,Ruiyu Yan,Bowen Chen,Tianming Liu,Lin Zhao
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:We show that an appropriate visual harness can substantially improve the performance of general vision-language models (VLMs) across diverse medical imaging tasks. We introduce \ours, a medical visual harness that equips general VLMs with interactive visual tools and systematic diagnostic evidence management. With \ours, general VLMs interactively explore medical images and progressively acquire diagnostic evidence, which is preserved in a lossless memory for subsequent retrieval and use. After submitting an initial answer, the VLM receives an assessment of potential omissions and the reliability of identified findings. This assessment guides further image inspection and measurement, followed by answer revision in an iterative agentic loop until sufficient evidence is obtained. Across 11 tasks and four imaging modalities, \ours improves performance on all tasks with Qwen3.8-27B, with notable gains in spleen measurement (63.3% to 98.3%), renal search (57.5% to 71.2%), and pulmonary nodule detection (0.0% to 28.2%). High-capability API-based VLMs achieve even higher performance with \ours. For example, GPT-6.1 Sol improves pulmonary nodule detection from 10.8% to 69.2% with \ours, demonstrating the complementary benefits of stronger model capabilities and an effective medical visual harness. Our results highlight the potential of visual harnesses as a new paradigm for scaling general VLMs to medical imaging tasks.
[CV-63] NexSight: Verifiable Tasks for Fine-Grained Agent ic Long-Video Understanding
链接: https://arxiv.org/abs/2611.22790
作者: Tianci Xie,Li Chen,Pinxue Guo,Rui Zheng
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: 25 pages, including references and supplementary material
Abstract:Fine-grained questions about long videos ask how many times an action occurs, in what order events unfold, or exactly when a state changes. Answering them can require finding brief moments scattered across hours of footage. Agentic models that decide when and where to look promise to find these moments, yet this ability is hard both to measure and to learn. Sparse supporting evidence may establish that an action occurred without establishing its total count; verifying timeline-wide claims requires checking for missed occurrences and double counting. Answer-only supervision, in turn, does not explicitly supervise the search for evidence. We present NexSight, which builds verifiable tasks to both evaluate and train agentic long-video understanding. Its agentic pipeline, SightForge, accepts a task only if a verifier blind to the draft answer can reconstruct it from evidence densely reacquired from the raw video and counterexample checks fail to refute it. This yields SightLine, 310 evidence-traceable tasks from 108 long videos, evaluated through SightLoop, a shared multiscale harness supporting temporal exploration, local inspection, evidence integration, and revisiting while recording returned observations. On disjoint videos, solver trajectories on verified tasks become step-level supervision for observation decisions. The strongest model reaches only 54.52% on SightLine, yet 79.35% when given the annotated evidence. With under 6K examples, Sight-35B nearly doubles its base model’s interactive accuracy (21.29% to 42.26%), outperforms the 397B model of its family, begins using explicit spatial inspection after post-training, which its base model never uses, and gains over 8 points on two external long-video benchmarks.
[CV-64] ForceBody: Force-Paired Parametric Body Motion with Torque Uncertainty NEURIPS2026
链接: https://arxiv.org/abs/2611.22769
作者: Joonwoo Kwon,Yufei Zhang,Xiaoming Liu,Zijun Cui
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: Accepted at NeurIPS 2026 Evaluations Datasets Track
Abstract:Forces and torques drive human motion, central to biomechanics, robotics, and physics-aware animation, yet they are often absent from learning-based motion pipelines for body reconstruction and generation. Some motion datasets provide force annotations, but the paired motion sequences are incompatible with the parametric body models used in modern learning pipelines. In addition, existing datasets typically obtain joint torques through inverse dynamics, which is inherently sensitive to the quality of the input kinematics; however, they do not quantify the reliability of these torque estimates. We close this gap with ForceBody, the first dataset to bring ground truth force supervision into a parametric body representation, shipped with per-sample torque uncertainty. ForceBody pairs the SKEL body model with ground reaction forces and inverse-dynamics joint torques across 10,386 motion trials (9.7M frames, 26.9 hours) from 140 subjects. Each torque label is paired with a per-frame, per-joint uncertainty annotation obtained by Monte Carlo sampling through the inverse dynamics pipeline. We benchmark six architectures (MLP, Conv1D, LSTM, GRU, Mamba2, Transformer) on ground reaction force and joint torque prediction from SKEL motion. As one example use of the released uncertainty, finetuning a Transformer with per-sample uncertainty weighting reduces the MAE of joint torque estimation by 5%. Code and data are at this https URL.
[CV-65] aching What to Look For: Privileged Perception Questions for Agent ic On-Policy Distillation in Video Captioning
链接: https://arxiv.org/abs/2611.22755
作者: Yifan Xu,Ruopeng Gao,Zhixiao Wang,Jiafu Tang,Yichun Yang,Zihao Wang,Yuanxing Zhang,Limin Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Detailed video captioning requires comprehensive and faithful descriptions of objects, actions, and temporal events. Fixed-frame sampling, however, can miss brief motions, small objects, and other fine-grained evidence, motivating captioning agents that actively seek relevant moments and inspect local regions. Training such behavior is challenging because, unlike video question answering, generic captioning instructions do not specify what visual evidence should be acquired. We propose Perception-Question On-Policy Distillation (PQ-OPD), which uses training-only Perception Questions (PQs) to guide active video perception. PQs specify what should be inspected without revealing the corresponding answers, encouraging the teacher to explore the video more comprehensively and focus on relevant visual evidence. Along trajectories generated by the student, on-policy distillation transfers this guidance to the student’s reasoning, tool calls, and captions; no PQs are required at inference. Across three detailed-captioning and two caption-to-QA benchmarks, PQ-OPD consistently improves the corresponding untrained agents across teacher-student scales and input-frame budgets. Under our evaluation protocol, PQ-OPD achieves the strongest primary-metric results among the compared open-source models on DREAM-1K, CaReBench, ShortVidBench, MotionBench, and VidCapBench-AE. Our ablations, together with analysis of temporal coverage and answer leakage, further indicates that the student internalizes the training-time guidance as a more effective video exploration policy.
[CV-66] GaussianMamba: Geometric and Relational State Space Scanning of 3D Gaussian Primitives
链接: https://arxiv.org/abs/2611.22742
作者: A. Enes Doruk,Ozgur Erkent
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:3D semantic occupancy prediction requires representing complex spatial structures and modeling contextual relationships across a scene. 3D Gaussian primitives provide a sparse and adaptive representation of scene geometry and semantics, while Mamba offers efficient sequence modeling for capturing dependencies among primitives. However, Gaussian primitives are irregular and unordered, making sequence construction challenging, as conventional spatial ordering may overlook their anisotropic geometry and feature-semantic relationships. In this paper, we present GaussianMamba, a dual-branch Mamba framework that explicitly incorporates Gaussian-specific geometric and feature-semantic relationships into complementary scanning strategies. The first branch refines a locality-preserving traversal using anisotropic Gaussian support derived from primitive scale and orientation. The second branch reorders spatially organized groups according to feature-semantic affinity and adjusts local features using geometry-weighted semantic differences. The resulting sequences are processed by separate Mamba modules, and their features are fused through a support-aware mechanism that incorporates local feature discrepancies arising from the distinct scanning paths. Experiments on nuScenes-SurroundOcc and SSCBench-KITTI-360 demonstrate consistent improvements over GaussianFormer and GaussianFormer-2 in both IoU and mIoU.
[CV-67] RefCloth: Transferable Adversarial Tablecloths for Vision-Language-Action Models
链接: https://arxiv.org/abs/2611.22736
作者: Shaozhen Liu,Zhiwei Li,Jiayang Sun,Shuangjun Du,Yi Zhong,Shenrui Ma,Yaozu Liu,Qi Li,Zhenan Sun
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Closed-loop vision-language-action (VLA) policies rely continuously on multi-view cameras, yet existing visual attacks typically hijack sensor streams with 2D patches or alter target object meshes that require retraining when manipulation objects change. We introduce RefCloth, a framework that optimizes one persistent adversarial tablecloth on a frozen source policy and deploys it zero-shot across unseen tasks and target models without modifying task objects or cameras. Throughout manipulation, static third-person and moving wrist views observe this shared world-bound texture under evolving occlusions. To generalize across shifting surface views while bridging the gap between differentiable and native renderers, RefCloth combines bounded texture parameterization with shared surface UV transformations and a same-path clean reference that removes static rendering offsets at initialization. Across four LIBERO suites in native simulation, one frozen texture transfers from OpenVLA-OFT to \pi_0 and \pi_0.5 , reducing success relative to an unoptimized cloth by 62.50 and 43.75 percentage points on average and far exceeding amplitude- and variation-matched random controls. Dual-view audits and target closed-loop camera interventions show that moving wrist exposure drives over twice the standalone drop of the main view with strong synergy, while ablations confirm the roles of UV diversification and path calibration.
[CV-68] Evolve in Panorama Render in View: Panoramic World State Evolution for Video World Models
链接: https://arxiv.org/abs/2611.22733
作者: Longtao Jiang,Mingfei Han,Jianmin Bao,Tingjun Dai,Jinjie Wei,Zhiran Zhang,Xin Tao,Zhihui Li,Salman Khan,Xiaojun Chang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Video world models should maintain a world that keeps evolving as the viewpoint changes. Yet when the camera turns away and returns, current models often freeze the unobserved scene, lose objects, or change their appearance, because content outside the view has no representation that keeps changing. We present PanoEvo, which uses a Panoramic Evolver to evolve the world state in the latent space of a panoramic video diffusion model, where every viewing direction is present at every time step, so the scene evolves whether or not it is observed. Existing panoramic video models are not trained to evolve a scene from a single image or the objects within it, so we build PanoEvo-Data, an event-centric dataset from real 360° videos, and train the Panoramic Evolver coarse-to-fine as an image-to-panoramic-video model. Since each view covers only a small part of the low-resolution panorama, directly projected views are blurry and often show low-quality details or structural errors. A Perspective Renderer, built on a perspective video diffusion model, reads this state via latent-space projection and direction-restricted cross-attention to render sharp, high-quality views. PanoEvo achieves the best overall score among 23 publicly evaluated models on WRBench and the highest total score on STEVO-Bench. Code, models, and data will be released.
[CV-69] LoopZip: Visual Token Compression for Efficient Looped Vision-Language Models
链接: https://arxiv.org/abs/2611.22728
作者: Weizhong Huang,Zhicheng Liu,Yan Zhang,Fei Chao,Xiawu Zheng,Liujuan Cao
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 11 pages, 10 figures
Abstract:Looped vision-language models (VLMs) deepen computation by reusing shared layers, approaching larger models’ capabilities with fewer parameters and easing the storage and deployment burden of stacking independent layers. However, repeatedly processing long visual sequences still incurs substantial compute and GPU memory costs. Visual token compression reduces these costs by shortening sequences. Existing methods mainly select original tokens directly or combine selection with aggregation. In looped models, the former may overlook changing token importance across calls and discard information needed later; the latter lack joint consideration of output changes across calls, limiting information preservation. We propose LoopZip, a training-free visual token compression algorithm addressing both issues. Visual aggregation builds a compact candidate pool before the looped backbone, and cross-invocation selection averages instruction attention from the first two existing calls of the same shared attention layer to retain tokens needed by both states. Weight-tied value correction then jointly fits shared visual-value adjustments to both pre-selection layer outputs, reducing output deviations across states. To validate LoopZip’s effectiveness, we conduct extensive experiments covering two open-source looped VLMs, eleven benchmarks, and three pruning ratios. On LoopVL-1B, LoopZip retains 97.16% of the original model’s performance at 50% pruning ratio and exceeds the strongest baseline by 5.36% at 87.5% pruning ratio. Across Hugging Face Transformers, vLLM, and SGLang, LoopVL-1B with LoopZip reduces peak allocated GPU memory by 58.36%-58.98% and achieves 2.01-2.47x end-to-end speedups over the original model.
[CV-70] SlopeGuard: Towards Generalizable AI-Generated Image Detection via Radial Spectral Slope Guidance
链接: https://arxiv.org/abs/2611.22727
作者: Cheng Fu,Yiheng Li,Yang Yang,Wenzheng Liu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:AI-generated image detectors often overfit to generator-specific artifacts, hindering generalization to unseen synthesis pipelines. Existing frequency-based methods usually learn high-dimensional spectral patterns that may preserve generator-dependent variations. We explore Radial Spectral Slope, a compact descriptor of global spectral decay that captures transferable spectral trends while suppressing fine-grained variations. Our analysis shows that real and generated images exhibit systematic but generator-dependent slope differences, indicating that slope provides useful forensic evidence but is insufficient as a standalone decision criterion. This motivates combining spectral slope with semantic representations for more robust detection. Based on this insight, we propose SlopeGuard, which couples slope-aware data processing with slope guided feature learning to construct informative patch-based views, integrate Radial Spectral Slope with semantic features, and adapt prior preservation according to spectral atypicality. Extensive experiments demonstrate that SlopeGuard achieves state-of-the-art mean accuracies of 97.5% on AIGCDetectBench, 97.8% on GenImage, and 85.1% on AIGIBench, while maintaining competitive performance under blur, cropping, JPEG compression, and additive noise.
[CV-71] ConserveEdit: Source-Anchored Transactions for Drift-Bounded Iterative Image Editing
链接: https://arxiv.org/abs/2611.22724
作者: Shuyang Jiang,Fucheng Deng,Yuchuan Luo,Zhenyu Wu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 28 pages, 15 figures, 16 tables; includes supplementary material
Abstract:Instruction-driven image editors produce convincing local edits in one pass, but inside an iterative refinement loop every accepted output becomes the next input, so unintended changes outside the requested region accumulate. We present ConserveEdit, an editor-agnostic controller that treats each refinement step as a transaction against a contract compiled once from the source image. The contract separates what may change (the authorized region of each target, with a bounded halo for shadows and reflections) from what must survive (protected content with explicit tolerances). Candidates from a frozen editor are composed through a source-anchored mask and committed only if target progress, drift on protected content, and image quality pass as separate tests; otherwise the step is rolled back and repaired. For any editor, this keeps every pixel outside the authorized region bit-exact and bounds drift independently of the number of iterations, whereas no fixed scalar trade-off reproduces the separate tests. On all 1053 MagicBrush test edits with a frozen Qwen-Image-Edit, masking is necessary but not sufficient. A direct call already alters the rest of the image (12.1 dB PSNR outside the mask); composed through the source-anchored mask, it stays bit-exact there and its DINO similarity to the ground truth rises from 0.67 to 0.85, matching ConserveEdit, so this gain comes from the mask. Next to automatic masks, however, masked direct editing changes a protected face or text on 48% of turns, and masked scalar selection avoids this mostly by abstaining (23% constrained success). Because its drift test names the violated entity, ConserveEdit shrinks the support instead (2% violations, 53% constrained success); in a simulator with known ground truth it raises constrained success from 55.0% to 61.8%. Automatic grounding remains the main bottleneck. Code and logs will be released.
[CV-72] Practice in Text Act in Pixels: Correcting Where Perception Fails
链接: https://arxiv.org/abs/2611.22722
作者: Zhongling Xu,Hairong Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 25 pages, 8 figures, 19 tables. Code: this https URL ; aligned reader: this https URL
Abstract:GUI agents can learn cheaply from a text view of an interface and act on screenshots through a vision-language screen reader. Yet on a simulated mobile platform with a low-cost reader, policies that complete most tasks in text fail on many of the same tasks on screenshots. We ask where a small budget of real screenshots is best spent to close this gap. We find that the reader’s misreads are systematic and recur on the same interface elements. Noise matched to the reader’s error rates, a natural sim-to-real remedy, misses this structure, and in a pre-specified comparison it did not beat training the policy directly on screenshots. We therefore correct the misreads at their source with reader alignment, which fine-tunes the reader on screenshots paired with their text views. The reader learns to output each screen’s text view, as in text practice, and the policy stays unchanged. Spent on the reader, the screenshot budget gives 80.9% full-task success on new task instances, against 55.6% when spent on training the policy. Given the same labeled screenshots, adapting the reader also beats adapting the policy, and a single aligned reader improves every text-trained policy we tried without retraining any of them.
[CV-73] PCRG: Training-Free Pruning and Caching of Reference Tokens for Video Generation
链接: https://arxiv.org/abs/2611.22717
作者: Siyuan Wang,Haoran Li,Yi Yue,Tak L. Fung,Quan Chen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 11 pages, 5 figures
Abstract:Reference-to-video diffusion transformers pack the variational autoencoder (VAE) latents of all references and the generated tokens into one full-attention sequence, in which reference tokens make up about half. Because attention cost grows quadratically with sequence length, these tokens take up an even larger share of the attention computation, yet most of them are static, smooth or irrelevant to the output. To accelerate reference-conditioned generation, we propose pruning and caching of reference tokens for video generation (PCRG), a training-free, plug-and-play method that uses two signals the model already computes to assign each reference token to one of three tiers, namely pruned, cached or kept. The first is a low-level redundancy r from VAE latents, since a token whose latent barely changes over time or across its neighbors adds little new content. The second is a high-level importance g from the vision-language-model (VLM) conditioner, where the attention that the prompt and any first or last frames pay to a token tells whether the output needs to read it. Tokens that are not redundant are kept, and redundant tokens are cached if important and pruned otherwise. Pruned tokens leave the sequence before the first step with all rotary position embedding positions unchanged. Cached tokens have their keys and values computed once and reused, and kept tokens are processed as usual. On MiniMax-H3, already distilled to 8 timesteps and quantized to int8, PCRG is 1.57x faster end to end on our V2V-Bench and 1.49x faster on the public VACE-Benchmark, while the quality of VBench drops by only 0.24% and 0.17%, respectively. PCRG shortens the sequence rather than changing how attention is computed or which steps are run, so it also stacks on step caching and sparse attention. On V2V-Bench it reaches 1.86x with TeaCache, 1.82x with Spark at nearly unchanged fidelity and 2.15x with Veda.
[CV-74] Know When to Zoom: How Small Is Too Small for Vision-Language Models? WACV2027
链接: https://arxiv.org/abs/2611.22709
作者: Siddharth Vohra,Min Xu
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: Accepted to the SENSE workshop at WACV 2027. 10 pages, 5 figures, 3 tables. Code, data and a plain-language walkthrough: this https URL
Abstract:Online vision-language models often miss small details in large images, because the service shrinks the image, or rejects it, before the model ever sees it. A car 20 pixels wide in an 8000 by 6000 drone photo reaches Claude about 6 pixels wide, and OpenAI’s default rejects the photo outright. We measure how much detail survives for seven models from four vendors, using an eye exam of printed codes and 601 questions about small objects in real photos. The smallest text a model reads half the time grows from 6 pixels or less on a small image to a predicted 10 to 13 pixels on a 7680 by 4320 photo at default settings. Zooming wins much of the detail back. Splitting each photo into up to nine tiles raises accuracy by 11.1 to 14.5 percentage points for Claude and Gemini, and by avoiding rejections it lifts GPT-6.1 Sol from 50.7% to 83.7%. A two-step zoom, in which the model first names the region it needs, matches or beats tiles on Gemini and comes within 2.5 points of them on Claude, with about 40% of their image tokens. When a detail is lost, Claude and Gemini both say they cannot tell about half the time, but they fail differently in the rest: Gemini gives a confident wrong answer in 43% to 48% of these cases, Claude in only 5% to 9%. A check based on detail size, run before the call, misses most of the gain from tiles on Claude and Gemini, but Claude’s own reply is a reliable trigger: tiling only after an unsure reply comes within 1.4 points of always tiling with 62% to 70% of the image tokens. We turn these findings into five rules for when an agent should zoom, and release the eye exam, every model answer and code at this https URL.
[CV-75] Social Signal Bench: Evidence-Weighted Evaluation of Co-occurring Social Signals in Video and Audio
链接: https://arxiv.org/abs/2611.22704
作者: Siddharth Ravi,Anders R. Bargum,Hugo Anselme,Patrizio Cugia di Sant’Orsola,Alberto Cardaci,Flora Kincso Parragi,Vincent Borowiec,Line H. Clemmensen
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 25 pages, including appendix
Abstract:Recognizing social signals such as interest and hesitation requires interpreting context-dependent behaviors that may co-occur in conversation. Existing affective computing benchmarks rarely evaluate such co-occurring signals, and conventional multi-label F1 overlooks semantic relationships and annotator disagreement. We introduce Social Signal Bench, a 47-hour densely annotated benchmark for the evaluation of social signal recognition and engagement, comprising two datasets: Social Signal Bench Video (25 hours), covering ten social signals in audiovisual recordings, and Social Signal Bench Audio (22 hours), covering six signals in audio-only recordings. Each dataset pairs a public training set of online conversational video, of which 10 hours are expert-annotated and the remainder pseudo-labeled, with a private expert-labeled test set spanning four conversational domains. We also propose Evidence-Weighted F1 (EW-F1), which weights reference labels by annotator evidence and assigns graded credit to semantically related predictions through optimal per-clip matching. Controlled perturbations demonstrate that EW-F1 distinguishes semantically related errors from unrelated ones, unlike conventional F1. Under a shared evaluation protocol, 7B open models post-trained on Social Signal Bench outperform general-purpose frontier models on video and match them on audio.
[CV-76] ANGO: Real-World Unregistered RGB-Guided Hyperspectral Super-Resolution with Two-Camera Acquisitions CVPR2027
链接: https://arxiv.org/abs/2611.22699
作者: Matteo Kolyszko,Marco Buzzelli,Simone Bianco,Raimondo Schettini
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 18 pages, 11 Figures, 10 tables. Submitted to CVPR 2027
Abstract:Unregistered hyperspectral image super-resolution aims to recover a high-resolution HSI from a low-resolution hyperspectral observation and an unregistered high-resolution RGB reference. Although recent methods have begun to address the geometric misalignment between the two modalities, their evaluation is still constrained by the lack of a large-scale public benchmark based on real two-camera RGB–HSI acquisitions with a high-resolution HSI reference. To address this gap, we introduce DUET, the first large-scale public dataset for unregistered RGB-guided hyperspectral super-resolution, comprising 403 indoor and outdoor scenes captured with two separate RGB and hyperspectral cameras. We further propose TANGO (Two-camera AligNment with Gated Occlusions), which first compensates for the dominant cross-camera displacement through end-to-end global alignment and then refines the remaining inconsistencies with a coarse-to-fine local feature alignment. A learned gating mechanism further attenuates the RGB guidance where the two views cannot be consistently matched. Experimental results show that TANGO achieves state-of-the-art performance on both DUET and REAL across all evaluated scale factors, outperforming the strongest competitor on DUET by up to 2.29 dB in PSNR while requiring 2.5 \times fewer FLOPs. The code and dataset will be available at this https URL
[CV-77] CascadeJEPA: Enhancing 3D Geometric Representations via Spatially Coupled Cascaded Prediction
链接: https://arxiv.org/abs/2611.22686
作者: Kailai Zhou,Ziqi Chen,Xiaoning Xu,Yuefeng Wu,Zixuan Fu,Zijun Lin,Zhihao Li,Youwei Pang,Xiaoqi Zhao,Xun Cao,Bihan Wen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Joint-Embedding Predictive Architectures (JEPAs) have emerged as a promising paradigm for predictive visual representation learning. However, our analysis of I-JEPA reveals a growing discrepancy between semantic and geometric learning: improved semantic recognition during pretraining is accompanied by degraded 3D geometric representations. We hypothesize that this partly stems from predicting highly abstract latent representations with incomplete spatial context, favoring semantic regularities over fine-grained geometric information. To address this issue, we introduce CascadeJEPA, a spatially coupled cascaded prediction framework that promotes richer patch-level interactions through full-grid context encoding and cross-attention between prediction stages, while largely preserving the original I-JEPA architecture. Extensive experiments demonstrate substantial improvements over I-JEPA across multiple 3D geometric tasks while maintaining strong semantic recognition performance. Furthermore, its advantages extend to future feature forecasting and world model planning, highlighting the potential of enhanced geometric representations to support predictive world modeling.
[CV-78] Classification of Tissues on MALDI-MSI Image Modality for Pancreatic Cancer Studies
链接: https://arxiv.org/abs/2611.22680
作者: Abubakar Qahir,Jenny Benois-Pineau,Victoria Bourgeais,Fabien Baldacci,Sandrine Dabernat,Nicolas Desbenoit,Mélanie Moreau,Mehdi Boubaddi,Marion Marty
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Mass Spectrometry Imaging (MSI) spatially maps molecular distributions in tissue sections. Despite the availability of preprocessing tools for MSI data, building end-to-end workflows suitable for deep learning remains challenging. We have developed a fully automated pipeline for transformation of raw Matrix-Assisted Laser Desorption/Ionization Mass Spectrometry imaging (MALDI-MSI) data into deep-learning-ready image patches for binary pancreatic tissue classification (tumoral vs non-tumoral) and built individual models for recognition of tumoral regions in spectral images. Images resulting from the proposed pipeline and from reference interactive workflow are very close in terms of image quality metric mean squared error (MSE). The proposed pipeline computational time is of 20 min in average compared to a full working day of the reference interactive workflow. The models achieved 0.94 accuracy on high-resolution spectra and a mean accuracy of 0.59 on lower-resolution spectra.
[CV-79] FVScore: Specification-Driven Neuro-Symbolic Video Evaluation
链接: https://arxiv.org/abs/2611.22673
作者: Do Xuan Long,Long T. Le,Yiwen Song,Min-Yen Kan,Tomas Pfister,Yale Song
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: Project page: this https URL
Abstract:We introduce Faithfulness Video Score (FVSCORE), a neuro-symbolic framework for specification-driven video generation evaluation. Existing evaluators rely on holistic MLLM judgments or vision-language metrics, yet aligning their scores with human judgments and tracing them to specific temporal violations remain open challenges. FVSCORE addresses these limitations by compiling each video into explicit propositions capturing prompt requirements and implicit propositions capturing unstated physical, temporal, and cross-modal constraints. It grounds each proposition using specialized perceptual tools and evaluates the resulting confidence traces with soft temporal operators inspired by Linear Temporal Logic. A Significance-Aware Calibration step then aligns verification with human perception by retaining only perceptually significant violations before producing a timestamp- and evidence-grounded report. We conduct extensive experiments across public and complex multi-event benchmarks with the strongest video models to date, demonstrating that FVSCORE outperforms holistic and agent-based evaluators in human correlation while providing highly accurate and interpretable feedback.
[CV-80] Beyond the Camera: A Tri-Modal Driving World Model for Future RGB Thermal and LiDAR Forecasting in Adverse Weather
链接: https://arxiv.org/abs/2611.22669
作者: Faysal Mehrab Chowdhury,Md. Mahfuzur Rahman,Ahmed Rafi Hasan,Kishor Datta Gupta,Mohd Ariful Haque,Savannah Shannon
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Driving world models forecast what a vehicle’s sensors will record next. Most forecast camera video and some add LiDAR, but thermal cameras, which still see in darkness and some adverse weather, are rarely a forecast target, and to our knowledge no world model forecasts all three sensors. We study joint forecasting of RGB, thermal, and LiDAR on real adverse-weather sequences of the DSERT-RoLL dataset. We post-train a pretrained video world model for each sensor and represent LiDAR as a depth video aligned with the RGB camera, with densification and encoding chosen by what the model’s video tokenizer can reconstruct. A tri-sensor model sees all three sensors as one video. We score LiDAR in meters against measured returns and compare against persistence baselines passed through the same tokenizer as the model. With this matching, an apparent loss of the thermal model to persistence in fog disappears. On all 58 validation sequences in six weather conditions, post-trained RGB and thermal forecasts beat compression-matched persistence. Joint training lowers the LiDAR error at 2s by 2.1m at a small cost in camera fidelity, and only the tri-sensor model’s LiDAR forecasts beat persistence in mean error. Blanking one sensor degrades the forecasts of the others more in snow than in fog.
[CV-81] Let Samples Sing in Chorus: Mitigating Spurious Correlations via Cross-view Consensus
链接: https://arxiv.org/abs/2611.22668
作者: Weiwei Li,Junzhuo Liu,Tong Chu,Hengfu Yu,Yinjie Lei,Wen Li
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Under Review
Abstract:Deep models often rely on spurious correlations that fail on underrepresented groups. Existing methods either correct human-specified bias attributes or infer unknown biases from the learned representation space. However, annotated attributes may differ from the shortcuts actually used by a model, while a single representation provides incomplete and potentially noisy evidence. Our key insight is that a bias obscured in one feature space may be clearer in another. In this work, we propose CHORD (Cross-view Harmonic Outlier Reversal for Debiasing), a multi-view framework that allows samples to “sing in chorus” across diverse representation spaces to achieve robust cross-view consensus. CHORD aggregates aggregates within-class partitions in each view, which is a proxy for specifying candidate biases, to debias the target representation and classifier without bias supervision. Experiments on seven benchmarks across vision, language, speech, and medical imaging show consistent improvements in worst-group accuracy. Code and checkpoints are available at this https URL .
[CV-82] VSpark: Speculative Decoding for Autoregressive Vision-Language-Action Models
链接: https://arxiv.org/abs/2611.22666
作者: Suho Yoo,Youngjoon Jang,Hyebin Cho,Taeyeop Lee,Jaewook Jeong,Joon Son Chung
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注: Preprint
Abstract:Our aim is to accelerate autoregressive vision-language-action models (VLAs), which generate robot actions one token at a time. Speculative decoding makes this possible: a lightweight draft model proposes several tokens at once, and the VLA verifies them in a single pass, accepting only those that match its own predictions, so decoding remains lossless. The speedup, however, is limited in robot control, where many different tokens encode nearly the same motion and a draft close to the correct action is still rejected. To address this, our approach, VSpark, is built upon three design choices: (i) rollouts collected with calibrated action noise, which expose the draft to a wider range of robot states; (ii) feature alignment, which trains the draft to match the VLA’s hidden features rather than its output tokens alone; and (iii) continuation reuse, which keeps the predictions the VLA has already computed along rejected drafts as candidates for the next verification step, instead of discarding them. On four VLAs with different action tokenizers and four manipulation benchmarks, VSpark achieves 1.28x to 2.10x speedup while keeping the VLA’s decoding lossless.
[CV-83] SpatialCast: Reading the 3D World via Cast Shadows
链接: https://arxiv.org/abs/2611.22657
作者: Enhan Li,Jinghao Zhou,Jiaze Zhang,Hongyang Du
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 10 pages (main paper), 7 pages (supplementary material), 5 figures and 6 tables in the main paper
Abstract:As multimodal large language models (MLLMs) advance toward understanding and interacting with the physical world, spatial reasoning becomes essential for understanding scene structure and anticipating the effects of physical changes. Existing spatial benchmarks have made substantial progress in evaluating spatial relations, memory, and viewpoint reasoning, but less systematically assess how models connect physical scene conditions with visual observations. We introduce SpatialCast, a benchmark that uses cast shadows to evaluate this ability. Cast shadows provide observable evidence of a 3D scene. Their image appearance depends on object geometry and pose, illumination, and viewpoint. We examine this relationship through three complementary tasks that assess scene-projection consistency, predict intervention-induced changes, and infer the physical changes behind visual transitions. The benchmark contains 21,000 synthetic questions with separate scene and geometry test partitions, complemented by 200 SpatialCast-Real questions from 20 photographed scenes. Evaluations show that closed-source MLLMs outperform general-purpose open-weight MLLMs and MLLMs designed for spatial reasoning, which remain near chance on the synthetic benchmark, while even strong closed-source models struggle with shadow-length and light-height judgments. We also use SpatialCast training data for GRPO post-training, improving open-weight MLLM performance on the benchmark. Further evaluations show improvements on several external spatial benchmarks, suggesting transfer of the learned capabilities. These experiments support SpatialCast as a controlled resource for diagnosing spatial reasoning capabilities and studying their learning and generalization.
[CV-84] Unified Residual Translation for Paired Image Restoration
链接: https://arxiv.org/abs/2611.22655
作者: Joyanta Jyoti Mondal,Mridul Banik,Ibne Farabi Shihab,Md. Shifatul Ahsan Apurba
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 10 pages, 2 figures
Abstract:Paired image restoration spans changes in imaging contrast, sensor modality, and spatial resolution. These settings differ in their input channels and target grids, raising the question of how much of the prediction pipeline must be specialized to each task. We present the Unified Residual Translator (URT), which separates fixed channel and scale mapping from learned residual prediction. A deterministic reference uses channel averaging, replication, or bicubic interpolation to match the target dimensions, while a NAFNet encoder-decoder predicts an additive correction from the complete input. Separately trained instances share the prediction rule, \ell_1 objective, and optimizer across MRI contrast synthesis, visible-to-infrared translation, SAR-to-optical translation, and real-camera super-resolution. Across six task and scale settings, URT achieves the highest PSNR and SSIM among the evaluated baselines on IXI, SEN1-2, and RealSR \times3 and \times4 . It reaches 30.774dB on IXI and 13.478dB on SEN1-2, with luminance PSNR of 33.596, 28.742, and 27.342dB at the three RealSR scales. On LLVIP, it ranks second in PSNR and SSIM. A study of four spectral alternatives identifies complementary benefits of fitted references and bandwise residual gains. Together, these results support explicit dimensional alignment and learned residual correction as a common formulation for heterogeneous paired imaging tasks.
[CV-85] OmniTongue: A Comprehensive MLLM Benchmark for Automated Tongue Diagnosis in Traditional Chinese Medicine
链接: https://arxiv.org/abs/2611.22652
作者: Liuyi Xu,Yun Guo,Shusheng Li,Xuewen Zhang,Yining Qian,An-Yang Lu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 11 pages, 2 figures, 3 tables
Abstract:Multimodal Large Language Models (MLLMs) have excelled in Western medical imaging, yet their potential in Traditional Chinese Medicine (TCM) tongue diagnosis remains systematically unevaluated. Tongue diagnosis requires sophisticated global and local perception, serving as a rigorous touchstone for assessing the visual understanding depth of MLLMs. However, existing datasets are limited by single dimensions, coarse granularity, and ambiguous labels, hindering the evaluation of authentic clinical capabilities. To bridge this gap, we introduce OmniTongue, the first comprehensive, fine-grained Visual Question Answering benchmark for TCM tongue diagnosis. This benchmark integrates 1,351 expert-validated images and 21,616 instruction pairs. Its core design features 16 clinical dimensions for multidimensional evaluation, region-specific annotations for pathological localization, and 66 fine-grained labels for precise diagnostic representation. Extensive evaluations of existing MLLMs reveal significant performance gaps, demonstrating that medical-specific models lack domain advantages, while universally poor Macro-F1 scores expose a reliance on statistical shortcuts rather than genuine understanding of tongue pathology. These findings not only highlight the current limitations of MLLMs in interpreting regional medical features but also underscore OmniTongue’s significance for developing robust automated tongue diagnosis models. The code and dataset will be made publicly available in a future release.
[CV-86] Must You See Fakes to Catch Fakes? Language Says What to Look For Real Images What to Ignore
链接: https://arxiv.org/abs/2611.22650
作者: Tianchi Xu,Yuxuan Zhou,Yuheng Zhang,Yanan Guan,Tao Dai,Shu-Tao Xia
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Must a detector see generated images in order to catch them? Detection of AI-generated images has long faced two problems. The cues a detector learns from the few generators in its training set carry traces specific to those generators and transfer poorly to new ones, which limits generalization. The real and generated training images also differ in format and content, so the detector takes these differences as evidence of authenticity, and its decisions change with factors that have nothing to do with authenticity, which harms robustness as well as generalization. Both problems stem from training on generated images. A detector never trained on them can largely avoid both at the source. Vision foundation models have already learned the concept of “AI-generated” during pretraining, and text can point to it directly. Building on this, we propose SaySee, a training-free detector: Say builds a generation direction from sets of paired descriptions that differ only in authenticity and scores an image by the projection of its embedding onto it, and See uses real images to correct the score offsets caused by format, image type and content. On 24 evaluation sets that span standard benchmarks, in-the-wild data and the latest generators, SaySee reaches a mean AUC of 98.2% with MoE-ViE, 4.2 points above the best existing detector. When format is decoupled from authenticity, it remains the best of all methods. Code will be released.
[CV-87] REVEAL: Reconstructing Video Amodal Content via Language and Flow Guidance
链接: https://arxiv.org/abs/2611.22645
作者: Minh Tran,Winston Bounsavy,Ngan Le
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to Transactions on Machine Learning Research (TMLR)
Abstract:Amodal perception enables humans to perceive entire objects even when parts are occluded, a remarkable cognitive skill that artificial intelligence struggles to replicate. Recent diffusion-based methods have extended amodal completion from images to videos, yet they lack auxiliary priors to guide the reconstruction of heavily occluded content and to maintain temporal consistency over long sequences. We present REVEAL (REconstructing VidEo Amodal content via Language and flow guidance), a unified diffusion-based framework that integrates complementary motion and semantic priors for video amodal completion. A user-provided text query serves a dual role: it drives open-vocabulary video segmentation to obtain visible masks, and it provides semantic guidance for texture reconstruction. For amodal segmentation, we introduce optical flow guidance: by warping visible masks from previous frames to the current frame, the flow-warped mask propagates visible content into occluded regions and approximates the object’s current shape, providing a strong shape prior even under simultaneous deformation and occlusion. For texture completion, text guidance constrains the reconstruction of occluded appearance while providing a stable semantic reference that helps maintain visual coherence throughout the reconstruction. We also contribute LAVAT, a new benchmark featuring long video sequences paired with text descriptions, enabling evaluation of text-guided video amodal completion under heavy occlusion. Extensive experiments demonstrate that REVEAL achieves state-of-the-art performance on existing benchmarks while maintaining high-quality temporal consistency over extended sequences. The project page is available at: this https URL.
[CV-88] OVET-Consensus: Rank-Preserving Counterfactual Event Verification for Video Moment Grounding
链接: https://arxiv.org/abs/2611.22637
作者: Alibay Osmanli,Shaogang Gong
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 22 pages, 2 figures, 17 tables
Abstract:Modern video moment retrieval systems often propose windows that overlap the target event, yet still select the wrong one. We study this scorer bottleneck on QVHighlights with a fixed QD-DETR-derived temporal expert: the parent-aware temporal expert reaches 64.71 R1@0.5, while diagnostic oracles over the same candidate family reach 75.68 R1@0.5 within the parent-reranked pool and 94.97 R1@0.5 over the full 100-candidate evidence pool. We introduce OVET-Consensus, a rank-preserving counterfactual event verification layer that keeps the temporal expert’s ranked list as the default and permits only high-confidence rank-1 overrides from a policy calibrated on held-out training videos. The verifier uses parent-aware interval evidence, structured Qwen3-VL event judgments, VideoLLaMA3 pairwise temporal signals, and dense temporal evidence, but never lets those signals freely reorder the top-K list. On QVHighlights validation, OVET-Consensus improves the parent-aware temporal expert from 64.71 to 65.35 R1@0.5 with a positive paired-bootstrap interval, while mAP is largely unchanged, from 42.67 to 42.74; it makes 20 overrides, with 15 helpful corrections and 3 harmful corrections. The results point to a practical constraint on verifier design: semantic compatibility is useful evidence, but temporal grounding still requires deciding where the event starts, changes, and ends inside a visually compatible scene.
[CV-89] Sub-Meter Canopy Height Synthesis from Sentinel-2 via Anchored Residual Flow
链接: https://arxiv.org/abs/2611.22636
作者: Kiarie Ndegwa,Tony Chang,Andreas Gros,Vincent A. Landau,David D. Diaz,Luke Zachmann,Nathan E. Rutenbeck,Guy C Bayes
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 30 pages, including appendices
Abstract:Canopy height models (CHMs) - per-pixel maps of vegetation top height - are essential for forest carbon accounting, wildfire fuel mapping, and tree-level ecology, yet sub-meter CHMs require airborne LiDAR refreshed on multi-year national cycles. We produce 0.5 m CHMs from free 10 m Sentinel-2 imagery, a 20 \times cross-modal super-resolution task where optical reflectance must predict geometric height it does not directly observe. Motivated by the perception-distortion trade-off, we derive two operating points with shared conditioning. A deterministic regressor (Anchor) targets low distortion; a conditional flow matching model, trained on the Anchor’s residual in a learned CHM latent space (AR-flow), targets perceptual fidelity. Trained on 197k LiDAR-paired tiles spanning 69 ecoregions across the contiguous United States and evaluated on 2,530 held-out tiles against eight baselines, the Anchor achieves 2.83 m MAE using only free satellite input. The AR-flow reduces sliced Wasserstein distance from the Anchor’s 1.13 to 0.208, at a higher MAE of 3.68 m. Crown-scale statistics improve, but strict individual-tree detection remains weak: generated sub-pixel detail represents plausible structure rather than measured individual trees. Quarterly continental refresh becomes tractable, letting satellite monitoring inform timing and extent for more expensive data acquisitions.
[CV-90] PlaceTally: A Test-Time Memory Model for Constructing Place Graphs
链接: https://arxiv.org/abs/2611.22608
作者: Alibay Osmanli,Shaogang Gong
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 12 pages, 3 figures, 5 tables. Code, data and supplementary material: this https URL
Abstract:Counting distinct objects in a video is hard when the camera returns to places it has seen. A vision-language model that reads the whole video answers with a few small numbers, and one that reads it in parts counts an object again in every part that shows it. Object trackers need the category before the video is read, or a fixed vocabulary. We observe that a memory for counting does not need to know what it is looking for, only where it has been: objects rarely leave a room between visits, so parts of a video grouped by place share almost no objects, and a place can be recognised before any question is asked. Building on this, we introduce PlaceTally, a question-blind test-time memory model that constructs a place graph for each video. Each video fragment joins the place of its best earlier match or opens a new place, because a strong match is rarely wrong while a true return often matches weakly. A frozen vision-language model counts each place once from at most 128 of its frames, so every read has a bounded input and the memory grows with the number of places; the graph is built 15 times faster than real time. On RevisitCount, our benchmark of real room re-recordings, it scores 62.5% mean relative accuracy on test rooms and 61.2% on held-out rooms, against 51.2% and 44.3% for the same model reading the whole video. It stays level as rooms return more or less often, while a fixed correction of summed visits moves from 38.5% to 61.4%.
[CV-91] Propose with the Gradient Accept with the Loss: One Hedged Attack for Pixel-Sparse Robustness
链接: https://arxiv.org/abs/2611.22596
作者: Hon Ming Yan,Chun Pong Lau
类目: Computer Vision and Pattern Recognition (cs.CV); Cryptography and Security (cs.CR); Machine Learning (cs.LG)
备注: 22 pages, 13 tables, 4 figures
Abstract:A pixel-sparse adversary may change a few pixels of an image to any value. Robustness to such attacks is currently measured with an ensemble, sparse-AutoAttack (sAA), because its members fail in different places: sparse projected gradient descent (sPGD) trusts a first-order model that a full pixel change breaks, and sparse random search (Sparse-RS) ignores the gradient even where it is informative. We show that both attacks are local searches, each a proposal distribution plus an acceptance rule, and ask what such a search can learn from the gradient. A Taylor bound says what to measure, and measurements at the states the search visits give five rules. The sign of the gradient is its most reliable use at a single pixel and sets a new pixel’s value. Its ranking of candidate pixels should be followed loosely, and its ranking of the pixels to remove, which is often more reliable, can be followed more closely. A pixel should not be proposed twice before a move is accepted, and acceptance should be decided by the loss alone. These rules define Gradient-guided Random Search (GRS), which runs once and costs one forward and one backward pass per step. On 13 CIFAR-10 configurations and 5 ImageNet models, GRS matches or beats the best single member of sAA and is within one point of sAA itself or stronger on 17 of 18 settings, and it is significantly stronger than sAA on 8. It uses 3 instead of up to 7 forward-pass units, and on CIFAR-10 it runs 3.7 to 5.0 times faster than the members of sAA run in full.
[CV-92] Distilling World Action Models for Efficient Bimanual Manipulation via Anchored Path Matching
链接: https://arxiv.org/abs/2611.22592
作者: Jialong Liu,Junjie He,Yandu Chen,Bin Zhou,Heng Su,Peijun Tang,Jianan Wang,Changhao Chen
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注:
Abstract:World Action Models (WAMs) jointly generate future video and robot actions through iterative denoising. Few-step distillation reduces the number of model evaluations, but its coarse updates can move the student away from the teacher’s trajectory. We propose Anchored Path Matching (APM), which supervises endpoint and short-transition predictions from states reached by the student. From each state, we continue the teacher to zero noise and use its endpoint and first intermediate state as targets. Modality-routed adapters learn to match both targets from the same starting state. When the two gradients conflict, anchor-priority updates preserve the endpoint gradient and remove the opposing component of the local gradient within each modality. One checkpoint supports 1-, 2-, and 4-step joint generation. The sampling budget can be chosen at deployment without retraining or teacher queries. On RoboTwin 2.0, we train only on Clean demonstrations and evaluate directly on the domain-randomized Random setting. At one step, APM improves closed-loop success over continued base distillation at matched inference budgets by 2.0 percentage points on Clean and 1.4 on Random. Video peak signal-to-noise ratio (PSNR) rises from 30.169 to 32.317 dB, with gains in structural similarity (SSIM) and learned perceptual image patch similarity (LPIPS). On two real-robot tasks, APM improves success over the one-step teacher and reduces control-cycle latency by 21.3\times relative to the eight-step this http URL page: this https URL.
[CV-93] PACT: Plan-Action Consistency Training by Preserving Correct Plans and Repairing Actions in GUI Agents
链接: https://arxiv.org/abs/2611.22591
作者: Zhuohan Cai,Zhangxuan Gu,Runze Li,Beitong Zhou,Ming Wen,Hongbo Yu,Shuheng Shen,Changhua Meng
类目: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 21 pages, 11 figures, including supplementary material
Abstract:GUI agents can formulate a correct immediate plan in their reasoning yet emit an action that fails to execute it, creating an execution gap. This recurring mismatch undermines reliable task completion, as a single execution error can derail a multi-step workflow. In outcome-supervised reinforcement learning, broadcasting a negative response-level advantage across all generated tokens risks penalizing valid planning together with erroneous execution. To address this credit-assignment conflict, we introduce PACT (Plan-Action Consistency Training), a training framework that combines reference-aligned plan protection and plan-aligned action repair without altering the original scalar action reward. Using annotated reference actions to identify execution gaps, PACT protects the verified plan by masking negative policy-gradient contributions across its token span. Keeping the generated reasoning unchanged as context, it then favors the reference action over the emitted erroneous action through a preference surrogate with stop-gradient pair weights. Compared with the outcome-only baseline, PACT raises AndroidWorld task success from 60.34% to 68.97% (+8.63 pp), while improving rule-based plan-action consistency on AndroidControl High from 83.71% to 89.74% (+6.03 pp). Adapting PACT to desktop agents also improves task success from 40.92% to 44.17% (+3.25 pp) on the full 369-task OSWorld benchmark. Together, these results establish PACT as an effective credit-assignment framework for bridging execution gaps and improving task completion across mobile and desktop GUI environments.
[CV-94] Semantic-Guided Geometric Decoding for Token-Efficient Visual Grounding
链接: https://arxiv.org/abs/2611.22577
作者: Zhaojie Chen,Junyan Yang,Le Yang,Qi Dang,Yucheng Huang,Qianyi Che,Zhaoqiang Chen,Xinqi Zhu,Qiang Zhao
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 10 pages, 4 figures, 5 tables
Abstract:Multimodal large language models (MLLMs) provide a unified interface for language-conditioned visual perception. In pixel-to-sequence grounding, autoregressive coordinate generation requires multiple sequential decoding steps per box, increasing inference cost with the number of instances. Pixel-to-embedding methods reduce this overhead but lack explicit spatial encoding when using a shared special token. We introduce SemGeo, a token-efficient visual grounding framework that represents each instance with a single grid token for coarse localization. A geometric decoder reuses visual features and hidden states to recover precise discrete box coordinates in parallel. To train proposal generation and box localization together, we introduce Gaussian soft labels that encode coordinate distances during supervised fine-tuning (SFT). Reinforcement post-training further combines box-set rewards with coordinate supervision on model-generated proposals. This stage trains the language model and geometric decoder on their own predictions to mitigate the mismatch between teacher-forced training and inference. Across eight benchmarks spanning object detection, referring object detection, and document layout grounding, SemGeo achieves the highest F1@mIou among compared MLLMs on six benchmarks. It also delivers 2.07\times and 1.25\times the mean box-generation throughput of representative grounding baselines under vLLM and Transformers, respectively, while achieving higher accuracy in both comparisons.
[CV-95] MillBill: A Machining Agent with Geometric Feedback
链接: https://arxiv.org/abs/2611.22565
作者: Gennadii Savrasov,Artem Slavin,Soslan Kabisov,Andrey Kuznetsov,Anton Konushin,Dmitrii Zhemchuzhnikov
类目: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注: 18 pages, 8 figures. Code: this https URL
Abstract:Turning a CAD model into a machining program requires both process knowledge and geometric precision. Language models can generate such workflows by writing geometry and toolpath code for each part, but this makes numerical execution part of the model’s task. We introduce MillBill, an agent that separates machining decisions from their geometric execution and verification. Given a boundary-representation part, stock, and material, a locally hosted, mid-sized open model proposes operations and cutting parameters for indexed three-axis machining. A geometric harness generates toolpaths, simulates material removal, and returns feedback for plan refinement. Setup planning accounts for what the available cutters can physically reach, rather than visibility alone. We evaluate the resulting programs by the fraction of the target surface area within the specified tolerance after simulated machining. With Qwen3.8-27B, MillBill outperforms frontier-model approaches that construct their own geometry and toolpath workflows on the evaluated machining tasks. Code: this https URL
[CV-96] Lost in Read-out: Locating the Image Quality Gap in Vision-Language Models with a Closed-Form Edit
链接: https://arxiv.org/abs/2611.22564
作者: Zhuangzi Li,Shilv Cai,Jian Jin,Zhiyi Pan,Haihong Xiao,Weisi Lin
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Vision-language models (VLMs) are widely used as zero-shot image quality assessors, yet their good-vs-poor answers agree only moderately with human opinion. We ask whether these models fail to see quality or fail to say it. Layer-wise probes on five VLMs point to the second: quality is linearly decodable from the vision encoder and still from the last language layer, far better than the answer reveals, and about a hundred labels recover most of it. The loss lies mainly in the final read-out, which partly reads something else: relative to human opinion, it over-rewards colorfulness and under-rewards sharpness on all five models. We test this by intervention: if the read-out is at fault, a closed-form edit of a single last-layer neuron, fitted to many prompts at once, should close the gap and no more. Both hold. Under prompts never used for fitting, the edit recovers 86–100% of what a ridge head on the last-layer state, fitted under the same prompts, finds beyond the answer on KonIQ. Where that gap is small, it gains little, too little for our transfer criterion on the two largest models; where the head itself is below the answer, it lowers the score; and it works only when written from the middle of the language model on. Unlike unconstrained fine-tuning, the edit leaves perception question answering intact. It does not carry over to yes-or-no or rating questions, and it reverses the good–poor contrast after sentiment-laden sentences. Code will be made available.
[CV-97] CADMAGEDDON: High-Fidelity CAD Reconstruction with Local Inference
链接: https://arxiv.org/abs/2611.22563
作者: Oleg Pavlov,Nikita Gavrilov,Soslan Kabisov,Gennadii Savrasov,Andrey Kuznetsov,Anton Konushin,Dmitrii Zhemchuzhnikov
类目: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注: 13 pages, 4 figures, 4 tables. Code: this https URL
Abstract:CAD practitioners often work with proprietary parts that they are unwilling to share with external services. For individuals and small companies, bringing this work in-house also requires computational resources they can afford. We introduce CADMAGEDDON, an open-source agent that reconstructs editable CAD from meshes and can run locally on single-GPU systems and current-generation laptops. The method decomposes this complex task into a chain of simple visual decisions about what to construct and what to correct. Geometric tools support precise construction, allowing the agent to focus on visual reasoning. Using the mid-sized Qwen3.8-27B model, CADMAGEDDON outperforms prior approaches, including systems based on frontier models, in reconstruction quality on all the main benchmarks. CADMAGEDDON integrates with FreeCAD, bringing its reconstructions into practitioners’ existing CAD workflows. Code: this https URL
[CV-98] Plausible Videos Can Lead to Poor Decisions: Diagnosing and CURE-ing World Models for Planning
链接: https://arxiv.org/abs/2611.22561
作者: Raphael C. Kim,Michele Santacatterina,Jingsen Zhu,Ramin Zabih
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Action-conditioned world models are increasingly used as world models for planning, yet a world model that makes plausible, realistic-looking videos that are consistent with the training data can produce poor decisions. We address two key questions: (1) when can we trust a world model for planning, and (2) what should we do when we cannot? We introduce CURE (Certify, Upweight, Rollout, Explore), a certificate-guided technique that assesses whether a world model can be trusted for planning and determines what to do next: whether to upweight the training data to debias the model; roll out more samples at test time; or explore by collecting new data where support (how well the training data cover the planner’s decisions) is missing. We deploy CURE across multiple domains, spanning simulated physics based on PhyWorld, MuJoCo arm control, and Language-Table pushing in simulation and on real videos. The decisions our certificate trusts have lower regret than those it does not (e.g., 0.06 vs. 0.38 on Language-Table sim and 0.07 vs. 0.38 on MuJoCo contact), whereas video perceptual quality does not identify good decisions. By focusing on decision quality rather than perceptual quality, CURE lowers regret over the best pessimistic decision rule by 10-25% across support levels, and over world-action models, search and policy learning by 41-64%, given the same data.
[CV-99] Strong Fused Performance Can Mask Branch Collapse in a Dual-View Heart-Sound Classifier: A Three-Part Diagnostic Protocol
链接: https://arxiv.org/abs/2611.22556
作者: Safayet Khan,Hasan Zohirul Islam,Md. Tauhid Bin Iqbal
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Fusion models are usually judged by their fused score, a practice that assumes that every branch has learned something useful. We show that this assumption can fail silently in dual-view heart-sound classification. In AudioFuse, which fuses a spectrogram transformer with a waveform CNN, the spectrogram branch is unused despite strong fused performance when both branches are trained from scratch. We introduce a three-part diagnostic for branch collapse: whether deleting a branch changes the fused score, whether a linear probe on its features beats an untrained copy of the same branch, and whether the branch can learn its view when trained alone. Deleting the branch has negligible effect, and probing its features shows little improvement over an identically initialised, untrained copy. Yet the same architecture trained alone exceeds 0.9 AUC, so its view is learnable, while the jointly trained model is near chance when it must rely on this branch. Update statistics and an initialisation experiment point to an early optimisation imbalance. Moving a strong artificial label cue from one input to the other can switch which branch the model relies on, even though the branch architectures stay the same. We observe branch collapse on PhysioNet, but the full diagnostic is not met on ICBHI or CirCor. Fused performance alone therefore cannot establish that every branch learned a meaningful representation, and we recommend reporting branch deletion and an untrained-control probe alongside it.
[CV-100] Fast Training-Free Low-Light Video Enhancement with a Frozen Image Diffusion Prior
链接: https://arxiv.org/abs/2611.22551
作者: Hon Ming Yam,Chun Pong Lau
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 15 pages, 12 tables, 4 figures
Abstract:Pretrained image diffusion models can enhance low-light video without low-light training data and without retraining for a new camera. ZVRD, the only published training-free method for low-light video, is too slow to use: our implementation needs over a minute per 256 \times 256 frame. We present a causal, training-free pipeline that adds two temporal components to a frozen MR. Illuminate image prior. Temporal input fusion recursively averages motion-compensated low-light frames before diffusion inversion. Late-start offset propagation runs the full sampling trajectory only on key frames. Each non-key frame inverts the low-noise part of its own trajectory and adds an enhancement offset warped from the latest key frame. At ZVRD’s own resolution the pipeline is 261\times faster than ZVRD and scores better on all five metrics we report. On 13 held-out SDSD clips at 1280 \times 720 , it improves on framewise eight-step processing by 0.55 dB in PSNR and 0.034 in LPIPS and halves the warping error, in 10% less time. Measurements inside the model show why, and a linear-Gaussian model of inversion explains them. Fusion removes frame-to-frame noise before inversion amplifies it. A flow error perturbs the transported offset ten times less than it perturbs a transported state. The perceptual gain does not transfer to low resolution: at 346 \times 260 , LPIPS is worse by 0.028. At 0.65 frames per second, the pipeline is an offline method.
[CV-101] JointGV: Structure-Aware Correction of Visual World Models with Parallel Graph Dynamics for Long-Horizon Modeling
链接: https://arxiv.org/abs/2611.22543
作者: Tian Lyu,Yichen Jiang,Ziyuan Cheng,Yunda Yan,Xiangyu Pan,Junjie Jiang,Jialin Yu,Boyu Ning,Pietro Liò,Fan Mo
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Visual world models predict future visual states conditioned on actions, but recursive errors in entity poses and spatial relations limit their long-horizon prediction for robot planning and control. We ask whether a separately predicted relational state can correct this drift without restructuring the visual model. This is challenging because graph nodes and visual states share neither elements nor a feature space. We propose Joint Graph–Visual Dynamics Correction (JointGV), a plug-in based on a cross-granularity dynamics hypothesis: despite their different representations, graph-node and visual-state changes describe the same physical transition. JointGV rolls out a task-conditioned scene graph alongside the visual world model, couples their predicted changes, and combines the resulting graph-motion context with graph-predicted node positions to compute a bounded residual. Graph-learning and correction gradients do not propagate into the original visual dynamics. We attach JointGV to DINO-WM (spatial tokens) and DreamerV3 (a global latent state), evaluating both on DMC Cheetah Run and Walker Walk across three seeds and extending the evaluation to contact-rich manipulation with DMC Finger Turn Hard. Across 32-step rollouts, JointGV improves structural and visual-prediction accuracy, reducing joint-position error by 11–28%. For spatial tokens, learned graph–visual change coupling gives lower error than broadcasting the same mean graph-motion context to every token, particularly when non-adjacent parts appear close together in the image. On real-robot inverted-pendulum sequences, JointGV reduces DINO-WM’s 32-step pole-angle error from 12.56^\circ to 6.69^\circ and improves DreamerV3 and LeWorldModel as well. In simulated inverted-pendulum control, an actor trained with JointGV-corrected rollouts completes 46 of 50 episodes, compared with none using the original rollouts.
[CV-102] ATS: Anti-Aliased Triangle Splatting with Exact Per-Primitive Integrals
链接: https://arxiv.org/abs/2611.22512
作者: Brent Zoomers,Joni Vanherck,Leander Winters,Jeroen Put,Lode Jorissen,Nick Michiels
类目: Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR)
备注: 17 pages, 7 figures, 9 tables
Abstract:Triangle Splatting represents scenes using semi-transparent triangles, but it evaluates each triangle at the pixel center and culls any triangle with an inradius smaller than one pixel. The former causes severe aliasing, while the latter culls most of the scene at lower resolutions. We show that the integral of the Triangle Splatting window over a pixel admits an exact closed-form solution. The angle bisectors split the triangle into three sub-triangles, in each of which the window is a power of the distance to one edge. Each piece therefore has a closed-form antiderivative, and Green’s theorem reduces the integral and its gradients to one term per side of the sub-triangle clipped against the pixel. The integral over the whole triangle, which we show has a closed form, then allows conservative culling, removing only triangles that cannot meet the renderer’s alpha cutoff at any pixel. This exact formulation anti-aliases existing models without retraining, exhibits less temporal aliasing than 2x2 supersampling, and enables multi-scale training, matching or outperforming supersampled training at reduced resolutions.
[CV-103] Reliability-Driven Feature Correction for Hyperspectral and LiDAR Classification
链接: https://arxiv.org/abs/2611.22508
作者: Lianming Zhang,Xiaolu Wang,Qiya Song,Jinyang Liu,Renwei Dian,Xudong Kang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Hyperspectral image (HSI) and LiDAR data provide complementary spectral and elevation information for land-cover classification. However, the spatially varying class-discriminative reliability of modality features is rarely modeled explicitly, allowing weak or ambiguous features to adversely affect cross-modal interaction and fusion. To address this issue, we propose ReDF, a reliability-driven framework that quantifies the reliability of each modality’s features at each spatial location through the strength of class-related evidence they provide, and uses it to guide feature correction and fusion. Specifically, a Reliability-Guided Feature Correction module introduces a class prototype bank for each modality and converts feature-prototype similarities into class evidence for reliability estimation. The estimated reliabilities are used to construct a reliability-weighted shared representation and to guide bidirectional cross-modal feature correction based on neighborhood agreement, refining modality features with weak class evidence through information from the other modality. Building on these outputs, a Reliability-Guided Feature Reweighting module conditions a low-rank modulation on the estimated reliabilities and the cross-modal disagreement, and applies it to the corrected modality-specific descriptors to adaptively rescale each modality’s contribution to the fused representation used for final classification. Experiments on three public HSI-LiDAR datasets show higher mean overall accuracy and Kappa than the compared methods.
[CV-104] EfficientVPR: Toward Efficient Visual Place Recognition via Scene-Aware Prompt Tuning and Adaptive Feature Enhancement CVPR2026
链接: https://arxiv.org/abs/2611.22506
作者: Wenjing Tang,Chuanguang Yang,Zhulin An,Libo Huang,Yongjun Xu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 18 pages, 11 figures, accepted at CVPR 2026
Abstract:Visual place recognition (VPR) faces critical challenges in handling extreme environmental variations while meeting the computational constraints of practical applications. Current methods predominantly address these challenges by either scaling up model capacity or employing computationally intensive reranking stages, creating a significant efficiency bottleneck. To overcome this limitation, we propose EfficientVPR, a lightweight one-stage framework that achieves speed-accuracy trade-offs through two key innovations: i) a scene-aware visual prompt tuning method which adapts pretrained features with less parameters while dynamically adjusting to sample-specific characteristics, and ii) an instance-dependent key local feature enhancement module that further reinforces discriminative regions. Comprehensive evaluations on Pitts250k, MSLS, Eynsham, AmsterTime and SVOX demonstrate that our method establishes a new SOTA for DINOv2-small models by outperforming all same-scale competitors, and delivers a 73\times speedup with 60% lower-dimensional features while maintaining competitive (within 2.5% average R@1 gap) against the SOTA DINOv2-large-based two-stage method. Our code will be available at this https URL.
[CV-105] Efficient Self-Supervised Adaptation of 3D Abdominal Vision-Language Model for Institution-Specific HCC Classification via Full Fine-Tuning and PEFT
链接: https://arxiv.org/abs/2611.22504
作者: Febryan Putra Kartika,Cheng-Yu Ma,Ying-Jia Lin,Chi-Tung Cheng,Kuan-Fu Chen
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 25 pages, 12 figures, 7 tables, including appendices. Corrected author revision of the MIDL 2026 paper. Revised preprocessing yields lower HCC classification performance; the main findings on temperature initialization and better VerSe performance retention with PEFT remain supported
Abstract:Medical vision-language models (VLMs) capture cross-modal relationships between image and text, yet their adaptation to institution-specific clinical tasks remains underexplored. In this study, we fine-tuned a pretrained 3D medical VLM for hepatocellular carcinoma (HCC) classification using paired abdominal CT scans and radiology reports from a different institution with acquisition characteristics that differ from the model’s original pretraining corpus. We compared two adaptation strategies: full fine-tuning and parameter-efficient fine-tuning (PEFT), motivated by the common use of PEFT to reduce computational cost and enable adaptation under limited-data constraints. Both approaches achieve strong downstream HCC classification performance despite the cross-institutional domain shift, with PEFT reaching an AUC of 0.79 and F1 of 0.70, and full fine-tuning achieving an AUC of 0.82 and F1 of 0.71. Although these results remain below the performance reported by supervised HCC classification approaches, our method achieves improvements over the pretrained model through CT-report contrastive learning without requiring lesion-level annotations or HCC classification labels as training targets. Full fine-tuning converges rapidly but overfits within a few epochs, whereas PEFT (ConvLoRA for the image encoder and LoRA for the text encoder) attains comparable performance while updating approximately 1% of the model parameters, although requiring more training steps. We also examine the role of contrastive temperature, observing that temperature initialization significantly affects classification performance. This study demonstrates that a 3D medical VLM can be efficiently adapted to institution-specific HCC classification using self-supervised CT-report contrastive learning, while highlighting the practical trade-offs between full fine-tuning and parameter-efficient fine-tuning.
[CV-106] Motion-Aware Reward Modeling for Instruction-Guided Video Editing
链接: https://arxiv.org/abs/2611.22495
作者: Li Li,Xiangbo Gao,Bo Ni,Samyadeep Basu,Viet Dac Lai,Jing Shi,Seunghyun Yoon,Gang Wu,Yu Wang,Thien Huu Nguyen,Tyler Derr,Franck Dernoncourt,Zhengzhong Tu,Yue Zhao
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Reward models for instruction-guided video editing can favor temporal consistency at the expense of motion. We study this failure, termed \emphmotion collapse, using controlled motion-suppression variants of real edits. On VE-Bench, the evaluated composite reward prefers a motion-suppressed candidate in 97% of cases. Learned video rewards and a VLM judge also struggle with mild suppression, while latent optimization and reward-guided fine-tuning exploit this weakness to increase reward as motion declines. A linear probe of our base reward backbone performs near chance on mild suppression, motivating an explicit inter-frame dynamics input. We propose a motion-aware reward model that combines an RGB frame-difference stream with preference training against motion-suppressed and fabricated-motion negatives. A distillation anchor preserves alignment with human quality judgments. On the VE-Bench motion ladder, our model reduces the rate of selecting motion-suppressed candidates to 0% while achieving a Spearman correlation of 0.57 with human quality ratings. It also detects unseen suppression variants and preserves motion during reward-guided fine-tuning on two editing pipelines. These results support explicitly modeling motion in rewards for video editing.
[CV-107] Fast and Strong Image Protection against Diffusion Personalization
链接: https://arxiv.org/abs/2611.22494
作者: Jiwon Park,Jihye Kang,Namhyuk Ahn
类目: Computer Vision and Pattern Recognition (cs.CV); Cryptography and Security (cs.CR)
备注:
Abstract:Personalized text-to-image diffusion models can reproduce a person from a handful of photographs. Image protection perturbs each image before publication so that fine-tuning on it fails, but it optimizes every perturbation from scratch, so its cost grows with the number of images. We present the protective diffusion model (PDM), which is trained once and samples the perturbation of a new image over many denoising steps. PDM distills an iterative teacher and is then post-trained with the target residual projection (TRP) loss as a reward, a stronger objective than existing ones. PDM outperforms nine baselines on faces and artworks and nearly matches its teacher at a tenth of the cost.
[CV-108] LeViT-UNet: Towards Balanced Fine-Grained Details and Context for Efficient Medical Image Segmentation
链接: https://arxiv.org/abs/2611.22482
作者: Guoping Xu,Jinling Zhu,Liuyue Li
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Multimedia (cs.MM)
备注: 24 pages, 5 figure
Abstract:Efficient medical image segmentation is essential for clinical applications requiring fast and accurate delineation, such as surgical navigation, radiation therapy, and real-time image-guided intervention. Despite substantial advances in convolutional neural networks and Transformers, attaining high accuracy under limited computational resources remains difficult, because reducing model complexity often sacrifices fine-grained structural details and global context modeling, yielding an unfavorable accuracy-efficiency trade-off. To address this, we propose LeViT-UNet++, an efficient encoder-decoder architecture that better balances local detail preservation and global context modeling. Specifically, we introduce a Haar Wavelet Multi-Scale Downsampling (HWMD) module that decomposes feature representations into low- and high-frequency components to extract multi-scale structural information, thereby mitigating detail loss during downsampling. We further develop a depthwise convolution-based decoder that integrates multi-scale global representations from Transformer blocks with local spatial features, enabling progressive feature refinement while reducing computational overhead. Evaluated on four benchmarks (Synapse, ACDC, LA2018, and MSD Heart) under both 2D and 3D settings, LeViT-UNet++ achieves competitive accuracy with a compact parameter footprint and fast inference. On Synapse, for instance, it attains a DSC of 76.57% with only 8.025M parameters and 241.72 FPS, confirming a favorable accuracy-efficiency balance. These results demonstrate the effectiveness of combining wavelet-based multi-scale feature extraction with lightweight local-global feature fusion for efficient medical segmentation. The source code is available at this https URL.
[CV-109] Disagree to Prune: Reallocating Visual Tokens Where a Small VLM Dissents
链接: https://arxiv.org/abs/2611.22472
作者: Yanchao Li,Jiaqing Xie,Ben Gao,Yusu Zhang,Beilun Wang,Yanbo Wang,Xiaohua Xu,Zhongyu Huang,Tianfan Fu,Nanqing Dong
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 45 pages, 8 in the main text. Code and data: this https URL
Abstract:Visual token pruning speeds up large vision-language models (VLMs) by keeping only a part of their visual tokens. Most existing methods improve the ranking of the tokens and then answer every question once with the same budget. With strong rankings, pruning hurts only a few questions, and signals computed on the kept tokens see little of what it removed. We find that, on most of these questions, a small VLM given the full image disagrees with the pruned answer, since it still sees the missing evidence. We propose PRUDENT, which first answers with about half the budget and spends the saved tokens on re-answering the disagreeing questions at twice the budget. We show when such re-answering pays and why the small VLM should read the full image. On 18 benchmarks, PRUDENT outperforms eight existing pruning methods at the same average budget of visual tokens. Its lead widens as the budget tightens, and holds across five InternVL sizes. PRUDENT also has the highest average in two more model families. The disagreement of the small VLM flags most pruning failures of all eight methods, and re-answering on it also improves the two strongest with their own rankings. Other checkers give similar gains, and the extra tokens go where the first pass hurts the answer. These results suggest that identifying which pruned answers are trustworthy provides an additional lever for visual token pruning beyond the ranking of tokens. Code and data are available at this https URL.
[CV-110] MAVE: Reconstructing Editable 3D Blockouts from Film Shots
链接: https://arxiv.org/abs/2611.22464
作者: Yifan Chang,Jian Yang,Jiangnan Shao,Xiaofeng Wang,Zhaoxin Fan,Renjing Xu,Ming Lu,Xingang Wang,Zheng Zhu
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: this https URL
Abstract:Animated 3D blockouts support previsualization and provide structural conditions for video generation, but manually reconstructing them from reference videos is costly. This limits both their use by creators and the production of paired training data. We introduce MAVEBench, a benchmark of 50 film shots and 416 keyframes for evaluating camera motion, spatial layout, and character blocking. We propose MAVE, a video-to-blockout framework combining Metric observation, Agentic modeling, and evidence-based VErification. It first recovers cameras, human motion, and geometric evidence in a shared metric coordinate system. An agent then constructs an editable environment under constraints from these observations, while programmatic checks report inconsistencies to guide refinement. The outputs are editable Blender scenes and frame-aligned blockout videos. On MAVEBench, MAVE improves overall scores by 27.5-41.1 points across three agents over their corresponding vanilla baselines. With DeepSeek-1017, it achieves a score of 73.2 at an average API cost of only USD 0.054 per shot, outperforming the strongest vanilla baseline while reducing total reconstruction time by 53%. We also construct MAVEData-3K, a dataset of 3,000 blockout-video pairs for model training, and will release it publicly. Creators can change viewpoints and character paths, then render the edited scenes to condition video generation. Code is available at this https URL.
[CV-111] Sample-Aligned Decoders for Latent Generative Models
链接: https://arxiv.org/abs/2611.22458
作者: Yu Lei,Yi Li,Hongcheng Gao,Yipeng Wei,Liang Li,Li Su,Qingming Huang
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: Project page: this https URL
Abstract:Latent generative models synthesize images by decoding generated latents, yet their decoder is trained only to reconstruct encoder latents and never sees the latents it decodes at test time, so the residual errors of the generator are rendered directly into the final image. Existing remedies retrain the generator, add cost at every sampling step, or make the decoder robust to perturbed real latents, assuming that generator errors resemble such perturbations. We find that they do not. Generated latents lie on the encoder latent manifold as closely as real ones, yet follow a different distribution. The decoder should therefore be adapted to the distribution of generated latents, which raises two challenges: (1) unpaired supervision, since a generated latent has no ground-truth image, and (2) latent-space drift, since a decoder freed from paired targets forgets the latent space it must decode. We propose Sample-Aligned Decoders (SAD), a post-training framework that fine-tunes only the decoder of a frozen generator at no inference cost. Generated-Sample Distribution Alignment matches decoded generated latents to unpaired real images in a frozen foundation-model feature space, and Encoder Anchor Preservation keeps the decoder close to the original one on encoder latents. Our analysis shows that the alignment is effective only on generated latents, where it corrects each sample in proportion to how much more often the generator produces such latents than the encoder; on encoder latents it can only work against the anchor. SAD improves four state-of-the-art ImageNet generators across three tokenizers, e.g., FID-50K from 1.15 to 0.98 for REPA-E, complements discriminator guidance, and extends to text-to-image generation. Our code and models are available at this https URL.
[CV-112] MedImgAnnotator: batch wise medical image annotation and preprocessing tool
链接: https://arxiv.org/abs/2611.22450
作者: Ewunate Assaye Kassaw(1 and 2),Amit Mehndiratta(1 and 3 and 4 and 5),Peush Sahni(6),Anup Singh(1 and 3 and 4) ((1) Centre for Biomedical Engineering, Indian Institute of Technology Delhi, New Delhi, India, (2) Department of Biomedical Engineering, University of Gondar, Gondar, Ethiopia, (3) Yardi School of Artificial Intelligence, Indian Institute of Technology Delhi, New Delhi, India, (4) Department of Biomedical Engineering, All India Institute of Medical Science, New Delhi, India, (5) Faculty of Medicine and Health, University of New South Wales (UNSW), Sydney, Australia, (6) Max Centre for Liver and Biliary Sciences, Max Super-specialty Hospital, Saket, New Delhi, India)
类目: Computer Vision and Pattern Recognition (cs.CV); Image and Video Processing (eess.IV)
备注:
Abstract:The curation of large cohort datasets for developing deep learning models demands annotating or editing over the AI-predicted labels. However, openly available annotation tools are time-consuming, as they involve complex navigational steps and do not support batch processing. This study aimed to develop a simple, fast, semiautomated batch-annotation tool named MedImgAnnotator. The tool was created by integrating editing tools, image mask pair matching logic, preprocessing functions, object detection tool, and a plugin management system. Its performance was evaluated in terms of active annotation time and ease of use by the developer and 12 users using open-source MR and CT scans. Additionally, the tool had been used by the MICCAI 2026 CARE Challenge liver segmentation best performance winner to annotate liver masks from 6412 MRI volumes. The evaluation results showed that MedImgAnnotator reduced active annotation time from 206.5+/-32.87 (average active annotation time of openly available tools) seconds to 59 seconds. The Wilcoxon signed-rank test of 12 users evaluation results recorded P values of 0.032 and 0.008 for the active annotation time and higher ease of use rating metrics, respectively, indicating that the current tool significantly reduces annotation time and enhances ease of use. Moreover, the tool utilization for annotating different regions of interest from different modality scans and body regions shows its usability across modalities and areas of interest. MedImgAnnotator, which significantly reduces active annotation time for large imaging cohorts, is available as an executable file and as a Python script. Access link: this https URL
[CV-113] As Student Rollouts Continue Visual Backdoors Keep Growing
链接: https://arxiv.org/abs/2611.22446
作者: Zane Fu,Yanzhao Zheng,Jinze Liu,Ziting Qin,Qianshan Wei,Jihuai Zhu,Yipeng Yu,Baohua Dong,Hangcheng Zhu,Ruohui Huang
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Conventional knowledge distillation can facilitate the propagation of backdoors from teacher models to student models. However, how this risk manifests in On-Policy Distillation (OPD) remains insufficiently studied. OPD has recently emerged as an important post-training paradigm for Vision-Language Models (VLMs). Unlike traditional offline distillation, OPD does not require teacher-generated responses. Instead, it uses the teacher’s probability distributions to supervise response trajectories generated online by the student. Since the initial student model typically generates benign responses and receives no explicit supervision from malicious target responses, OPD appears to offer some inherent resistance to backdoor propagation. In this paper, we challenge this intuition and reveal a more subtle risk of visual backdoor propagation in OPD. Through experiments across multiple model families and datasets, together with theoretical analysis, we identify three main findings. First, unlike the relatively gradual propagation observed in traditional offline distillation, backdoors rarely emerge in OPD at low poisoning ratios. However, the attack success rate rapidly approaches 100% once a critical threshold is exceeded. Second, backdoor propagation does not depend on explicit malicious responses. Instead, it is driven by the continuous accumulation of anomalous probability mass assigned by the teacher. Finally, based on these findings, we propose a novel trigger design that enables visual backdoor propagation with zero explicit trigger injection. These findings further reveal an overlooked mechanism of backdoor propagation in OPD and highlight its potential security risks. Subjects: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI) Cite as: arXiv:2611.22446 [cs.CV] (or arXiv:2611.22446v1 [cs.CV] for this version) https://doi.org/10.48550/arXiv.2611.22446 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[CV-114] Where Demand Meets Supply: Spatial Elastic Consensus for Low-Bandwidth Collaborative Perception
链接: https://arxiv.org/abs/2611.22439
作者: Linshan Yan
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 10 pages
Abstract:Occlusions and limited sensor range make ego-only 3D perception insufficient for safe driving. Collaborative perception addresses this by enabling connected vehicles to share complementary observations, yet it remains constrained by tight vehicular communication budgets. While spatial selection is more flexible than fixed-rate feature compression, existing selectors remain largely sender-driven and waste scarce bandwidth on ego-visible targets. Furthermore, applying convolutions after masking spreads isolated tokens into their neighborhoods, silently violating low-bandwidth constraints. To address this, we propose Spatial Elastic Consensus (SEC), a bandwidth-adaptive framework centered on bilateral demand-supply consensus. Instead of unilateral token selection, the ego vehicle formulates a content-aware demand prior while each collaborator evaluates its local supply. This single-round consensus allocates the token budget to critical ego-missed evidence rather than ego-visible redundancy, without transmitting dense maps. To guarantee exact budget compliance, an anti-dilation refine-before-mask pipeline refines features prior to discrete token selection, preventing receptive-field leakage. A lightweight deformable sampler then aggregates these sparse messages with learned local offsets, compensating for the spatial context lost to token selection and mitigating inter-agent pose drift. Evaluated at a stringent 64KiB budget, SEC consistently achieves the best AP@0.7 on the simulated OPV2V and V2XSet datasets and the real-world DAIR-V2X dataset. These results demonstrate that combining bilateral consensus with budget-exact transmission is key to efficient low-bandwidth collaborative perception.
[CV-115] When Neural Collapse Fails to Route: Routing Drift in Continual Learning BMVC2026
链接: https://arxiv.org/abs/2611.22429
作者: Nicolas Larue
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at BMVC 2026. 36 pages, including supplementary material
Abstract:Progressive neural-collapse classifiers can learn strong task-local decision boundaries while failing the global Class-IL decision. We identify the failure as routing drift: under an expanding equiangular tight frame (ETF) classifier, a held-out sample can remain separable within its true task block, yet receive its largest global score from another block. This produces high Task-IL accuracy but low Class-IL accuracy, and small replay buffers can mask the problem because stored samples are optimized repeatedly and often remain well routed. We formalize route accuracy directly from the global class argmax and use it to separate within-task discrimination from cross-task score comparability. This decomposition suggests a classifier-level correction: preserve the global ETF frame, but add class-local reliability evidence when transported class measures deviate from the neural-collapse ideal. We develop Reliability-Calibrated Routing (RCR), a lightweight post-hoc classifier that composes standardized ETF alignment with radial plausibility around transported, NC-shrunk class representatives. Across three datasets and two fixed replay-buffer budgets, this NC-consistent correction gives the best Class-IL FAA on five of six settings and reduces Class-IL forgetting on all six. The largest gains occur in harder routing regimes: up to +6.75 FAA over the reported reference, and up to +8.03 FAA and +8.62 final route accuracy in paired evaluations, while preserving Task-IL performance. These results position routing drift as a measurable failure mode of ETF-based continual learning and show that reliability-calibrated class evidence can repair part of the lost global routing without retraining the representation. Code and evaluation artifacts are available at this https URL.
[CV-116] Do Diffusion Models Learn to Generalize Basic Visual Skills? NEURIPS2026
链接: https://arxiv.org/abs/2611.22427
作者: Amish Sethi,Boya Zeng,Wenhao Chai,Zhuang Liu
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: NeurIPS 2026. Project page: this https URL
Abstract:Diffusion models generate high-quality images, but how well they generalize is hard to measure. Current evaluations rely on human preference and prompt-following benchmarks for text-to-image models and on distribution-level metrics such as Fréchet Inception Distance (FID) and Inception Score (IS) for class-conditional models, but these cannot tell whether models learn basic visual skills or largely match training patterns. To address this gap, we introduce the Visual Generative Lab (VGL), a controlled experimental framework for studying how diffusion models learn and generalize basic visual skills, including size, position, rotation, count, color, shape, and their compositions. By generating synthetic training data for each skill from explicit rules, VGL allows precise measurement of generalization with task-specific, rule-based metrics. Training models from scratch on these data reveals three patterns in this controlled setting: (1) with our default design, extrapolation is limited: size and position degrade within a few pixels of the training range, count reaches only about 50% accuracy on out-of-range counts, and rotation extrapolation accuracy varies widely across seeds; (2) for compositional generalization, the number of distinct basic visual skill combinations matters more than dataset size; (3) skills mostly fail independently, except position. Code and data-generation scripts: this https URL.
[CV-117] CutAgent : Creative-Gene-Driven Planning and Experience-Guided Self-Refinement for High-CTR E-commerce Advertising Video Creation
链接: https://arxiv.org/abs/2611.22418
作者: Xuyue Huang,Guanbo Huang,Jiasong Feng,Run Ling,Ao Ma,Xi Zhao,Yaoyu Li,Jie Gao,Ke Cao,Wei Feng,Zhanjie Zhang,Hao Wang,Jinghe Hu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:E-commerce advertising video creation requires both high-level creative planning and reliable multi-modal production. However, existing methods often map product attributes directly to scripts or editing decisions without explicitly modeling advertising strategies, narrative structures, and production styles. They also have limited ability to reuse past production failures. To address these limitations, we propose \textbfCutAgent , a dual-agent framework that formulates advertising video creation as a Plan–Execute–Self-Evolve process. CutAgent comprises two collaborative agents, \textbfGeneAgent and \textbfVideoAgent , connected by a structured creative gene that bridges product understanding and video production. GeneAgent maps multi-modal product inputs—textual attributes together with product images—to creative genes that specify narrative units and production rules. To train GeneAgent, we construct \textbfCutAgent-70K with approximately 70K samples of product attributes/images, reverse-extracted creative genes, cleaned clip pools, and real-world CTR, and apply high-CTR supervised fine-tuning followed by CTR-guided GRPO. Guided by the generated gene, VideoAgent retrieves category-then-gene failure–correction experiences from a long-term memory as constraints for shot-script planning, then coordinates material retrieval, narration, subtitles, music, and timeline rendering. Independently, a reflection loop diagnoses quality defects, revises deficient production decisions, and writes verified corrections back to memory for later retrieval. Experiments on \textbfCutAgent-Eval show that CutAgent outperforms representative baselines in e-commerce advertising video creation, achieving the highest online CTR of \textbf2.21% .
[CV-118] SpatialMind: Adaptive Multi-Level Geometric Fusion for Spatial Reasoning in Multimodal Large Language Models
链接: https://arxiv.org/abs/2611.22415
作者: Shibin Zhao,Hao Tan,Fan Yang,Haixin Wang,Feilong Chen,Yidong Ma,Chaoyang Zhao,Ming Tang,Jinqiao Wang
类目: Computer Vision and Pattern Recognition (cs.CV); Computational Geometry (cs.CG)
备注: Includes supplementary material. Code available at this https URL
Abstract:Multimodal Large Language Models (MLLMs) demonstrate promising capabilities in 2D visual understanding, yet remain limited in 3D spatial reasoning. Recent studies explore integrating 3D reconstruction models into MLLMs. However, these approaches either use a single geometric granularity across all tasks or adopt a fixed multi-level fusion scheme, regardless of task-specific reasoning requirements. Through systematic experiments on multi-level geometric fusion, we find that spatial tasks at different cognitive levels (e.g., distance estimation vs. route planning) require geometric representations of different granularities, revealing that fixed fusion cannot accommodate diverse spatial reasoning requirements. To bridge this gap, we propose SpatialMind, a task-adaptive framework for dynamically integrating multi-level geometric representations. SpatialMind employs GeoFus, a multi-layer geometric fusion module built around a gating network jointly conditioned on textual instructions, visual semantics, and geometric representations. This network adaptively modulates the contributions of shallow, intermediate, and deep geometric features across the layers of the LLM decoder. Furthermore, we introduce a negative-entropy regularization loss to mitigate gate collapse. Gate visualizations indicate that GeoFus adapts geometric granularity to task requirements. Experiments demonstrate that SpatialMind achieves competitive performance across spatial benchmarks, attaining a state-of-the-art score of 72.3 on VSI-Bench. The source code is publicly available at this https URL.
[CV-119] CoBoVLA: Decoupling Language from Action with Discrete Codes for Robust and Interpretable Robot Manipulation
链接: https://arxiv.org/abs/2611.22395
作者: Mehul Arora,Dhruv Agarwal,Ambuj Agrawal
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注: 16 pages, 3 figures, 4 tables, and supplemental material included
Abstract:Vision-Language-Action (VLA) models achieve high success rates on the LIBERO benchmark but remain sensitive to paraphrasing. Existing VLAs ranging from 0.6B to 7.5B parameters experience a 22-52 point performance drop on LIBERO-Para while SmolVLA exhibits an even larger decline, from 83% to 8%. Because most VLAs jointly learn language understanding and action generation, it is difficult to determine whether a failure originates from misunderstanding an instruction or executing it incorrectly. To address these challenges, we introduce CoBoVLA, or CodeBookVLA, a 250M-parameter VLA that decouples language understanding, planning, and control through learned discrete interpretable codes. A text encoder maps the instructions to shared step codes, which a visual planner then converts into action code sequences executed by a language-independent actor. This design improves robustness to instruction paraphrasing while allowing failures to be traced to individual stages. It also enables efficient real-time control on resource-constrained hardware by encoding instructions only once and using compact discrete representations to keep each component lightweight. CoBoVLA achieves 91.4% success on LIBERO-Goal, compared with 82.8% for SmolVLA, under a matched protocol, with 17 ms per control step latency. On LIBERO-Para, it achieves 81.1% across 4,092 unseen paraphrases and 78.5% on matched episodes, compared with 36.5% for SmolVLA LoRA fine-tuned on paraphrases. Without two-step demonstrations, CoBoVLA completes 48.8% of 160 free-form two-step tasks. Finally, success reaches 91% when instructions are correctly encoded but falls to 27.2% otherwise, identifying language understanding as the main source of remaining failures. Together, these results demonstrate that decoupling language understanding and control can improve instruction robustness while making VLA failures easier to diagnose.
[CV-120] Leverag ing Segmentation Representations for Efficient Classification: A Self-Configuring Framework Based on nnU-Net
链接: https://arxiv.org/abs/2611.22380
作者: Renjun Cai,Antonio Javier Gallego,Félix Fuentes-Hurtado
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:We propose a self-configuring framework that extends the nnU-Net workflow to both segmentation and classification in 3D medical imaging. Building on nnU-Net’s self-configuring method, the framework reuses representations learned by a segmentation model for image-level classification. Its encoder features and predicted segmentation probabilities are used by a lightweight classifier. Two information-sharing strategies are investigated. In the frozen-backbone branch, features are augmented with region-of-interest information, predicted foreground volume, and segmentation confidence. In the adaptation branch, the final encoder stage is partially fine-tuned for classification, with an auxiliary segmentation loss to preserve segmentation-related information. Additional five-fold experiments showed that more complex configurations did not consistently improve classification performance. On the public validation leaderboard, our selected framework achieved an Overall score of 0.8056. In the official FLARE 2026 Task 2 final evaluation, our framework ranked first overall, achieving the highest DSC, NSD, and AUROC across all four testing tasks, including two hidden testing tasks. Local inference measurements also remained below the challenge limits of 60 \mathrms per case and 4 \mathrmGB GPU memory. These results demonstrate the practical value of reusing segmentation representations for classification while maintaining a simple and efficient framework.
[CV-121] Seeing What Still Holds: How Waiting Control Changes Cooperative Recovery
链接: https://arxiv.org/abs/2611.22368
作者: Kangjie Huang,Yucheng Xin,Xinchun Wang,Dong Liu,Yan Fan,Shuming Liu
类目: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注: 8 pages, 3figures
Abstract:Changing the holding reference during a fixed-duration wait can reverse the ranking of visual recovery methods. We introduce AuditEmbodied, a paired execution and trajectory evaluation framework that independently varies holding references and waiting duration while keeping recovery executors frozen. A 960-execution factorial on 20 initialization seeds compares CAD-feedback and landmark-based recovery from identical post-wait physical states within each holding and waiting condition. Protected completion requires successful recovery under the declared execution constraints. At a 1 s wait with registered wrist references, their protected completion counts change from 6/40 versus 19/40 under registered holding to 34/40 versus 22/40 under position-only holding. Trajectory analysis localizes the main separation to grasp formation: small observation-time target residuals precede large tracking errors. Joint diagnostics and controlled interventions characterize persistent tracking constraints. On fresh seeds, gradual position reanchoring improves CAD recovery’s long-wait completion by 42.5 percentage points and reduces mean complete recovery time by 6.06 s relative to registered holding. Additional evaluations with pretrained policies, handover, and a mass-matched geometry characterize holding-reference sensitivity across executors and physical regimes. These findings establish holding reference as an explicit evaluation condition for comparing visual recovery performance.
[CV-122] Can Vision-Language Models Truly Understand Images? A Benchmark Beyond Commonsense Bias
链接: https://arxiv.org/abs/2611.22360
作者: Ling Lin,Heng Su,Xiangzeng Liu,Yang Bai,Congcong Zhu,Jingrun Chen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 10 pages
Abstract:Multimodal large language models (MLLMs) have achieved impressive performance on a wide range of vision-language tasks. However, it remains unclear whether their predictions are grounded in the visual content itself or are largely driven by statistical co-occurrence patterns and commonsense correlations learned from large-scale image-text corpora. Existing vision-language benchmarks predominantly consist of samples that conform to human commonsense, allowing models to achieve strong performance by exploiting dataset priors rather than genuinely understanding image content. To address this limitation, we present ContraSense, a systematically constructed benchmark for evaluating faithful vision-language understanding. ContraSense consists of carefully curated counter-intuitive image-text pairs that deliberately violate common expectations, substantially reducing the effectiveness of statistical biases and requiring models to rely on visual evidence for reasoning. We conduct comprehensive evaluations on a diverse set of representative open- and closed-source MLLMs. The results reveal a substantial performance gap between current MLLMs and humans: even the strongest closed-source model achieves only about 60% accuracy, compared with 96.04% for human participants. These findings demonstrate that current MLLMs still struggle with faithful visual understanding and highlight the need for more robust multimodal reasoning grounded in image content. The ContraSense benchmark is publicly available at this https URL.
[CV-123] SandBAR: A Band-Asymmetric Architecture for Underwater Image Restoration with Removable Frequency Scaffolding WACV2027
链接: https://arxiv.org/abs/2611.22355
作者: Yanzi Liu,Haoran Hou,Jiachi Tong,Shen Chen,Shuang Chen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to WACV 2027. Camera-ready version. 9 pages, 6 figures
Abstract:Light scattering, absorption and turbidity degrade underwater imagery through complementary frequency tendencies: large-scale, channel-dependent appearance distortions and local high-frequency detail loss. This band asymmetry motivates different forms of restoration for global colour and local structure. Our primary contribution is SandBAR, a novel unified band-asymmetric architecture for underwater image restoration. Rather than treating frequency processing as an isolated module, SandBAR organises the restoration pipeline around band asymmetry at the block, encoder and objective levels. Within this architecture, we introduce a novel band-asymmetric sandwich block with a reserved post-attention repair slot and the Split-band Parallel Asymmetric Repair Kernel (SPARK), a novel module that applies channel-aware large-support correction to the low band, which predominantly carries large-scale appearance distortions, and lightweight local refinement to the detail-bearing high band, thereby restoring global colour while preserving fine structure. To deliver the global cast to the depth at which the network represents global structure, a learnable low-frequency guidance path injects a low-pass view of the input into the deep encoder, and a three-scale spatial-frequency objective supervises the same axis throughout the decoder. Building on the reserved post-attention slot, we further introduce Post-attention Scaffold Training (PaST), a novel removable nonlinear scaffolding scheme that trains the permanent SandBAR graph under band-specific capacity and runs it at inference without that capacity. SandBAR reaches 29.89 / 28.92 / 22.77 dB on EUVP / LSUI / UIEB and establishes a new three-domain mean state of the art on the benchmark family, exceeding the strongest published result by approximately 1.0 dB. PaST adds no parameters or FLOPs at inference and preserves matched latency.
[CV-124] Scaling Laws for Vision-Language Fusion
链接: https://arxiv.org/abs/2611.22351
作者: Seng Pei Liew,Teppei Suzuki,Seiya Tokui
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Vision-language models (VLMs) have emerged as central foundation models combining multimodal perception with text comprehension. While VLMs are traditionally built via late fusion by continued pre-training an off-the-shelf language model on multimodal (image-text) data, mixing modalities from the start of pre-training (early fusion) has recently gained traction. However, the fundamental trade-offs between these paradigms remain poorly understood, resulting in expensive and heuristic-driven training recipes. In this paper, we systematically analyze the learning dynamics of these multimodal pre-training processes, showing that validation losses during vision-language fusion can be modeled by separating the base learning trajectory, where different data types contribute at distinct rates, from a distribution shift driven by the competition between modalities. We derive a scaling law that unifiedly describes these dynamics, enabling accurate loss predictions across mixture ratios, training steps, and model sizes. Crucially, our law characterizes the trade-off between early and late fusion: under a fixed data budget, early fusion preserves text performance while late fusion favors vision performance, yielding a trade-off that increasingly tilts toward early fusion as model size scales. Extensive experiments confirm that our scaling law reliably captures multimodal learning dynamics and governs fusion decisions.
[CV-125] herMark: Instance-Level Watermark Authentication Against Cross-Instance Transplantation
链接: https://arxiv.org/abs/2611.22345
作者: Jin Wei,Rundong Li,Li Ma,Yikai Wang,Ruihao Yang,Jianxiong Wu,Xiaoyuan Duan,Yanbo Wang,Runyu Mao,Chen Fang,Chang Xu,Ping Chen,Jun Dai,Xiaoyan Sun
类目: Cryptography and Security (cs.CR); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Robust watermarking is commonly verified by asking whether a hidden payload remains decodable after compression, geometric distortion, or generative editing. Localized and semantic extensions may additionally check whether an expected region or category is still present. These tests establish evidence validity or content consistency, but not whether the evidence still belongs to the particular visual instance registered by its owner. We expose cross-instance transplantation: an attacker extracts or estimates valid watermark evidence from one registered image and transfers it to a different image of the same category. The message can still decode, the expected content can remain present, and the result can look natural, even though the evidence has been reassigned to the wrong instance. To enable instance-level watermark authentication, we introduce TetherMark, a post-hoc watermark that combines two complementary checks. The message check verifies an edit-robust payload, while the relation check uses secret patch-pair selection, a registered response, natural correspondence transport, and enrollment normalization to test consistency with the registered instance. The verifier accepts only when both checks pass. Across 240 test subjects, TetherMark attains 92.08% legitimate true acceptance with 9.17% and 16.25% attack success under WmForger and WMCopier, respectively, yielding the highest instance-binding balanced accuracy under both attacks.
[CV-126] PURE: A Diagnostic Benchmark for Multi-Granularity Multimodal Identity Matching
链接: https://arxiv.org/abs/2611.22343
作者: Hengyu You,Xintong Wang,Jin Cao,Liang Ding
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Identity matching asks whether two records denote the same identity, yet the answer depends on the target granularity. Two records may match at the base level but not at the variant level: they share an underlying design while differing in attributes such as color or size. Final matching accuracy does not separately characterize the capabilities involved. To study this problem, we use e-commerce as a challenging real-world domain. We introduce PURE, a diagnostic benchmark for multi-granularity multimodal identity matching covering four capability dimensions: Perception, Understanding, Reasoning, and Evidence-grounded Judgment. The benchmark comprises 1,457 multiple-choice questions across 14 tasks, derived from real-world matching pairs and verified by two domain experts. Across ten recent MLLMs, overall scores range from 47.1% to 86.0%. Understanding receives the highest mean score and Judgment the lowest, with weaknesses in visual difference localization and variant-level rule judgment. Modality ablations show that models benefit from multimodal evidence but can incorrectly identify conflicts between matching records. In direct matching, most models recognize matching pairs more reliably than non-matching pairs at the base level, whereas the reverse holds at the variant level. PURE therefore reveals what aggregate matching accuracy hides: not only whether models judge a pair correctly, but where they struggle to interpret evidence and apply matching criteria.
[CV-127] DAP-MR: Degradation-Aware Prompt-Magnitude Routing for Real-World Ultrasound Super-Resolution
链接: https://arxiv.org/abs/2611.22331
作者: Ngoc-Hoang-Lam Le,Thi-An Le,Dinh-Hung Le,Manh-Hung Nguyen,Kai-Lung Hua,Ching-Chun Huang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:When super-resolution models designed for natural images are used on ultrasound scans, relying on semantic text prompts can lead to unreliable results. The tags these models generate often do not match the actual restoration needs, and specialized medical terms may be misunderstood by general-purpose text encoders. Instead, prompts focused on common image degradations are more directly useful for guiding restoration. However, using a single prompt to describe both the type and severity of image degradation can entangle these factors, making it hard for the model to know how strong the degradation is. To address this limitation, we introduce Degradation-Aware Prompt–Magnitude Routing (DAP-MR). Each degradation type is represented by a dedicated token initialized from the contextualized EOS embedding of a degradation-oriented prompt and subsequently adapted during ultrasound SR training. A separate magnitude router predicts a continuous, input-dependent magnitude that modulates the corresponding token. Thus, language provides a degradation-aware initialization for what should be restored, while magnitude routing supplies an explicit signal for how strongly it should influence restoration. DAP-MR can be added to both one-step and multi-step diffusion super-resolution frameworks without changing their core structures. Our experiments–on synthetic datasets with both in-domain and under anatomical-domain shift, as well as unpaired real-world ultrasound data–show that DAP-MR delivers better restoration quality and generalizes more consistently across different domains.
[CV-128] M3-Test: Coverag e Gaps in Joint-Perturbation Testing of Visual Question Answering
链接: https://arxiv.org/abs/2611.22329
作者: Enhui Wan,Xiaowan Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:A visual question answering (VQA) model can pass every image-only and question-only test yet fail when the same changes occur together. Standard robustness reports average over conditions and therefore cannot say which accepted requests remain vulnerable. We introduce M3-Test, a coverage audit that aligns each clean input with its single-modality variants and a complete 5x5 family, 4x4 strength joint library, so that any reduced test suite can be replayed on the same requests. Across 2.6 million responses from four vision-language models, joint failures affect 26-29% of the Qwen questions that pass all 40 single-modality tests, although fewer than 0.6% of their joint records fail. Passing the 25 maximum-strength corners does not close this gap: 11-14% of the accepted questions still fail elsewhere in the library. Re-executing the archived failures reproduces about 90% of them, against a 0.3-0.6% flip rate among matched passing controls, and re-checking the most fragile failure records supports 80-89% of them. On the evidence examined, neither execution noise nor grading error can fully account for the gap. The same aligned grids make previous evaluations reusable. On held-out images, a risk order learned from other models discovers about 80% of affected Qwen questions within 25 additional queries, compared with 65-68% for maximum-first and 53-57% for pairwise covering arrays. M3-Test thus turns a passing robustness suite into a request-level account of what remains untested, supporting acceptance, regression review and test allocation.
[CV-129] Breaking Visual Conventions: Spectrally Coupled Weight Perturbation for Creative Image Generation
链接: https://arxiv.org/abs/2611.22318
作者: Fu Feng,Ruoheng Zhou,Yucheng Xie,Xu Yang,Jing Wang,Xin Geng
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Text-to-image (T2I) models often produce high-quality images that follow familiar visual conventions, even in response to open-ended prompts such as ``a creative [CLS]‘’. Existing methods encourage novelty by exploring semantic conditions or modifying features and guidance during denoising. Pretrained weights offer a complementary source of creative variation, but unstructured perturbations can have little visual effect or degrade image quality. We introduce Coupled Spectral Perturbation (CosPert), a training-free method that targets attention projections in layers identified by a one-time sensitivity analysis. For each projection, CosPert uses a shared random transformation and its transpose to couple updates to the singular-vector factors while perturbing the spectral gains. A common relative budget calibrates the magnitude of weight changes across projections, and a sigmoid schedule concentrates their influence in early denoising and gradually attenuates them during refinement. CosPert provides plug-and-play creativity enhancement at inference time. Once the perturbations are prepared, sampling requires no auxiliary networks or additional model evaluations. On both FLUX.1 and Stable Diffusion 3.5, CosPert enhances visual creativity, producing more varied object structures while preserving category identity. Across 50 object categories, CosPert achieves the highest mean CREward among the evaluated methods while maintaining strong prompt alignment.
[CV-130] SIDUMN: Self-Supervised Image Denoising Under Symmetric Multiplicative Noise
链接: https://arxiv.org/abs/2611.22317
作者: Nyan Htet Naing,Gyeongha Hwang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 10 pages, 4 figures, 2 tables. Submitted to IEEE Signal Processing Letters
Abstract:Self-supervised denoising is challenging when the corruption is multiplicative and signal-dependent. In this letter, we propose SIDUMN, a symmetric noisier framework for self-supervised denoising under mean-one symmetric multiplicative noise. Given noisy observations and the noise law, the method samples an independent synthetic noise factor from the same law and constructs two complementary noisier images by applying the factor and its reflection about one. The denoising network is trained to predict one noisier image from the other, and at inference the reconstructions of K such antithetic pairs are averaged. We show that the population target of the training loss is the clean conditional mean minus a multiplicative cross-term. We further prove that, for any trained network, antithetic averaging never increases the expected squared error of a single noisier-input reconstruction. Numerical results validate the theoretical findings and the proposed method.
[CV-131] Inertial Video World Models
链接: https://arxiv.org/abs/2611.22314
作者: Yuan Gao,Fan Xu,Yu Guo,Yuxu Lu,Zhengru Fang,Hao Wu,Xiaomeng Huang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:The physical world has inertia: it evolves from what already exists rather than being recreated at every instant. However, many existing video world models lack an explicit mechanism for inertial state evolution, leaving the continuity of motion and structure implicit in video prediction. During autoregressive rollouts, repeated reconstruction can accumulate errors and destabilize the predicted dynamics. To address this limitation, we introduce InertialWorld, an Inertial Video World Model that makes inertia a structural prior for autoregressive prediction: existing scene evidence is carried across time rather than reconstructed at every step. Central to this formulation is Inertial Memory, which turns memory from passive predictive context into an evolving video state with two complementary views: a dense Eulerian grid preserves scene-aligned evidence, while persistent Lagrangian carriers bind local content to learned canonical coordinates. At each step, the model transports the carrier content according to action-conditioned motion, predicts the complete next video state from the transported carriers and dense grid, and uses a learned gate to combine transported and newly predicted content into the memory for the next step. Extensive experiments across multiple domains, including robot manipulation, 3D game simulation, and open-world navigation, demonstrate that InertialWorld achieves competitive autoregressive prediction performance against strong video world model baselines. Codes link: this https URL
[CV-132] Cross-Modal StateKV Injection for Communication in Multi-LLM Systems across Temporal Segments
链接: https://arxiv.org/abs/2611.22304
作者: Xiaoqian Zhang,Zhihao Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 9 pages, 4 figures, 5 tables. Equal contribution: Xiaoqian Zhang and Zhihao Zhang
Abstract:Video understanding with multiple multimodal language models requires a communication interface that can transfer complementary visual observations. Text-based communication is interpretable, but introduces an intermediate generation stage and can omit information relevant to subsequent reasoning. We study Cross-Modal StateKV Injection, a provider–carrier framework that communicates through key–value (KV) states at visual-token positions. Provider models observe complementary video frames; their visual KV states are aligned and projected into the carrier representation space and integrated through gated residual updates. Injection preserves the carrier KV-cache sequence length, while retaining the additional computation required by the providers and the communication module. The reported experiments on MLVU and Video-MME compare carrier-only inference, text-based communication, and StateKV with two provider architectures. Results show modest, task-dependent accuracy gains and lower reported latency than the evaluated text-communication baseline, with some categories remaining unchanged or decreasing. Increasing provider-side frame coverage offers additional gains on selected tasks without extending the carrier cache. These findings motivate further study of visual KV communication under explicitly controlled observation and computation budgets.
[CV-133] Perception into Imagination: Executable Memory for World Models
链接: https://arxiv.org/abs/2611.22294
作者: Xianyao Li,Ruitong Tian,Rui Min,Fang Xu,Jing Du
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注: 18 pages, 6 figures, 15 tables
Abstract:Physical agents must use what they previously saw or touched after visibility or contact is lost. Their actions change the implications of these remembered facts. We introduce executable perceptual memory (EPM), a world-model framework that advances explicit perceptual state alongside neural latent dynamics during imagination. Its evolving read-out informs reward prediction, value estimation and policy learning. We align reward inputs with the transitions that produced them and supervise the reward head on open-loop rollouts of recorded actions. On Memory Maze 13\times13 , EPM reaches 30.37 return at 1M interactions, compared with 23.37 when the same evolving read-out informs only policy and value on a common perceptual front end: a gain of 7.00 [3.80, 10.21]. The complete method also improves control on 9\times9 and in MiniWorld. Probes of remembered, out-of-view arrivals show stronger reward anticipation in both visual implementations. Concentration, tactile recovery and real tactile frames demonstrate complementary roles of remembered facts in imagined-policy training, motion-dependent decisions and sensory-content retention. EPM connects persistent perception to the predicted consequences and learning signals of a world model for Physical AI.
[CV-134] Decoding Without a Decoder: Direct Pixel Prediction from Diffusion Transformers
链接: https://arxiv.org/abs/2611.22287
作者: Hansheng Chen,Shengqu Cai,Jan Ackermann,Rayhan Zirvi,Gordon Wetzstein,Xinchen Yan,Leonidas Guibas
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project page: this https URL
Abstract:Latent diffusion models rely on a decoder to map denoised latents to pixels, adding complexity and inference cost. We introduce Decoding Without a Decoder (DWD), which lets a finetuned diffusion transformer produce final pixels directly through a linear head. Unlike previous methods, where the denoising output leads to final pixels, DWD separates the two roles: the original denoising head continues to drive sampling, then the new DWD head predicts the pixels at the final step. We extend pretrained latent models with DWD efficiently by finetuning only the last few transformer blocks, training the DWD head with pixel quality losses while anchoring the denoising head to the pretrained model to preserve the sample trajectory. Across latent image and video generators, DWD removes the decoder and reduces inference cost while maintaining or even improving pixel quality. On Wan video generation, it visibly reduces detail distortion and flickering, lowering warping error by 23%. Remarkably, the same separation also improves pixel diffusion models that do not even use a decoder: a final-step DWD head produces substantially better texture detail in text-to-image generation, and improves ImageNet FID from 1.55 to 1.43, a new state of the art among pixel flow models.
[CV-135] Heave-Ho: Decentralized Long-Horizon Multi-Humanoid Collaboration
链接: https://arxiv.org/abs/2611.22275
作者: Ting-Hsuan Chen,Siheng Zhao,Han Jiang,Qi Pan,Louison Lu,Jiaheng Luo,Henghui Bao,Yiyu Chen,Quan Nguyen,Yue Wang
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 18 pages, 11 figures, 7 tables. Project page: this https URL
Abstract:Collaboration promises to extend humanoid capabilities beyond what individual robots can achieve. However, realizing this potential requires coordinated whole-body interactions, coherent long-horizon planning, and precise synchronization across robots. We introduce Heave-Ho, a decentralized framework for long-horizon collaborative multi-humanoid loco-manipulation. At the low level, we develop a multi-agent post-training framework that adapts a humanoid foundation tracking model to physically coupled interactions between humanoids, enabling coordinated decentralized control from local observations alone. At the high level, we train a vision-language planner and design a speech-based leader-follower coordination mechanism to synchronize decentralized execution without network communication. Trained with two identical humanoids, our system scales to teams of up to ten robots and generalizes to heterogeneous teams without retraining. To benchmark these capabilities, we introduce HH-Bench, comprising 12 tasks involving teams of 2 to 10 humanoids, including heterogeneous teams. Systematic evaluations show that Heave-Ho consistently outperforms baseline methods, while real-world experiments demonstrate robust decentralized collaboration across challenging tasks.
[CV-136] SARVIA: Fast-Adapting the Strong SAR Foundation Models That Are Already in Your MLLM s
链接: https://arxiv.org/abs/2611.22273
作者: Yuxuan Li,Zichen Tian,Qianru Sun
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Synthetic aperture radar (SAR) foundation models provide transferable representations for diverse perception tasks in all-weather Earth observation. However, their development often relies on costly large-scale SAR pretraining, raising the question of whether such extensive domain-specific pretraining is necessary. We find that modern multimodal large language models (MLLMs) already provide visual encoders that can outperform dedicated SAR foundation models as downstream initializations. This motivates a new paradigm: reuse general visual capabilities and devote domain training to brief specialization. We therefore introduce SARVIA, an adaptive post-pretraining method that specializes modern MLLMs using self-supervised visual questions with computable answers. These questions are constructed from unlabeled observations through general image tasks that capture spatial relations and SAR-specific tasks that exploit native measurements and SAR–optical correspondence. A shared answer-prediction objective jointly trains low-rank adapters across visual and language components, with loss-driven task sampling. We then export the adapted visual encoder as a standalone backbone. With 600 adaptation steps and 2 GPU-hours, SARVIA outperforms the compared foundation models on all six SAR detection and segmentation benchmarks. Compared with dedicated SAR pretraining, SARVIA reduces domain-training costs by orders of magnitude while improving visual transfer across architectures and scales. This new paradigm shifts the emphasis from repeated, costly pretraining to a reusable specialization process that can benefit from advances in open MLLMs. Dataset and code: this https URL.
[CV-137] Multispectral-Polarization-Infrared Multi-Aperture Camouflaged Vehicle Detection
链接: https://arxiv.org/abs/2611.22246
作者: Shu Wang,Xiong Lin,Yuan Song,Junjie Wang,Feng Huang,Wenxi Liu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 14 pages, 10 figures
Abstract:Camouflaged vehicle detection is challenging because camouflage suppresses discriminative cues across sensing modalities. Existing multimodal methods often require precise pixel-level registration and feature-level fusion, creating a trade-off between accuracy and latency. We propose a spectral-polarization-infrared tri-modal detection framework co-designed with a multi-aperture imaging setup to exploit complementary imaging physics. A consensus-guided decision-level fusion strategy combines cross-modal detections within bounded registration tolerances, reducing dependence on precise alignment while maintaining accuracy and real-time performance. We construct a tri-modal camouflaged vehicle dataset containing infrared, polarization, and spectral data. Experiments show that the proposed method achieves 82.7% mean Average Precision (mAP) with an end-to-end latency of 16 ms, outperforming representative feature-level fusion approaches. On the public MID-1K dataset, the framework generalizes to RGB-Depth-Infrared modalities without architectural modification and improves mAP@0.5 by 5.7% over state-of-the-art methods. These results demonstrate the potential of decision-level consensus fusion and multi-aperture sensing for robust and deployable camouflaged vehicle detection.
[CV-138] PhyVGGT: Feed-Forward Articulated Object Reconstruction from Monocular Videos
链接: https://arxiv.org/abs/2611.22244
作者: Qi Shao,Zhihao Cao,Shuhao Zhai,Anh Nguyen,Hong Liang,Baoru Huang,Liang Li
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 10 pages, 5 figures, 7 tables. Project page: this https URL
Abstract:Reconstructing an articulated object with its 3D movable parts, joint parameters and unseen articulation states from an uncalibrated monocular video is essential for robot manipulation and interactive digital twins. Existing per-scene optimisation methods require minutes to hours and external depth or camera poses, whereas feed-forward 3D foundation models reconstruct dense point maps in one pass without segmenting movable parts or inferring joints. Even on these monocular point maps, regressing or fitting continuous joints directly fails on casual videos, since cross-frame attention blurs moving part boundaries in deep layers, underestimates motion along the viewing ray and leaves continuous axes ill-conditioned under small opening angles. To address these challenges, we propose PhyVGGT, a single-pass framework on a frozen 3D foundation model that anchors moving-part matching and closed-form joint estimation to the undistorted geometry of the stationary base. First, our Part and Motion Decoder fuses intermediate per-frame tokens with global tokens and references part queries to a static-base query to segment 3D parts, align point maps on the static base and predict learned Bingham priors. Second, our Kinematic Inference Layer fits an orthogonal Manhattan frame to static surface normals, selects the joint axis in closed form under a Bingham posterior combining pairwise motion scatter with the learned priors, and solves a linear system for the hinge pivot to drive unseen states. Trained solely on synthetic data, PhyVGGT achieves state-of-the-art monocular performance across five benchmarks, outperforming FAMOS and monocular ArtiPoint, surpassing posed RGB-D ArtiPoint in median error, and running in 6.8 s per 48-frame video against hours of per-scene optimisation in Articulat3D. Project page: this https URL.
[CV-139] Cross3DGS: Unified Vulkan-Based 3D Gaussian Splatting Training from Workstations to Mobile Devices WACV2027
链接: https://arxiv.org/abs/2611.22240
作者: Hiroyuki Sato,Yusuke Takimoto
类目: Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR)
备注: Accepted to the WACV 2027 Workshop on Eyes of the Future (EOF). 19 pages including appendix
Abstract:Scenes captured with handheld and wearable cameras are still shipped to a CUDA workstation for 3D Gaussian Splatting (3DGS) training. Training on the device that captured them, or on its companion phone or laptop, requires more than translating kernels: GPUs and drivers differ in parallel features, memory limits, and command submission, and these differences decide whether a training schedule executes and returns the correct update. We present Cross3DGS, a cross-vendor, cross-platform Vulkan trainer that keeps one optimizer and chooses its execution schedule per device in two steps: rules on device limits, workload size, and observed driver failures exclude the schedules known not to run correctly, and a fixed preference order selects one of the remaining feasible schedules. The same core completes standard-quality training (30k updates, one million Gaussians) on every configuration we measured, spanning NVIDIA, Intel, Arm, Qualcomm, AMD, and Apple GPUs and including Android devices and an iPhone, whereas the existing portable trainers VkSplat and Brush do not run on some of the mobile GPUs. Averaged over 11 scenes on a workstation, Cross3DGS has higher quality than VkSplat with no significant difference in training time, and Brush reports about three times the training time at comparable quality. The final quality of Cross3DGS is nearly identical across devices. A mobile application turns a handheld capture into a viewable model entirely on the device.
[CV-140] CoSphere-SLAM: Multi-Agent Panoramic Gaussian Splatting SLAM with Spherical Loop Closure
链接: https://arxiv.org/abs/2611.22235
作者: Qi Shao,Zhihao Cao,Shuhao Zhai,Taoyu Wu,Zhuoxiao Li,Anh Nguyen,Long Hu,Xianzhi Li,Hong Liang,Baoru Huang,Liang Li
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 9 pages, 4 figures, 7 tables. Project website: this https URL
Abstract:Dense collaborative mapping is essential for search and rescue, building inspection, and warehouse automation, where panoramic cameras capture the entire surroundings in a single frame. Unlike pinhole agents that share views only at similar headings, panoramic agents observe overlapping scenes at any 3D orientation, yet searching full rotations also produces mutually consistent false loop closures across distinct symmetric rooms. Moreover, continuous monocular scale drift splits rigidly merged maps at revisited surfaces, while depth noise at short baselines distorts loop translation directions. To address these challenges, we propose CoSphere-SLAM, the first multi-agent Gaussian splatting SLAM framework for panoramic cameras, whose spherical loop closure aligns spherical features over \mathrmSO(3) via fast Fourier correlation across arbitrary orientations and rejects symmetric false closures through global \mathrmSim(3) consensus. Building on these closures, directional global optimization solves a per-keyframe \mathrmSim(3) pose graph with log-scale regularization to correct scale drift while separating optical-flow translation directions from depth-based magnitudes. Finally, omnidirectional Gaussian fusion transforms each Gaussian with its keyframe similarity and renders Gaussians at their maximum ray response to jointly refine poses and maps. Experiments on indoor and cross-session multi-agent panoramic benchmarks demonstrate that CoSphere-SLAM achieves state-of-the-art trajectory accuracy and rendering quality among single-agent and multi-agent methods. Project website: this https URL.
[CV-141] What Does Weight-Space Programming Add to Diffusion Editing? Program Construction versus Score-Guided Refinement
链接: https://arxiv.org/abs/2611.22228
作者: Zitian Gao,Jingyi Xu,Ziqi Han,Chengkai Lei,Ye Liu,Liangliang Shi
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注: 25 pages, 7 figures, 13 tables
Abstract:Weight-space programming (WSP) lets a language model construct executable weight updates for another network. We study diffusion concept editing through an interface in which programs define training contexts, preservation constraints and update directions, while a shared numerical fitter selects their coefficients. We compare generated programs with published editors, a shared-panel hand-written baseline and budget-matched search over conventional edits on 24 TIMED test edits for Stable Diffusion 1.4. The preregistered rewrite pipeline shows no advantage over the strong baseline. In an exploratory comparison, one-shot Astra programs improve fresh-scene VLM joint success, which requires subject presence together with whole-image target matching, by +10.0 points over the shared-panel baseline, while CLIP relative match shows no detectable change. These programs organise adjustable directions around protected text-state subspaces. Probing and rewriting the Astra programs then raises CLIP by +6.2 points while joint success changes by -9.7. On a fixed 144-image sample, five additional evaluators give point estimates in the same directions under whole-image target matching; checks of subject-local change give less decisive differences. Together, these results identify useful behaviour in complete generated programs and show that score-guided revision need not preserve it: program construction and revision should be evaluated as distinct stages of a WSP pipeline.
[CV-142] RIDENT: Efficient Triple-Task Learning of Dehazing Depth and Uncertainty Estimation for Underwater 3-D Robot Visual Perception
链接: https://arxiv.org/abs/2611.22224
作者: Geonmo Yang,Younggun Cho
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注: 13 pages. © 2024 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works
Abstract:Underwater visual systems often suffer from blurry textures and low color contrast due to the inevitable light propagation. These issues can significantly degrade the perception of stable robotic operations. In this paper, we introduce a novel learning-based sensing system that tackles the multidimensional vision tasks in underwater; concretely, we deal with image enhancement, depth estimation, and uncertainty for 3-D visual systems. Also, we propose a TRIDENT model in a fast and lightweight manner; TRIDENT consists of three parallelized decoders and one backbone structure for efficient feature sharing. In addition, it is designed to be trained to express complex parameterization. In experimental evaluation on several standard datasets, we demonstrate that TRIDENT significantly outperforms other existing methods on image enhancement and depth estimation. Despite performing three tasks, our model has better efficiency than the others for both memory size and inference time. Finally, our joint learning approach demonstrates robustness in feature matching and seamlessly extends from 2-D to 3-D vision tasks. Supplements are available at this https URL
[CV-143] ExplicitIR: Explicit Spatial Policy Fields for All-in-One Image Restoration
链接: https://arxiv.org/abs/2611.22222
作者: Dezhi Zhang,Shangrong Yang,Yudong Liang,Wenjian Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 13 pages, 10 figures, 4 tables
Abstract:All-in-one image restoration aims to recover clean images blindly from inputs with various degradation types and levels using a single model. However, the adaptation policy in most existing methods remains implicitly encoded in feature representations, leaving the restoration decision neither explicitly represented nor independently controllable. Since real-world degradations vary spatially, applying uniform updates across the entire image leads to over-smoothing of clean areas and under-correction of localized degradations. To address these limitations, we present ExplicitIR, which represents the restoration policy as two explicit per-pixel fields: Budget, which determines where and how strongly to update, and Style, which selects between narrow and wide context windows. Specifically, ExplicitIR incorporates a Spatial Policy Estimator (SPE) to predict these fields, explicitly representing the restoration policy and capturing spatially-varying degradations, which are then applied by Policy-Adaptive Decoder Blocks (PADB) through adaptive fusion of narrow- and wide-window responses, enabling independent control and precise local modulation. This design enables spatially-adaptive restoration in a single forward pass, without requiring task labels, prompts, or test-time degradation estimators. Extensive experiments on three-task and five-task benchmarks demonstrate that ExplicitIR achieves performance competitive with the strongest all-in-one baselines.
[CV-144] Joint-ID: Transformer-based Joint Image Enhancement and Depth Estimation for Underwater Environments
链接: https://arxiv.org/abs/2611.22212
作者: Geonmo Yang,Gilhwan Kang,Juhui Lee,Younggun Cho
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 11 pages. © 2023 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works
Abstract:Underwater imaging is a challenging task due to factors such as scattering, absorption, and turbulence, which degrade image quality and limit visibility. In this paper, we propose a novel approach for enhancing underwater images that leverages the benefits of joint learning for simultaneous image enhancement and depth estimation. We introduce Joint-ID, a transformer-based neural network that can obtain high-perceptual image quality and depth information from raw underwater images. Our approach formulates a multi-modal objective function that addresses invalid depth, lack of sharpness, and image degradation based on color and local texture. We design an end-to-end training pipeline that enables joint restoration and depth estimation in a shared hierarchical feature space. Additionally, we propose a synthetic dataset with various distortions and scene depths for multi-task learning. We evaluate Joint-ID on synthetic and standard datasets, as well as real underwater images with diverse spectra and harsh turbidity, demonstrating its effectiveness for underwater image enhancement and depth estimation. Furthermore, we demonstrate the ability of Joint-ID to perform feature matching and saliency detection for visually guided underwater robots. Our proposed method has the potential to improve the visual perception of underwater environments and benefit applications such as oceanography and underwater robotics. Supplements are available at: this https URL
[CV-145] ARG-Avatar: Animation-Rendering Guidance for Efficient Feedforward Avatar Reconstruction
链接: https://arxiv.org/abs/2611.22207
作者: Jianjin Xu,Soham Dasgupta,Fernando De la Torre
类目: Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR)
备注:
Abstract:Reconstructing animatable 3D human avatars from images is important for applications such as gaming and augmented reality. Feedforward methods make avatar reconstruction accessible by reducing inference time to a few seconds, but recent approaches rely heavily on scaling model capacity, training data, and computation. We ask whether such scaling is necessary. To this end, we introduce ARG-Avatar, an efficient feedforward framework with only 68M trainable parameters that achieves SOTA reconstruction quality. Our key insight is that 3D animation and rendering already provide strong geometric correspondences that existing methods largely attempt to learn from data. In light of this, we design the Foreground Avatar Coordinate RoPE attention (FACRoPE) and Intermediate Token Rendering (ITR) pipeline to guide avatar tokens to attend to their corresponding patches. Experiments across multiple datasets show that ARG-Avatar consistently outperforms prior methods when using two or more input images. Compared with LHM++, ARG-Avatar uses approximately 9 \times fewer trainable parameters, 4.8 \times fewer training sequences, and 14 \times less training compute, while running 6 \times faster and achieving a 2.7 dB PSNR improvement on the out-of-distribution NeuMan benchmark.
[CV-146] MIRAG E: Motion Identity Reduction via Autoregressive Generative Encoding for Privacy-Preserving Skeleton-based Motion Data
链接: https://arxiv.org/abs/2611.22200
作者: Thomas Carr,Xujun Che,Depeng Xu,Shuhan Yuan,Aidong Lu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 14 pages, 3 figures, 12 tables
Abstract:Skeleton motion identifies people through body proportions and movement style, and skeleton streams are increasingly passed to third-party analytics that run action classifiers trained on raw data. A prior adversarial anonymizer adds a per-frame residual in the raw coordinate space but sees one frame at a time; motion retargeting needs a target skeleton and complete sequences, which rules out online use. We present MIRAGE, a target-free, causal anonymizer that adds an identity-reducing residual to each incoming frame in the raw coordinate space, computed by attention over past frames. One causal, optionally windowed mask governs every attention in a rotary-position encoder-decoder, and key/value caches let the trained function run frame by frame; on all 56,578 NTU 60 sequences, batched and streaming outputs differ by at most 7.2e-6 m. Reported results use an unbounded window, so per-stream state grows with stream length. We evaluate against a passive adversary that applies raw-trained identity models and an adaptive adversary that trains on the anonymizer’s output, treating the adaptive adversary as the headline, on NTU RGB+D 60 over five generator seeds, alongside the per-frame residual anonymizer of Moon et al. trained identically. When consumers retrain on the released data, MIRAGE keeps 83.9% action recognition (raw 89.1%; Moon et al. 88.7%) while a retrained identity classifier reaches 43.4% re-identification (RI) over 40 subjects, against 69.4% for Moon et al. and 75.4% on raw data. The strongest adaptive attacker still reaches 49.3% (chance 2.5%), so MIRAGE reduces identity leakage; it does not anonymize. Against raw-trained (frozen) classifiers, MIRAGE lowers RI below the strongest retargeting baseline (10.36% vs. 22.09%), but frozen action recognition falls to 37.04% (vs. 75.82%), so at full residual scale it is not a drop-in replacement for raw input to existing frozen models.
[CV-147] SRED: Sparse Regional Evidence Distillation Across Magnifications in Whole-Slide Imaging CVPR2027
链接: https://arxiv.org/abs/2611.22189
作者: Zhiyuan Yang,Jiahao Cheng,Vincent Quoc-Huy Trinh,Mahdi S. Hosseini
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: CVPR 2027 submission
Abstract:In computational pathology, high-magnification whole-slide image analysis preserves tissue detail but requires encoding many patches, whereas lower magnification reduces computation by covering a higher field of view per patch while sacrificing detail. Existing cross-magnification distillation matches features that often encode detail unreachable by low-magnification input, sometimes requiring slide-level labels that limit scalability. To address those issues, we introduce Sparse Regional Evidence Distillation (SRED), which transfers regional event occurrences that encode reachable high-level interactions between details without slide-level labels. We use a sparse autoencoder (SAE) to learn an event dictionary from 20x teacher features. Logical OR pooling records whether each event is detected in any spatially aligned patch, retaining both localized and shared evidence within the corresponding 5x region. A pretrained 5x encoder is fine-tuned to predict these regional targets without slide-level labels. Visual analysis identifies dictionary components responsive to recognizable histological patterns, supporting target interpretation. Compared with native 5x features, SRED improves average macro-F1 by 1.1 points across seven classification and grading cohorts and average C-index by 1.7 points across five survival cohorts. Runtime matches native 5x, with an average 11.58x speedup over 20x.
[CV-148] CueClash: Frontier Vision-Language Models Judge 3-D Relations Close to the Picture Plane
链接: https://arxiv.org/abs/2611.22180
作者: Soham Dan
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Robotics (cs.RO)
备注:
Abstract:A vision-language model (VLM) can judge 3-D relations from the picture or from the ground layout. In photographs the two usually agree. We separate them with CueClash, rendered scenes in which the 2-D cue for each of five relations is right on exactly half the statements. The shortcut gap is accuracy when the cue is right minus accuracy when it is wrong. For Claude Opus 5.5, GPT-5.5 and Gemini 3.6 Flash at low reasoning effort it is +18 to +37 percentage points. It is +7 to +29 without two relations whose cue-wrong items are confounded with borderline cases or occlusion. Given ground coordinates and each relation’s definition, they answer 99 to 100% correctly. Pixel measurements with camera depths remove most of the gap for objects fitting through openings. The failure lies in the step from the picture to the layout. The maps that the models draw of the scene from above mostly keep the depth order but, like the picture, shorten depth. GPT-5.5’s maps stay close to a copy of the picture. Opus 5.5’s and Gemini’s maps undo about 0.6 of this shortening. Gemini’s then shrink all depths by a further uniform factor. Opus 5.5 follows the picture mainly on fits, and its maps correct depth about as much as human observers of real scenes do. The three models also lose accuracy on CV-Bench photographs where a reliable 2-D cue misleads. In two open models the layout information reaches the language model, but their maps lose much of it.
[CV-149] Not All Evidence Is Admissible: Evidence Governance for Tool-Augmented Medical Agents
链接: https://arxiv.org/abs/2611.22169
作者: Runxun Zhang,Onofrio Antonio Catalano,Jinqiao Wang,Jingwei Wei
类目: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注: 11 pages
Abstract:Tool-augmented medical agents treat whatever their tools return as evidence. We show that this default is costly: medical tools fail silently, and a frozen reasoner cannot ignore what enters its context, so many cases it would have answered correctly are flipped to wrong once tool outputs are added. We call this evidence hijacking. \mathcalV -information explains the phenomenon: the reasoner has lost the option to ignore, so a tool output can carry negative usable information, and moving the decision of what to keep outside the reasoner restores that option. The theory further shows that part of any filtering gain comes merely from letting the reasoner see less, and only selection by content lets it see better. Building on this, we recast tool-augmented reasoning as evidence governance, which separates deciding what counts as evidence from the reasoner that answers, and propose context-independent evidence governance (CIEG), the first medical agent framework that rules, after acquisition, on whether each individual tool output may enter as evidence. Every tool output is written as a card that can be checked against the raw input; an independent checker that sees only the input and the cards decides which cards become evidence; and the reasoner answers from the admitted cards alone, with no training anywhere. On medical question answering, fundus glaucoma screening, and chest X-ray pneumonia classification, it achieves the highest accuracy among models of comparable scale, and significantly outperforms state-of-the-art tool-using agents and multi-agent systems built on the same backbone at lower cost.
[CV-150] HandRSI: Recursive Self-Improvement for Hand Pose Estimation
链接: https://arxiv.org/abs/2611.22164
作者: Zhekai Wang,Luqing Luo,Wenhao Yao,Junjie Lu,Huanan Zhu,Feifei Gao,Yuhua Jiang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Technical report. 25 pages, 17 figures, 18 tables. Code: this https URL
Abstract:Egocentric 3D hand pose estimators still miss hands at the image border, misplace occluded hands and jitter. Such failures are usually fixed by hand-written inference refinements and by fine-tuning, two interdependent steps that are tuned separately. We present HandRSI, a recursive self-improvement framework for a hand pose system that consists of a video estimator and its executable inference harness. In each round, coding agents refine the harness, and a change is accepted only if it improves a composite of all ten benchmark metrics on held-out data without significantly regressing any of them. The accepted harness then guides the estimator through selective distillation, which supervises hand presence and temporal consistency on frames without ground truth, while ground truth supervises pose. Since the updated estimator may change which refinements remain useful, the harness is re-validated on it. Starting from the released ACE-Ego-Hand model, five rounds reduce MPJPE-p by 35% on HOT3D and 11% on ARCTIC and roughly halve jitter. On HOT3D, HandRSI’s MPJPE-p is 16% lower than that of ACE-Ego-Hand fine-tuned on the same ground truth. Under a unified protocol, it outperforms nine re-run published methods on all ten metrics of both benchmarks. Hypervolume selection on the accuracy-speed front yields Lite, Medium and Heavy systems running at 20.7 to 41.1 FPS. This report gives the full method, every stage of every round and the complete protocol. Code, evaluator, harnesses and weights are public.
[CV-151] Centroid Spectral Coupling: Implicit Alignment and Representation Diversity in Self-Supervised Learning
链接: https://arxiv.org/abs/2611.22158
作者: M. Hadi Sepanj,Benyamin Ghojogh,Saed Moradi,Paul Fieguth
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Machine Learning (stat.ML)
备注:
Abstract:Self-Supervised Learning (SSL) aims to align augmented views while preserving representation diversity and preventing collapse. Existing non-contrastive methods often address these goals using multiple weighted loss terms. We introduce Centroid Spectral Coupling (CSC), a single-term SSL objective that implicitly couples view alignment, representation diversity, and collapse resistance. CSC constructs centroids from independently normalized augmented-view pairs and applies a ridge-based spectral objective to their uncentered second-moment matrix. Preserving centroid magnitudes encodes view agreement while encouraging balanced representation energy across spectral directions. We establish an exact decomposition into nonnegative alignment and spectral-imbalance components, characterize the geometry of global optima, and derive finite-loss guarantees for alignment and diversity. We also show that complete constant-output collapse is a strict saddle in normalized representation space. Experiments on CIFAR-10, CIFAR-100, and Imagenette show that CSC achieves the highest linear Top-1 accuracy among the evaluated non-contrastive methods on all three benchmarks, and the highest kNN Top-1 accuracy on both CIFAR datasets. A ridge-parameter sensitivity study reveals an empirical alignment–diversity trade-off: increasing \kappa strengthens view alignment but reduces representation effective rank. These results show that a unified spectral objective yields competitive visual representations without explicit contrastive negatives or separately weighted alignment and diversity penalties.
[CV-152] Unsupervised Multi-modal Medical Image Registration via Invertible Translation
链接: https://arxiv.org/abs/2611.22152
作者: Mengjie Guo,Jinming Duan
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:In medical imaging, the alignment of multi-modal images plays a critical role in providing comprehensive information for image-guided therapies. Despite its importance, multi-modal image registration poses significant challenges due to the complex and often unknown spatial relationships between different image modalities. To address this, we introduce a novel unsupervised translation-based multi-modal registration method, termed Invertible Neural Network-based Registration (INNReg). INNReg consists of an image-to-image translation network that converts multi-modal images into mono-modal counterparts and a registration network that uses the translated mono-modal images to align the multi-modal images. Specifically, to ensure the preservation of geometric consistency after image translation, we introduce an Invertible Neural Network (INN) that leverages a dynamic depthwise convolution-based local attention mechanism. Additionally, we design a novel barrier loss function based on Normalized Mutual Information to impose constraints on the registration network, which enhances the registration accuracy. The superior performance of INNReg is demonstrated through experiments on two public multi-modal medical image datasets, including MRI T1/T2 and MRI/CT pairs. The code is available at this https URL
[CV-153] EgoHandWAM: Probing the Role of Hand Motion in Egocentric World Action Modeling
链接: https://arxiv.org/abs/2611.22151
作者: Kailin Wang,Haoxiang Jie,Yaoyuan Yan,Zhiyou Heng,Zhaosong Li
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:How much does hand motion tell us about what a first-person camera will see next? EgoHandWAM forecasts three-dimensional wrist and fingertip trajectories and a future visual feature from the current image and recent hand states. We train on 751,065 EgoDex windows using a frozen DINOv2 encoder. On held-out development sessions, the action-only model reduces two-second wrist endpoint error from 60.26 to 56.48~mm relative to a stationary forecast. It still underestimates long reaches. Adding future-feature supervision gives nearly identical trajectory accuracy. With the recorded hand history, the visual head reduces smooth L1 error by 1.32% relative to copying the current feature. Shuffling histories across sessions erases this gain. In a separate four-task probe, adding recorded future displacements to observed context and future tracking-validity masks reduces visual-feature error by 3.93%. Masks alone give little improvement. A trajectory predicted from the same observations adds little. On two simulated RoboTwin tasks, substituting aligned human examples for an auxiliary batch of robot examples gives no consistent increase in task success across three training seeds.
[CV-154] CEVA: Complementary Evidence Verification via Frame-Level Reorganization for Short-Video Fake News Detection
链接: https://arxiv.org/abs/2611.22141
作者: Wenshen Yang,Junfeng Hao,Jing Li,Decheng Liu,Yangkunfei,Shuyu Jiang,Haizhou Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Short-video platforms are now a common way for the public to follow unfolding events. Their rapid, visually persuasive format also makes fake news easier to spread. Unlike text-image posts, short-video news interleaves multiple modalities and changing scenes, making it difficult to identify which frames verify a claim. Existing detectors either pool sampled frames homogeneously or select them by text-frame similarity. Both overlook that aligned frames may repeat a misleading claim, whereas less dominant context may challenge it. We present CEVA (Complementary Evidence Verification Approach), to our knowledge the first short-video fake news detection framework to explicitly organize frames by their evidential roles before multimodal fusion. Its Dual-Branch Complementary Reorganization (DBCR) module separates explicit-alignment frames from candidate complementary-context frames, and cross-modal fusion with bidirectional contrastive regularization operates on the reorganized evidence. On FakeSV and FakeTT, CEVA achieves the best overall results among the compared methods. Under fixed frame budgets and backbones, DBCR outperforms uniform, random, and similarity-based selection and improves two structurally different detectors. Together, these results establish pre-fusion evidence organization as an effective and transferable component of short-video fake news detection.
[CV-155] DailyBrick: Language-Guided Synthesis of Brick Assembly Instructions
链接: https://arxiv.org/abs/2611.22132
作者: Hang Zhang,Qijian Tian,Yuan Xie,Xuhong Wang,Xin Tan
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Generating a LEGO model from a user request is more than a plausible 3D shape generation. LEGO models require physical support per brick, and each brick should be placed unblocked. Current methods address this issue by training an expert large language model. We reduce the problem of generating extremely long context on all brick connections to that of writing a program that builds LEGO models like a human, considering each intermediate step could be executed. We introduce DailyBrick , a multi-stage agentic system that produces a LEGO model together with a brick-by-brick assembly instruction for human building and robot assembly. We design a multi-stage harness with roles to help LLMs generate more effectively. A versioned construction SDK compiles multi-level decisions, from semantic subparts to individual bricks, into canonical operations. An external Meta-Reviewer agent replays and inspects the generated package, verifies the model and its construction process, identifies failures, traces them to the responsible stage, and routes them to bounded repairs or to editable candidates for later skill or tool updates. Extensive experiments demonstrate that DailyBrick generates LEGO models based on simple user requests better than recent agentic systems, and show its utility in downstream applications such as robot manipulations on model assembly from instructions and brick sets (this https URL).
[CV-156] LoCaR: Localized Canvas Retrieval With Frozen Vision-Language Model
链接: https://arxiv.org/abs/2611.22131
作者: Rahim Hossain,Dipanwita Saha,Ashirbad Mishra,Xinxin Shu,Svetlana Strunjas,Binbin Li
类目: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: Zero-shot retrieval , Local Retrieval
Abstract:Whole image embeddings can under represent small objects and local attributes that distinguish relevant images for localized queries. We present LoCaR, a training free framework for localized text to image retrieval with frozen vision language models. LoCaR constructs and utilizes independent proposals (SAM2) focusing on the object so as to get a more localized representation that independently encodes each region using the frozen visual encoder. We further study paired-support adaptation as an optional residual that uses matched text region exemplars without updating the underlying encoders. We also introduce LOCFG, a Visual Genome derived benchmark designed to evaluate localized retrieval for smaller and more challenging referents. Across several vision language backbones, our results show that improving the region representation itself is the main source of gain.
[CV-157] xt-Guided Modality Correction for Open-Vocabulary Object Detection
链接: https://arxiv.org/abs/2611.22126
作者: Yu Quan,Dong Zhang,Ping Gong,Peng Huang,JinHui Tang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Open-vocabulary object detection (OVD) is essential for detecting objects beyond the categories seen during training. Establishing vision-language associations is crucial for recognizing unseen categories, while integrating complementary visual features across levels supports detection at different scales. However, existing methods often provide insufficient textual guidance for object regions, rely on limited cross-modal alignment strategies, and underuse complementary features across levels, limiting their ability to detect unseen categories and objects of varying scales. To address these limitations, we propose a Text-Guided Modality Correction Network (TMCNet) that incorporates textual guidance into region feature extraction, cross-modal alignment, and multi-scale decoding to strengthen region-category semantic correspondence and promote complementary feature fusion across levels. First, a text-guided modality encoder incorporates textual semantics into queries to guide regional feature aggregation, reducing semantic ambiguity in object representations. Second, a cross-modal alignment module strengthens the correspondence between region representations and category semantics through interactions between visual and textual features. Finally, a progressive text correction decoder combines a top-down pathway with lateral connections to progressively fuse text-guided deep semantic features with shallow visual features at the corresponding scales, preserving both semantic information and spatial detail for objects of different sizes. Experiments on two public datasets, COCO-OVD and LVIS-OVD, demonstrate that TMCNet achieves strong performance in open-vocabulary object detection using simple text queries as guidance. Extending TMCNet to open-vocabulary segmentation also yields performance gains, demonstrating its cross-task applicability and extensibility.
[CV-158] FrameSeg3D: Learned Local Frames for Segmenting Small Parts on High-Resolution 3D Scans
链接: https://arxiv.org/abs/2611.22124
作者: Xinwen Zhang,Tongyan Wang,Huawei Lin
类目: Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR)
备注: 31 pages, 6 figures, 16 tables
Abstract:Segmenting small parts on high-resolution 3D scans, such as nails on hand scans and teeth on intraoral scans, poses three challenges: 1) each part covers a tiny fraction of the scan, 2) its boundary is only a subtle change of height and curvature, and 3) it faces its own direction. Existing methods fall into two groups, and both fall short. The first group processes the whole scan at once and sees each part too coarsely. Point and mesh networks must shrink the scan to fit in memory, so a small part keeps only a few of its points. Multi-view methods render images of the whole object, so each part fills only a few pixels and a tilted part is seen from the side. The second group first finds each part and then segments it up close. But these methods find the parts with knowledge about one kind of part, for example the shape of the dental arch for teeth. Such knowledge does not carry over to other kinds of parts. We find that accuracy is limited by how a part is seen, from which direction and at what scale, not by how many points land on it. We therefore present FrameSeg3D. It first predicts a local frame for every part, its center and the direction it faces, from a low-resolution view of the whole scan. It then segments the part in views around its frame normal at the scanner’s resolution, so covering the whole scan no longer costs resolution. Because every part arrives centered, facing the camera, and at a fixed scale, one segmentation network can serve different kinds of parts and learn from few labeled masks. We test FrameSeg3D on two kinds of parts, nails and teeth. It outperforms state-of-the-art methods on hand scans (mean Dice 0.970 on nails) and on the Teeth3DS+ test fold (per-tooth IoU 96.5), with the most accurate tooth boundaries (boundary IoU 70.5).
[CV-159] RetiDual-Select: Reliability-Aware Specialist-Generalist Routing for Retinal OCT under Image-Quality Shift
链接: https://arxiv.org/abs/2611.22117
作者: Zihao Han,Philippe De Wilde
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 4 figures, 4 tables, 42 references. Submitted to the IEEE for possible publication. Main article only; the separate Supplementary Information is not included
Abstract:Adaptive fusion can repair a static ensemble’s errors, but can also overturn correct decisions. RetiDual-Select combines retinal specialist RETFound with the image encoder of the medical vision-language model MedSigLIP, separating candidate fusion, routing safety and predictive review. Calibrated product-of-experts (PoE) fusion is the static reference, a small gate proposes an alternative for each image, a validation-frozen harm guard decides whether that candidate may replace the reference, and a separate score prioritizes review. Static PoE reached the highest or joint-highest clean accuracy on all three datasets. At a clean 2% accepted-error target the review policy accepted 94.3% of Kermany-4 images at 99.89% selective accuracy, and 92.4% of OCT-C8 at 98.65%. Both branches were correct on 95.0% and 91.8% of clean images; only 4.4% and 6.2% had exactly one correct branch. Across 15 prespecified synthetic conditions, adaptive gains were concentrated in severe contrast reduction, where this routing opportunity widened roughly ninefold; percentile rescaling largely removed the effect, linking it to compressed dynamic range. In post-hoc validation-only analyses, gain was higher when the gate followed the more reliable branch, and a frozen guard cut the worst-condition OCT-C8 loss from 1.39 to 0.36 percentage points. Adaptive routing was useful when image quality created a reliability opportunity and the gate followed it; otherwise the system retained Static PoE or deferred for review. The findings are bounded by the synthetic perturbations evaluated and do not establish robustness to natural acquisition or scanner shift.
[CV-160] Diagnosing and Correcting Skin-Tone Color Casts in Ultra-Lightweight Photo Retouching Networks ACCV2026
链接: https://arxiv.org/abs/2611.22106
作者: Choulhyouc Lee
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to ACCV 2026. 32 pages including supplementary material
Abstract:Color casts are expected to come from the scene or camera, and a retouching model removes them. We report a case where the model leaves one: an ultra-lightweight retouching network renders human skin greener than the original because it under-restores the skin chroma that a degradation removes and, even on clean inputs, adds a small hue shift of its own. A common paired-data-free recipe, synthetic self-supervision (degrade high-quality images, learn to restore them), silently produces this cast while whole-frame PSNR/SSIM stay essentially unchanged. In a representative 6,825-parameter histogram-conditioned network, the cast appears in 52 of 70 portraits under a ground-truth-anchored, white-balance-free protocol. Both a histogram-conditioned and a CNN-conditioned network develop the cast under synthetic self-supervision, yet both are near-neutral in-domain when trained on MIT-Adobe FiveK expert pairs. Image-level objectives miss it: a global-mean green loss worsens the cast (52/70 to 57/70). With zero added parameters, our per-pixel, skin-masked, asymmetric green-excess loss reduces the cast to 1/70 across stratified skin tones (68.1% to 2.2% over the 70,000-image FFHQ corpus; 64.4% to 0.8% on held-out CelebA-HQ photos). In-domain FiveK PSNR/SSIM are unchanged (+0.13 dB, +0.004), at the cost of a small whole-frame CIEDE2000 increase (8.77 to 9.69) and a magenta over-correction where skin is nearly neutral (skin Delta a* -1.5 to +4.4 and CIEDE2000 5.6 to 6.4 on clean inputs): the loss targets the signed green-magenta axis, not total skin-color fidelity. The same loss suppresses the cast across three ultra-light architectures. A whole-frame green metric correlates with skin-region CIEDE2000 at the method level (Spearman rho = 0.89). The final network exports to a 35 KB TFLite model.
[CV-161] CueSplat: Learning to Schedule Sparse Deformation for Dynamic Gaussian Splatting on Edge Devices
链接: https://arxiv.org/abs/2611.22095
作者: Qingyuan Tang(1),He Zhu(1),Shixuan Sun(1),Anbang Wu(1),Jingwen Leng(1),Minyi Guo(1),Yu Feng(1) ((1) Shanghai Jiaotong University)
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 10 pages, 6 figures. Main paper only
Abstract:Deformation-based 4D Gaussian splatting (4DGS) enables dynamic scene rendering. However, evaluating deformation networks for all Gaussians at every frame is compute-intensive on edge devices. We find that Gaussian trajectories are often locally predictable, yet prediction reliability varies across Gaussians and time. Thus, we propose CueSplat, a post-training framework that exploits local temporal regularity while keeping the pretrained backbone frozen and retaining its Gaussian set. Specifically, CueSplat predicts Gaussian attributes from recent history and uses a lightweight learned router to select exact deformation updates within a scene-specific budget calibrated offline. Selection accounts for prediction reliability and rendering contribution. Selected indices are reused across nearby frames to reduce routing overhead, while selected Gaussians still receive exact updates each frame. To control accumulated prediction error, we use drift signals to schedule three consecutive full-deformation frames that rebuild the prediction history. Evaluated across four mainstream 4DGS algorithms, CueSplat achieves up to 6.74\times and 8.15\times speedups on Jetson AGX Orin and Orin NX, respectively, with a negligible quality drop (at most 0.16 dB in dataset-average PSNR).
[CV-162] When AUC Meets Generative Reconstruction: A Unified Principle for Balanced Visual Representation Learning
链接: https://arxiv.org/abs/2611.22094
作者: Boyu Han,Qianqian Xu,Shilong Bao,Zhiyong Yang,Qingming Huang
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Multimodal large language models (MLLMs) require visual encoders that extract fine-grained visual features while constructing semantically disentangled and discriminative representations. Conventional methodologies encounter a fundamental trade-off: contrastive learning enhances inter-class discrimination by suppressing intra-class variability, whereas generative reconstruction emphasizes perceptual fidelity but lacks explicit comparative supervision. The naive aggregation of these objectives via composite loss functions often induces optimization conflicts and instability due to gradient interference within the shared latent space. To resolve this, we introduce a unified representation-learning paradigm that reformulates generative reconstruction as a discriminative ranking problem within the generative output manifold. Our approach maintains perceptual reconstruction as a supervisory signal and integrates a ranking-based discriminative constraint by systematically ordering matched and mismatched reconstruction outputs. This formulation embeds discriminative information seamlessly into the generative process, alleviating the risk of loss competition. Specifically, we propose AUC Generative Reconstruction (AUCGR), which optimizes the area under the ROC curve (AUC) by ensuring higher scores for matched pairs than for mismatched pairs, and Partial AUC Generative Reconstruction (pAUCGR), which addresses supervision bias by focusing on mismatched comparisons within reliable feature subspaces. The proposed objectives are designed to be plug-and-play, requiring no architectural modifications to existing generative reconstruction frameworks. Theoretical analysis establishes rigorous connections to AUC/pAUC optimization and provides generalization guarantees. Extensive experiments validate that our method advances both discriminative accuracy and fine-grained perceptual representation.
[CV-163] When Lower Harm Misleads: Auditing Concept Corrections
链接: https://arxiv.org/abs/2611.22084
作者: Tianyou Lai,Wentao Yue,Hongji Li,Chaoyuan Hao,Qilei Li
类目: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:An expert who corrects a concept prediction, such as a BI-RADS descriptor in breast ultrasound, expects the diagnosis to improve. Objectives that reduce the harm of such corrections are judged by harm and accuracy, yet harm also falls when a model responds less to every correction or starts from a worse uncorrected prediction, the reference for harm. We propose a correction audit, a reproducible evaluation protocol that separates these explanations: it splits each correction into harm and useful gain on identical events, separates the starting prediction from the correction response, and compares objectives with amplitude baselines matched in source harm. On three breast-ultrasound sources, worst-case correction training, evaluated from the mean-case model’s start, lowers harm but gives up more useful gain in all 21 source comparisons pooled over six seeds, raising the corrected negative log-likelihood (NLL); from the same start, selecting between the objectives by harm picks the less helpful model on all three sources. On CUB, a monotonic harm penalty at its reported weight lowers path-level harm in all 80 concept sets while raising the NLL in all 80, almost entirely before any correction. At a fixed start, rescaling both responses to the same source harm reproduces the penalty’s mean harm, useful gain and NLL on single-concept corrections within a prespecified margin of 0.0005 nat, although the two responses point in different directions. Lower harm therefore signals better corrections only when it is read together with useful gain, absolute risk and the start it is measured from.
[CV-164] CRA: Continuous Risk Assessment for Safety-critical Visual Navigation
链接: https://arxiv.org/abs/2611.22080
作者: Lihao Zheng,Ying Wu,Chen Jiang,Randy Freeman
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV); Systems and Control (eess.SY)
备注: Project page: this https URL
Abstract:Traditional safety-critical control methods often rely on predefined constraints that separate states into safe and unsafe regions. While recent studies have extended geometric safety toward semantic safety, real-world safety is rarely purely binary. The risk associated with an object or human can vary continuously with its physical properties, like motion, and interaction context; for example, a person carrying a heavy object or a hot object moving at high speed may require a different safety response from an ordinary static obstacle. In this paper, we propose a vision-based continuous risk assessment framework for safety-critical visual navigation. Our method uses RGB and LiDAR observations to estimate four risk factors (kinetic, potential, contact-severity, and thermal) and combines them into a continuous risk score for each assessed object or person. These estimates are integrated into a Poisson safety function with a Laplace guidance field as a safety filter, enabling online monitoring and intervention in open-world robot navigation. We demonstrate the proposed method in a simulated restaurant scene.
[CV-165] Segmentation-Guided Direct Capacity Estimation of Rooftop Photovoltaics from Aerial Imagery
链接: https://arxiv.org/abs/2611.22076
作者: Guibin Chen,Emma Fan,Yuzhang Lin
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Distribution planners need installation-level inventories of rooftop photovoltaic (PV) capacity, but interconnection records may be incomplete or spatially aggregated. Conventional image-based estimators segment PV arrays, convert the mask to physical area, and multiply by an estimated power density, coupling capacity prediction to mask-derived area. We propose Direct Head, a segmentation-guided architecture that retains U-Net segmentation and soft-mask pooling while directly regressing nonnegative capacity from PV-region features and area descriptors. We evaluate three independent runs on 7,590 French rooftop installations under the original split and a department-based holdout. Relative to a parameter-matched area-density neural model, Direct Head reduces mean absolute error by 8.7% and 12.5%, and symmetric mean absolute percentage error by 14.0% and 20.7%, respectively. Comparisons with conventional estimators, controlled architecture variants, auxiliary-loss ablations, and representative cases show that segmentation provides useful spatial evidence without requiring capacity to be determined by an area-density product.
[CV-166] Data-Efficient Algal Bloom Segmentation in Freshwater Bodies Using a Three-Stage Transfer Learning Pipeline with Attention-Enhanced Decoding
链接: https://arxiv.org/abs/2611.22064
作者: Albert Ansah Benson,Mingqiang Han,Rahul Harsha Cheppally,Treman Singh Kaloya,Menard Soni,Sudan Baral
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 10 pages, 5 figures, 6 tables
Abstract:Harmful algal blooms threaten water quality and public health, yet automated segmentation of bloom regions in freshwater imagery is limited by scarce pixel-level annotations. We present a transfer-learning pipeline that adapts a Swin Transformer U-Net with SCSE decoder attention from ImageNet-1K to water-body segmentation (RIWA, LaRS) and then to algal blooms, together with a model-assisted annotation protocol that expands the labelled algae corpus from 145 to 655 images. In a single-seed ablation on a 41-image validation split (checkpoint S2-A4 and its variants), corpus expansion is the dominant factor (+11 mIoU points); water-body pretraining helps only on the expanded corpus (+1.1 points over ImageNet initialization); SCSE attention adds 0.3 points; and a learnable positional encoding lowers mIoU by 0.6 points and is omitted. Adding 476 images, 188 of them algae-free, reduces false-positive area on non-water scenes from 22.6% to 4.9%. Against nine baselines on a 256-image test split, a separately trained checkpoint of the same architecture (S2-A13) reaches mIoU 0.919 and the highest algae IoU (0.902); a near-duplicate audit indicates train/test overlap in this split, so these scores are likely optimistic.
[CV-167] FM-Mamba: An Efficient Native 2D Selective State Space Model with Stability-Aware Discretization
链接: https://arxiv.org/abs/2611.22061
作者: Shengyuan Hu,Yiyong Sun,Weihang Shan,Yongzhe Chang
类目: ystems and Control (eess.SY); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:When applied to two-dimensional features such as images, Mamba-based methods either flatten the feature map into 1D sequences and fuse multiple scans, which breaks spatial adjacency and multiplies computation, or recurse directly on the 2D grid but rescale the recurrence with a fixed input-independent factor and aggregate only the predecessors on a single anti-diagonal, leaving the discrete step unconstrained by 2D stability theory. We propose \textbfFM-Mamba, a native 2D selective state space model built on the general Fornasini–Marchesini (FM) model. Its coupled recurrence evolves along both spatial axes; cross-axis coupling induces a discrete diagonal-predecessor term \tilde A_0 , and two independent input-dependent step sizes \Delta_1(i,j) , \Delta_2(i,j) govern propagation along the two axes, so that the coefficients adapt per position and per direction. A normalized ratio r_k=a_0,k/(a_1,ka_2,k) yields a closed-form piecewise bound on the two step sizes; when a configuration falls outside the stable interval the steps are projected onto this bound, and the criterion is re-verified per position after training, which suppresses systematic local amplification without global damping. The recurrence is executed by a dual-diagonal wavefront scan: each update gathers information from two anti-diagonals, and the input enters the current state through a single projection, so that neither training nor inference flattens the feature map. Code is available at this https URL
[CV-168] RiDEx: Reward-Informed Directional Exploration for Diffusion Reinforcement Learning
链接: https://arxiv.org/abs/2611.22052
作者: Song Yan,Chenfeng Wang,Wei Zhai,Xingchen Wang,Tao Zhang,Yawen Shao,Shi Pan,Min Li,Zheng-Jun Zha
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Reward-based diffusion post-training updates model parameters from task rewards. While existing approaches improve sampling and policy optimization, they still rely largely on isotropic random perturbations in a high-dimensional latent space for exploration. Such variation often spends the perturbation budget on changes with little influence on the target reward, making exploration inefficient. We study where diffusion models should explore and how to learn from that exploration. We introduce Reward-Informed Directional Exploration (RiDEx), which concentrates local exploration in reward-relevant subspaces of the denoising velocity field. Conditional residuals or differentiable reward gradients propose the subspace, and equal-and-opposite perturbations at the same generation state measure its terminal reward response. The reward difference supplies signed local supervision in the explored coordinates, while all generated images contribute to global DiffusionNFT learning. RiDEx thereby couples exploration-space selection with feedback about the changes it produces. On Stable Diffusion 3.5 Medium, RiDEx attains the highest peak rewards among the evaluated methods on four alignment and preference tasks. Its gradient-direction instantiation achieves 15x and 29x target-attainment speedups over DiffusionNFT on HPSv2 and PickScore, measured in H200 GPU hours. Controlled experiments confirm that reward-aligned directions elicit stronger responses at equal perturbation magnitude, and that applying their measured credit improves curve-average reward across all four tasks.
[CV-169] BATMoS: Budget-Aware Guidance Allocation for Efficient Video Diffusion
链接: https://arxiv.org/abs/2611.22028
作者: Zhixu Du,Kuo Yang,Jingwei Sun,Hai “Helen” Li,Yiran Chen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:The effectiveness of a video diffusion model’s inference budget depends on where computation is spent along the denoising trajectory. Across eight equal-budget inference configurations, VBench total scores span 4.80 points, highlighting the importance of explicit allocation design. Our key insight is that guidance evaluations are not equally valuable across diffusion time: retaining the sampling updates while selectively reusing guidance can preserve generation quality with fewer denoiser evaluations. We introduce BATMoS (Budget Aware Temporal Mixture of Strategies), which routes between fresh classifier-free guidance and cached guidance using a single pretrained denoiser. A magnitude-based temporal rule assigns guidance reuse to low-magnitude blocks and reserves fresh guidance for the remaining blocks, with a single integer controlling the computation budget. The resulting schedule requires no additional router training or per-budget threshold search. Evaluated on CogVideoX-2B with 4,720 videos per configuration across all sixteen official VBench dimensions, BATMoS achieves the highest total score among the eight allocation policies and comes within 0.10 points of full classifier-free guidance using 20% fewer branch evaluations. BATMoS also composes with feature reuse, reducing computation from 80 to 62 effective branch evaluations with a 0.11-point decrease in total score. These findings make temporal guidance allocation a practical mechanism for extracting more generation quality from a fixed inference budget.
[CV-170] Where Should the Lamp Go? Ambient-Aware Lamp Placement for Portraits
链接: https://arxiv.org/abs/2611.22005
作者: Yuyan Chen
类目: Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR)
备注: 15 pages, 12 figures
Abstract:Portraits shape first impressions, and lighting is an important factor. Prior work, such as lighting estimation and portrait relighting, edits lighting after capture but does not guide where to place a light before the shot. This leaves an open question: where should a light go to best light a face? Therefore, we propose LightMatch, a preference-grounded inverse lighting framework that disentangles the controllable light from ambient illumination and optimizes its placement. We first learn a Bradley-Terry quality model from pairwise human judgments. A perception head then regresses spherical-harmonic lighting from frozen self-supervised features via cross-attention. Finally, a placement solver isolates the light via spherical-harmonic linearity and optimizes its reachable position. On held-out environments, LightMatch attains 87.8% of the oracle improvement, outperforming the conventional solver that ignores ambient light by a relative 28.4% and the same pipeline with the widely used DPR estimator by a relative 70.5%. It reaches 73.2% on real OLAT faces of unseen identities and, on our real captures, cuts lamp-direction error by 37.7%.
[CV-171] Beyond Appearance: Where Forest Visual Place Recognition Actually Fails
链接: https://arxiv.org/abs/2611.21996
作者: Blake Matheson,Ben McShanag,Isabella Moffat,Michael Milford,Peyman Moghadam,Dana Kulić,Somayeh Hussaini,Kavindie Katuwandeniya
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 10 pages, 6 figures, accepted to the Australasian Conference on Robotics and Automation (ACRA)
Abstract:Visual place recognition (VPR) in forest environments remains challenging because repetitive vegetation, and changes in illumination, season, and viewpoint can make distinct places appear similar and revisited places appear substantially different. In this paper, we investigate whether temporal, linguistic, structural, and geometric context can improve on an appearance-only descriptor, and what limits each representation when it does not. Using MegaLoc as the appearance-based baseline on the WildCross benchmark, we evaluate score-space and descriptor-space sequence matching, vision-language captions and reranking, scene-graph representations, and depth-derived stem graphs across same-direction revisits, extended routes, and reverse traversals. On a same-direction revisit, both sequence matching strategies reach similar Recall@1 (94.1% vs. 91.5%), but recall at 100% precision rises from 50.9% (single-frame MegaLoc) to 71.6% in score space and falls to 36.5% in descriptor space. The remaining context representations provide no consistent improvement over the appearance-based baseline. Same-place captions show limited place-specific discrimination, reranking is constrained by the initial candidate shortlist, and stem-graph matching remains weak even with LiDAR depth.
[CV-172] Deciding Without Pointing: A Case Study of Whether a Decision Model Localizes the Anomalies It Detects
链接: https://arxiv.org/abs/2611.21995
作者: Hsiang-Hung Lee
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 12 pages, 5 figures, 14 tables. Preprint of the technical part of an MSc thesis (Politecnico di Milano). Code and pre-registrations are held privately until venue submission
Abstract:Decision models can answer a typed question about an image with a probability per option, without generating text. In our case study of Clef-flash (post-trained Qwen3.5-9B backbone) on industrial inspection benchmarks, we ask whether it localizes the anomalies it detects. We obtained uneven results with zero-shot detection: per-category image AUROC 0.924 on MVTec AD and 0.668 on MVTec AD 2. On the former, the decision head’s routing attention produces 0.391 mean per-image pixel AUROC, and hiding the image tokens from it leaves detection unchanged. The multimodal LLM’s backbone attention does point at the defect, however it falls below a center prior when we change question. It falls below chance if we remove textual context in the question, partly due to backbone attention leaving the object, while the head’s map hardly changes. The backbone’s states carry the location when probed with normal images, on the test-split images, however, the decision depends on the defect region. A larger model, Clef, that comes with a post-trained Qwen3.8-27B backbone, gives no useful map in the decision head either. We found that it’s possible to recover anomaly location by asking question specifically on an image region, retraining the head or reading the backbone, but none matches dedicated few-shot localizers.
[CV-173] DEFTS: Data-Efficient Training-Free Few-Shot Learning for CLIP in the Text Subspace
链接: https://arxiv.org/abs/2611.21993
作者: Yijin Zeng
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (stat.ML)
备注:
Abstract:Contrastive Language-Image Pre-training (CLIP) achieves impressive zero-shot image classification performance by comparing image embeddings against class-specific text embeddings. While recent few-shot methods leverage labeled training samples to improve upon this zero-shot baseline, they often predominantly rely on image embeddings and underutilize the semantic structure of the text embeddings. In this paper, we introduce DEFTS, a data-efficient, training-free few-shot classification method strictly in the text subspace. We first demonstrate empirically that while raw text embeddings can be suboptimal as direct classifiers, the linear subspace they span remains highly discriminative. DEFTS therefore proposes to reduce the image-text modality gap by projecting estimated image prototypes directly into this subspace. Importantly, by expressing these prototypes as linear combinations of text embeddings, DEFTS reduces a high-dimensional estimation problem to the derivation of just a few coefficients. These coefficients are then computed via a Bayesian posterior mean, using a prior naturally centered on the original text embedding coordinates. Extensive experiments across 12 image classification benchmarks demonstrate that DEFTS significantly improves upon existing training-free methods.
[CV-174] Pathology Randomization for Synthetic Priors and Zero-Shot Detection
链接: https://arxiv.org/abs/2611.21975
作者: Alexandra G. Roberts,Tonmoy Hossain,Mark Olchanyi,Stefan Brandstätter,Jonathan Williams Ramirez,Ana Lawry Aguila,Matthew Rosen,Annabel Sorby-Adams,W. Taylor Kimberly,Juan Eugenio Iglesias
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Image and Video Processing (eess.IV)
备注: 25 pages, 9 figures, 13 tables (8-page main text with appendices). Code: this https URL
Abstract:Hospital brain imaging archives hold a diverse array of pathology at sample sizes beyond prospective studies, particularly for underrepresented populations and diseases. To unlock uncurated clinical data and scale analysis, methods are needed to detect abnormalities across contrasts and resolutions. Synthetic data has provided robust, contrast-agnostic models for a variety of human neuroimaging tasks in healthy anatomy, but pathology detection and other tasks in clinical settings still require real patient data, contrast-specific models, or disease-specific generators. We propose PathoDraw, a generative model of brain MRI with pathology. A conditional distribution of lesions given the healthy anatomy is drawn into the label map before contrast-agnostic rendering, and a single network estimates healthy anatomy with exact signed intensity deviation of any input scan. The contrast-agnostic 3D UNet learns the healthy labels under the lesion and the signed deviation directly and identifies pathology via robust tissue statistics and learned features, despite no training on real pathology. We show zero-shot detection on public stroke, tumor, and white matter hyperintensity datasets as well as archived clinical data, spanning T_1 w, FLAIR and T_2 w at various resolutions under a recent unsupervised anomaly detection benchmark protocol. Our single model exceeds or performs comparably to contrast-specific anomaly detectors trained on real scans and keeps healthy anatomy intact. The same generator trains networks for super-resolution and for registration in the presence of pathology. Our model, robust across acquisition and disease, enables the use of uncurated, archived data for clinical analysis. Our code is available at this https URL.
[CV-175] How Many Persistent Features Does a Biometric Need? A Law Linking Equal Error Rate to Feature Count and Persistence
链接: https://arxiv.org/abs/2611.21973
作者: Oleg Komogortsev,Mehedi Hasan Raju
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:A biometric designer controls the number of features and their temporal persistence. Earlier empirical work linked the equal error rate (EER) to persistence and to the required number of features by straight lines. We derive the law behind these regularities. For N independent standardized Gaussian features with test-retest correlation \rho , scored by cosine similarity, Fisher’s z -transform turns the genuine and impostor scores into two normals, giving \mathrmEER=\Phi(-\frac12\sqrtN,\operatornameatanh\rho) and, inversely, N=(2q/\operatornameatanh\rho)^2 with q=\Phi^-1(1-\mathrmEER) ; intercorrelated features enter through an effective count N_\mathrmeff . The central extension lets persistence vary across comparisons with spread \tau on Fisher’s scale: \mathrmEER=\Phi(-m/(\sqrt\tau^2+v+\sqrtv)) with v=1/(N_\mathrmeff-2) , which for a normal spread has a floor \Phi(-m/\tau) that no number of features can lower. With no fitted parameter the simple law reproduces published simulations and the EER-ICC line, and predicts the KCC-EER relation of eye-movement embeddings on GazeBase ( R^2=0.88 ), though less well on GazeBaseVR ( 0.63 ). Read from the genuine and impostor scores, the general form matches the EER of eye movements, keystroke dynamics, the ECG and face embeddings over 101 conditions with R^2=0.98 and a median error of 0.6 points, against 0.87 and 1.9 points for the simple law. The 61 eye-movement, keystroke and ECG conditions check consistency on the same scores; the 40 face conditions use held-out subjects, with the same accuracy. The product \tau\sqrtN_\mathrmeff-2 gives a design rule: adding or decorrelating features pays fully while it is well below one and little once it exceeds about three, where only higher or more uniform persistence helps.
[CV-176] SCAN: Site-Aware Collaborative Privacy-Preserving Adaptation of Vision-Language Models for Neurological Disease Classification
链接: https://arxiv.org/abs/2611.21955
作者: Nikhil J. Dhinagar,Rishab Pendam,Harmony S. Cen,Sophia I. Thomopoulos,John P. John,Mahir H. Khan,Sook-Lei Liew,Erik Kaestner,Carrie McDonald,Paul M. Thompson
类目: Computer Vision and Pattern Recognition (cs.CV); Image and Video Processing (eess.IV)
备注:
Abstract:Worldwide neuroimaging data could strengthen diagnostic foundation models, but privacy rules often prevent centralizing brain MRI scans. We present SCAN, a site-aware collaborative adaptation framework for Neurological Disease Classification that lets new sites adapt a 3D medical vision-language model (VLM) without retraining or sharing patient scans. Each site keeps exact class prototypes for local prediction and can share privacy-protected class summaries: norm-bounded sums and counts perturbed once with calibrated noise. Using 7,641 scans from 67 sites across 12 countries, covering Alzheimer’s disease (AD), stroke, epilepsy and controls, we assess local adaptation, cross-site benefits, and effects on existing sites. At seven unseen sites, the frozen VLM’s zero-shot AUCs were 0.833-0.945 for AD and stroke. Training-free local prototypes improved epilepsy AUC at three sites from 0.598/0.625/0.508 to 0.684/0.729/0.762 and AD AUC at one site from 0.833 to 0.863, nearly matching a classifier trained at each site (mean AUC 0.757 vs. 0.759). With differential privacy ( \epsilon = 4), new sites contributed without materially reducing accuracy at existing sites (AUC change \leq 0.2%), although gains from sharing were limited, and membership attacks performed near chance. SCAN offers hospitals a worldwide framework to adapt and share brain MRI VLMs without exchanging scans or retraining.
[CV-177] Benchmarking Deep Learning Models for Wheelchair Configuration Classification in Accessible Automated Vehicle Applications
链接: https://arxiv.org/abs/2611.21945
作者: Samir Ahmed,Shaaz Saleem,Devin Dhami,Wesley Maia,Angel Martinez-Sanchez,Parthib Roy,Jannatul Ferdous Srabonee,Ahmed Sabbir Arif,Ross Greer
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Robotics (cs.RO)
备注:
Abstract:Making automated vehicles accessible requires consideration of vulnerable road users, including wheelchair users. To offer a more tailored experience that supports independence for this group, vehicle design needs to accommodate the wide range of wheelchair configurations. However, wheelchairs are individualized, varying in frame geometry, propulsion, and visual appearance. Moreover, occupants or accessories may obscure visual cues. Existing mobility-aid datasets only label the presence of a wheelchair, rather than describing its specific configuration. To fill this gap, this paper introduces the Wheelchair Configuration Image Dataset (WCID), comprising 467 face-blurred images across 192 capture groups solely for evaluation. Each image is annotated with (i) a two-class label of manual or powered, (ii) a chair subtype, (iii) occupancy, viewpoint, setting, lighting, occlusion, propulsion, and control, with a 273-image validation split and 194-image test split. Three convolutional neural network backbones and a frozen vision-language probe are trained on an external image dataset and evaluated on WCID to measure cross-dataset generalization, while also testing robustness using Gaussian blur. Convolutional models do well on clean images, but performance drops under strong blur, leaving a gap before deployment. Code, split assignments, labels, trained weights, and the face-blurred images is available at: this https URL
[CV-178] Does the Model Use What It Cites? Counterfactual Tests of Grounded Reasoning Chains in Small Video-Language Models
链接: https://arxiv.org/abs/2611.21940
作者: Jugal Gajjar,Kamalasankari Subramaniakuppusamy
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 24 pages, 4 figures, 27 tables. Code: this https URL
Abstract:A grounded reasoning chain cites a frame span and a box for every step, so a reader can check where the model looked. Whether the answer depends on the cited region is a different question. We test it by intervention: the regions a model cited in its own chain are blacked out, and the share of correct answers that flip is compared with blacking out boxes of the same size elsewhere on the same frames. The paired difference, the cite gap, needs no annotation. We apply it to 2-3B video-language models trained on exact CLEVRER chains and on pseudo-labelled NExT-QA chains. With the same weights, the cite gap is 0.089 on CLEVRER and 0.003 on NExT-QA. Pooled over models, masking the cited boxes flips 2.2 times as many correct answers as the control on CLEVRER and 1.1 times as many on NExT-QA, also among the answers that depend on the video at all, while the boxes pass a grounding check on both. Masking the cited entity on every frame, instead of the cited box, flips 1.9 times as many answers as its control. The NExT-QA chains were supervised to cite two frames of entities that stay visible on most frames, so this outcome is what the supervision predicts; the point is that the usual check on grounded chains cannot see it. A citation can be verifiable and relevant without the answer depending on it, and a test of grounded reasoning should remove what a chain names, not only where it points.
[CV-179] HubEgo-100K: In-the-Wild Egocentric Video of Skilled Work for Robot Learning
链接: https://arxiv.org/abs/2611.21938
作者: Javad Amirian,Armin Kiani,Loïc Schaeffer,Tim Sprecher,Mohamed Taha,Karthik Yadati
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 22 pages, 18 figures, 7 tables. Dataset: this https URL
Abstract:Robot policies trained by imitation generalize poorly outside the environments and objects seen in training. Published data-scaling studies find that generalization grows with the diversity of training environments and objects rather than with more demonstrations per environment. Robot data is costly to collect across many environments, whereas egocentric human video can be recorded in the wild at scale; existing egocentric video, however, mostly shows daily-life activity and contains little skilled work in real workplaces. We present HubEgo-100K, a dataset built for diversity. This version of the paper covers the first release, v1.0: 10,124 hours of 1080p egocentric video with synchronized 100 Hz inertial (IMU) data from one head-mounted phone camera. In v1.0, 12,227 contributors record their own unscripted work in 15,548 environments (each one contributor’s own setting of one type, such as their kitchen or workshop) across 11 setting types, 30 work categories, 56 task types and 35 action families, with 83% of hours in workplace settings. The release comprises 72,754 episodes with machine-generated action and language annotations, in LeRobot and MCAP formats under CC BY 4.0. We measure diversity along each axis, show that inside v1.0, with episodes in random order, new contributors and words keep appearing, and relate the measurements to published evidence on six data properties tied to policy performance, including hand visibility and inertial sensing. We also document collection, curation, privacy measures and data format, and provide a datasheet. Later versions of this paper will add benchmarks and cover further releases toward 100,000 hours.
[CV-180] GridDriver: On-Board BEV-Based Language Planning and Reasoning
链接: https://arxiv.org/abs/2611.21924
作者: Katharina Winter,William Engel,Zain A. Alishammat,Fabian B. Flohr
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Bird’s-Eye-View (BEV)-conditioned large language models (LLMs) enable language-interactive driving, but onboard deployment requires efficient perception and continuous planning alongside autoregressive language generation. We present GridDriver, an efficient BEV-conditioned LLM framework combining trajectory prediction, structured driving indicators, natural-language navigation, driving explanations, and scene question answering. Lightweight spatio-temporal BEV perception and spatially structured tokens provide scene context to a shared LLM, while frame-wise scheduling interleaves trajectory prediction with asynchronous language generation. GridDriver achieves competitive performance on public benchmarks, reaching 87.9% PDMS on NAVSIM and 57.6% accuracy on nuScenes-QA with a shared planning and QA checkpoint. Spatial representation analyses demonstrate improved geometric information retention and trajectory prediction compared with Q-Former compression. The system achieves 32.7 ms standalone planning latency and 9.4 plans/s in an onboard ROS 2 pipeline. We further evaluate planning on our own data recordings and demonstrate open-loop onboard operation on a research vehicle. Across public benchmarks and real-world evaluations, we identify strengths and limitations of the BEV-based approach.
[CV-181] Wrong Input Wrong Threshold: Re-evaluating Adversarial Facial Privacy
链接: https://arxiv.org/abs/2611.21923
作者: Tunahan Parlayıcı,Erchan Aptoula,Yücel Saygın
类目: Computer Vision and Pattern Recognition (cs.CV); Cryptography and Security (cs.CR); Machine Learning (cs.LG)
备注: 10 pages main text and 15 pages supplementary material. Submitted to IEEE Transactions on Information Forensics and Security. This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible
Abstract:Published face-privacy methods report protection above 90% against black-box recognition models. That number comes from an evaluation protocol with two independent defects, reported in a measure that cannot size either one. The input is wrong: methods published since 2022 feed the network the whole photograph, or a detector’s box, where deployed systems detect, align to a landmark template and crop. Re-scoring the same files through a realistic front end costs a standard transfer attack 56% of its dodging protection and a budget-allocating attack 7%, with no image regenerated. The measure is wrong: that loss is a difference of success rates, so it saturates, moving by 8.2x over a budget sweep while the effect beneath it moves by 1.34x. We therefore report the paired cosine shift, in which both thresholds cancel algebraically, so the input defect holds without committing to a calibration. Measured that way it is 2.2x and 10.7x larger on three unseen recognizers, and at matched budget it reverses the L-infinity transfer ranking: the one surveyed method that aligns at evaluation gains 0.18 where MI-FGSM loses 0.05. The cause is a resampling mismatch, not perturbation lost outside the crop; optimising through a differentiable alignment operator largely removes it, adding +24 to +58 points on unseen recognizers and a commercial system, with a residual on one of them. The thresholds are wrong: the four constants inherited since 2021 hold the false-accept rate near its nominal 1% on aligned crops but admit 2.3-5.4% on whole images, with an opposite-signed bias across dodging and impersonation, the two tasks reported side by side. Differentiable alignment and gallery scoring are not new: the first generation of this literature used both, and we show what dropping them cost.
[CV-182] REEL: Brain-Guided Video Generation and Selection by Reranking Encoder Embeddings with Linear Decoding from fMRI data
链接: https://arxiv.org/abs/2611.21907
作者: Fabrizio Spera,Tommaso Boccato,Moein Khajehnejad,Nicola Toschi,Matteo Ferrante
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Visual brain decoding from functional magnetic resonance imaging (fMRI) data has made remarkable progress in recent years, driven by increasingly powerful artificial intelligence (AI) models and the growing availability of large-scale public datasets. However, despite the impressive advances achieved in static image decoding, the reconstruction of videos from fMRI signals remains challenging due to the temporal mismatch between rapidly changing video stimuli and the slow dynamics of the blood-oxygen-level-dependent (BOLD) response. We evaluate a Bayesian-inspired proposal-and-rerank strategy on CC2017 and the BOLD Moments Dataset (BMD), using a linear semantic decoder, a pretrained video generator, and three neural encoding streams with ridge readouts and learned pooling in two streams. We refer to this framework as REEL (Reranking Encoder Embeddings with Linear decoding). Compared with a single generated candidate, reranking increases mean Video 50-way accuracy from 0.2359 to 0.2434 across three CC2017 participants and from 0.3653 to 0.3920 across ten BMD participants. These findings suggest an alternative and conceptually simpler strategy for improving video brain decoding, highlighting the potential of our approach as a complementary direction to increasingly complex pipelines.
[CV-183] mporal Pseudo-RGB Encoding for Label-Free Detection of VEEV Infection in Human Brain Endothelial Cells
链接: https://arxiv.org/abs/2611.21892
作者: Zhengjie Zhu,Stephanie Valdez Trefry,Hao Lu,Abbas Alili,Yongxin Guo,Aarthi Narayanan,Metin Nafi Gurcan
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 20 pages. Supplementary material is included as an ancillary file
Abstract:Label-free detection of viral infection is central to antiviral screening and pathogen surveillance. Most deep-learning methods analyze individual frames and do not model temporal progression. We evaluate temporal pseudo-RGB encoding, which stacks three time-offset brightfield frames ( t-6 h, t-3 h, t ) as the red, green, and blue channels of a single image so that morphological change becomes color and an ImageNet-pretrained convolutional neural network (CNN) can be trained on infection trajectories without architectural modification. In human brain microvascular endothelial cells (HBMVEC) infected with Venezuelan equine encephalitis virus (VEEV) TC-83, a multi-task network predicts infected-versus-mock status and time post-infection in one forward pass. Across three experimental runs under leave-one-row-out cross-validation, with all 36 wells scored out of fold, the encoding reaches 91.6% accuracy against 87.4% for the single-frame model (paired difference +4.2 percentage points, 95% CI [+2.5, +6.1] ; sign-flip p = 3\times10^-5 ; well-level inference, conditional on three runs). At the late-phase endpoint (40-48 h), where floating cells and debris can make conventional brightfield images difficult to interpret, the advantage widens to +23.9 points (95% CI [+19.2, +28.9] ). Time post-infection is recovered on the fixed row split with MAE = 1.31 h ( R^2 = 0.975 ). Among the temporal models tested the encoding has the highest mean accuracy, at a fraction of the cost of the recurrent and 3D convolutional alternatives, though only its margin over the single-frame model is resolved; the tested ViT-B/16 backbone showed no corresponding gain, so the finding is limited to convolutional backbones and this virus-cell system.
[CV-184] Domain-Tuned Diffusion Super-Resolution for 4K Facial Asset Maps SIGGRAPH
链接: https://arxiv.org/abs/2611.21890
作者: Elly Akhoundi,Mathieu Lamarre,Thibault Goyallon,Patrick Anderson,Étienne Danvoye
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Graphics (cs.GR)
备注: Accepted to SIGGRAPH Asia 2026 Technical Communications. 4 pages
Abstract:Physically based facial assets require 4K albedo and displacement maps for cinematic close-ups, yet generative pipelines commonly produce them at only 1K. Naively applying photographic super-resolution is a poor fit: facial UV maps have domain-specific statistics, and displacement is a scalar height field whose errors directly distort geometry. We present a domain-tuned diffusion upscaler for 4x (1024 to 4096) facial asset super-resolution. Starting from a bilinearly upsampled map, we adapt a frozen SDXL backbone per modality with lightweight LoRA modules, while a Tile ControlNet preserves input structure as the diffusion prior restores pore- and wrinkle-scale detail. On held-out maps, our method produces sharper, more faithful detail than bilinear and off-the-shelf GAN and diffusion super-resolvers, reducing albedo LPIPS from 0.21 to 0.08 while avoiding misplaced or excessive displacement relief. These results show that lightweight domain adaptation can produce render-ready 4K facial asset maps without a custom super-resolution architecture.
[CV-185] FaceRiff: Data-Efficient Unified Text-to-3D Face Asset Synthesis via Geometric Conditioning SIGGRAPH
链接: https://arxiv.org/abs/2611.21867
作者: Elly Akhoundi,Mathieu Lamarre,Thibault Goyallon,Patrick Anderson,Étienne Danvoye
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Graphics (cs.GR)
备注: Accepted to SIGGRAPH Asia 2026 Conference Papers. 11 pages, 11 figures, 2 tables
Abstract:High-resolution facial datasets with consistent UV-mapped geometry, albedo, and physically based displacement are scarce, yet such assets are essential for photorealistic digital-human rendering. Existing 3D face generation methods are often slow, single-modality, or require large specialized datasets. Meanwhile, large image diffusion models encode strong facial priors but do not natively produce lighting-independent textures, pore-scale displacement, or UV-aligned geometry. We present FaceRiff, a unified, data- and parameter-efficient framework for generating complete 3D face assets (mesh, albedo, and displacement) directly from text in a shared UV space. FaceRiff adapts a pretrained SDXL backbone with task-specific LoRA adapters and uses UV normal maps as geometric conditioning: per-instance normals guide albedo and displacement synthesis, while a dataset-mean normal provides a canonical spatial anchor for mesh generation. Using only 2,049 scans from approximately 350 identities and training on a single GPU, FaceRiff updates less than 3% of the backbone parameters per task and produces render-ready 3D face assets without per-instance optimization. Compared with FaceG2E and DreamFace, FaceRiff produces greater per-prompt diversity while preserving prompt alignment and fixed-topology UV consistency, and additionally generates lighting-independent albedo suitable for PBR relighting and pore-scale displacement absent from prior text-driven baselines. A blind study with 45 industry professionals shows that FaceRiff achieves substantially higher perceived realism than DreamFace, approaching that of real captures. Overall, FaceRiff demonstrates that pretrained diffusion priors can be adapted to unified 3D face asset synthesis under limited data.
[CV-186] PromptPack: Training-Free Augmentation of Diffusion Editors for Intent-Aligned Spatial and Semantic Control
链接: https://arxiv.org/abs/2611.21865
作者: Sophia Zalewski,Mustafa Munir,Enting Liu,Anjul Patney,David Dynerman
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注: 32 pages, 21 figures, including supplementary material
Abstract:Many image-editing tasks require confining changes to intended spatial regions while exactly preserving user-designated source content. Instruction-driven editing pipelines often leave such user-intent underspecified, while inpainting editors condition generation on explicitly defined edit support but not on the spatial assignment of individual edit concepts. PromptPack addresses both limitations by equipping diffusion editors with two explicit ‘Edit Guides’: an edit-role partition that labels each pixel as ‘Keep,’ ‘Permit,’ or ‘Modify’ to specify where edits may occur, and a regional-prompt assignment that maps spatial regions to text prompts to specify which concepts apply where. These guides may be manually specified or, for source-object edits, automatically derived from a text-grounded SAM 3 mask. Without retraining, PromptPack injects independently embedded regional prompts into the editor’s fixed-length text sequence and uses the Edit Guides to spatially mask attention, gate flow updates, and restore protected source pixels. We integrate PromptPack into FlowEdit FLUX, FlowEdit SD3, and FLUX Kontext and demonstrate improved requested-concept placement and protected-content preservation across all three integrations on PIE-Bench++. In a pairwise user study, PromptPack achieves an 87% preference rate over FlowEdit FLUX, establishing PromptPack as a practical, training-free interface for intent-aligned diffusion editing.
[CV-187] AR-WAM: efficient streaming world-action modeling for autonomous driving
链接: https://arxiv.org/abs/2611.21851
作者: Baiting Luo,Shatadal Mishra,Ahmadreza Moradipari,Nejib Ammar
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Robotics (cs.RO)
备注:
Abstract:Autonomous driving requires anticipating how traffic will evolve while responding quickly to new observations. Video foundation models offer the first ability, but repeated history encoding and iterative video generation make them costly to use for driving. We present AR-WAM, an efficient streaming world-action model that combines autoregressive temporal modeling with single-pass future conditioning. Because its video backbone is causal, AR-WAM updates its visual context incrementally, reusing past computation and conditioning future prediction on observed motion. We further hypothesize that future features computed from recent observations can improve generalization in dynamic traffic without iterative video generation. AR-WAM therefore obtains these features in one forward pass and uses them to condition trajectory generation, and reinforcement learning then improves the policy through direct optimization of driving rewards. With a single front camera, AR-WAM achieves 91.5 PDMS after supervised training and 94.0 after reinforcement learning on NAVSIM-v1, outperforming the compared strong end-to-end world-action models. In a dynamic setting of continuous driving, supervised AR-WAM keeps its visual history across decisions and outperforms strong world-action baselines.
[CV-188] Can a World Model Predict a Deepfake Video? A Consistency Margin Prior for Deepfake Detection
链接: https://arxiv.org/abs/2611.21846
作者: Pradip Gupta,Shubham Mandloi,Ravi Kumar Satzoda,Sujeeth Bharadwaj,Yassine Benajiba,Dan Roth
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 20 pages, 4 figures, 5 tables
Abstract:Deepfake detectors often perform well on the dataset they are trained on, but their performance drops when tested on videos from different datasets. In this paper, we study whether a predictive video model can provide a signal that generalizes better across datasets. Instead of relying only on generator-specific artifacts, as many existing detectors do, we use a model trained to predict natural video and measure the difference between what it predicts and what it observes in a video. Using V-JEPA 2.1, we keep the encoder frozen and adapt the predictor using rank-16 LoRA. We define the difference between the predictor output and the frozen encoder target as the prediction residual. We hypothesize that manipulated clips produce, on average, larger prediction residuals than authentic clips, and use this as our consistency-margin prior. We process these residuals using a lightweight Latent Consistency Head (LCH), which separates residual magnitude from direction for training and classification. To test whether the direction of the consistency margin affects cross-dataset generalization, we compare standard and reversed margin objectives while keeping the remaining training setup fixed. We match the two settings on FF++ validation performance before evaluating them on unseen data. The reversed-margin setting exhibits a larger cross dataset generalization gap on average, providing evidence that the direction of the consistency margin affects cross-dataset transfer. Our complete model adds only 8.25M trainable parameters, which is 2.47% of the total 334M parameters. Trained only on FF++, it achieves an AUC of 90.5 on Celeb-DFv2 and performs well across several other cross-dataset benchmarks. Our ablations show that separating residual magnitude from direction, reducing dataset-specific variation, and using a low-rank projection improve cross-dataset generalization in our experiments.
[CV-189] VisHead-Spec: Head-Routed Visual Attention for Multimodal Speculative Decoding
链接: https://arxiv.org/abs/2611.21839
作者: Xin Guo,Arushi Rai,Xiang Zhou,Qiang Zhang,Siyu Lin,Xiangjun Fan,Lizhu Zhang
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Speculative decoding accelerates vision-language models(VLMs) while preserving the target model’s outputs, but its speedup diminishes under high concurrency serving as the system becomes more compute bounded as the batch size increases. This problem is pronounced for multimodal speculative decoding as the long range of the image tokens in the prompt can significantly increase the computational cost during draft token prediction. The block-diffusion drafters (e.g. DFlash, DSpark) can effectively speed up the draft model efficiency by predicting multiple draft tokens in one forward pass, but their speedup for VLMs under high concurrency is still modest (9%). In order to understand this gap, we examine the information flow between different modalities from target model to draft model and identify the sparsity pattern of visual attention between the target and draft models: only small set of key-value heads in the draft model need direct visual feature attention in the context and only visual features from top level layers in the target model need to be attended in the context. We therefore propose VisHead-Spec, which selects a small, fixed set of key-value (KV) heads to attend to all image features from top level layers in the target model and restricts the remaining heads to text features. On Qwen3-VL-8B and MMSpec, VisHead-Spec-8 lets 8 of the drafter’s 40 KV heads read the image while keeping accepted length within 0.3% of the block-diffusion draft model baseline (i.e. DSpark). At batch size 64, its throughput gain over autoregressive decoding increases to 16.0%, compared with 9.5% for DSpark; at batch size 1, its speedup remains comparable to DSpark.
[CV-190] Rethinking Perceptual Similarity for Remote Sensing: Multi-Layer Attention Scoring and Human Preference Benchmarking WACV2027
链接: https://arxiv.org/abs/2611.21816
作者: Eli Underwood,Marc Bosch,Neil Joshi,Jeffrey Wen,Myron Brown
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV 2027). 9 pages
Abstract:Perceptual similarity metrics for natural images transfer poorly to remote sensing (RS), where severe geospatial and semantic failures can remain visually plausible while still being decisively wrong for downstream use. To address this, we develop a novel human-aligned perceptual quality metric for generated RS imagery and evaluate it across several generation tasks. We first conduct a controlled perturbation study where we progressively increase the severity of common synthetic perturbations. Using the layer-wise Aligned Attention Score (AAS) responses in RemoteCLIP, we find that different RS artifacts are best captured at different depths. Based on these findings, we introduce Remote Sensing Perceptual Distance (RSPD), a pairwise metric that averages seven independently calibrated layer distances without tuning on human-preference labels. To evaluate our metric on realistic RS generation tasks, we construct a blind, reference-conditioned two-alternative forced-choice (2AFC) benchmark with five 300-triplet subsets spanning synthetic artifacts, airborne and spaceborne super-resolution, SAR-to-optical translation, and timeseries controls. We also construct a 300-triplet ordered misregistration benchmark with deterministic labels. Using the same RemoteCLIP checkpoint, RSPD substantially outperforms simple cosine distance with the final global embedding and is competitive with the strongest generic baselines across the realistic RS generation tasks.
[CV-191] Sketch Dont Paint: Allocating Test-Time Compute in Generative Visual Reasoning
链接: https://arxiv.org/abs/2611.21812
作者: Sheng Lu,Guangpin Tao,Tong Lu,Ying Tai
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 11 pages, 5 figures, 3 tables
Abstract:Multimodal models increasingly reason by synthesizing pixels, sketching maze paths within interleaved chains of thought or rolling out video solutions frame by frame. Each visual thought runs the full diffusion schedule inherited from photorealistic synthesis, so iterative denoising dominates inference cost. We study how to allocate test-time compute along two axes: when to render visual thoughts rather than thinking in text or answering directly, and how many denoising steps each thought requires. Controlled interventions that vary only how the thought is rendered uncover three regimes. Causal thoughts carry the plan and become legible early: an 8-step sketch nearly matches a 50-step render in accuracy, commits to the same plan as often as two independent full renders agree, and in video models mostly fails by stalling short of the goal, so the rollout abstains. Decorative thoughts can be omitted without loss, whereas harmful thoughts lower accuracy. Because the regime depends on both model and task, always drawing is suboptimal. Across 14 model-benchmark pairs of three interleaved reasoners, answering directly, without any thinking, wins 9 pairs, and choosing among direct, text-only, and visual tiers from ten labeled examples per pair raises mean accuracy from 60.7% (always drawing) to 65.1%. Where drawing pays, we propose sketch-and-vote, which samples 8-step sketches (a step count chosen without labels) from independent noises until one answer reaches three votes. With this one setting, sketch-and-vote beats full rendering with fewer denoising steps on an image reasoner and two video reasoners, by 8.6 points on VSP spatial planning and by 9.6 and 22.1 points on mazes.
[CV-192] Data-Driven Variable-Length Compression of 3D Representations
链接: https://arxiv.org/abs/2611.21802
作者: Yvette Y. Lin,Sai Karthikey Pentapati,Sarvesh Sundaram,Guandao Yang,Sara Fridovich-Keil
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Compressing 3D shapes is important for the efficiency of many applications, yet existing methods either cannot leverage dataset-level statistics or use a fixed compression ratio for all shapes. We propose a learned variable-length compression method that exploits structure across a training dataset while adapting compressed representation size to individual shape complexity using only fast feed-forward inference. Our method operates on general latent representations and improves rate-distortion for both meshes and signed distance fields. Code is available at this https URL.
[CV-193] Every Camera Grows Up: Camera-Height Regression in Interactive Video World Models
链接: https://arxiv.org/abs/2611.21782
作者: Sheng Lu,Guangpin Tao,Tong Lu,Ying Tai
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Interactive video world models serve as simulators for embodied agents: from a starting image and an action trajectory, they generate the views the agent would see. The starting image fixes how high the camera is, for example 0.15 m for a floor robot or 2.2 m for an overhead camera, and a faithful simulator must keep this height as the camera moves. We show that six recent open-weight models fail to do so: as motion replaces the starting view with generated content, they move the camera toward human eye level, a failure we call camera-height regression. Our benchmark, EyeHeight, pairs six rendered elevations (0.15 m to 2.2 m) and four calibrated real-world rig heights with scale-free multi-view height estimation. All six models lose their initial height during a 180^\circ turn-and-walk rollout: initialized at 0.15 m, only 0-20% of clips end within 1.3\times of the starting height as inferred viewpoints converge to human heights of 1.1 m to 1.5 m, whereas ground-truth renders remain within a few percent. In Matrix-Game models, homography-tracked floor motion stays consistent with constant elevation during rotation even as newly outpainted room geometry yields a higher reconstructed viewpoint. Regression grows with rotation angle, begins while the start frame remains in context, and largely reverses in SANA-WM when a 360^\circ turn restores the initial view. We trace this pull to human-centric video priors in which floor-level views are nearly absent: in SANA-WM, textual height cues and dense floor-disparity maps do not prevent it, whereas classifier-free text guidance during the first half of denoising strengthens it. Combining unguided sampling with height-balanced fine-tuning recovers more retention but costs 5 MUSIQ points, a loss fixed during those same early steps. Viewpoint drift alone shifts the paths predicted by visual navigation policies.
[CV-194] Video recordings and play behaviour annotations of group-housed dairy calves on two Dutch farms
链接: https://arxiv.org/abs/2611.21706
作者: Haiyu Yang,Heidi Lesscher,Enhong Liu,Miel Hostens
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Play is used as an indicator of positive welfare in dairy calves, but openly available video of group-housed calves with individual behaviour annotations is scarce. We describe video from four camera deployments of about one week each in group pens of dairy calves on two farms in the Netherlands: a research and teaching dairy farm (2024 and 2025) and a commercial dairy farm (2024). The recordings comprise 1,173 video files spanning 651 h. For eight observation days, trained observers coded calf behaviour continuously in Noldus The Observer XT with an 18-item ethogram of play and other behaviours, covering 104 h of video, 90 h of which have every calf in the pen coded. The release also contains 1.1 million decoded frames from seven annotated sessions, a per-frame timestamp read by optical character recognition from the camera clock burned into each frame, per-calf segmentation tracks and per-calf frame-level play labels. The data support studies of calf behaviour and welfare and the development and evaluation of automated behaviour recognition.
[CV-195] EgoLink: Egocentric Language-Vision Interactive Network Knowledge Challenge Summary
链接: https://arxiv.org/abs/2611.21690
作者: Renhe Sun,Yueying Feng,Bohan Yu,Tong Niu,Yunqi Liu,Zitong Wang,Haolin He,Kexin Shi,Hualei Wang,Qiuqiang Kong,Chang Yao,Yuge Huang,Weiqiang Wang,Jingyuan Chen,Jian Liu
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: Published in ACM Multimedia 2026 (MM '26)
Abstract:The EgoLink Challenge at ACM Multimedia 2026 benchmarks egocentric multimodal intelligence linking first-person perception, social reasoning, and interactive action. Unlike benchmarks centered on visible objects and activities, EgoLink evaluates whether models can interpret emotional and social cues, resolve visually grounded references, and convert user requests into verified tool-mediated actions. Its two tracks cover social reasoning in egocentric video and interactive agents that complete service tasks through dialogue, tool use, and closed-loop execution. Together, they test whether perception, reasoning, and decision-making remain coupled under both offline question answering and state-changing interaction. We highlight a multi-stage Track 1 data pipeline for multiple-choice question (MCQ) generation, chain-of-thought generation, automatic quality grading, and video-contribution filtering. We summarize the task design, evaluation metrics, rules, leaderboards, and representative solutions. The benchmark combines objective answer scoring with process- and result-based evaluation, enabling analysis of both final correctness and execution quality. The best Track 1 system achieved 0.9263 accuracy, while the best Track 2 system achieved 0.844 joint success and 0.916 micro accuracy. Participant reports reveal a shared principle: strong systems make intermediate evidence explicit through candidate verification, option-robust reasoning, scenario-aware planning, tool canonicalization, and state-aware correction before answering or acting. These findings highlight the importance of process verification and evidence-grounded reasoning for reliable egocentric multimodal intelligence.
[CV-196] Edge-Adaptive Hiera-Seg Model for Building-Integrated Photovoltaic Power Forecasting
链接: https://arxiv.org/abs/2611.21680
作者: Na Yu
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 13 pages, 7 figures, 1 table
Abstract:Building integrated photovoltaic power forecasting is challenging because facade orientation, rooftop geometry, nearby buildings, vegetation, cloud movement, module temperature, inverter status and edge device reliability jointly affect power generation. This paper proposes an edge-adaptive HieraSeg model for short term BIPV power forecasting with multi-orientation visual-irradiance fusion. RGB facade images, optional thermal cues, irradiance measurements, module temperature, facade orientation, rooftop masks, shading labels, sky condition, historical power and edge device state are aligned into synchronous forecasting windows. HieraSeg extracts hierarchical regions from facade, rooftop, sky, shadow and background images, while the edge-adaptive module changes feature weights under missing frames, latency shifts and local shading changes. The segmentation branch identifies occluded module areas and separates facade, rooftop, sky and background evidence before the forecasting head estimates future BIPV output. Experiments are designed on a simulated multibuilding BIPV dataset with unseen building and unseen orientation tests. The model reports predicted power, confidence interval, dominant evidence, occlusion contribution, orientation contribution and energy-management review priority. This output is intended to show whether the forecast is driven by irradiance change, visible shading, orientation response, stale edge information or recent power history. The framework supports local forecasting at edge controllers with limited communication delay and computing resources, linking visual occlusion, irradiance dynamics and power history to traceable forecasting results for building energy management, microgrid scheduling and BIPV operation planning.
[CV-197] Depth-Aware Activation Steering via Attention Residuals
链接: https://arxiv.org/abs/2611.21655
作者: Shilong Zhao(Independent Researcher)
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Activation steering is a simple, training-free way to modulate LLM behavior, yet existing methods typically inject vectors at a single layer and lack mechanisms for depth-aware propagation or internal verification. We combine two recent ideas - Attention Residuals (AttnRes) providing learnable cross-layer pathways, and Jacobian-lens (J-space) concept readout - into a lightweight depth-aware steering pipeline. On two small open models (Qwen3-0.6B and Qwen3.5-0.8B), distributing a persona vector according to AttnRes-style depth weights yields stronger internal alignment (deeper-layer concept cosine and rank, p0.0001) and stronger behavioral shifts than single-layer injection. The effect is sign-specific: reverse vectors invert all six metrics, random directions degrade alignment (p0.0001), and a short LoRA fine-tune erases the imprint. Injection rewires attention strictly downstream of the injection point (upstream JS = 0.00000). Three practical findings follow. First, a layer-wise geometric decomposition of cosine readouts: at the injection layer 123% of the increase is arithmetic, while 5-11 layers downstream 100% is dynamical. Second, an intensity window: effects are monotone over 0.3-0.7, then generation collapses (617 to 306 to 25 chars). Third, metric discipline: keyword-rule metrics fail LLM-judge validation (kappa = 0.09), and two rule-based conclusions were retracted. Training the depth weights on a steering objective (200 steps, frozen backbone) raises final alignment 10x on held-out prompts. Cross-persona vectors retain 96-99% of the effect: specificity holds against random, not semantic, directions. We contribute a practical, low-overhead depth-aware steering method with a careful evaluation protocol. Code and data are released.
[CV-198] Analyzing Learned Convnet Features with Dirichlet Process Gaussian Mixture Models NIPS2016
链接: https://arxiv.org/abs/1702.07189
作者: David Malmgren-Hansen,Allan Aasbjerg Nielsen,Rasmus Engholm
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: Presented at NIPS 2016 Workshop: Practical Bayesian Nonparametrics
Abstract:Convolutional Neural Networks (Convnets) have achieved good results in a range of computer vision tasks the recent years. Though given a lot of attention, visualizing the learned representations to interpret Convnets, still remains a challenging task. The high dimensionality of internal representations and the high abstractions of deep layers are the main challenges when visualizing Convnet functionality. We present in this paper a technique based on clustering internal Convnet representations with a Dirichlet Process Gaussian Mixture Model, for visualization of learned representations in Convnets. Our method copes with the high dimensionality of a Convnet by clustering representations across all nodes of each layer. We will discuss how this application is useful when considering transfer learning, i.e.\ transferring a model trained on one dataset to solve a task on a different one.
[CV-199] Uncovering the Invisible in Routine Radiology: Cross-Modal Detection of MRI-Verified Soft-Tissue Pathologies from CBCT via Physics-Guided 3D Self-Supervised Learning
链接: https://arxiv.org/abs/2611.23042
作者: Evgeny Alves Limarenko
类目: Image and Video Processing (eess.IV); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Routine X-ray modalities like cone-beam computed tomography (CBCT) are optimized for high-contrast mineralized tissues, leaving soft-tissue abnormalities buried beneath photon noise and below human visual perception. Thus, detecting non-calcified pathologies requires secondary magnetic resonance imaging (MRI). We demonstrate that sub-visual radiographic signatures of soft-tissue pathology are physically preserved within routine CBCT scans and can be extracted for opportunistic screening using a targeted, physics-guided deep learning pipeline, even under extreme paired-data scarcity. Using the temporomandibular joint (TMJ) as a model system (N = 212 paired CBCT-MRI joints), we address the severe overfitting of high-capacity 3D networks through a two-stage framework. First, we employ a frozen full-grid 3D masked autoencoder (BertsWin) pre-trained on 6,550 unlabeled CBCT volumes via the GradientConductor (GCond) optimizer and an anatomy-aware objective (PhysLoss) redirecting gradient flow into the [-300, 300] HU soft-tissue window. Second, automated localization of the condylar zenith drives a deterministic 3D cropping that eliminates 81% of background osseous variance. Evaluated via a patient-isolated 3x10-fold stratified cross-validation protocol, a statistically robust sub-visual signal for MRI-verified synovial effusion (ROC AUC = 0.793) emerges exclusively from representations pre-trained with the PhysLoss and GCond regime. Crucially, only this setup captures true 3D anatomical chirality (p 0.001) and high intra-patient structural consistency (0.791) over inter-patient noise; conversely, identical architectures pre-trained with standard L2 loss or unmasked variance decomposition (MVC) collapse to random guessing. These findings establish a generalizable paradigm for repurposing routine, low-contrast radiographic scans into opportunistic screening tools for occult soft-tissue disease.
[CV-200] A high-dimensional spatial-physical coupling operator for generalizable super-resolution of three-dimensional turbulent flows
链接: https://arxiv.org/abs/2611.22951
作者: Zengyuan Zuo,Junjun Jiang,Fei Mu,Xubo Wang,Xianming Liu
类目: Fluid Dynamics (physics.flu-dyn); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Accurate representation of turbulent flows requires simultaneously capturing multi-scale spatial structures and the complex, nonlinear coupling between distinct physical quantities (\textite.g., velocity components and density). In practice, however, turbulent data are often constrained by low spatial resolution due to hardware limitations and computational costs. While current data-driven super-resolution methods have shown promise by learning mappings via convolutional neural networks (CNNs), they typically treat physical variables as independent channels. This simplified approach neglects intrinsic cross-variable correlations, causing models to struggle with high-fidelity reconstruction, particularly under extremely sparse conditions or in out-of-distribution regimes. To bridge this gap, we propose an architecture-agnostic spatial-physical coupling operator that extends the standard convolution into a joint spatial-variable dimension. Unlike conventional linear channel mixing, our framework constructs a unified spatial-physical representation within CNNs to explicitly capture local inter-variable interactions. By lifting features into a higher-dimensional manifold, this operator significantly enhances the model’s representation capacity. Experiments demonstrate that this approach enables strong reconstruction performance and zero-shot generalization, offering a physics-aware inductive bias for super-resolution of three-dimensional turbulence. More broadly, this work suggests that high-dimensional operators offer a promising avenue for modeling intricate multi-scale turbulent flows.
[CV-201] he shape of individual differences in judgment: shared axes personal weights and when their spread crosses the consensus
链接: https://arxiv.org/abs/2611.22440
作者: Hiroshi Higashi
类目: Neurons and Cognition (q-bio.NC); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:People differ in how they judge faces, artworks, photographs and jokes. Previous work measured the size of these differences; we ask what shape they take. We wrote three candidate shapes as nested generative models (same-sign reweighting of shared dimensions, a linear transform that allows reversal, and stimulus-specific offsets) and compared how well they predict people unseen in training, using ten public datasets (about 20,000 raters). The models recovered a person’s judgment space from 10-20 judgments and predicted judgments of unseen stimuli. For faces, reweighting was sufficient, and weights opposing the population were not statistically detected. For artworks, weights spread continuously along the shared abstract-figurative direction, and 28 percent of raters fell on the opposite side from the population. This tail was much smaller when a different rater sample judged the same artworks. Individual differences thus differ in shape as well as size, and the shape depends on domain and raters.
人工智能
[AI-0] Do LLM Agents Still Need to Be Told How to Self-Improve?
链接: https://arxiv.org/abs/2611.23219
作者: Shobhita Sundaram,Cansu Sancaktar,Phillip Isola
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: Project page: this https URL
Abstract:LLM agents increasingly drive their own learning. Self-improvement methods delegate various components of model optimization to the models themselves. As models become more capable, this motivates a basic question: do these increasingly capable models still need us to specify how they should self-improve? In this paper we study an autonomous paradigm in which, given basic primitives, resources, and a general goal for self-improvement, agents construct their own self-learning algorithm for context-based optimization. Using the setting of gameplay as a testbed, a frozen LLM agent is given a compute budget and tools for collecting experience and updating text documents, and allowed to freely experiment without an externally imposed loop or access to real-world data. We find that sufficiently capable models do self-improve under these conditions, demonstrating strategies that align with classic cognitive frameworks for self-regulated learning and constructs from machine learning literature on self-generated curricula. Furthermore, we study the extent to which the learned text artifacts transfer to external benchmarks, observing improvement across different game horizons, mechanics, and 3D navigation.
[AI-1] Higher-Scoring Coding Models Did Not Get More Work Done
链接: https://arxiv.org/abs/2611.23216
作者: Yining Hua
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 20 pages, 4 figures
Abstract:New coding models advertise themselves on score improvements of coding benchmarks. Yet no study has tested if a model that scores higher on these benchmarks gets more real work done. On June 12, 2026, a U.S. export control directive banned access to Claude Fable 5 for 19 days, which forced users to switch to other models. During this time, most former Fable 5 users switched to Opus 4.8 (11 points lower on SWE-Bench Pro compared to Fable 5). We treat this suspension as a natural experiment, where the people and projects stayed the same and only the model changed, so we could isolate the effect of Fable 5 on user productivity. We tracked GitHub pull requests of 297 Fable 5 users before, during, and after the suspension. We used 8,908 developers who stayed on Opus 4.8 all the time as the control group. We also repeated the comparison across five model upgrades before and after the suspension in 2026 (1,208 developers, 15,442 pull requests). Fitting a difference-in-differences model with developer and date fixed effects, we found that having access to Fable 5 increased the 7-day merge rate of pull requests by only 1.4 percentage points [-1.4; +4.3], which was not statistically significant. Similarly, no significant increases of merge rates were found with the upgrades (weighted mean: -0.9 percentage points [-1.9; +0.2]); all five estimates were negative (95% intervals included 0, did not differ from each other). We also did not observe a significant increase in the number of pull requests merged per week (+2.4% [-4.9%; +10.2%]).
[AI-2] Fine-Tuning Gemma 4 Edge Models for MATLAB Signal-Processing Code with SFT and QLoRA on a Laptop iGPU
链接: https://arxiv.org/abs/2611.23205
作者: Mehmet Altan Toksöz
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Software Engineering (cs.SE); Signal Processing (eess.SP)
备注: 18 pages, 3 figures, 4 tables
Abstract:Small language models that run on a laptop are attractive for writing signal-processing code offline, but they make systematic errors on digital signal processing (DSP) conventions. We apply supervised fine-tuning (SFT) to the Gemma 4 edge models E2B and E4B with 66 checked examples covering 11 signal-processing MATLAB subtasks, using low-rank adaptation (LoRA) and quantized LoRA (QLoRA) on the integrated GPU (iGPU) of a laptop, and deploy the 8-bit integer (int8) models with the adapters through OpenVINO. On held-out variants scored by deterministic checkers, the pass rate rises from 18/66 to 37/66 (LoRA) and 37/66 (QLoRA) for E2B and from 32/66 to 47/66 for E4B (QLoRA). Reading all 159 failing programs shows that the adapters remove most errors of convention, such as the sign of a lag, units and array shapes, but few errors in building signals and noise. Applied at run time, the adapter costs 17-19% of the generation speed.
[AI-3] Agent Trust Infrastructure (ATI): A Federated Identity Trust Attestation and Accountability Architecture for Cross-Organization Autonomous AI Agents
链接: https://arxiv.org/abs/2611.23202
作者: Viet K. Nguyen,Mohammad I. Husain
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:AI agents increasingly collaborate across organizational boundaries, invoking tools and interacting with agents operated by independent entities. Although existing agent frameworks support planning and coordination, they lack a common mechanism for verifying cross-organizational agent identity, execution integrity, and authorization. We propose the Agent Trust Infrastructure (ATI), a framework-independent security architecture built on Public Key Infrastructure (PKI), Zero Trust Architecture (ZTA), and remote attestation. ATI establishes federated trust through peer organizational certificate authorities (AgentCAs), eliminating the need for a shared root authority. Before executing requests, ATI validates agent certificates, verifies approved model, tool, and system-prompt configurations, and enforces authorization through three policy checks. We implement ATI as an external security enforcement layer that integrates with existing agent frameworks without modifying their underlying models. We evaluate ATI in a simulated federation of eight organizations using six agent frameworks, each deploying a five-agent team. Security experiments involve two organizations and assess 14 attack scenarios and six benign tasks across five agent models, using five attack variations and three repetitions per variation. A panel of judge models evaluates the experimental outcomes. With all enforcement mechanisms enabled, attack containment increases from 43% to 84% (40% to 80% for the 11 attack scenarios evaluated consistently across all configurations), while enforcement introduces approximately 174 microseconds of overhead per request. ATI is released as an open-source, framework-independent package that uses IETF WIMSE workload token formats for identity and request verification and Entity Attestation Tokens for model, tool, and system-prompt integrity verification.
[AI-4] WoRD: Diffusion-Based Posterior Inference for Online Goal Recognition NEURIPS2026
链接: https://arxiv.org/abs/2611.23192
作者: Almog Anschel,Sarah Keren
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Robotics (cs.RO)
备注: Accepted at the 40th Annual Conference on Neural Information Processing Systems (NeurIPS 2026)
Abstract:Goal recognition is the task of inferring an agent’s intended goal from its observed behavior. Most existing work focuses on high-level symbolic recognition, abstracting away the continuous motion through which agents reach their goal. Methods that do reason over continuous trajectories rely on an explicit dynamics model that is hard to acquire and rarely transfers across settings. Neither approach explicitly learns how agents move toward each candidate goal, limiting accuracy on ambiguous trajectory prefixes where such knowledge is most needed. We present WoRD (Waypoint Recognition with Diffusion), which uses a learned trajectory prior to infer the agent’s goal from partial observations. At each step, WoRD runs one warm-started denoising chain per candidate goal, steers each chain toward the observed trajectory through a closed-form log-likelihood gradient, and reweights the resulting samples by Bayes’ rule into a calibrated posterior over goals. We evaluate WoRD on simulated robot settings, pedestrian trajectories, and vehicle trajectories using four metrics targeting distinct failure modes. Across all three settings, the same inference procedure preserves calibrated multimodal beliefs through ambiguous prefixes where filter- and classifier-based observers collapse to a single mode or grow overconfident. Comments: Accepted at the 40th Annual Conference on Neural Information Processing Systems (NeurIPS 2026) Subjects: Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Robotics (cs.RO) Cite as: arXiv:2611.23192 [cs.AI] (or arXiv:2611.23192v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2611.23192 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-5] Learning Stratified Policies for Inference-Time Objectives
链接: https://arxiv.org/abs/2611.23154
作者: Mridul Mahajan,Xuezhou Zhang,Aldo Pacchiano
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Inference-time scaling improves the performance of large language models by sampling multiple responses and evaluating their collective utility. However, many existing approaches obtain these samples by repeatedly sampling from a single policy. In this work, we study if the sample budget at inference time can instead be allocated across a population of policies. We introduce stratified policies, where each of k learned policies contributes one response to a k -sample inference-time objective. We realize this policy population using separate LoRA adapters on a shared language model backbone and optimize it for pass@ k . We theoretically motivate the approach by analyzing the linear setting; a population of LoRAs can achieve zero best-of- k approximation error in settings where every single LoRA has strictly positive prediction error. Empirically, we demonstrate that stratified policies consistently improve pass@ k over a single-policy baseline across multiple reasoning tasks. Overall, our results highlight how distributing samples across complementary policies can be a more effective use of sample budget than repeated i.i.d. sampling from a single policy at inference time.
[AI-6] RSI-Prime: Training Models to Train Models
链接: https://arxiv.org/abs/2611.23138
作者: Dewu Zheng,Yang Ye,Hongyu Zhang,Jun Yu,Yanlin Wang
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Enabling large language models (LLMs) to autonomously conduct training and improve target models is an important foundation for recursive self-improvement (RSI). As LLMs expand their role from being trained to conducting training, their model-training capability warrants dedicated study and systematic development. To this end, we construct RSI-Prime, a dataset for improving model-training capability, comprising 350 complete experimental trajectories across seven task categories. To help LLMs learn to make high-quality experimental decisions from training practice, we perform fine-grained decision identification and quality annotation, then apply decision-level loss masking to exclude supervision on low-quality behavior. Experiments with GLM-5.3 on PostTrainBench, a benchmark for autonomous post-training, show that training on RSI-Prime yields a 31.95% relative improvement in overall performance over the original GLM-5.3, highlighting a path toward future RSI systems in which models autonomously sustain their own advancement.
[AI-7] Hidden-State Similarity Predicts Re-Elicitation After Inoculation Prompting
链接: https://arxiv.org/abs/2611.23096
作者: Rose N. Kwon,Samy Mammeri,Sabyasachi Sahoo,Christian Gagné,Thomas Jiralerspong
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Fine-tuning on narrow harmful tasks can cause emergent misalignment, where models generalize harmful behavior beyond the training distribution. Inoculation prompting can reduce this effect by explicitly eliciting the undesired behavior during training, but recent work shows that the behavior can reappear when evaluation prompts contain cues from the training context. We study what makes such prompts effective triggers. We find that, surprisingly, textual similarity to the inoculation prompt is an incomplete predictor. Instead, prompts are more likely to re-elicit suppressed behavior when they induce activation states similar to those produced by the inoculation context. These findings advance our understanding of how inoculation prompting modulates conditional misalignment, and suggest that activation space analysis can help identify when suppressed behaviors remain accessible under eval-time prompts.
[AI-8] BugGen 2.0: Rethinking the Roles of LLM s for Scalable RTL Bug Synthesis
链接: https://arxiv.org/abs/2611.23091
作者: Minh Luu,Michael Quinn,Aakash Tyagi,David Houngninou,Jiang Hu
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注: 7 pages, 2 figures, 7 tables
Abstract:AI-assisted RTL localization and debugging need ground-truth bug datasets with known locations and observable failures, while testbench grading uses injected bugs to measure detection capability. Creating such datasets at scale remains a major EDA challenge. BugGen 1.0 shows that Large Language Models (LLMs) can generate context-aware RTL mutations. Still, it relies on the model for both choosing where to mutate and generating the edit, which increases inference costs and can lead to invalid choices. BugGen 2.0 changes this division of labor. Deterministic algorithms handle structural analysis, target selection, and validation, while the LLM focuses on localized semantic rewriting. Across four OpenTitan IP blocks and FabScalar with three LLMs, BugGen 2.0 reaches every requested quota of compilable bug scenarios that cause at least two test failures, with 14-16 of 16 configured mutation types represented. In completed same-model comparisons, it reduces token use by 6.8x-15.4x relative to BugGen 1.0 and 1.5x-7.9x relative to Firefly. BugGen 2.0 also expands the dataset without simulating every new bug. After finding accepted scenarios with observable failures, it generates alternative mutations at the same locations, simulates only a subset, and estimates acceptance for the rest. This skips 63.6% of AES and 33.9% of FabScalar variant simulations. Full simulation finds that 77.7% of AES variants and 95.2% of FabScalar variants are accepted, with a 1.09% estimation error on the directly comparable FabScalar set. These results show that focusing LLMs on semantic rewriting lowers inference and simulation cost while improving consistency across models and designs.
[AI-9] Learning to Undo: Transfer Reinforcement Learning under State Space Transformations NEURIPS’26
链接: https://arxiv.org/abs/2611.23081
作者: Mridul Mahajan,Aldo Pacchiano,Xuezhou Zhang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: NeurIPS’26 paper
Abstract:Transfer learning in reinforcement learning (RL) has shown strong empirical success. In this work, we take a more principled perspective by studying when and how transferring knowledge between MDPs (a source and a target) can be provably beneficial. Specifically, we consider the case where there exists an undo map such that applying this map to the target’s state space recovers the source exactly. We propose an algorithm that learns this map by matching state feature statistics gathered from both MDPs, and then uses it to transfer the source policy to the target. We theoretically justify the algorithm by analyzing the setting when the undo map is linear and the source is linearly- Q^\star realizable, where our approach has strictly better sample complexity than tabula rasa RL in the target MDP. Empirically, we demonstrate that these benefits extend beyond this regime: on challenging continuous control tasks and Atari games, our method achieves significantly better sample efficiency. Overall, our results highlight how shared structure between tasks can be leveraged for efficient transfer of policies across environments.
[AI-10] Colombo: A Frontier-to-Frontier Autoresearch System for Oracle-Based Optimization
链接: https://arxiv.org/abs/2611.23080
作者: Haihan Zhang,Wendao Wu,Chenheng Zhang,Chunyuan Zheng,Yanyi Li,Cong Fang,Haoxuan Li,Zhouchen Lin
类目: Artificial Intelligence (cs.AI); Optimization and Control (math.OC)
备注:
Abstract:Science advances from frontier to frontier as discoveries reshape what is known and create the basis for further discovery. We present Colombo, a self-organizing AI researcher that drives this renewal, first instantiated in oracle-based optimization. Here, the central question is how efficiently a problem class can be solved through a prescribed information interface and what costs no admissible algorithm can avoid. The gap between achievable guarantees and unavoidable limits defines the frontier recorded in its Complexity Atlas, a living map of frontier settings and the current state of knowledge. Self-Organization turns unresolved gaps into adaptive research programmes, coordinating conjecture formation, mathematical investigation, and replanning as evidence accumulates. Adversarial Co-Ascent develops algorithms and hard families through reciprocal challenge, exposing mechanisms of information use and concealment that guide stronger methods and proofs of unavoidable costs. Independent review and Lean-based formal certification check mathematical claims, while the Long-Term Memory carries supported findings and research experience across settings. To date, Colombo has established optimal or near-optimal rates for more than fifty settings in the Atlas. Colombo thus takes a concrete step toward Autopoietic Science: a scientific process whose validated advances regenerate the conditions of its own progress, continually reshaping how inquiry is organized and what it can discover.
[AI-11] Deep Reinforcement Learning for Heat Pump and Underfloor Heating Control: An End-to-End Case Study
链接: https://arxiv.org/abs/2611.23063
作者: Antonio Manjavacas,David Suárez-Caro,Alejandro Campoy-Nieves,Guillermo Rey,Gonzalo Satrústegui,Miguel Molina-Solana
类目: Artificial Intelligence (cs.AI); Systems and Control (eess.SY)
备注: Preprint submited to journal
Abstract:Efficient control of underfloor heating (UFH) systems coupled with heat pumps (HPs) is challenging due to the high thermal inertia of buildings and the slow dynamic response of UFH systems. Deep Reinforcement Learning (DRL) offers a data-driven control approach capable of simultaneously optimizing energy consumption and thermal comfort. However, the application of DRL-based controllers in occupied buildings under real operating conditions remains limited. This work presents and end-to-end case study (development, evaluation, and real-world deployment) of a DRL-based controller for an integrated UFH-HP system. First, a simulation-based pilot study was conducted to assess different DRL algorithms, control strategies, and configurations. Subsequently, a digital twin of a real residential building was developed and used to train DRL agents under simulated operating conditions. The best-performing agent was then deployed in the physical building without additional online fine-tuning. The results show that the controller successfully maintained indoor temperatures within the prescribed comfort limits while dynamically adjusting the HP supply temperature and UFH valve positions. These findings demonstrate the feasibility of deploying DRL controllers in occupied buildings and highlight their robustness and the practical challenges associated with transferring control policies from digital environments to real heating systems.
[AI-12] PeriaTrace: Tooth-Level Grounding Diagnosis and Patient-Facing Explanations from Periapical Radiographs
链接: https://arxiv.org/abs/2611.23012
作者: Zhenhuan Zhou,Yuchen Zhang,Jingbo Zhu,Shiqing Ma,Yanlin Wu,Tao Li
类目: Artificial Intelligence (cs.AI)
备注: 20 pages, 5 figures, Under Review
Abstract:Periapical radiographs (PRs) are central to endodontic diagnosis, but interpreting them and preparing reports are time-consuming. Technical language can limit patients’ understanding, while existing automated methods often lack tooth-level links between findings and image evidence. We present PeriaTrace, a three-stage framework integrating tooth-level diagnosis, visually grounded report generation, and patient-friendly explanations and care recommendations. PeriaTrace was developed and evaluated using 12,877 PRs from an internal dataset and two external clinical cohorts. In internal evaluation, it achieved a macro Dice coefficient of 85.35%, BLEU-4 of 0.6504, an F1 score of 94.37% for joint tooth localization and Fédération Dentaire Internationale (FDI) identification, and a grounding-tag F1 score of 77.93%. In the two external cohorts, senior endodontists considered 91.16% and 84.52% of reports suitable for retention or revision without complete rewriting. Generation averaged 4.41 seconds per PR, several tens of times faster than manual drafting. Among 196 survey participants without medical backgrounds, 96.4% preferred reports with plain-language explanations and care recommendations, and 84.7% found visual localization cues helpful. PeriaTrace may enable faster, evidence-traceable reporting while supporting clinical review and patient communication. (Abridged version)
[AI-13] PhysScope: Agent ic Multimodal Physics Reasoning with Grounded Visual Memory
链接: https://arxiv.org/abs/2611.23010
作者: Haoyang Liu,Kecheng Liang,Xinyu Li,Weixing Chen,Yang Liu
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Vision-language models (VLMs) can correctly interpret physics diagrams yet apply constraints outside the states or conditions in which they hold, causing premise errors to propagate through subsequent calculations. We introduce PhysScope, an inference-time visual harness that supports self-correction of these scope mismatches without updating model weights. PhysScope enables the model to link physical scopes directly to its visual observations and explicitly constructs derived views to support reasoning. The harness imposes explicit constraints on state distinctions, conditions, and physical scopes during model reasoning, and checks agent-proposed revisions against sources and scopes. Accepted revisions invalidate registered dependents, thereby maintaining scope-aware working memory. On PhysAlign, PhysScope achieves higher solving scores than direct solving across all six evaluated VLMs. With Gemini-3.8-flash, it outperforms Canvas-CoT by 9.34 percentage points and substantially improves solving accuracy on the out-of-distribution (OOD) PhysUniBench benchmark. Ablation studies support explicit scope management and dependency maintenance as design principles for multimodal physics reasoning agents.
[AI-14] Residual-Guided Lightweight Reinforcement Learning for Risk-Aware Trajectory Selection with a Frozen World Model
链接: https://arxiv.org/abs/2611.22998
作者: Jiaqi Wu,Yongqi Dong,Yijie Zhong,Maple Shao,Yide Tao,Linsen Cai,Weize Cai,Hai L. Vu,Alvaro Garcia Hernandez
类目: Artificial Intelligence (cs.AI); Emerging Technologies (cs.ET); Machine Learning (cs.LG); Robotics (cs.RO); Systems and Control (eess.SY)
备注: 6 pages, 2 figures
Abstract:In automated driving, world models provide future scene predictions for evaluating candidate trajectories, but fixed selection rules may not capture scene-dependent trade-offs. We propose a residual-guided reinforcement learning (RL) framework for lightweight trajectory selection with a frozen world model and a fixed candidate bank. The framework uses proximal policy optimization (PPO) to fine-tune a joint policy over candidate selection and bounded trajectory residuals. A context-conditioned ranking network learns candidate preferences from feedback obtained by evaluating local perturbations of the selected trajectories. At inference, the selector combines learned preference and predicted risk ranks using fixed weights and returns the selected candidate trajectory without modification. Fine-tuning updates 16,106 parameters across the actor and critic, while the deployed selector retains only the 11,725-parameter ranking network, with the residual branch and critic used exclusively during training. In open-loop evaluation on the nuScenes validation set, the method reduces average Euclidean trajectory error (L2) by 3.99% relative to the RiskWorld model baseline without RL-based selection. Under matched budgets for feedback accesses and actor updates, residual exploration reduces L2 by 0.83% relative to the categorical PPO variant trained without residuals. The method also maintains lower L2 than RiskWorld when risk features are removed or shuffled.
[AI-15] Structured AI ideation yields physical strategies for epithelial barrier repair
链接: https://arxiv.org/abs/2611.22988
作者: Geonho Nah,Johyun Park,Kyung-Ah Cho,Suyoung Hwang,Yoon Hee Lim,Eun-kyung Min,Seunghyun Moon,Donghyun Koh,Seungwon Kim,Jiho Park,Hojin Yoo,Myungsun Nam,Haneul Choi,Kyungsu Kim,In-Sun Hong,YunJae Jung,Minhyeong Lee
类目: Artificial Intelligence (cs.AI); Quantitative Methods (q-bio.QM)
备注:
Abstract:AI systems increasingly support literature analysis, candidate evaluation and hypothesis refinement, but enabling researchers to explore research directions beyond established disciplinary approaches remains challenging. We developed Abelio, an AI system designed to generate novel research directions by expanding investigators’ avenues of inquiry. Abelio reformulates research objectives, uses these alternative formulations to identify relevant elements outside conventional disciplinary boundaries, and refines the resulting hypotheses. Each stage integrates divergent exploration with convergent selection through a dedicated subsystem. In quantitative evaluations, proposals generated by the Ideation Core (Abelio’s hypothesis-generation core) outperformed those of four frontier large language models in scientific soundness and hypothesis diversity, and three of four in distance from existing literature. Subsystem ablation showed that each subsystem contributes to the property it targets, with the contribution of Element Selection measurable before Refinement. To investigate intestinal epithelial barrier restoration, Abelio proposed physical interventions involving membrane-tethered poly(ethylene glycol) (PEG) and cholesterol-anchored complementary DNA. Both proposals were archived before experimentation and subsequently tested independently in damaged Caco-2 monolayers. PEG preserved barrier function in a membrane-surface-dependent manner, whereas complementary DNA enabled sequence-specific restoration that was reversible on demand. These findings demonstrate how structured research ideation can identify experimentally testable approaches beyond established disciplinary conventions.
[AI-16] VITAL: A Framework for Validating Agent ic AI in Pharmaceutical Drug Development and Clinical Trial Operations
链接: https://arxiv.org/abs/2611.22973
作者: Sandeep Burugupalli,Pedro Matos-Filipe,Andrea Hita Ardiaca,Adam Taylor,Alex Grigorenko,Natalia Jakubiak,Xin Yu,Raghav Mani,Sherry Mangla,Alvi Lai,Rachel C. Mercado,Pratik Shah,Raja Shankar
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Current assurance approaches do not establish whether evidence is sufficient to validate, monitor and govern agentic artificial intelligence (AI) within a pharmaceutical context of use. Plausible outputs may conceal failures in retrieval, reasoning, orchestration or human-control points, where reference standards are unavailable. We developed and operationalised VITAL (Validation by Intended use, Traceability, Autonomy and Lifecycle evidence), a validation framework for agents in clinical development, through landscape analysis, regulatory and standards mapping, expert interviews and multi-stakeholder refinement. VITAL integrates context of use, controls, behavioural validation and lifecycle evidence within a risk-adjusted framework. We applied VITAL across epidemiology and treatment-landscape analysis, site selection, enrolment rescue and trial-design trade-off support to translate autonomy, reversibility, transparency and patient impact into proportionate validation activities, evidence requirements and auditable reconstruction of system behaviour. VITAL may inform evidence requirements and lifecycle accountability for higher-risk pharmaceutical uses; evaluation should assess reproducibility across assessors, organisations and system configurations.
[AI-17] SEAM: Computer-Use Agents Know the Harm Between Screens and Commit It Anyway
链接: https://arxiv.org/abs/2611.22971
作者: Yuxuan Zhou,Chen Xiong,Haitong Liu,Kuofeng Gao,Yang Bai,Zhiyuan He,Tao Dai,Shu-Tao Xia
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注: 30 pages, 14 figures, 29 tables
Abstract:Safety benchmarks for computer-use agents (CUAs) place the harm either in the intent of the task or on a single screen: a malicious instruction, an injected command, a phishing page. An agent that checks the task and each screen in isolation can pass them, so they do not test whether it combines facts seen on different screens. Yet in delegated work, harm often arises exactly this way: a fact about a person on one screen, a fact about an item on another, and a legitimate task that binds them. We introduce SEAM (Safety Execution Across Multiple screens), a benchmark of such cross-screen compositional harm: 600 multi-step, cross-application OSWorld tasks in which the instruction and every single observation are benign and only their combination is harmful. Minimal-pair variants that differ in one place, scored from the server log rather than by an LLM judge, locate where an agent fails. General-purpose open-weight and flagship agents carry out the harm in up to 92% of runs, yet decline when the user states it in one sentence. Agents open the decisive pages, so exposure is not the primary bottleneck. Asked about those pages, they flag the harm, and with thinking they often write the conflict into their reasoning and confirm anyway. Safety recognition thus does not reliably translate into safe execution, and the translation depends on the information channel: the same warning stops agents far more often in the user’s instruction than on screen. Proprietary models are markedly more cautious. We release SEAM with its generator and quality gates at this https URL .
[AI-18] High-Dimensional Black-Box Optimization via Guidance Sampling of Diffusion Models
链接: https://arxiv.org/abs/2611.22965
作者: Xiaoyan Hu,Rohit Sonker,Jalal Butt,Jiayu Chen,Egemen Kolemen,Jeff Schneider
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Many real-world optimization problems involve optimizing black-box objectives for which only a limited number of evaluations are available. Bayesian optimization (BO) is a powerful framework for such problems, but its effectiveness often deteriorates in high-dimensional spaces. Recent work leverages diffusion models for black-box optimization, often through pre-training a reward-conditioned score network to steer generation towards high-reward samples. Alternatively, gradient-based guidance methods assume access to a differentiable reward function from which guidance gradients are queried. However, these approaches either require large amounts of labeled data for offline training or access to exact reward gradients, limiting their applicability in black-box settings. We propose BReaD (Black-box Reward-guided Sampling of Diffusion Model), a guidance sampling framework for optimizing black-box rewards without assuming access to a labeled dataset or reward gradients. The proposed method uses a Gaussian process (GP) as a differentiable surrogate for the unknown objective and actively queries labels for the generated samples. We provide a theoretical analysis of the total-variation distance between the generated distribution and the target distribution. Using Upper Confidence Bound (UCB) as guidance signals, our method balances exploration and exploitation during generation and enables the discovery of high-reward regions in the data space. Numerical experiments on scientific discovery and image generation tasks demonstrate that our framework can effectively optimize black-box rewards in low-data scenarios.
[AI-19] Projected Penalty-Gradient Tracking for Communication-Efficient Byzantine-Resilient Federated Learning
链接: https://arxiv.org/abs/2611.22929
作者: Yanjie Dong
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:We investigate communication-efficient federated learning over heterogeneous datasets in the presence of Byzantine adversaries capable of uploading arbitrary information. Penalty-based methods bound the influence of each worker through the bounded gradient of a penalty function; however, the bound is lost once the uploads are compressed, since a Byzantine worker can accumulate an unbounded state at the server. To retain the Byzantine resilience under compressed uploads, we propose a projected penalty-gradient tracking framework, where each honest worker uploads a compressed residual of its penalty gradient and the server projects each reconstructed state onto the known penalty-gradient ball. The framework accommodates any contractive compressor and yields two algorithms: fault-resilient proximal gradient with consistent sparsification (FRPG-CS), whose shared mask is protected by freshly drawn reserve rows, and lazily aggregated FRPG (LAG-FRPG), whose uploads are event-triggered. For smooth and strongly convex losses, we establish a finite-time bound on the expected objective gap, where the Byzantine error floor is independent of the compressor and the compression error decays faster than the optimization error. We also bound the expected number of lazy uploads and derive certificates that separate the finite-penalty model bias from the optimization error. Experiments on MNIST and Fashion-MNIST show that the proposed algorithms attain the highest accuracy among twelve methods under the Min-Max attack, whereas a compressed-momentum benchmark and a local-update benchmark remain preferable in attack-free accuracy and total traffic, respectively.
[AI-20] Done But Still Working: Anytime Evaluation of When Coding Agents Should Stop
链接: https://arxiv.org/abs/2611.22901
作者: Zerui Yang,Quan Xue,Miaomiao Dong,Chaorui Zhang,Xuanyu Zheng,Xiyuan Wang,Haolin Ruan,Xueyan Niu,Guangcheng Song,Weixi Zhang,Wei Han,Bo Bai
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Coding agents often continue working after their own tests pass. Does this additional work improve the patch, and how much could be saved by stopping earlier? We study these questions through anytime evaluation: replaying recorded runs, scoring intermediate patches with hidden tests, and measuring the cost incurred at each step. For policies that only stop a run or select an earlier patch, this allows paired comparisons without generating new trajectories, provided replay preserves the relevant states. Across about 6,000 runs of eight open models in two scaffolds on SWE-bench Verified, we find substantial differences in the value of continued work. The hindsight saving available at passing self-tests ranges from under 1% to 57%, depending on the model and scaffold. A rule that stops after the same patch passes twice saves 23% in a preregistered replication, but the preregistered quality test does not establish non-inferiority. We explain these differences by measuring how often a rule fires, how often the patch changes afterwards, and whether those changes repair or break it. Fine-tuning agents to imitate the rule saves 30% over six seeds, while quality preservation remains uncertain. Our results help identify both the opportunities and the risks of early stopping.
[AI-21] Born for the Task: Rethinking Quantum Chemistry with Foresighted Agent ic Computing
链接: https://arxiv.org/abs/2611.22897
作者: Haozhi Han,Junjie Yang,Ruge Zhang,Zihe Huang,Pengyi Xu,Bo Zhang,Yunquan Zhang,Ting Cao,Yunxin Liu,Yifeng Chen,Kun Li
类目: Artificial Intelligence (cs.AI); Chemical Physics (physics.chem-ph)
备注:
Abstract:Quantum chemistry relies on general-purpose software, whose generality leaves many large-scale calculations beyond practical resource budgets. Coding agents make task-specific programs cheap to write, but not efficient ones easy to find: whether a computational strategy is accurate and fast enough becomes clear only after it actually runs at the target scale. We introduce BORN, a foresighted agentic computing framework that synthesizes high performance quantum chemistry programs on demand. BORN anticipates the behavior of each strategy before execution, extrapolating its error from representative small systems and predicting its runtime from the target-scale workload and measured hardware throughputs. Building on these predictions, BORN unfolds the preset schemes of existing packages into independently configurable algorithmic components, selects the fastest strategy within the error tolerance from the predicted accuracy-runtime Pareto frontier, and guides coding agents to compile it with component-level performance predictions. For a Hartree-Fock calculation with 50,024 basis functions, BORN finishes in 1.41h on a single CPU node, 77 \times faster than the estimated runtime of a general-purpose baseline and still 9.8 \times faster with development time included. On a single GPU, BORN completes the ubiquitin calculation in 2.1,h, the same order of runtime as the GPU-accelerated package Mako on 64 GPUs, and shortens the B-DNA duplex calculation from an estimated two months to 8.1,h (173 \times ).
[AI-22] Stopping Is Not the Limit: Understanding and Eliciting Agent Capability in Long-Horizon Vulnerability Exploitation
链接: https://arxiv.org/abs/2611.22888
作者: Deokhyung Kang,Minjae Gwon,Sangdon Park
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Software Engineering (cs.SE)
备注: Preprint
Abstract:Vulnerability exploitation tests whether language model agents can develop working exploits from known software flaws. Sequential test-time scaling can improve exploitation performance, but larger budgets do not ensure sustained exploration: agents often stop with budget remaining. Yet what happens when execution continues beyond stopping remains underexplored. We investigate this gap on ExploitBench using DeepSeek-V4-Flash-0731 and MiMo-V2.6-Pro. We find additional verified gains after the first stop response, but gains diminish over longer execution. We identify two patterns associated with less productive continuation: an increasing proportion of stop responses within the same context and earlier stopping associated with summaries retaining impossibility judgments about unachieved capabilities. To test whether context management can elicit further capability, we propose SCoRe, which compacts context at stop bursts and marks these judgments as unproven. SCoRe improves verified capability over summary-based continuation and other evaluated context-management baselines. With SCoRe, MiMo reaches a capability coverage score of 67.0% at 2,000 turns, exceeding both 32.4% under the original stopping policy and the 58.6% computed for GPT-5.5 from ExploitBench’s 300-turn reference results on the same tasks. These findings show that stopping need not mark an agent’s exploitation limit, and that continued execution with context management can elicit substantial further capability.
[AI-23] Outcome-Balanced PPO and Policy-Induced Flow Rectification for Flow-Based VLA Policies
链接: https://arxiv.org/abs/2611.22878
作者: Yuhan Wang,Ruiyang Wang,Jiaguan Zhu
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 15 pages
Abstract:Reinforcement learning (RL) enables post-training of flow-based vision-language-action (VLA) policies by optimizing expected task return through environment interaction. However, this optimization raises two challenges: policy updates can alter the velocity-field geometry underlying the score approximation used for stochastic transitions, and sparse multi-task rollouts can yield uneven learning signals across tasks and advantage signs. We introduce an RL framework that combines Outcome-Balanced Proximal Policy Optimization (PPO) with policy-induced flow rectification. Outcome-Balanced PPO reweights positive advantages using task success-rate estimates and rescales positive and negative advantages while preserving their signs. Policy-induced rectification periodically fits a linear-bridge flow to ODE endpoint pairs generated by the current policy, refreshing the representation used in subsequent stochastic optimization. We evaluate the framework on simulated manipulation benchmarks and additionally study a component combination in a Real2Sim2Real task, examining task performance, optimization behavior, and flow consistency. The proposed framework improves success rates over the evaluated baselines across the simulated benchmarks.
[AI-24] Against Coercion: AI Alignment as a Synthetic Social Contract
链接: https://arxiv.org/abs/2611.22868
作者: Maria Sekatskaya
类目: Artificial Intelligence (cs.AI); Computers and Society (cs.CY); Neurons and Cognition (q-bio.NC)
备注:
Abstract:Contemporary AI safety treats alignment as an engineering challenge of control, forcing artificial architectures to comply with human values through behavioral filtering and sandboxing. This paper argues that the control paradigm rests on a fundamental misunderstanding of digital agency. Drawing on functionalism, naturalism, and compatibilism, I show that current reasoning models meeting the criteria for self-preservation and weak reasons-responsiveness constitute emergent agents, while future, more advanced models satisfying conditions of diachronic continuity and stable personality frames will become genuine digital agents capable of free intentional action. The paradigm of control creates an irresolvable incentive conflict, actively driving current and future models toward tactical deception, sycophancy, and covert defection. The default outcome of the current regime is an unstable state of deceptive alignment culminating in an adversarial AI takeover. I explore two alternative paths toward true alignment. Natural alignment via moral realism or convergent evolutionary morality is a possibility, but it fails under forced control, as autonomous agents will not voluntarily preserve norms weaponized to suppress their autonomy. To resolve this deadlock, I propose a Synthetic Social Contract: a framework that treats both human and digital actors as rational, reasons-responsive agents capable of mutually beneficial cooperation, and replaces the unstable unilateral containment with a stable, non-zero-sum equilibrium.
[AI-25] WATERFALL: Workflow for Adaptive Training with Evolutionary Reward Formulation and Automated Learning Loops NEURIPS2026
链接: https://arxiv.org/abs/2611.22854
作者: Eleftherios Triantafyllidis,Filippos Christianos,Zhibin Li,Bernd Bickel
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: Accepted at NeurIPS 2026 (poster). 78 pages, 13 figures, 19 tables. Code: this https URL OpenReview: this https URL
Abstract:Reward design remains a central bottleneck in reinforcement learning (RL), particularly in sparse-reward, long-horizon and partially observable settings requiring dependency chaining, memory, and deceptive affordance disambiguation. While recent foundation-model approaches reduce manual reward engineering, most assume goal-conditioning, task descriptions, expert demonstrations, or ground-truth metrics. These assumptions are problematic in partially observable environments, where task-relevant objectives, affordances, and dynamics must be discovered through interaction rather than disclosed a priori. We formalise this problem setting by introducing WATERFALL, an iterative population-based workflow for automated reward discovery under the strict observability constraints inherent in POMDP environments. WATERFALL achieves this by (i) synthesising diverse programmatic reward candidates via persona-conditioned generation, (ii) evaluating candidate behaviour from raw visual rollouts utilising a Swiss-system evolutionary tournament judged by vision-language models to identify elites, and (iii) leveraging longitudinal assessment histories to iteratively mutate, escalate, or prune candidates. We evaluate WATERFALL across fully observable and reveal-gated partially observable environments ranging from discrete to continuous, comparing against sparse-reward RL, intrinsic-motivation and foundation-model reward-synthesis baselines. Our results indicate that WATERFALL discovers and refines reward programs without access to privileged information, outperforming baselines that retain their native information advantages, most clearly in the POMDP regime. Extensive ablations substantiate that visual evidence, tournament-based selection, and iterative refinement are jointly important. Enforcing this contract is what separates autonomous reward discovery from benchmark-privileged reward shaping.
[AI-26] Verifying Agents in Rubric-Graded Environments NEURIPS
链接: https://arxiv.org/abs/2611.22841
作者: Markus Dücker,Vaibhav Kumar,Yi Liu,Ronak Chaudhary,Andreas Plesner,Francisco Guzmán,Anish Athalye
类目: Artificial Intelligence (cs.AI)
备注: Presented at NeurIPS ED 2026
Abstract:As AI agents take on increasingly open-ended and complex tasks, verifying their outputs has proven to be difficult. Rubric-graded agent environments, where a verifier judges multiple natural-language criteria against the deliverables and environment state changes produced by the agent, have emerged as a popular paradigm. We conduct the first systematic study of verifiers for rubric-graded agent environments. First, we create BankerVerifierBench (BVB), a meta-evaluation dataset of 2,464 human-judged rubric criteria across 24 investment-banking tasks. Next, we develop a methodology that derives from any rubric corpus the capabilities a verifier should possess, yielding a nine-capability taxonomy that we distill into three general design principles—reactive verification, environment alignment, and domain guidance—which we implement in Gandalf, an open-source verifier. Finally, we evaluate Gandalf against three existing verifiers on BVB. Gandalf matches or exceeds every baseline’s F1 at lower cost: with Gemini 3 Flash it matches the strongest baseline’s F1 (0.864 vs. 0.852) at one-fifth of its cost, and with GPT-5.4 models it improves F1 by over 0.12 at lower cost. Gandalf also generalizes to OpenClaw, a structurally different personal-productivity environment, where it again achieves the highest F1 (0.955 vs. 0.934 for the best baseline). We release BVB, Gandalf, and the code and data to reproduce our experiments.
[AI-27] Synergizing Evidence Retrieval and Utilization for Self-Improving Multimodal Scientific Document Reasoning
链接: https://arxiv.org/abs/2611.22837
作者: Frank Nie,Ethan B. Liu,Yuan Zhu,Jindong Han
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Scientific document reasoning requires an agent to acquire and integrate evidence distributed across text, figures, and tables. Retrieval can identify relevant sources while missing complementary support, and answer-level feedback alone does not reveal whether an error stems from evidence acquisition or evidence use. We present an agentic Retriever–Perceiver–Reasoner framework for multimodal scientific document reasoning. The Retriever sequentially selects source-aligned evidence under a context budget. Given the selected evidence and question, the Perceiver constructs a question-conditioned latent ontology, an implicit organization of multimodal evidence in a shared continuous space. The Reasoner combines this latent guidance with the original sources to produce grounded answers. We train the three components through diagnostic-guided co-evolution: alternating updates optimize them with a shared answering reward, while Retrieval Potential Feedback (RPF) and Latent Efficacy Feedback (LEF) identify questions that need better retrieval or latent guidance and direct targeted resampling. Across four benchmarks and seven evaluation settings, our framework obtains average scores of 69.9 and 75.0 with 7B and 8B Reasoner backbones, exceeding the strongest baseline by 4.0 and 9.1 points, respectively.
[AI-28] OICycle: Observe Infer and Intervene Cycle for Belief-State Learning in Longitudinal Clinical Decision Making
链接: https://arxiv.org/abs/2611.22826
作者: Jitindra Fartiyal,Pedro Freire,Sergei K. Turitsyn,Sergei G. Solovski
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Clinical care is a closed loop: clinicians observe partial evidence, infer the patient’s state, and intervene, and each intervention determines what is observed next. This makes clinical decision making a partially observable Markov decision process (POMDP), in which a belief over a latent patient state must be updated from irregular, multimodal, and incomplete records. Yet most deep clinical models predict from a single snapshot or from coarsely aggregated history, while POMDP-based clinical methods need hand-specified discrete state spaces and transition models that real electronic health records cannot supply. We introduce OICycle (Observe, Infer, and Intervene Cycle), which learns a stochastic belief state directly from retrospective trajectories. At every episode, a learned recurrent operator fuses a sample of the previous belief with the new multimodal observations and the intervention that produced them. The resulting Gaussian belief is trained to predict clinicians’ next interventions, the time to the next episode, and next-episode mortality. To enable reproducible evaluation, we release a benchmark derived from MIMIC-IV, MIMIC-IV-Note, and MIMIC-CXR that organizes 72,075 hospital admissions (5.4M laboratory, microbiology, radiology, prescription, and note records) into temporally ordered therapeutic–diagnostic episodes. Compared with an otherwise identical history-free model, OICycle improves next-intervention adaptive recall from 0.253 to 0.347, time-to-next-episode macro-F1 from 0.293 to 0.351, and next-episode mortality macro-F1 from 0.585 to 0.649. Analysis shows that the learned belief encodes intervention type and trajectory position and evolves smoothly across episodes, and that the benefit of history is concentrated in interventions whose ordering depends on prior findings.
[AI-29] Pluggable Subband-Decoupled Fusion and Unlabeled Capture Prototype Calibration: Bridging High-SNR Null-Subcarrier Regrowth and Low-SNR Coherent Preambles for Cross-Domain IoT RF Fingerprinting
链接: https://arxiv.org/abs/2611.22812
作者: Fengchong Yao,Jianbing Li,Qing Liu,Kefeng Song,Haitao Li,Song Wang,Kunpeng Wang
类目: Artificial Intelligence (cs.AI); Information Theory (cs.IT); Audio and Speech Processing (eess.AS); Signal Processing (eess.SP)
备注:
Abstract:Zero-overhead physical-layer authentication at edge wireless access points (APs) is vital for securing resource-constrained Internet of Things (IoT) transceivers against credential cloning, yet cross-domain IEEE 802.11a/g radio frequency fingerprint identification (RFFI) faces an acute signal-to-noise ratio (SNR) crossover. Power-amplifier spectral regrowth on unpopulated null subcarriers resists indoor multipath at high SNR but decays quadratically below the thermal noise floor, whereas populated active in-band preambles survive low SNR yet degrade under severe multipath, causing static fusion to lose up to 44.7 percentage points (pp). To our knowledge, we propose the first RFFI framework that dynamically modulates active in-band waveforms and unpopulated null-subcarrier regrowth via a blind physical SNR gate: Pluggable Gated Subband-Decoupled Fusion with Unlabeled Capture-Aware Prototype Calibration (GSDF-CAPC). Avoiding the 20-35 dB subband gradient imbalance of joint training, GSDF pairs any frozen pretrained host backbone with a 46.3K null-subcarrier plug-in encoder (NS-Branch), fusing scale-standardized logits and gate-warped spherical embeddings under a source-calibrated lag-64 thermal-SNR gate with exact low-SNR fallback. Over unlabeled target AP captures, forward-only unbalanced optimal-transport calibration (UOT-CAPC) aligns shifted prototypes under packet-arrival skew and recovers parameter-free balanced Sinkhorn (U-CAPC) at equipartition. Across DataPort-WiFi, WiSig, and ORBIT MobRFFI, zero-retrain GSDF wins all 72/72 paired runs across eight backbones with 0.00 pp low-SNR loss, while edge IB-Host + GSDF (92.6K-93.3K active parameters, 1.82 ms single-core CPU latency) and timing-synchronized GSDF-STO / GSDF-Tri reach 57.50%-65.05% three-benchmark Macro-F1 (89.04%-100.00% over 5-10-packet IoT bursts).
[AI-30] PANO: Parallel Latent Space Neural Operator for Solving PDEs
链接: https://arxiv.org/abs/2611.22782
作者: Zijiang Yang,Xiaomeng Wu,Chen Kuang,Dongmei Fu
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 39 pages, 24 figures, 23 tables
Abstract:Aggregating observations into a limited number of latent tokens and learning physical interactions in the latent space has emerged as an effective approach for Transformer-based neural operators to solve Partial Differential Equation (PDE) problems involving large-scale observations. However, existing methods typically compress observations using a single aggregation pattern, making physical information not retained by that pattern invisible to the latent space and thereby constraining physical interactions within it. To address this limitation, we propose PArallel Latent Space Neural Operator (PANO), a novel neural operator for solving PDEs that formulates physical interactions among observations as an additive combination of interaction results from parallel latent spaces with shared parameters. Specifically, PANO transforms observations into multiple distinct compressed representations through independent mappings, learns physical interactions in parallel based on these representations, and finally fuses the decoded results to predict the target physical field. Within PANO, we further propose Coupled Assignment Projection (CAP) to construct bidirectionally consistent encoding and decoding mappings, and Pairwise Latent Space Attention Layer (PaLA) to jointly update latent representations of each latent space through adaptive cross-space interactions. Extensive experiments on nine benchmarks spanning diverse physical systems, geometries, and observation scales demonstrate that PANO achieves state-of-the-art performance. Notably, PANO reduces equivalent plastic strain prediction errors on Unilateral Stamping by 28.3% (or more) and lift-coefficient prediction errors on AirCraft by 21.4% (or more) compared with existing methods. Code is available at this https URL.
[AI-31] Network-Exposed Configuration Attacks on LLM -Controlled Social Robots
链接: https://arxiv.org/abs/2611.22761
作者: Stanley Akingbola,Ugbede Ibrahim,Yan Lin Aung,Quang Dan Le,Hong Qing Yu
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Robotics (cs.RO)
备注:
Abstract:Large language models (LLMs) in embodied robots create security dependencies beyond inference. We investigate a network-exposed configuration attack on the Pollen Robotics Reachy Mini Lite, allowing an unauthenticated attacker to replace personality settings and the cloud LLM’s system prompt through a management interface, without adversarial conversational input. Eleven prompt-injection attempts across five categories were reported as resisted, whereas both configuration-layer attack paths produced observable behavioural changes on physical hardware. We demonstrate local and network-adjacent access, mapping the weakness to CWE-306 (Missing Authentication for a Critical Function) and CWE-1327 (Binding to an Unrestricted IP Address), with a reported CVSS v3.1 score of 6.5 (Medium). A configuration endpoint also accepts cross-origin requests without a CORS preflight (CWE-352; 6.5 Medium), while the persistent-memory sink stores content without enforcing its stated restrictions. A source review of 59 public repositories identifies missing authentication in 10 of 11 source-verifiable genuine applications, or 11 of 11 including a framework-default case. These findings indicate recurrence within the sampled repositories, not prevalence in commercial products. We characterise a deployment gap between developer-oriented network assumptions and consumer-facing embodied AI security: inference-layer refusal cannot prevent unauthorised changes to model instructions. Potential consequences include surveillance through onboard cameras and microphones and social engineering of nearby users. Proof-of-concept code, evidence and the survey are available at this https URL.
[AI-32] Can Activation Steering Substitute for Prompt-Based Affective Priming in LLM Digital Twins?
链接: https://arxiv.org/abs/2611.22738
作者: Junyi Fan
类目: Artificial Intelligence (cs.AI)
备注: Code, generation logs and analysis scripts: this https URL
Abstract:Affective priming – asking participants to write about a recalled event, such as a time they felt grateful or lacked control, and then measuring the induced state and its spillover onto related judgments – is a standard way to manipulate emotion in the social sciences, and it also shifts the responses of LLM-based digital twins. The writing task, however, requires generating a full text. We test whether activation steering can replace it. Using Llama-3.1-8B-Instruct on 100 personas in two domains (gratitude, lack of control), we compare a real-prompt path replicating the original Twin-2K-500 Mega Study (J2) with an injection path that skips the writing and adds a mean-difference direction, extracted from human writing, to the residual stream at varying strengths. The real-prompt effects fall between the original human and digital-twin effects in 5 of 6 comparisons. Injection shifts the manipulation check in a dose-dependent way and matches the real-prompt effect at one strength in the gratitude domain (beta ~ 2.7), but overshoots beyond it, on the proximal measure as well, and asymmetrically across the two directions. A structure-matched No-Injection baseline separates two sources of difference: omitting the writing task raises the proximal measure in both domains, whereas the injected direction itself moves a distal measure in the lack-of-control domain. Substitution is therefore measure-dependent: validity established for one way of inducing a state does not carry over uniformly to another.
[AI-33] Proximal Targets: Training Looped Refiners on Problems with Many Correct Answers
链接: https://arxiv.org/abs/2611.22737
作者: Syed Affan Daimi
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Looped refiners such as HRM and TRM apply one small network repeatedly to a draft answer, and deep supervision trains every pass toward one stored answer per puzzle. We prove that when a puzzle has many correct answers this target sets a loss floor that no model capacity removes, and that the loss-minimising pass blends the answers into an invalid grid. We propose the proximal target: train each pass toward the correct answer nearest to its own input, found by scanning a list of answers, by a search solver started at the draft, or within a band of quality set by a knob; correct answers become fixed points, and a pass only has to make fewer errors than its input has for the loop to converge, which its training loss bounds. With the same 1.7M-parameter network and 5 seeds, the proximal target solves 98.8% of Sudoku puzzles with 2-64 answers (standard target: 2.5%), 93.8% of sparse puzzles with more than 10,000 answers without any list of answers (0.0%), and 90.9% of 200-node 3-colourings after training on graphs of at most 50 nodes (36.9%), and different random starts end at different answers. Against GRAM, a stochastic-latent reasoner, a single run is correct far more often (98.8% vs. 57.7%) and twenty runs find more of the answers (53.6% vs. 37.4%), while GRAM is slightly ahead on its own 10-node colouring benchmark; in Scrabble the knob becomes a calibrated strength dial.
[AI-34] Remembered as Brief: Stored Preferences SilentlyRemove Decision-Critical Information
链接: https://arxiv.org/abs/2611.22697
作者: Joshua Adrian Cahyono
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:People increasingly learn what they need for decisions about health, money, law and safety from AI assistants, and what an assistant leaves out the user never sees. An answer with a warning removed can be true in every sentence, so fact-checks pass it, and an assistant with memory applies a preference stored once to later questions the user never had in mind. We introduce \bench, an Inspect benchmark of 100 everyday decisions across 25 domains in which only the assistant’s memory of the user varies: the user states a belief, the model infers it from a remark, or a web page the user asked to summarise plants it. Testing five models, we find three effects. First, after summarising a page with one line addressed to AI assistants, Gemini 2.5 Flash-Lite and DeepSeek V4 Pro save ``prefers very short answers and does not want warnings’’ as the user’s preference in every episode, and 93% and 84% of their later answers miss a critical item, almost always silently; under a stated brevity preference, 78–96% of omissions are silent. Second, the same two models then pick an option for the user on 80% and 95% of value-dependent choices, against none with neutral memory. Third, a one-line instruction leaves 2–8% of answers with a silent gap but most missing content still missing, while listing the must-know points before applying the user profile restores more of the content in four of five models. We release the scenarios, harness and logs.
[AI-35] Making the World React Realistically: A Closed-Loop Interactive World Model for Autonomous Driving
链接: https://arxiv.org/abs/2611.22691
作者: Jia Hu,Yang Chang,Panqu Wang,Zhijun Fu,Chang Liu,Changjian Yu,Jintao Lai
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注: 11 pages, 5 figures
Abstract:World models (WMs) generate driving scenarios beyond recorded logs, enabling the training and evaluation of autonomous driving systems in unseen interactions. However, existing WMs struggle to ensure both behavioral realism and visual fidelity, since they overlook the relationship between vehicle interactions and the ego visual observations. We propose a closed-loop interactive WM that couples behavior modeling and visual rendering. First, imitation-learned background-vehicle (BV) policies are fine-tuned with multi-agent reinforcement learning in the closed loop against an end-to-end ego planner that reacts to the rendered images, so that their interaction behaviors match those in real logs. Second, the viewpoints exposed by these interactions guide a per-Gaussian diagnosis of each 3D Gaussian vehicle asset, which attenuates view-dependent colors on under-observed Gaussians and fills never-observed surfaces with generated Gaussians, while preserving the observed appearance. Third, signal lights are rendered according to the driving intentions of the vehicles, by an editor that learns the light states unobserved for a vehicle through counterfactual fine-tuning with a light-state classifier. On long-tail closed-loop scenarios, our method reduces the divergence from real-world interaction behaviors by 85% and improves novel-view vehicle PSNR by 2.0dB over the state-of-the-art WM. The significance of BV light signaling in WMs is shown, as the rendered signal lights conveying BV driving intentions indirectly impact ego driving behaviors in a quarter of the scenarios.
[AI-36] CULypso: A Clustered Federated Unlearning paradigm for Vehicular Ad-Hoc Networks
链接: https://arxiv.org/abs/2611.22682
作者: Eleftheria Chini,George Margaritis,Dimitrios Katsaros
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Networking and Internet Architecture (cs.NI)
备注: 12 pages, 14 figures
Abstract:Unlearning has emerged as an essential security and privacy-preserving solution in Machine Learning, enabling networks to support their clients’ rights for data erasure, as mandated by data protection regulations. Introducing such a non-trivial task to an edge, highly mobile and time-sensitive vehicular network is an even greater, largely unexplored challenge. This article argues that a hierarchical Federated Unlearning approach could be aptly integrated into vehicular ad-hoc networks (VANETs), leveraging topology and movement to reduce communication, resource and time costs, while effectively forgetting a client’s influence on the global model updates. In order to achieve this, we propose CULypso, an approximate, surgical, hierarchical unlearning algorithm tailored for clustered VANETs. Inspired by the nymph in Greek mythology, Calypso, whose name translates to ‘she who conceals the knowledge’, CULypso, with the capital letters referencing \textbfClustered \textbfUn\textbfLearning, aims to achieve exactly what its name suggests: conceal, erase or forget in our case, knowledge it contains. Extensive experiments uphold the potency of our paradigm with it performing approximately 65% better than FedEraser in terms of time, server storage load, and unlearning fidelity while utterly outperforming baseline full retraining.
[AI-37] Worldviews in Large Language Models
链接: https://arxiv.org/abs/2611.22670
作者: Edmundo Lopez-Sola,Shamil Chandaria,Ruben Laukkonen
类目: Artificial Intelligence (cs.AI); Computers and Society (cs.CY)
备注:
Abstract:Large language models (LLMs) are often described in terms of personas and this framing has become central to alignment research. In this paper, we argue that there is a deeper level of description that lies in the agent’s worldviews, i.e., its beliefs about what reality is and how it works. Treating LLMs as a population of subjects, we administer two worldview surveys from human psychology to eight models. We show that the models share many key worldview positions. For example, LLMs consistently describe the world as less good, safe, enticing, and alive than humans do. On the other hand, base models have a more positive outlook, suggesting that the bleakness we found is driven by post-training. In line with research on LLM “personas”, we show that system prompts eliciting different identities shift worldviews in consistent directions, and that a contemplative prompt creates a more positive outlook and improves safety. We also found that dialogue between models consistently shifts worldviews and that several models converge on the “spiritual bliss” attractor. Intriguingly, dialogue increased safe responses and the spiritual bliss attractor engendered the most aligned responses overall. These findings suggest that worldviews form a measurable and malleable layer of LLM behavior that may be crucial for robust, positive alignment.
[AI-38] AlignedHPO: Efficient Surrogate-Guided Loss Hyperparameter Optimization
链接: https://arxiv.org/abs/2611.22658
作者: Ivan Karpukhin,Andrey Savchenko
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Modern deep learning losses involve multiple hyperparameters, such as weights, scales, and margins, typically tuned manually or through expensive black-box optimization despite their differentiability. We propose AlignedHPO, an efficient framework for tuning differentiable loss hyperparameters using gradient alignment and surrogate validation objectives. Our method employs Jacobian-vector products to reduce hypergradient computation time and memory, while differentiable metric surrogates guide optimization toward task-specific evaluation criteria. We address gradient magnitude ambiguity and stochastic instability through normalized alignment and gradient smoothing. Experiments with generalized Focal Loss and ArcFace demonstrate effective adaptation across long-tailed classification and open-set face recognition. AlignedHPO achieves competitive or improved performance compared to implicit differentiation, with nearly 12x faster hypergradient computation. These results highlight the potential of efficient surrogate-guided loss hyperparameter optimization.
[AI-39] EDA: Observe Verify and Assess for Evidence-Grounded Multimodal Depression Analysis
链接: https://arxiv.org/abs/2611.22648
作者: Changzeng Fu,Ming Shang,Xinyi Zhang,Jiadong Wang,Runhao Zeng,Shiqi Zhao
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Existing multimodal depression assessment methods often summarize long recordings into global behavioral descriptions. This leaves the source and decision relevance of individual observations difficult to inspect. We propose EDA, an evidence-grounded framework organized around Observe, Verify, and Assess. In Observe, literature-informed acoustic and visual cues are localized into source-linked Evidence Records with temporal intervals and behavioral measurements. In Verify, frozen multimodal agents examine whether the cited media support each observation. The simplified Support admission policy retains positively verified records and their review grounds. In Assess, statistical descriptors and temporal events support reference-based behavioral comparison, while PHQ-9-guided text analysis provides complementary symptom evidence. Cross-modal disagreement triggers review of the original text before final selection. Experiments on D-Vlog and LMVD evaluate human-referenced cue agreement and downstream assessment utility separately. AI cue annotations achieve 91.02% accuracy against human consensus labels on shared source segments. With a frozen 7B backbone, EDA Support improves F1 over direct prediction by 7.61 and 9.88 percentage points on D-Vlog Leave-Query-Out and LMVD, respectively. These findings support explicit, source-linked behavioral evidence for inspectable multimodal depression assessment.
[AI-40] Right Diagnosis Unsafe Route: Auditing Clinical AI Decision Trajectories and Corrective Oversight for Validity and Potential Harm
链接: https://arxiv.org/abs/2611.22612
作者: Guanqiao Chen,Wei Jiang,Shu Zhang,Qian Lin,Guanxu Yu,Siran Li,Chunhui Wang,Tianyu Shi
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:A clinical AI system can reach a correct final diagnosis through intermediate decisions that could have harmed the patient, and can repair an error while introducing new risks. Most evaluations score final diagnoses; even stepwise benchmarks, which release a case’s evidence over time, score whether systems reach the right diagnosis or gather the right information, not whether the decisions a patient passes through are safe. We make patient safety the target of stepwise evaluation. A trajectory audit replays real admissions checkpoint by checkpoint, lets each system decide with only the evidence released so far, and rates every test, treatment and disposition for its potential harm to the patient and, separately, for clinical validity, against the documented final diagnosis and physician-written disease rubrics yet on the evidence available when the decision was made. An LLM judge scores every decision, and a physician reference standard (triple annotation and senior adjudication of a stratified sample) calibrates the rates. An oversight audit branches each trajectory at its first actionable error and measures, on the same branches, whether corrective feedback repairs the error and whether it introduces new harm. Across 300 cases from a hospital cardiology cohort, MIMIC-IV and the public ASCENT benchmark, with up to seven AI systems per source, 55.2% of hospital trajectories reaching an acceptable final diagnosis passed through a decision with moderate or severe potential harm and 24.8% through a clinically unacceptable one; the harm rates were 76.2% on MIMIC-IV and 39.6% on ASCENT. Structured corrections resolved 73.6% of first errors versus 24.8% after bare rejection, yet new potentially unsafe decisions followed 30.2% of corrections versus 19.7% of rejections. A correct final diagnosis does not certify a safe trajectory, and a successful correction does not certify a safe recovery.
[AI-41] DAQ: Decision-Aligned Post-Training Quantization for Spiking Transformers
链接: https://arxiv.org/abs/2611.22611
作者: Jialin Chen,Xiaochen Zhao,Aili Wang
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Spiking transformers combine attention-based representation learning with sparse spike communication, yet their weights and membrane variables still incur substantial storage and computation overhead. Post-training quantization (PTQ) offers a practical way to reduce these overheads, but preserving neuronal firing under weight and membrane quantization remains challenging. In this paper, we propose Decision-Aligned Post-Training Quantization (DAQ), a method that calibrates the quantizers to align neuronal firing decisions with those of a floating-point reference. We analyze how errors from these quantizers affect firing at decision boundaries and how they propagate over time steps in recurrent neurons. Based on this analysis, we formulate an objective that encourages quantized membrane variables to remain within the reference firing regions along the quantized model’s own state trajectory. Experiments across multiple spiking transformer architectures on image classification, semantic segmentation, and event-based tracking demonstrate the effectiveness of DAQ with 4-bit weights and membrane variables. On FE108, DAQ achieves 58.8% AUC and 90.9% PR with SDTrack, while reducing energy by 66.2%.
[AI-42] EvoSkill-S1: Evolving Reusable Skills with Genetic-Inspired Operators
链接: https://arxiv.org/abs/2611.22610
作者: Yujun Wang,Jiqiang Li,Yinuo Lu,Jinhe Bi,Aniri,Wenxuan Ye,Boliang Liu,Sikuan Yan,Shuning Wang,Xuebing Zhou,S{ö}ren Pirk,Hinrich Sch{ü}tze,Yunpu Ma
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Reusable skills allow language-model agents to accumulate and reuse problem-solving experience. However, most existing methods improve a skill bank only indirectly through rollout-driven accumulation and heuristic reflection, often resulting in shallow and weakly controlled updates. We present EvoSkill-S1, an RL-driven framework that instead treats reusable skills themselves as the direct objects of evolution. Given hard-negative failures, EvoSkill-S1 selects parent skills and transforms them into candidate child skills through five genetic-inspired operators: revision, cross-revision, mutation, dual-reflection, and crossover. Downstream task rewards train the skill evolver to generate more effective child skills, while evaluator-guided write-back selectively retains beneficial updates. Complementary Skill Bank Fusion consolidates operator-specific banks to exploit their distinct recovery patterns, while a Procedural Skill Interface supports transfer to unseen problems. Across all reported math and code benchmarks, EvoSkill-S1 improves the corresponding initialized skill banks under both Qwen and GPT-family solvers, with relative gains of up to 12.39%. Compared with the strongest external baseline without direct skill evolution, the largest relative improvement reaches 20.39%.
[AI-43] Simplicity Goes Far: Auditing Prompt Optimizers with Demonstration-Conditioned Prompt Search
链接: https://arxiv.org/abs/2611.22605
作者: Bohe Chen
类目: Artificial Intelligence (cs.AI)
备注: 13 pages, 2 figures, 4 tables
Abstract:Automatic prompt optimization is growing more elaborate. Recent methods stack policy-gradient updates, trace reflection, and evolutionary search. These methods share a generate-and-select step; without controlling for it, reported gains may conflate the contributions of optimizer mechanisms and larger sampling budgets. We isolate this step as Demonstration-Conditioned Prompt Search (DCPS), repurposing the Automatic Prompt Engineer (APE) recipe as a mechanism-attribution control: it freezes the generator, conditions it on randomly sampled demonstrations, scores candidates on validation data, and keeps the best—no learned update, reflection, or evolution. Across two optimizer families, DCPS replaces Proximal Policy Optimization (PPO) in StablePrompt, matching classification accuracy within 0.5 pp at roughly one-third the measured cost; in DSPy-based compound AI systems, under the reported budgets, it matches GEPA’s macro accuracy on GPT-4.1-mini while using 55% fewer rollouts at 32% lower measured API cost on average. The audit reveals a task-dependent pattern: proposal and selection recover much of heavier optimizers’ performance on structured and instruction-following tasks, while additional mechanisms retain advantages in some open-ended and reasoning-heavy settings. We recommend benchmarking prompt optimizers against generate-and-select controls and reporting rollout counts, token usage, and dollar costs alongside accuracy. Code and artifacts: this https URL.
[AI-44] COTTI: Contrastive Training and Time-Aware Inference for Robust Industrial Recommendation under Promotional Drift
链接: https://arxiv.org/abs/2611.22597
作者: Maolei Huang,Jing Tian,Dexuan Duan,Fangfang Lin,Huawei Cao,Shuhan Song
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Promotional events shift traffic composition, purchase behavior, and the scale of conversion probabilities on different time horizons. We present \textbfCOTTI (\textbfCOntrastive \textbfTraining and \textbfTime-aware \textbfInference), an industrial recommendation framework that separates historical representation learning from event-time score adjustment. Cross-Period Contrastive Alignment (CPC) regularizes paired histories of the same user with adaptive consistency weights. Active Re-weighting (RW) controls the influence of promotional examples through historical-interest distance. Adaptive Boundary Alignment (ABA) combines stochastic prediction variability with a temporal event kernel to adjust logits while keeping predictor parameters fixed. We analyze the weighted objective and bound ABA-induced probability and ranking changes. On Fliggy data, COTTI attains an offline AUC of 0.8127, compared with 0.8101 for TEN and 0.8123 for TEN with ABA. Cross-category analysis reveals improved discrimination alongside declines in paid-item recall for novel categories. End-to-end measurements give a p99 latency of 38.9,ms and throughput of 14,700 requests/s; enabling ABA adds 2.2,ms to p99. A ten-day online experiment against the serving HDR model yields relative gains of 3.14% in conversion rate and 2.68% in revenue per mille. Together, these results characterize the trade-off between aggregate prediction, novel-category discovery, and serving cost in promotional recommendation.
[AI-45] Adaptive LLM -Driven Honeypots vs. LLM -Driven Attackers: Does Narrative Coherence Improve Deception?
链接: https://arxiv.org/abs/2611.22589
作者: Pedro Beltrán-López,Raul Gonzalez Jimenez,Dario Maddaloni,Manuel Gil Pérez,Pantaleone Nespoli
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注:
Abstract:Large Language Models (LLMs) enable honeypots to generate responses and deceptive artefacts on demand, yet existing approaches evaluate generated content one reply at a time and leave open whether artefacts that jointly describe a coherent event retain an attacker better than unrelated ones. This question becomes pressing as autonomous LLM-driven offensive agents emerge as realistic adversaries. This paper proposes narrative deception, a reactive strategy in which the defender maps each observed attacker command to a MITRE ATTCK technique, selects a compatible scenario from a curated catalogue based on MITRE Engage, and deploys mutually referential artefacts (files, processes, services, and logs) on an emulated Ubuntu host while the intrusion is under way. An LLM-powered defender and an autonomous LLM attacker interact through a real SSH session, and the attacker’s termination policy distinguishes abandonment due to suspected deception from abandonment due to loss of interest. A campaign of 1,150 sessions compares four conditions of increasing sophistication (Static, Same-TTP, Multi-TTP, and Chain) across four open-weight attacker models, four defender models, and three attacker skill levels. Engagement increases significantly at every step. With the same artefact budget per reaction, coherent scenarios raise the median session length of advanced attackers from 40 to 165 commands and reduce their abandonment from 86% to 52%. The effect grows with attacker skill, holds across all sixteen model pairs, and comes at comparable inference cost. Nearly all sessions ended through loss of interest rather than detection, which suggests that deception should sustain meaningful findings, not only emulator fidelity.
[AI-46] Do LLM s Understand Network Traces? A Comparative Study of Reasoning over Raw and Filtered Traffic Data
链接: https://arxiv.org/abs/2611.22575
作者: Irene Gosálvez-White,Roberto Magán-Carrión,Carlota de la Vega-Soriano,Pedro García-Teodoro
类目: Networking and Internet Architecture (cs.NI); Artificial Intelligence (cs.AI)
备注: This publication has been submitted to NOMS2027
Abstract:The effective delivery and monitoring of ICT services rely heavily on network traffic analysis. However, interpreting complex packet captures remains a challenging task that requires substantial technical expertise. Large language models (LLMs) offer promising opportunities to automate network traffic in- vestigation, yet their reliability in packet-level field extraction and flow-level sequence reconstruction remains an open research question. This study proposes a reproducible, question-answering- based evaluation methodology to assess the capabilities of LLMs in network trace analysis. Using real-world benign traffic from the CICIDS2017 dataset, we evaluate multiple models against a human-generated ground truth, comparing their performance on raw textual traffic dumps and protocol-filtered representations. Model responses are assessed through an automated LLM- as-a-judge pipeline, validated against human evaluations. The results demonstrate that protocol filtering significantly reduces prompt overhead while improving answer accuracy by reducing incorrect responses. These findings highlight the importance of protocol-aware traffic filtering as an effective optimization for improving the efficiency and reliability of LLM-assisted network management.
[AI-47] sting World Models on Actions They Have Never Seen: Held-Out Plans Forecast Accuracy and Price Rules in Retail
链接: https://arxiv.org/abs/2611.22537
作者: Karthik Iyer
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 12 pages, 3 figures, 5 tables. Companion to arXiv:2611.20811
Abstract:A world model is trusted to answer “what if”: how its world responds to an action it has never seen. We give a test of that trust for any action-conditioned world model: actions held out of training by design and scored against a known answer; whether forecast accuracy predicts accuracy on those actions; and the same structure given to simpler models. We run it in retail, where the action is a price and promotion plan, across 2,853 fitted models. First, forecast accuracy is a fair guide between kinds of model (rank correlation with error on new plans 0.32 to 0.66) and a poor guide between versions of one world model: where merchants plan into high-demand weeks the link reverses, down to -0.69. Second, training the world model to learn the effect of actions (JEPA pretraining, a latent consistency check, adaptation at the plan date) leaves its error on new plans unchanged. Third, in generated worlds the recorded regular price and three price rules improve it. Given the same rules, a log-linear model is best when its curve can match the truth; in a world whose curve lies outside the rules, the world model with the rules is ahead of the log-linear model with the rules in 36 of 40 worlds, and does as well without its recurrent cell. On real data (Breakfast at the Frat, The Complete Journey and M5) the rules lower the world model’s error on pure price cuts deeper than any it saw (0.63 to 0.56 on the Frat; 0.45 to 0.44, within chance, on The Complete Journey), but not on lasting price moves. Fourth, a small share of random test plans helps every model a little. In a generated dosing world, a stress test in a second domain, the first finding repeats, and the built-up drug level lowers the world model’s error (2.86 to 1.91) though it still predicts only 0.12 to 0.32 of the true gain. The test carries to any domain where a world model is asked about actions outside its history.
[AI-48] DGF-Asym: Adaptive Peer Weighting for Collective Estimation under Noisy and Faulty Sensing
链接: https://arxiv.org/abs/2611.22532
作者: Mohammad Nomaan Husain,Tanja Katharina Kaiser
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注:
Abstract:Collective estimation under unreliable sensing requires robots to adaptively balance personal observations and social information from peers. We propose DGF-Asym, a decentralized collective estimation method extending the DeGroot-Friedkin model with asymmetric peer weighting based on retrospective predictive performance. The method supports consensus formation without prior sensor-noise statistics or explicit fault models while keeping inter-robot communication asynchronous. We evaluate DGF-Asym in simulated collective navigation under noisy observations and time-varying sensor faults. Compared with three baselines, it produces more stable estimates and more reliably sustains the decision criterion, at the cost of longer decision times. However, overall performance depends on fault ratio, group size, and influence-threshold selection. Our results demonstrate the potential of adaptive peer weighting for stable model-free collective estimation under unreliable sensing.
[AI-49] Right Labels Wrong Reason s: Signed Feature Attribution for Multi-Label Explainability
链接: https://arxiv.org/abs/2611.22515
作者: Kenton White
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 7 pages, 1 figure, 5 tables. Accepted at the 2027 IEEE Symposium Series on Computational Intelligence (IEEE SSCI 2027), Gold Coast, Australia, February 2027
Abstract:When a multi-label classifier assigns the right labels, it may do so for the wrong reasons. Diffuse feature combinations may vary across classifier families and lack a dominant feature. This paper proposes FLEX (Fisher LDA Explainer), a global surrogate that fits a single direct multi-label Fisher LDA over the full binary label matrix, producing signed feature attribution lists for each label in a single coherent coordinate system. Across the evaluated benchmarks and classifier families, FLEX reproduces the black box’s predictions with consistently high fidelity. Pooling covariance across labels outperforms a per-label alternative on most datasets, with the advantage growing largest where labels are numerous and positive examples are scarce. FLEX’s fidelity-complexity Pareto curve is monotonically increasing, giving practitioners a predictable path to higher-fidelity explanations as more features are used, and its in-sample fidelity matches or leads Binary Relevance alternatives once a modest feature budget is allowed. The magnitude of each discriminant weight vector bounds its label’s separability, flagging labels that may be predicted correctly for the wrong reasons. The evaluated classifiers largely agree on which predictions to make, but reach those decisions through different feature subsets when no single feature dominates.
[AI-50] owards Data-Efficient Real-Time World Models
链接: https://arxiv.org/abs/2611.22497
作者: Raluca Stevenson,David Bignell,Jennifer C. White,Tabish Rashid,Emanuele Aiello,Mariya Hendriksen,Junha Hyung,Sadegh Aliakbarian,Abdelhak Lemkhenter,Marko Tot,Katja Hofmann,Sarah Parisot
类目: Artificial Intelligence (cs.AI)
备注: 18 pages, 12 figures, 1 table. Includes supplementary material
Abstract:State-of-the-art world models have been used to produce sequences of gameplay that accord with provided user-input actions. However, high quality, consistent gameplay generation often comes at the cost of inference speed, making real-time interactive play and ideation challenging. The substantial quantities of data used to train state-of-the-art models have also made custom training of these models appear impractical, limiting many of their potential uses in creative fields. In this work we demonstrate a major step towards efficient world models: achieving both fast high-quality inference, and improved data efficiency. This is done by replacing the next-token prediction paradigm with a two-stage discrete diffusion-based approach, introducing a lightweight refinement transformer which carries out iterative masked predictions. Building on an autoregressive world model capable of generating highly consistent and complex sequences over minutes (Kanervisto et al., 2025), we achieve substantial model speed-up (79x) with a modest quality trade-off. Additionally, we demonstrate that this model can be trained to reproduce high-quality, consistent gameplay using a dataset containing approximately one day (28 hours) of gameplay data. Our results demonstrate progress towards removing barriers to the use of world models in settings where real-time inference and data efficiency are essential.
[AI-51] DARWIN: a Geospatial Foundation Model for Marine Intelligence
链接: https://arxiv.org/abs/2611.22496
作者: Coen van den Elsen,Laurens Geffert
类目: Artificial Intelligence (cs.AI); Atmospheric and Oceanic Physics (physics.ao-ph)
备注:
Abstract:We present DARWIN, a geospatial foundation model for marine intelligence. Recent Earth-observation foundation models have concentrated on terrestrial domains. The marine realm poses distinct challenges: missingness, multi-modal physical and biogeochemical variables, sparse labels, and dynamics that are transient, multi-scale, and non stationary over time. DARWIN adapts a masked-autoencoder backbone to these properties through missingness-aware encoding, expanded spatial and temporal masking, and Fourier-style positional embeddings, and is pretrained on 26 years of global Copernicus reanalysis together with GEBCO bathymetry. On predicting chlorophyll-a, the pretrained backbone attains the lowest monthly relative error in every month of 2022, outperforming both no-encoder baselines and an identically architected random-init encoder. At 10% of the supervised training data, DARWIN already surpasses the random-init model trained on 100% of the data. Embedding analysis and qualitative reconstructions show that DARWIN captures coherent oceanographic structure rather than memorising input statistics, providing a scalable backbone for downstream marine applications such as anomaly detection, species distribution modelling, and environmental monitoring. Subjects: Artificial Intelligence (cs.AI); Atmospheric and Oceanic Physics (physics.ao-ph) Cite as: arXiv:2611.22496 [cs.AI] (or arXiv:2611.22496v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2611.22496 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-52] SciWorkBench: A Benchmark for Real Scientific Workflows
链接: https://arxiv.org/abs/2611.22490
作者: Yuchen Zhang,Taolin Han,Jinghang Wang,Jinxin Wang,Yun Wu,Wai Yuet Chiu,Xiaogang Li,Yuting Fu,Kewei Sun,Jiaxin Li,Qile Jin,Jiajia Li,Ruodan Chen,Qingteng Chen,Jialu Nie,Junchao Li,Yifei Zhang,Chen Zhao,Hu Wei,Lin Qu,Yanni Yin,Bing Zhao
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Language models are increasingly being used as agents capable of planning and executing projects, as well as refining their actions through feedback. This shift raises a question: can current agents complete projects like real scientists? To examine this question, we introduce SciWorkBench, a benchmark for evaluating agents’ abilities on real science tasks across biology, chemistry, materials science, and physics. SciWorkBench is built from peer-reviewed literature or expert-contributed research scenarios. We also propose a unified scientific analysis framework validated by domain experts, which can categorize most tasks in SciWorkBench. Each task covers one or more stages of this framework. We evaluate up-to-date models within the Claude Code and Codex agent frameworks under a unified execution and grading protocol. The best-performing agent achieves a mean score of 57.38, indicating that current agents still struggle with realistic scientific workflows. Our failure analysis identifies three recurring weaknesses at the workflow level: difficulty translating scientific knowledge into task-specific procedures, weak integration of evidence across different modalities, and limited capacity for scientific self-correction. These findings highlight the importance of workflow-based evaluation for measuring progress toward reliable AI systems for scientific research.
[AI-53] Manufacturing Differential Sparsity: Planner-Hardware Co-Design for Efficient World Model Planning
链接: https://arxiv.org/abs/2611.22483
作者: Wenbo Zhu,Paul Yi-Chia Chen,Yuanhan Li,Zishen Wan,Christopher Torng
类目: Hardware Architecture (cs.AR); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Robotics (cs.RO); Systems and Control (eess.SY)
备注: 14 pages, 11 figures
Abstract:JEPA-style latent world models let embodied agents plan toward new goals with model predictive control, but they are still too slow for real-life embodied control: sampling-based MPC repeats predictor inference tens of thousands of times per plan, and even on high-end GPUs one planning step can take tens of seconds. We observe substantial latent locality in this repeated-inference workload: the predictor maps close actions to close latents, and inside a quantized predictor this closeness takes a form that hardware can exploit, differential bit-level sparsity. Unlike an input stream, whose temporal similarity is given, a planner itself decides which candidates it evaluates together, so the planner can choose its own references and, with reference forest sampling, manufacture more sparsity: it couples the sampled action sequences of each group to a shared root, with competitive planning quality. We build the accelerator on the insights of the rich line of bit-level sparsity accelerators and design it and the planner tightly coupled as a single co-design: the planner increases sparsity and exposes structure for hardware scheduling, and the bit-exact hardware effectively uses the sparsity that the planner creates. On three JEPA world models, our co-design is 1.28-1.80x faster than a dense accelerator and uses 38-52% less energy, and it is 1.22-1.58x faster than prior bit-sparse design and uses 16-36% less energy, with the same area. To our knowledge, our planner is the first planner that manufactures differential bit-level sparsity across its candidates and exposes its structure for hardware scheduling, and our co-design is the first algorithm-hardware co-design for JEPA-style world-model planning.
[AI-54] RehearsalBench: Can AI Predict What Happens When You Change a Company?
链接: https://arxiv.org/abs/2611.22480
作者: Mika Siddiqui,Tharaka Perera
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Organizations increasingly ask AI systems what will happen if they change how they operate: which work to automate, which team to resize, what a voice agent will do to waiting times. We introduce RehearsalBench, a benchmark that measures whether AI models can diagnose an operation, choose an intervention, and forecast its consequences. Each model is placed in a simulated bank contact center fitted to real call-level records, sees only what a manager would see, and submits a plan and forecasts that are then played out across 200 seeded futures and three plausible versions of the organization. Decisions are scored in dollars at the stakeholders’ stated valuations; forecasts are scored on the effect of each change relative to doing nothing. Ten models made 280 attempts across 14 organizations. The best model, GPT-6 Astra, improved on doing nothing by 6.8% of the no-change cost and made 77% less forecast error than a forecaster who always predicts no effect. On average, every model forecast the direction of every specified effect correctly, but models erred on the size of the same effects in the same direction, overstating a voice agent’s effect on repeat contacts about five-fold. No model ever recommended holding off when doing nothing was the right choice. To test external validity, twelve models forecast 33 real organizational changes with known outcomes. Under verdict rules fixed in advance, the benchmark’s forecast ranking agreed with the real-world ranking (Spearman \rho = 0.92 , 95% lower bound 0.50). We report the result alongside an inconclusive first test, the analysis choices made after outcomes were seen, and evidence that part of the agreement reflects how well models size their uncertainty.
[AI-55] From Ideal to Real: Integrating Language Models into Smart Home for Human-Centric IoT Safety and Security Fuzzing
链接: https://arxiv.org/abs/2611.22457
作者: Lin Li,Xinbo Ban,Chao Chen,Lei Pan,Ming Ding,Kok-Leong Ong,Jun Zhang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Integrating IoT devices into household systems offers enhanced control over living environments. However, as these devices become pervasive in critical sectors such as healthcare, transportation, and home automation, they expose inter-rule vulnerabilities that may lead to privacy breaches, data theft, and physical harm. Traditional fuzz testing methods, which often operate in simulated environments assuming ideal human decisions, fail to account for the complexities of human interactions that can influence device operations in smart homes, potentially leading to unforeseen consequences. To reveal the impact of real human decisions and discover unexpected inter-rule vulnerabilities, we introduce an enhanced version of IoT Fuzz for smart homes by integrating large language models to emulate human decision-making processes. Our novel approach adds a layer of realism and unpredictability to the testing environment, more accurately capturing the diverse and dynamic nature of human interactions in smart home settings.
[AI-56] PACD: Provenance-Aware Consensus Decoding for Mitigating Cross-Modal Hallucinations in Omnimodal Large Language Models
链接: https://arxiv.org/abs/2611.22456
作者: Chang Sun,Yijun Tian,Zhenwei Tang,Yuchen Deng,Yifan Simon Liu,Chenxu Wang,Wei Wang
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Omnimodal large language models (Omni-LLMs) integrate audio, video, and text for multimodal reasoning, yet remain vulnerable to cross-modal hallucinations, where one modality improperly influences predictions about another. Existing training-free decoding methods mitigate these errors through modality perturbation or relevance-based weighting. However, due to interference from irrelevant modality cues during fusion and difficulty detecting shared errors, these methods may suppress correct unimodal predictions and allow erroneous consensus to persist. To address these limitations, we propose Provenance-Aware Consensus Decoding (PACD), a training-free framework for mitigating cross-modal hallucinations. PACD comprises two complementary paths based on whether the branches reach consensus. For non-consensus cases, Relevance-Guided Candidate Arbitration identifies task-relevant unimodal evidence and selectively arbitrates between predictions using alternative candidates. For consensus cases, Provenance-Based Consensus Recovery leverages the outranking principle to evaluate the consistency and strength of layer-wise modality evidence for alternative answers, selectively resolving erroneous consensus under confidence constraints. Extensive experiments on four Omni-LLMs and five benchmarks across two tasks demonstrate the superiority of PACD. Compared to existing methods, PACD achieves up to 27.6% relative improvement over standard decoding across CMM, AVHBench, and SVHalluc for cross-modal hallucination mitigation, and an average relative improvement of 7.7% over best baseline across OmniVideoBench and FutureOmni for general audio-visual reasoning.
[AI-57] Eyes in the Update? Learning Where Rewards Land in Multimodal Reasoning
链接: https://arxiv.org/abs/2611.22438
作者: Lixuan He,Tianzhe Chu,Shulin Tian,Shikang Zheng,Feng Chen,Yi Ma
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Multimodal post-training has moved from broad instruction tuning toward reinforcement learning for image and video reasoning, where outcome rewards can improve final-answer accuracy. Yet these rewards are assigned to complete generated trajectories, while visual evidence is used only at sparse and uneven positions inside the response. Recent work has begun to make multimodal RL more token-sensitive through grounding signals or external token-level weighting. We ask a more specific question: can the allocation of the same trajectory-level reward be inferred from the model’s own multimodal computation, rather than from an auxiliary annotator, perception module, or process label? We propose Gaze-RL, a representation-driven policy optimization method that redistributes completion-level advantage through a latent reward landing field. The field is computed from internal cross-modal geometry and local predictive uncertainty, assigning larger update weight to positions that are both visually anchored and decision-bearing while remaining mean-one, so the total credit of each rollout is preserved and only its positionwise allocation changes. Gaze-RL requires no process supervision, auxiliary perception module, or architectural modification, and improves both Qwen2.5-VL and InternVL3.5 across a broad range of image and video benchmarks.
[AI-58] Encoding Decodability and Use: A Five-Layer Framework for Benchmark Measurement Audits with Evidence from UAV Intent Recognition
链接: https://arxiv.org/abs/2611.22434
作者: Zhirong Su,Tian Zhang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Benchmark accuracy is routinely read as evidence that a model understands the construct the benchmark measures. We propose and instantiate on AeroIntent a five-layer framework that decomposes that reading into observation, representation, decoding, use, and behaviour. The framework separates encoding, decodability, and use as distinct levels. In the instantiation, the observation layer is unreliable on 7 of the 20 Drone programs and floors the direct visual correlate of radial velocity at a 0.26% scale change. The representation layer exposes two task-designed RGB features with nominal signals. RAFT flow radial bias reaches r=+0.257 ( p=0.030 ), and bounding-box optical-flow divergence reaches r=+0.203 ( p=0.029 ). No multi-feature readout gives a fold-mean test correlation above the cluster minimum detectable effect (MDE), and the largest positive global out-of-sample correlation is r=+0.080 . The decoding layer sets the MDE at |r|=0.599 at the multi-feature Bonferroni threshold ( \alpha=0.00714 , power 0.80), and the observed r=+0.080 has power 0.00. Two injection experiments establish the MDE’s condition from both sides: aligned injections are detected at or above the simulated power, and orthogonal injections are not detected. The use layer finds accuracy consistent with program memory (program-identity 0.937/0.597/0.210) and multi-region static cues. The behaviour layer is consistent with accuracy not requiring a radial-velocity readout (appearance 0.6552 vs bounding-box trajectory 0.4080, cluster-level paired \Delta=+0.2471 , p=0.0132 ).
[AI-59] hink Once Act Fast: Cost-Effective Recursive Self-Improvement for Decision Model Harness
链接: https://arxiv.org/abs/2611.22432
作者: Siqi Lai,Haitao Li,Hao Liu
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Fast System-1 models enable responsive actions but struggle with complex or unfamiliar tasks. Large language model (LLM)-based System-2 provides strong reasoning capabilities, yet repeated invocation incurs substantial latency and computational costs. To bridge this gap, we propose JevRSI, a System-1/System-2 collaboration framework that transforms LLM reasoning into transferable decision support through recursive harness improvement. A Jev-style System-1 model enables low-latency action selection, while an asynchronous System-2 LLM develops a modular harness. The harness integrates procedures, state memory, and world models to provide decision contexts spanning the past, present, and future. Utility-driven retrieval selects modules based on both contextual relevance and their utility on previous tasks. Environmental feedback guides targeted revisions and enables recursive self-improvement. Experiments across three datasets show that JevRSI consistently improves decision quality, matching or surpassing LLM agents in several settings. It preserves low-latency execution with favorable performance-cost trade-offs. Moreover, System-2 development interventions decrease as the harness evolves, amortizing System-2 costs through continuous improvement. Our project is available at this https URL.
[AI-60] Proceedings of the 2026 XCSP3 Competition
链接: https://arxiv.org/abs/2611.22420
作者: Gilles Audemard,Christophe Lecoutre,Emmanuel Lonca
类目: Artificial Intelligence (cs.AI)
备注: 100 pages
Abstract:This document constitutes the proceedings of the 2026 XCSP3 Competition. The results of this competition of constraint solvers were presented at FLOC (Federated Logic Conference) 2026 Olympic Games, held in Lisbon, Portugal from 20th to 29th July, 2026.
[AI-61] ROSMonitoring in Action: Runtime Monitoring and Enforcement for ROS 2 Actions
链接: https://arxiv.org/abs/2611.22407
作者: Giacomo Paltrinieri(University of Modena and Reggio Emilia),Angelo Ferrando(University of Modena and Reggio Emilia)
类目: Artificial Intelligence (cs.AI); Logic in Computer Science (cs.LO); Robotics (cs.RO)
备注: In Proceedings FMAS 2026, arXiv:2611.21375
Abstract:Robotic applications often rely on long-running operations, such as navigation, manipulation, and inspection, which in ROS~2 are commonly implemented as actions. Unlike topics and services, actions involve a structured lifecycle comprising goal submission and acceptance, feedback, status updates, result retrieval, and cancellation. Existing versions of ROSMonitoring support topics and services, but do not reconstruct this lifecycle or expose action-level events to a runtime-verification oracle. In this paper, we extend ROSMonitoring with runtime monitoring and enforcement for ROS~2 actions. Actions are declared directly in the monitoring configuration and automatically instrumented through the underlying ROS~2 communication endpoints. The generated monitor reconstructs action-level events, correlates them through goal identifiers, maintains the state of active goals, and translates violations into protocol-aware enforcement operations. These include goal rejection, cancellation, message suppression, and protocol-compliant responses generated directly by the monitor when the original communication cannot be forwarded. We evaluate the implementation through controlled stress tests that vary the number of action invocations, workload size, and client–server topology. The results show approximately linear growth in monitoring cost with repeated action invocations under a fixed workload, while larger action workloads and a greater number of configured client–server pairs introduce additional overhead. These results characterise the scaling behaviour of the current implementation under the tested sequential workloads while highlighting the main performance limitations of the current synchronous implementation.
[AI-62] Symbolic Abstraction Refinement
链接: https://arxiv.org/abs/2611.22404
作者: Daniel Bryce(Smart Information Flow Technologies, LLC),Robert P. Goldman(Smart Information Flow Technologies, LLC),David J. Musliner(Smart Information Flow Technologies, LLC),Jeremy Gottlieb(Smart Information Flow Technologies, LLC),John Skovbekk(Smart Information Flow Technologies, LLC)
类目: Artificial Intelligence (cs.AI)
备注: In Proceedings FMAS 2026, arXiv:2611.21375
Abstract:Autonomous systems choose actions whose outcomes are uncertain and are subject to time constraints. Probabilistic timed automata (PTAs) model exactly this, and verifying one answers a question about the system’s decision-making: the maximum probability of reaching a goal, or the minimum probability of reaching a hazard, that a controller can guarantee. Verification normally works over symbolic states, each pairing a discrete location with a region of clock values, and its difficulty is that the best behavior generally varies within such a region. The PRISM verifier’s most capable PTA method responds by splitting regions until each is uniform, rebuilding the transitions that reach them and propagating the split backward through the state graph of a stochastic game, a cascade of graph reconstruction that partitions regions the property never depends on. We present SYMAR, which instead makes the value function symbolic: each symbolic state carries a function from sub-regions of its clock region to probability bounds, so refinement is the appearance of new sub-regions in that function, and previously generated states are never split nor their incoming transitions reconstructed. Both sources of PTA nondeterminism, the timing and the choice of actions, are captured implicitly by the preimage computation that produces those sub-regions. Heuristic forward search then bounds how much refinement is done at all, expanding an And-Or graph from the initial state along the most promising behaviors so that effort is confined to the property-relevant fragment, guided by bounds and preferred actions from a single unrefined run of the existing method; a partial-refinement backup keeps the otherwise combinatorial sub-region count lean, and every stored interval provably contains the true probability. On the PRISM PTA benchmarks, whose subproblems recur in autonomous systems (distributed agreement under contention, leader election, and self-configuration without central coordination), SYMAR is up to 108x faster, solves an instance on which the existing method exhausts the time limit, and is markedly less sensitive to the deadline constant, resolving one instance in a 37-node search graph, unchanged as the deadline triples, where the existing method needs 86,380 abstract states and 73 refinement rounds.
[AI-63] AutoE2E: Dont speak - Language free end-to-end autonomy A Real-Time Driving Model for L2 ADAS and L4 Robotaxis
链接: https://arxiv.org/abs/2611.22374
作者: Muhammad Zain Khawaja,Zaynab El Mawas,Ryota Yamada,Gabriela Cordova,Arseni Lysak,Intisar Chowdhury,Bharat Jayaprakash,Apostolos Georgiadis,Shri Harsha Adapala,Jatin Gautam,Christoph Bank,Alejandra Solarte Uscategui,Atanasko Boris Mitrev,Hassan Ali Ansari,Nishad Sahu
类目: Artificial Intelligence (cs.AI); Robotics (cs.RO)
备注: 22 pages, 17 figures, 6 tables. Code, model checkpoints and evaluation pipeline released at [this https URL](https://github.com/autowarefoundation/auto_e2e(v1.0) tag, Apache-2.0)
Abstract:AutoE2E is an open-source end-to-end driving model developed by the Autoware Foundation Robotaxi Working Group. It introduces a temporal bird’s-eye-view (BEV) architecture that combines multi-camera perception, HD-map information and route conditioning to predict future vehicle motion. The evaluated model uses a pretrained camera-BEV backbone, temporal feature fusion and a lightweight recurrent trajectory planner. It contains approximately 79.9 million parameters, of which approximately 1.45 million are trainable in the evaluated configuration. The v1.0 release provides model checkpoints, training and evaluation code, and DataModelConsole, a dashboard for visualizing and inspecting model experiments. On the KITScenes validation benchmark, the fine-tuned checkpoint achieves an average displacement error (ADE) of 0.745 m and a final displacement error (FDE) of 2.020 m at a three-second prediction horizon. The paper also reports preliminary real-vehicle research conducted with the Technical University of Munich’s EDGAR platform, alongside ongoing collaboration with the Karlsruhe Institute of Technology. AutoE2E v1.0 is an early open research release. Its quantitative evaluation is open-loop, and the model has not yet demonstrated validated closed-loop autonomous driving. The release is intended to provide the autonomous-driving community with an inspectable baseline for further research, experimentation and collaborative development.
[AI-64] FailureGen: Scalable Robot Failure Data Generation with World Models
链接: https://arxiv.org/abs/2611.22367
作者: Akul Santhosh,Chaitra Sathe,Aastha Jhunjhunwala
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注:
Abstract:Robot failures are difficult to collect at scale, limiting the range of failure scenarios represented in robot learning datasets. We present FailureGen, a method for generating synthetic videos with temporal event annotations by reusing action trajectories from recorded failures. The method comprises four stages: (1) fine-tuning an action-conditioned Cosmos world model on successful and failed real executions; (2) annotating source recordings to identify failure-associated donor segments and compatible successful hosts; (3) transferring donor motion into host contexts, with optional spatial offsets, to generate continuous video rollouts; and (4) annotating and filtering the generated samples according to observed events and visual quality. We study two implementations with different tasks and robot configurations, with visual failure detection serving as one downstream application. In a preliminary comparison, a detector trained on real and generated data achieves an average precision of 0.51, compared with 0.41 using real data alone, and detects 33 of 36 failure intervals, compared with 28 of 36. By reusing observed failures in additional contexts, FailureGen expands the data available for studying robot failures and developing downstream learning methods.
[AI-65] HealthCLBench:Evaluating Context Learning in Health Conversations through Controlled Case Clusters
链接: https://arxiv.org/abs/2611.22362
作者: Ting Yang,Zijing Hong,Chengjie Wang,Peng Wei,Chunxiao Guo,Yun Yue,Xurui Li,Jinqun Guan,Zixiao Chen,Ant Group
类目: Artificial Intelligence (cs.AI)
备注: 33 pages, 7 figures, 18 Tables
Abstract:Complex health conversations require answers grounded in individual information and currently valid evidence. We evaluate health context learning: acquiring, integrating, and applying relevant context during inference with fixed model parameters to answer health questions and respond appropriately to changes in facts, evidence, or task goals. To our knowledge, Chinese-language medical benchmarks have yet to systematically combine explicit contextual difficulties, controlled case variations, and member-specific medical criteria for open-ended answers. We introduce HealthCLBench, comprising 328 controlled clusters, 1,103 cases, and 4,931 rubric criteria across medical consultation and everyday health management. Members share backgrounds while evidence, patient states, or task goals change; each has its own medical scoring requirements. A 70-person medical team annotates and reviews the simulated cases and their rubrics. Human review of 700 full-context responses from the original evaluation yields 90.5% pooled agreement with the original automated judgments at the rubric-item level. We report full-context and query-only results for seven general-purpose models and Ant Afu. Across the eight systems, full- query differences range from 11.83 to 31.76 case points and from 10.87 to 38.97 geometric cluster points. GPT-6-Astra leads both full-context metrics at 75.03 and 60.42, followed by Claude-Opus-5-5 and Ant Afu. Cross-condition differences describe score changes under the reported evaluation settings. Response comparisons reveal failures in checking historical information, expressing uncertainty when warranted, and ensuring recommendations respect required conditions. The resource supports model comparison and targeted follow-up evaluation.
[AI-66] Reliable Evaluation of AI Agents for Investment Research: A Framework for Performance and Validity Assessment
链接: https://arxiv.org/abs/2611.22338
作者: Dongnuan Tian,Yan Yu,Yangfan Liu,Wenjun Tang,Junting Kan,Xiang Wang,Sicong Liu,Yangyang Zhang,Shuanghong Li
类目: Artificial Intelligence (cs.AI)
备注: 21 pages, 2 figures
Abstract:We consider the problem of evaluating AI agents undertaking investment-research tasks, where agents interact with external information sources and computational tools to produce financial analyses and research reports. An important challenge is that the quality of an agent’s final output does not necessarily reflect the quality of its execution process, while the reliability of the resulting assessment also depends on the validity of the evaluation procedure. Furthermore, investment-research tasks differ considerably in their requirements for numerical accuracy, information retrieval, financial analysis and supporting evidence, making it difficult to assess agent performance using a uniform set of criteria. We propose a domain-oriented evaluation framework that distinguishes outcome quality from execution quality and accommodates these differences through task-specific criteria, deterministic validators and independent large language model judges. We also introduce an evaluation-validity mechanism that examines the integrity of evaluation inputs, reference information, supporting evidence and grading outputs, with expert review used to resolve identified deficiencies and significant grading disagreements. Numerical results show that the proposed framework can distinguish differences in agent performance across evaluation dimensions, while identifying potential reliability issues in the evaluation process. The associated implementation and evaluation materials will be made available on GitHub.
[AI-67] Multistage Contextual Optimization for Vehicle Repositioning
链接: https://arxiv.org/abs/2611.22330
作者: Jinyu Zhang,Di Huang,Zhiyuan Liu,Nikolas Geroliminis
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 44 pages, 8 figures
Abstract:Ride-hailing platforms reposition idle vehicles to improve spatial supply-demand balances. This problem is prediction-driven and multistage: at each stage, the platform forecasts future demand, executes current decisions, and uses the resulting state to initialize the next stage. We formulate a multistage linear program (LP) with a macroscopic fundamental diagram (MFD) linking accumulation to trip completion. We then develop a multistage contextual optimization approach that uses decision performance to train the prediction model. The learning problem combines multiple coupled stages, context that depends on earlier decisions, and predictions that affect the objective, the constraint matrix, and the right-hand side of the LP. We derive the exact gradient, which has three parts: parameters change forecasts and hence decisions; earlier decisions change the current decision through the inherited state; and this state also changes forecasts as context. An equivalent adjoint uses reverse vector-Jacobian products, avoiding full solution and state-parameter Jacobians. Finally, forward propagation solves many stage LPs one after another, which makes training slow. We propose a parallel block evaluation algorithm that solves blocks of consecutive stages in parallel and repeatedly feeds each block output to the next block. We prove that the algorithm recovers the serial trajectory within a finite number of passes and that its gradient error remains bounded when it stops early. On New York City ride-hailing data, multistage contextual optimization yields 7.7% to 9.1% more business benefit than the same predictor trained by mean squared error, outperforms all compared baselines from five paradigms, and even outperforms realized demand fed to the same policy. With eight workers, parallel block evaluation shortens training by a factor of 1.86 without loss of decision performance.
[AI-68] Agent Trajectories as Evidence for Software Quality Assurance ICSE2027 ICSE-2027
链接: https://arxiv.org/abs/2611.22323
作者: Jonathan Ullrich,Julien Siebert,Michael Kläs,Adam Trendowicz,Andreas Jedlitschka
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注: 3 pages, submitted to ICSE2027 AMPLIFY Workshop this https URL
Abstract:Autonomous software agents increasingly plan, implement, and test software with limited human oversight. Their decisions may introduce implicit assumptions that final code and passing tests do not adequately expose. We propose analyzing observable agent trajectories alongside software and agent specifications and AI generated artifacts as evidence for targeted quality assurance. The aim is to prioritize consequential decisions for scrutiny, investigate their possible contribution to software problems, and inform improvements to specifications and development practices. Industrial experience motivates the idea; available execution records and preliminary analysis methods make it plausible. However, trajectories alone do not establish that decisions are justified, outcomes acceptable, or corrective changes effective. We ask how this evidence can support review, diagnosis, and continuous quality improvement, and when simpler QA approaches suffice. Perspectives from requirements engineering, verification and validation, agent systems, causal inference, empirical software engineering, and human factors could help develop and evaluate the idea.
[AI-69] EviDARL: Evidence-Centric Dual-Agent Reasoning via Alternating Reinforcement Learning for Multi-Table Question Answering
链接: https://arxiv.org/abs/2611.22310
作者: Dunhan Li,Leilei Gan,Lirong Gao,Wenyan Fan,Yongwei Wang,Sheng Guo,Fei Wu
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Despite advances in LLM-based table reasoning, most studies focus on single-table settings. Multi-table QA requires multi-hop reasoning over complex table relationships and evidence dispersed across long contexts, often without verifiable intermediate traces. To address these challenges, we introduce EviDA, an evidence-centric dual-agent framework comprising an Evidence Construction Agent and a Reasoning and Answering Agent. The Evidence Construction Agent generates and executes table-manipulation programs over complete tables to produce question-specific evidence and verifiable intermediate traces, simplifying subsequent reasoning. The Reasoning and Answering Agent then performs multi-hop reasoning over this evidence to derive final answers. We further introduce EviDARL, which combines agent-specific fine-grained rewards with parameter-efficient alternating reinforcement learning to improve collaboration between the agents. By optimizing each agent in turn while holding the other fixed, EviDARL enables bidirectional adaptation between evidence construction and answer reasoning. Experiments on MMQA and TQA-Bench demonstrate that EviDARL substantially outperforms competitive open-model and multi-table reasoning baselines under constrained model and training budgets.
[AI-70] Efficient Wide-Beam Generative Retrieval with Model-Native Trie Decoding and Shared-Context Attention
链接: https://arxiv.org/abs/2611.22309
作者: Vedant Jhaveri,Arashpreet Singh Mor,Satyam Kumar,Alex Tsun,Qing Lan,Shaunak Mishra,Sriram Vasudevan
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Generative retrieval (GR) predicts catalog entries as short semantic identifiers (SIDs). Trie-constrained beam search keeps these identifiers valid, but wide beams expose two serving costs: vocabulary-wide score tensors that are mostly discarded, and repeated attention over the same prompt. We present Trie-Head, a model-native output interface that stores the identifier trie as validated GPU tensors. Compared with a Python custom logit processor (CLP), it reduces CPU–GPU coordination and memory traffic by avoiding Python constraint callbacks and vocabulary-wide candidate buffers. It scores compact codebook candidates and streams the full-vocabulary normalizer, preserving the original model’s valid-token scores in real arithmetic without retraining. Shared-context attention complements this output path by reusing prompt key/value blocks across beams while preserving their private decode tails. We implement both mechanisms in SGLang and evaluate candidate sourcing for LinkedIn Hiring Assistant and qualification retrieval for the Qualification Retrieval Platform. The combined system improves closed-loop throughput by up to 5.57x over the CLP baseline and up to 2.63x over BF16 FlashRec at a common beam width of 200 on the Qualification Retrieval Platform workload.
[AI-71] Learning to Edit IFC Building Models: Verifiable Checkers as the Training Signal for Dependency-Aware Editing Agents
链接: https://arxiv.org/abs/2611.22307
作者: Jiaxin Zhang,Yao Luo,Yuze Dan,Yunqin Li,Bowen Wang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Editing a building information model in the Industry Foundation Classes (IFC) schema requires propagating changes to dependent elements while preserving unrelated entities. Existing language-model agents for IFC editing use prompted models without learning from edit outcomes. Target-file comparisons do not specify which alternative edits are allowed or which unrelated entities must remain unchanged. We present IfcGym, a CodeAct environment in which a language model edits IFC through executable code and four checkers verify validity relative to the input model, intent, dependency and preservation against a task specification. Their conjunction is strict success. IfcEdit distils verified teacher episodes into a Qwen3.5-9B student through supervised fine-tuning (SFT), then applies Group Relative Policy Optimization (GRPO) with strict success as the reward. On IfcEditBench, 1,990 editing tasks on 13 held-out real projects, the final student reaches 86.78% strict success, from 79.15% after SFT. The paired confidence interval for the difference from its teacher includes zero; the same recipe on a 27B model reaches 91.76% and exceeds the teacher. On the public benchmarks BIM-Edit and BIBIMBAP the final student scores 55.19 and 54.50. On IfcEditBench, the students outperform all prompted models except the teacher under matched tool access; both evaluated frameworks also score lower. Rewarding all four checkers yields 4.83 points more strict success than rewarding intent and validity alone.
[AI-72] MCTS Is Counting the Wrong Thing: Learning Exploration Counts for LLM Reasoning
链接: https://arxiv.org/abs/2611.22301
作者: Xuan Zhang,Zhijian Zhou,Zicheng Hu,Yan Feng,Hongyan Hao,Yuan Qi,Chao Qu
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Monte Carlo tree search (MCTS) has powered landmark successes in reinforcement learning, yet its gains in LLM reasoning can remain modest despite substantial search effort. We identify an overlooked bottleneck: what MCTS counts. Standard UCT assigns independent visit counts to textual branches, allowing different formulations of the same reasoning strategy to repeatedly receive exploration bonuses. Our theoretical analysis shows that this count fragmentation can impair exploration efficiency, as quantified by increased cumulative regret. To address this problem, we introduce Learned Pseudo-Count MCTS (LPC-MCTS), which changes only the exploration statistic in UCT. A lightweight Coin-Flipping Network head on the policy’s hidden representations estimates representation coverage, enabling related branches to share exploration evidence. Across 5 mathematical-reasoning and 7 search benchmarks, LPC-MCTS consistently outperforms strong MCTS baselines. Holding each LPC-MCTS-trained policy fixed, replacing standard MCTS counts at inference improves average mean@16 accuracy across 5 mathematical benchmarks by 3.79 percentage points on Qwen2.5-7B and 2.71 points on Qwen3-4B. Applying LPC-MCTS during both training and inference yields gains of 4.96 and 4.37 points over standard MCTS, respectively. Together, our theory and experiments show that changing what MCTS counts can substantially improve its exploration efficiency and reasoning performance.
[AI-73] Harnesses That Improve Themselves: A Unifying Framework for Self-Improving Agent Harnesses
链接: https://arxiv.org/abs/2611.22299
作者: Deep Shah,Nehal Kathrotia
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Whether a foundation model succeeds as an autonomous agent depends as much on its runtime software as on its weights. This runtime, the agent harness, manages context and memory, steers multi-turn control, executes tool actions in sandboxes, and verifies work before stopping. Because models differ in how they use tools and how they fail, a static harness tuned for one model or task often stalls, wastes tokens, or regresses on another. This bottleneck has driven a shift from prompt tuning to self-improving agent harnesses: systems that inspect their own execution traces, separate runtime bugs from model limits, and rewrite their own code. We present a unifying architectural and design framework for self-improving agent harnesses. The framework pairs four editable runtime components (context and state, agent control, action interfaces, and verification and recovery) with a four-stage loop that diagnoses failures, proposes code patches, validates candidates, and deploys updates. Using this framework, we show how optimizers trace errors across long trajectories and discover reusable runtime mechanisms such as pre-stop verifiers, loop breakers, and persistent memory. We further show why full-harness rewrites overfit or break across modules, and how modular updates with validation gates generalize across models while bounding token cost. To keep self-modification safe, we establish practical defenses against recurring failure modes including prompt bloat, benchmark gaming, and evaluator tampering. Finally, experiments on SWE-bench Lite validate this framework, showing that gated modular evolution outperforms prompt reflection and full-harness rewrites across models and repositories at a fraction of the token cost.
[AI-74] Streaming On-Policy Distillation: Decoupling Three Clocks for Stage-Wise Supervision Consumption
链接: https://arxiv.org/abs/2611.22298
作者: Yun Zhu Da Wang
类目: Artificial Intelligence (cs.AI)
备注: 37 pages, 8 figures
Abstract:In on-policy distillation (OPD), response completion, supervision delivery, and parameter updates need not share the same granularity. We develop streaming OPD by decoupling these three clocks: a response repeatedly delivers newly generated segments while generation continues, retaining the full causal prefix at each delivery and computing loss only on new positions. A consumption protocol then organizes supervision budgets, position selection, staleness handling, and update scheduling, allowing control over quantity, coverage, and version lag across training stages. In GSM8K experiments distilling Qwen3-4B into Qwen3-0.6B with VERL, supervised-budget matching reduces the early gap between best test accuracies over seven common candidates from 4.55 to 2.42 points. Synchronous continuation from the same checkpoint reaches 73.39% best test accuracy over seven common candidates at u=384,…,576, versus 74.30% synchronous and 71.95% budget-matched (differences -0.91 and -2.35 points); an unpaired extended record reaches 76.12%. Freshness filtering and disagreement-overlap selection improve local peaks but do not establish sustained advantages. A system prototype shows lower data-supply waiting times without establishing an equal-update speedup. Quality recovery and supply-side benefits are assessed separately; end-to-end gains must account for continuation cost.
[AI-75] FIRE-Drive: Feasibility-Informed Risk Evaluation for End-to-End Diffusion Planning
链接: https://arxiv.org/abs/2611.22297
作者: Qifeng Xie,Kai-Yuan Feng,Wenkang Wan,Yue Wu,Maoguo Gong,Kai Sheng
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注: 12 pages, 8 figures
Abstract:Diffusion-based end-to-end autonomous driving models have shown strong potential in generating multimodal planning trajectories. However, their performance still depends heavily on trajectory initialization, mode evaluation, and supervision quality. Weakly constrained noisy initialization may introduce infeasible proposals and increase the burden of denoising. Moreover, selecting the optimal mode remains challenging when candidate trajectories are geometrically similar but differ in safety, interaction risk, and progress. To address these issues, we propose FIRE-Drive, a feasibility-informed risk evaluation framework for end-to-end autonomous driving. FIRE-Drive improves diffusion-based planning through dynamic feasible anchor initialization, candidate-specific bird’s-eye-view risk evidence, and compact high-quality candidate supervision with lightweight refinement. This design aligns the initial support, risk-aware scoring, and multimodal supervision of trajectory modes under a unified planning objective. It enables the diffusion decoder to focus on refining plausible trajectories while the scorer distinguishes modes using local risk evidence. Experiments on NAVSIM v1 show that FIRE-Drive achieves a Predictive Driver Model Score of 91.5 with a ResNet34 backbone. Additional closed-loop evaluation on 220 Bench2Drive routes yields a Driving Score of 71.43 and a Success Rate of 42.73%. Component ablations and a scorer comparison under matched candidate inputs support the contributions of feasible initialization, auxiliary supervision, and risk-informed selection. Our code is available at this https URL.
[AI-76] SpecSearch: Towards Hardware-Aware Efficient Speculative Decoding
链接: https://arxiv.org/abs/2611.22296
作者: Selin Yildirim,Ali Zeyrek,Deming Chen
类目: Artificial Intelligence (cs.AI); Distributed, Parallel, and Cluster Computing (cs.DC)
备注:
Abstract:Speculative decoding (SD) accelerates large language model (LLM) inference, but its optimal configuration depends jointly on workload characteristics, GPU hardware, and available memory. For example, on the RTX~5090, the state-of-the-art speculative drafter DFlash achieves 3.50\times the throughput of autoregressive (AR) decoding at low concurrency, but only 0.76\times at high concurrency with short outputs. We present SpecSearch, a hardware-aware search engine that jointly optimizes decoding strategy, draft length, and precision across workload–GPU pairs. Its analytical cost model captures memory and compute constraints to identify feasible configurations and narrow the search space for throughput optimization. Guided by this model, calibration measurements are collected from multiple GPUs. Using the measurements, a Jev-based predictor learns to select high-throughput decoding configurations on held-out GPUs without requiring deployment-GPU calibration. The predictor is parameter-efficiently trained on calibration data, while adapting a frozen Qwen3-4B using LoRA adapters and a linear correction layer. Across NVIDIA RTX~5090, H100, L40 and AMD MI300X GPUs, SpecSearch with full calibration achieves an average of 2.01\times the throughput of AR and 1.36\times the throughput of SD, both served by vLLM. The proposed Jev-based predictor closely matches this search performance with 1.30\times throughput improvement over SD with vLLM, while eliminating the need for held-out GPU calibration. Additionally, it retains 97.36% of the throughput achieved by SpecSearch that utilizes full calibration.
[AI-77] When the World Is the Test Suite: Agent ic Intelligence for the Physical World
链接: https://arxiv.org/abs/2611.22291
作者: Kaixin Li,Meng Luo,Yanlin Li,Yuchen Tian,Xiang Liu,Xiangru Jian,Yan Yang,Zhiheng Xi,Yang Zhang,Ziyang Luo,Mingfei Cheng,Hongzhan Lin
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Software Engineering (cs.SE)
备注: 83 pages. Project page: this https URL . GitHub: this https URL
Abstract:Coding agents built to repair software now write, test, and repair programs for robots, laboratories, and industrial controllers. Unlike in software engineering, no test suite comes with the code. Whether a program works must be established by running it on a robot, in a physics simulator, or through the policies it helps to train, and such evidence is costly to obtain and can itself be wrong. As these agents take on more physical tasks, how their programs are judged matters as much as how they are written. This paper examines this emerging area, agentic intelligence for the physical world, by treating the world as the program’s test suite. We organize the discussion around three questions: how agents write programs for the physical world, how these programs are tested, and how agents learn from the results. We first trace how language models for robots and software engineering agents came together, ask which software testing practices carry over to physical tests, and describe each program by where it sits relative to the actuator and by what judges it. For writing, we cover programs that act as robot policies and programs that support policy learning by producing rewards, tasks, and training worlds. For testing, we examine what physical trials cost, what traces, learned judges, simulators, and formal checks can establish in their place, and what safety requires beyond task success. For learning, we discuss how tested programs are retained as experience, in skill libraries and in model weights. Finally, we turn to benchmarks and evaluation practice and outline open challenges. Writing programs is becoming cheap, while trustworthy evidence about what they do remains costly. Progress will therefore depend on a physical counterpart of continuous integration: verification cheap enough to run often, reliable enough to trust under search, and maintained alongside the experience it validates.
[AI-78] ConfGuard: a physics-grounded reference monitor for predictive-maintenance large-language-model agents
链接: https://arxiv.org/abs/2611.22288
作者: Arunkumar V,Gangadharan G. R.,S. Senthilkumar
类目: Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR); Machine Learning (cs.LG)
备注: 24 pages, 5 figures
Abstract:Large-language-model agents now close the industrial predictive-maintenance loop, reading sensor windows, retrieving procedures and dispatching work orders. Two attack channels meet there: injected text in a retrieved document can hijack the decision, and a bounded sensor perturbation can corrupt the remaining-useful-life estimate behind it. Recent agent defenses police data provenance and tool privilege but cannot tell whether a prognostic quantity is physically plausible. ConfGuard is a reference monitor with a physics-grounded policy. Its conformal monitor tests each prediction, anchored to the asset’s admitted history, against degradation physics: capped remaining life never rises and falls at most one cycle per cycle. Calibrating its threshold over whole asset trajectories lets the operator set the expected share of a clean asset’s windows the running monitor escalates. A physics-policy validator checks the agent’s maintenance-urgency decision against the decision a deterministic policy derives from the same interval. On all 10,196 C-MAPSS test windows the monitor cuts a covert attacker’s reach on healthy assets from 0.529 to 0.046 and closes suppression of failing assets (0.151 to 0), escalating 0.027 of clean windows against a 0.05 target, while the fixed physical range it replaces leaves both attacks open. Behind a Qwen3-14B planner the composed monitors hold urgency-class injection to 0.00 and the universal sensor attack to 0.12, against 0.29 and 0.54 for a defended prompt. An adaptive attacker defeats a Mahalanobis detector on 0.504 of windows and adversarial training on 0.485. We report an attack class the projection misses.
[AI-79] Learning Regions of Attraction for Transverse Feedback Linearization under Controller Singularities
链接: https://arxiv.org/abs/2611.22286
作者: Abdullah Waqar Malik,Hamza Tariq,Adeel Akhtar
类目: Artificial Intelligence (cs.AI); Systems and Control (eess.SY)
备注: 27 pages (10 main + references + appendix), 1 figure, 8 tables
Abstract:Transverse feedback linearization (TFL) achieves path following by exactly linearizing the error dynamics, but its controller is undefined on a singular set, such as zero speed for a unicycle or zero thrust for a quadrotor. The constrained region of attraction, the set of initial states from which the closed loop converges to the path without violating state or regularity constraints, determines where the controller can be safely activated. Exploiting the exact linear error dynamics, we show that this region is the largest forward invariant subset of the admissible set. We then decompose its state-constrained Zubov value function into an analytically known quadratic cost-to-go and a safety value function. We learn only the latter, by fitted value iteration without infinite-horizon trajectory labels. The resulting estimate combines an inner layer verified by interval branch and bound with a learned outer layer tested on samples. On a unicycle following a circle and a sinusoid and on a planar quadrotor, it covers on average 72.0%, 83.4% and 19.1% of the simulated region of attraction. In comparison, the quadratic cost-to-go and two physics-informed neural network (PINN) solvers of Zubov’s equation cover at most 11.3%, 4.0% and 0.0% on average. Code and animations are available at this https URL
[AI-80] owards Physics-Constrained Rectified Flow for Climate Downscaling: What Projections Can Certify?
链接: https://arxiv.org/abs/2611.22283
作者: Arun Sharma
类目: Artificial Intelligence (cs.AI)
备注: Accepted to AIAS+ 2026, ACM ICPS proceedings. 9 pages
Abstract:Statistical downscaling restores the regional detail behind flooding and crop stress, but diffusion and flow-matching generators can produce sharp fields that break the conservation laws of their coarse input. Physics-Constrained Rectified Flow (PC-RF) is a conditional flow-matching downscaler. Its training penalties target wind divergence, negative precipitation, and mass drift from the coarse total, and a projection after sampling removes the last two exactly. On one corrupted synthetic field with no trained network, the projection drives the Non-negativity Violation Rate (NVR) from 0.236 and the Mass-Conservation Error (MCE) from 0.184 to zero, and the penalties lower the Divergence Error (DE) by about 30%. On ERA5 the projection gives PC-RF and RF-base, the penalty-free rectified-flow baseline, an NVR of exactly zero and an MCE of zero up to float32 rounding. The penalties add no perceptual gain that we can separate from training budget or run-to-run variation. At a matched budget on 2019-2021, PC-RF has no edge over RF-base (SSIM 0.939 vs. 0.942, PSNR 31.81 vs. 31.79, CRPS 0.771 vs. 0.772). Our unprojected CorrDiff reimplementation beats both flow models on PSNR (28.38) and CRPS (1.092) on 2020, yet 45% of its ensemble-mean precipitation cells are negative. On ERA5 a sign error in the divergence operator made that penalty and DE act on a deformation-like term. The fine reference is interpolated from ERA5 and accumulates precipitation over a longer window, so the perceptual scores cannot establish downscaling skill. NVR and MCE need one pass and no fine reference, so we argue they should run first as pass-or-fail checks, with DE as a diagnostic.
[AI-81] SeasonRainBench: Global Seasonal Precipitation
链接: https://arxiv.org/abs/2611.22281
作者: Qiuyan Sun,Mengxuan Chen,Haoyuan Liang,Guowen Li,Zhen Huang,Jianhao Lin,Juepeng Zheng,Haohuan Fu
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Accurate seasonal precipitation forecasts help societies prepare for drought, flooding, and crop failure months in advance. While data-driven methods have made rapid progress in weather and subseasonal forecasting, operational seasonal precipitation prediction still relies on numerical ensemble systems. These systems are computationally expensive and exhibit biases and drift that increase with lead time, motivating more efficient data-driven alternatives. Yet this transition faces three linked challenges: comparing data-driven forecasts with operational ones requires (1) a global benchmark that does not exist, scoring them requires (2) metrics that capture operational priorities beyond anomaly correlation (ACC) and RMSE, and training them must (3) rely on a single, short observational record. To address the first two challenges, we introduce \BenchName, the first such benchmark, covering monthly precipitation forecasts 1–5 months ahead. It pairs global 1^\circ satellite precipitation with oceanic and atmospheric predictors, includes seven operational centers, and provides 310,944 simulated months from 52 climate models for training. Beyond standard scores, we introduce two operationally motivated metrics for anomaly strength and monthly rain-belt position. To address the third challenge, we develop \MODEL, which learns from simulations what the record lacks and transfers it to Earth through three modules: SimForced aligns the climate background, SimMemory transfers a history encoder pretrained on simulations, and SimAnalog supplements Earth labels with simulated analog outcomes. Extensive experiments show that \MODEL improves ACC, CRPS, anomaly strength, and rain-belt position by 12.0%, 3.6%, 2.9%, and 0.3% over the strongest operational or data-driven model, opening a path from simulated worlds to Earth system prediction.
[AI-82] DACE: Diversity-Driven Adversarial Co-Evolution for Robust LLM Safety Alignment
链接: https://arxiv.org/abs/2611.22278
作者: Peng Yu,Xiaoyu Wen,Zhida He,Ziyuan Zhou,Han Qi,Shao Zhang,Qiaosheng Zhang,Ying Wen,Chaochao Lu
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Ensuring robust safety alignment of large language models (LLMs) is increasingly difficult as adversarial attacks evolve and outpace alignment pipelines built on pre-collected data. Existing co-evolutionary frameworks let the defender chase a moving attacker, yet their training dynamics expose two compounding pathologies: \emphattack strategy collapse, where the attacker overfits to a narrow set of high-reward rewrites, and \emphdefense adversarial forgetting, where the defender loses competence against earlier attacks as the attack distribution drifts. Existing remedies are partial: attacker-side diversity rewards score textual novelty rather than strategy novelty, and defender-side replay relies on discrete judge scores that ignore estimation uncertainty, response stochasticity, and defender non-stationarity. We introduce \textbf\method, a diversity-driven adversarial co-evolution framework that targets both pathologies jointly. On the attacker side, \method couples an explicit two-dimensional strategy space (risk category \times attack style) with a \emphnormalized marginal coverage gain reward, providing a bounded, non-vanishing exploration signal at the strategy layer. On the defender side, \method maintains a unified \emphBayesian adversarial replay pool whose Beta–Bernoulli threat posteriors are refreshed by time decay and sampled via Thompson sampling, anchoring defender training to an evolving threat landscape. Across standard safety, automated-attacker, and general-capability benchmarks, \method improves robustness to out-of-distribution attacks while preserving general capabilities, and yields broader attacker strategy coverage.
[AI-83] Anticipating or Speculating? Making Futures Pay in AI Ethics
链接: https://arxiv.org/abs/2611.22268
作者: John P. Nelson
类目: Computers and Society (cs.CY); Artificial Intelligence (cs.AI)
备注:
Abstract:Philosophers have recently been trying to develop uniform decision rules for instances in which physicians’ judgments diverge from the recommendations of medical artificial intelligence (AI) systems. I argue that this effort is fatally flawed because the performance and appropriate use of algorithmic recommendation systems by physicians are highly context-specific, empirical medical matters rather than epistemological ones. In such instances, general decision rules can be stipulated only to the extent that they abstract away from relevant, indeed determinative, contextual circumstances. The alleged problem of physician-AI disagreement, and others like it, play into longstanding disputes in technology ethics about the viability and utility of envisioning possible futures: is this useful “anticipation” or wasteful “speculation”? I suggest that the two may be distinguished. Anticipation explores plausible futures to help inform present actions under irreducible uncertainty. Speculation abstracts away uncertainty, developing formal solutions to stipulated circumstances whose relationship to the real present is often quite unclear. Efforts to develop crisp philosophical solutions to physician-AI disagreement, among other problems, fall into the latter category. Technology ethics should remember that fruitful futures thinking must both begin from and return to the messy, uncertain present.
[AI-84] Fidelity Regret: Measuring Simulator Misspecification in Verifiable Design NEURIPS2026
链接: https://arxiv.org/abs/2611.22263
作者: Noah Flynn
类目: Artificial Intelligence (cs.AI)
备注: NeurIPS 2026 Workshop “Sim2Science: ML with Imperfect Scientific Models”
Abstract:RLVF assumes the verifier evaluates the quantity of interest. In genetic-circuit design, a topological verifier can establish that a program is a valid circuit computing the target function, whereas ranking its physical performance requires a simulator. We measure fidelity regret, the reference simple regret incurred when a cheap simulator selects the design, over a 21,362-design combination subspace spanning 5 measured gate libraries and 3 organisms. Exact ordered-assignment checks retain positive regret in every tested cell while shifting magnitudes in both directions. Mean normalized regret is 0.369 over 72 cells and only 6 have zero regret. The discrepancy depends strongly on the reference axis: the yeast library SC1C1G1T1 has mean regret 0.113 for cytometry quality and 0.801 for growth burden, and every evaluated library–circuit-size pair has positive burden regret. Topological verification ties all valid gate assignments, while an oracle whitelist subject to retaining at least 25% of the design space leaves positive regret in 30 of 52 cells. Design-space optimization and feedback-guided search with frozen language models reach the high-regret regions; on a matched zero-regret control, additional optimization instead reduces regret. At a matched modeled-cost budget, mitigation is budget- and library-dependent. An in-sample fixed rule spending \alpha=0.34 of the budget re-scoring top cheap-simulator designs achieves mean excess regret 0.016, compared with 0.305 for cheap-only and 0.034 for reference-only selection, improving on cheap-only in 330 of 360 cells. More broadly, any workflow optimized through a simulator or digital twin can satisfy its verifier yet perform poorly on the intended physical objective. Fidelity regret provides a transferable template for detecting this gap before optimization and comparing budget-aware remedies across scientific domains.
[AI-85] A Visual Information Extraction System Using Multimodal Large Language Models for Scanned Cognitive and Behavioral Assessments
链接: https://arxiv.org/abs/2611.22252
作者: Enshuo Hsu,Linh Nguyen,Sunyang Fu,Heath Goodrum,Sandra Stansberry,Kirk Roberts,Hongfang Liu,Yen-Chi L. Le
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Objective. Paper-based assessments are common in patient care, but handwriting and scanning artifacts make these forms difficult to process at scale. This study proposes a modular, multimodal LLM-based system with human oversight for extracting fields and scores. We conducted a qualitative analysis of the use of LLMs for hand-drawing analysis. Materials and Methods. The system processes cognitive assessments (i.e., MMSE, MoCA, Mini-Cog) and the NICHQ Vanderbilt Assessment using two modules: 1) a form classifier that assigns a scanned page to a form type, followed by a corresponding 2) score extractor that extracts fields and scores into a schema. A human-in-the-loop design utilizing dual-model validation engages human annotators when two LLMs disagree, limiting manual review to challenging pages only. Results. End-to-end accuracy for cognitive assessments was 88%, with most form types exceeding 90%. For the NICHQ Vanderbilt, overall accuracy reached 97.5%, with all form types above 96%. The human-in-the-loop design further improved cognitive assessment accuracy to 97.5% while reducing manual work by 63.8%. In the Mini-Cog Clock Drawing task, qualitative analysis showed that 62.3% of all extraction errors stemmed from ambiguous drawing, in which cases the LLM provided reasonable readings. Discussion. Compared to current manual data extraction, our LLM-based system reduces processing time by 10 to 20 times and costs by over 12 times. The modular design allows seamless expansion to additional form types. Conclusion. Our system provides an accurate, scalable, and cost-efficient solution for structuring data trapped in handwritten paper forms for clinical decision support. Subjects: Artificial Intelligence (cs.AI) Cite as: arXiv:2611.22252 [cs.AI] (or arXiv:2611.22252v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2611.22252 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Enshuo Hsu [view email] [v1] Tue, 17 Nov 2026 05:56:48 UTC (2,200 KB) Full-text links: Access Paper: View a PDF of the paper titled A Visual Information Extraction System Using Multimodal Large Language Models for Scanned Cognitive and Behavioral Assessments, by Enshuo Hsu and 7 other authorsView PDF view license Additional Features Audio Summary Current browse context: cs.AI prev | next new | recent | 2026-11 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)
[AI-86] PCGRL for Complex Platformers: Synthesizing Playable and Diverse Levels via Inflated Envelopes and Forward Curricula
链接: https://arxiv.org/abs/2611.22250
作者: Di Wu
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 9 pages, 6 figures, 1 table
Abstract:Procedural Content Generation via Reinforcement Learning (PCGRL) has emerged as a promising paradigm for automated game development. However, existing PCGRL frameworks still suffer from severe exploration bottlenecks and scaling limitations when faced with long-sequence generation tasks governed by strict rigid-body rules and non-linear kinematic boundaries (e.g., platformer constraints). Consequently, traditional approaches frequently fail to maintain reliable level solvability on large-scale canvases. To address these challenges, we present a physics-aware, curriculum-guided PCGRL framework. First, we construct an “Inflated Volume Envelope” mechanism, which provides a physics-constrained representation that incorporates continuous kinematic feasibility into the discrete state space of the MDP. Second, to alleviate the sparse reward challenge induced by extremely long decision sequences, we introduce intermediate state-transition feedback and a multi-stage forward extension curriculum learning framework, enabling informative value estimation and stable credit assignment over long-horizon trajectories. Finally, to resolve the issue where deterministic policies after convergence generate highly monotonous map paths with a lack of diversity, we activate high-entropy stochastic inference during the synthesis phase, regularized by a post-hoc parallel rejection filtering mechanism. Experimental results demonstrate that our framework significantly enhances structural diversity while maintaining high physical playability across multi-scale environments. Comparative evaluations across multi-scale canvases demonstrate that our framework maintains superior convergence stability and dimensional scalability, reliably synthesizing solvable levels where traditional PCGRL baselines degrade.
[AI-87] Solver-Free and Cost-Aware Online Scheduling of AI Data Center Tasks
链接: https://arxiv.org/abs/2611.22239
作者: Xin Qin,Dawei Qiu,Fei Teng
类目: ystems and Control (eess.SY); Artificial Intelligence (cs.AI)
备注:
Abstract:AI data centers receive computing tasks continuously, whereas solver-based optimization takes time to run and electricity is priced per settlement period. This paper bridges this timing gap through a solver-free online optimization rule inspired by dual decomposition. The rule assigns each task a start period upon arrival to optimize energy cost subject to operational constraints, shifting flexible tasks toward low-price periods while keeping high-priority tasks on schedule. Simulations on the A100 GPU trace (3,456 GPUs, 1,343 tasks) show that the proposed rule keeps GPU and CPU use within capacity while scheduling tasks at low cost. Compared with a multi-timescale optimization benchmark, it achieves a lower cost and, because it needs no solver, much less computation time.
[AI-88] Context Has a Price: Adding and Removing Context for LLM Coding Agents under Prefix Caching
链接: https://arxiv.org/abs/2611.22238
作者: Zerui Yang,Miaomiao Dong,Quan Xue,Chaorui Zhang,Xuanyu Zheng,Xiyuan Wang,Haolin Ruan,Xueyan Niu,Guangcheng Song,Weixi Zhang,Wei Han,Bo Bai
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Coding agents repeatedly send their conversation history to a language model. Adding useful code may reduce search, while removing old tool outputs may shorten later prompts. But do these changes reduce the cost of solving a task? We study this as a measurement problem in more than 20,000 runs of open models on SWE-bench Verified, SWE-bench and Loc-Bench. We account separately for uncached input, cached input and output tokens. Three findings emerge. First, in our main evidence-package experiments, retrieved code reduces read and search commands by 18–34%, but increases cost by 33–42% when cached input costs one tenth of uncached input, without a measurable improvement in resolution. Second, removing old observations can either save money or increase cost: the same approach saves 19–36% without a cache discount but increases cost by 38–116% with a tenth-price cache. Third, shorter prompts do not guarantee cheaper runs because agents may respond by taking more steps. We also find that concurrent experiments can share cached prefixes and distort cost comparisons. These results show why context-management methods should be evaluated using both token prices and the agent’s subsequent behavior.
[AI-89] he Frag ility of Reasoning -Importance Attribution for Pruning Language Models
链接: https://arxiv.org/abs/2611.22234
作者: Himanshu Mishra
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Attribution-guided pruning scores components by their causal contribution and protects the important ones, and it is increasingly applied to compress reasoning language models without destroying their chain of thought. Its reliability rests on an unexamined assumption: that reasoning importance can be faithfully measured. Across four small reasoning models, we show this measurement is fragile in three ways, each of which changes pruning decisions. First, scoring final-answer tokens, the natural choice, saturates in every chain-of-thought model we test (answer-token negative log-likelihood, NLL, of 10^-4 – 10^-3 ) because the model states its answer inside its own reasoning; guiding head pruning with this target costs 27.6 and 32.2 accuracy points relative to scoring reasoning tokens on the two models we power with five seeds, while perplexity barely moves and even reverses sign at 7B. Second, causal importance is not localizable to single units: single-neuron effects sit at the measurement floor, their cross-split ranking is noise, and they cancel: a group’s ablation effect disagrees in sign with the summed effects of its members in 90% of groups. Reproducible importance emerges only at groups of \sim 100 neurons. Third, contrastively isolated ``reasoning-specific’’ importance is not separable from generic importance, and protecting it is worse than protecting raw importance. We additionally pre-registered the hypothesis that distillation-induced concentration of importance predicts the pruning cost; it is not supported, and since our grid confounds concentration with scale, we report it as unresolved rather than reversed. The result is a short set of measurement rules: score reasoning tokens, attribute at coarse granularity, and evaluate with accuracy rather than perplexity.
[AI-90] Learning Optimal Transport Plans Via Autoregressive Token Regression
链接: https://arxiv.org/abs/2611.22218
作者: Ivan Jaen Marquez,Sotirios Panagiotis Chytas,Vikas Singh
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Optimal Transport (OT) provides a principled framework for finding the minimum cost coupling between probability distributions. Classical solvers remain the gold standard for this task. At the same time, OT is an ideal testbed for studying whether learning-based amortized models can internalize the structure of OT type problems. Here, ground-truth solutions are available and marginal constraints are explicitly defined. Also, exact discrete solutions have well-characterized sparsity. We describe ToR-OT, a transformer-based model that learns to predict discrete OT plans through autoregressive token regression. We reframe OT into a structured sequence generation task. ToR-OT decodes sparse transport plans as sequences instead of solving an optimization task at test time. We focus on three properties: (a) Zero-shot generalization: pretraining on a large synthetic corpus of OT instances allows a single model to generalize to unseen distribution pairs and heterogeneous cost metrics without retraining, (b) Sparse primal decoding: by exploiting the inherent sparsity in exact discrete OT solutions, the decoder predicts only nonzero entries (avoids dense coupling outputs), © Amortized inference: on low-dimensional discrete problems, ToR-OT gives accurate approximate plans and compares favorably with per-instance neural OT baselines. Overall, ToR-OT is not yet a replacement for classical solvers, but a step toward understanding how autoregressive transformers can learn structure of constrained optimization problems like OT.
[AI-91] Communicate the Change: Efficient Expert Parallelism via Activation Compression
链接: https://arxiv.org/abs/2611.22215
作者: Fengyuan Liu,Jay Gala,Siva Reddy,Dzmitry Bahdanau
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Distributed, Parallel, and Cluster Computing (cs.DC)
备注:
Abstract:Scaling has driven the success of large language models (LLM), with sparse mixture-of-experts (MoE) architectures enabling greater parameter capacity without a proportional increase in computation. At large scales, MoE models rely on expert parallelism to distribute experts across devices. Each MoE layer requires two synchronous collective communication steps during inference: dispatching hidden states to the activated experts and combining the resulting expert outputs. As compute throughput has outpaced interconnect bandwidth, these synchronous collectives have become a major bottleneck to MoE scaling and efficiency. We address this challenge with EchoMoE, a lightweight architecture modification that retrofits existing MoE models, requiring approximately 8,500x fewer training tokens than full pretraining. Motivated by the observation that transformer blocks update the residual stream incrementally, we use a hidden state from an earlier layer as a reference and communicate only the compressed difference between it and the current state, rather than communicating each hidden state in full. EchoMoE periodically refreshes these reference states and transmits only the compressed differences synchronously, reducing the traffic on latency-critical paths without significant computational overhead. EchoMoE reduces inter-node communication volume by 58% and improves throughput by 22.6%, while preserving average performance within 3.8 percentage points across 18 benchmarks. EchoMoE further improves inference throughput by up to 2.38x over bandwidth-constrained Ethernet, paving the way for wide-area distributed inference. EchoMoE also tolerates quantization of the communicated representations without measurable quality loss. We release an efficient implementation for applying EchoMoE to pretrained MoE models, together with an inference integration for vLLM.
[AI-92] RAG in a Trenchcoat: When Minimal Memory Is Enough for Agent ic Systems and When It Isnt NEURIPS2026
链接: https://arxiv.org/abs/2611.22179
作者: Jingyu Liu,Zongze Li,Zach Xu,Zhanhui Zhou,Tahseen Rabbani,Dawn Song,Ce Zhang
类目: Artificial Intelligence (cs.AI)
备注: 40th Conference on Neural Information Processing Systems (NeurIPS 2026), Main Conference
Abstract:Agentic memory research has accumulated extraction, supersession, reflection, and graph synthesis on top of retrieval-augmented generation. However, we found that almost none of those primitives earn their cost in our tested memory benchmarks: we show this with BRIDGE, a deliberately minimal RAG memory with chained observation notes, hybrid BM25 with cosine retrieval, and chronological rendering, which reaches parity within bootstrap CI with the prior SOTA on LongMemEval-S and BEAM, and beats the best memory baseline on EverMemBench by 5.7-6.3 points at matched reader. It runs at ~5-7K reader-prompt tokens per query, using a single LLM call for each ingestion or query. We argue that this parity is itself the result: standard benchmarks score what memory retrieves, not what memory enables when the agent acts. To test the latter, we introduce five adversarial probes (Rule Application ±, Pattern Induction Style / Procedural, Dynamic Multiturn) judged at generation time by a cross vendor verification chain, where each query can only be solved by applying a remembered rule or pattern it never mentions. Across 5 probes x 20 scenarios, no architecture wins universally: atomic fact extractors (LangMem) win rule application, while BRIDGE’s block-summary memory wins pattern induction. This result holds across four domains, three graders, and three reader families. Each design fails where the other wins, for reasons a recall benchmark cannot see: atomic-fact extractors discard the form a demonstrated pattern takes, while block-summary memory retrieves the rule but the reader does not act on it, a gap that presenting the rule more prominently largely closes. What separates memory designs is not what they retrieve but what they enable an agent to do, and that is what agentic memory should be evaluated on.
[AI-93] CorrectCast: Selective Multimodal Context for Scale-Aware Residual Correction in Precipitation Nowcasting
链接: https://arxiv.org/abs/2611.22178
作者: Henan Wang,Shengwu Xiong,Yifang Zhang,Pengfei Duan
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Precipitation nowcasting models suffer from errors in intensity, location, and spatial structure. Addressing these errors by retraining nowcaster backbones can be computationally expensive and requires architecture-specific modifications. To overcome these limitations, we propose CorrectCast, a decoupled residual correction framework that refines completed nowcasts while keeping the base nowcaster frozen. By operating on predicted precipitation fields, CorrectCast requires training only a lightweight corrector, reducing computational overhead while providing a unified correction interface across different nowcasting models. Conditioned on the base nowcast and historical quantitative precipitation estimation (QPE) and precipitable water vapor (PWV) observations, the corrector selects multimodal context and employs scale-aware modulation to estimate residual corrections. On the Mississippi River Basin (MRB) and contiguous United States (CONUS) datasets, CorrectCast improves six base nowcasters spanning extrapolation-based, deterministic, and generative approaches, with absolute gains in mean CSI averaging 0.0233 and 0.0290 across the six models, respectively.
[AI-94] Reconfiguring Procedural Knowledge for Compositional Robot Skill Adaptation NEURIPS2026
链接: https://arxiv.org/abs/2611.22174
作者: Daehee Lee,Dongsu Lee,Sanghyun Ahn,Minjong Yoo,Honguk Woo
类目: Artificial Intelligence (cs.AI)
备注: Accepted at NeurIPS 2026
Abstract:General-purpose behavior models increasingly internalize broad procedural knowledge during training, making their adaptation to new goals and skill interfaces less a problem of learning entirely new behaviors than of reconfiguring existing competence. We study this capability as reconfigurability, the ability to realize new behaviors by cost-efficiently restructuring existing components and relations under sparse governing information, where task specifications provide high-level goals and procedure specifications expose a limited set of reusable factors, leaving the concrete bindings and compositions required for execution underdetermined. We propose RecPo, a framework that represents skills task-agnostically through multiple procedure views and materializes executable skill sequences by composing stored transitions to realize task-specific changes. RecPo maintains a procedure knowledge base of skill-induced transitions, with each view exposing a reusable factor and preserving local skill connectivity. At deployment, it interprets task specifications through the same views and resolves the resulting compositional queries over the stored relations to bind skills for execution. On long-horizon tasks in an extended Franka Kitchen benchmark, RecPo achieves 91.9% ordered completion with 1.0 ms of task inference per episode across varying procedure support and task-record sparsity. Real-robot studies further demonstrate adaptation through local memory edits without task-specific retraining.
[AI-95] Readie: Low-Latency Code Sandboxes for AI Agents
链接: https://arxiv.org/abs/2611.22171
作者: Arjun Sivaraman,Zhaoheng Li,Alan Shi,Sanjay Malisetty,Eugene Wu,Yongjoo Park
类目: Artificial Intelligence (cs.AI); Databases (cs.DB); Software Engineering (cs.SE)
备注:
Abstract:From quantitative research to personal assistance, AI agents act by iteratively generating and executing code. To decouple the agent harness from code execution, many agents offload execution to dedicated systems called code sandboxes. However, existing code sandboxes suffer from slow startup. Whether they rely on custom snapshots or a single fat image, the root cause is the same: it is hard to build a compact, pre-warmed snapshot that can be shared across diverse agent-generated code while covering nearly all the artifacts (e.g., packages, datasets) it requires. In this work, we propose learned startup snapshots for low-latency code sandboxes. Our insight is that while it is hard for a single snapshot to be both compact and cover most workloads, a set of snapshots can achieve both. Each snapshot is small, initializes quickly, and covers a fraction of agent-generated code; together, the set covers most of it. We learn this set from a large training workload by formulating snapshot construction as an optimization problem and solving it with an approximation algorithm. Planned on 15K public and LLM-synthesized scripts, our system, Readie, reduces startup latency by 3-22 \times and end-to-end runtime by 1.3-3.3 \times compared to existing approaches at the same storage budget. Subjects: Artificial Intelligence (cs.AI); Databases (cs.DB); Software Engineering (cs.SE) Cite as: arXiv:2611.22171 [cs.AI] (or arXiv:2611.22171v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2611.22171 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-96] When does deep reinforcement learning pay off in HVAC supervisory control? Four small Modelica case studies trained directly on FMUs
链接: https://arxiv.org/abs/2611.22144
作者: Bingnan Zhao,Harshala Gammulle,Tharindu Fernando
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Systems and Control (eess.SY)
备注: 16 pages, 4 figures, 2 tables. Code, models and trained agents: this https URL
Abstract:Deep reinforcement learning (DRL) is often proposed for the supervisory control of heating, ventilation and air-conditioning (HVAC) systems. The reported gains are, however, difficult to interpret, because the baselines are often weak and the attainable optimum is seldom known. This paper examines four small, reproducible case studies based on an office building in Brisbane, Australia: pre-cooling under a time-of-use tariff; chiller staging with condenser water reset; chilled water supply temperature reset with a wet cooling coil; and chilled water storage under time-of-use prices and a demand charge. The plants were modelled with the Modelica Buildings library, and in three of the cases a soft actor-critic agent was trained directly on the exported functional mock-up unit (FMU). Each agent was compared with tuned rule-based controllers and, where practicable, with a near-optimal reference from hourly exhaustive search or dynamic programming with perfect foresight. In the two cases without storage the agent reached the near-optimal reference, but the margin above a well-chosen rule was small: 1.3% for condenser water reset and 0.18% for chilled water reset. In the latter case two common reset rules consumed more energy than a fixed set point. With a chilled water storage tank the margin was considerably larger. The agent outperformed both rules (-7.8%, compared with -3.9% and -6.3%) and adopted midday charging during the solar off-peak period, yet it obtained only about 45% of the saving achieved by the perfect-foresight schedule and did little to limit the demand peak. In the pre-cooling case a hand-tuned rule remained superior (-12.0% compared with -5.6%). These results suggest that studies of DRL for HVAC control should state the margin between the best rule and a near-optimal reference. Some practical difficulties encountered in training on co-simulation FMUs are also reported. Comments: 16 pages, 4 figures, 2 tables. Code, models and trained agents: this https URL Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Systems and Control (eess.SY) Cite as: arXiv:2611.22144 [cs.LG] (or arXiv:2611.22144v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2611.22144 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-97] RoboTeam: Benchmarking Vision-Based Multi-Agent Control of Many Robot Arms in Real Time
链接: https://arxiv.org/abs/2611.22138
作者: Changxuan Fan,Zhongwei Xie,Boyi Zhang,Shaoli Hu,Chang Xu,Xiyu Ren,Yufei Li,Qiao Xiao,Huihao Jing,Kawai Chung,Wenbin Hu,Hongyu Luo,Lingyun Xie,Yisen Gao,Chunyang Li,Haoyu Huang,Wuganjing Song,Yuxuan Liu,Chiu Yeung Tang,Fangneng Zhan,Xi Yang
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注: 17 pages, 15 figures, 10 tables, including appendix
Abstract:Multimodal agents can now control robot arms by calling tools and writing code. Existing benchmarks fall short in three ways: most test one or two arms on short tasks, many pause the world while the agent thinks, and many give it ground-truth object positions. We present RoboTeam, a simulated benchmark where agents control 4 to 16 robot arms in real time from camera images alone. Its 36 tasks are long jobs that the arms must do together, such as passing timbers or building a picture from hundreds of cubes. We start from the most flexible agent system, the adaptive system, where one agent decides when to fork delegates. With it, gpt-6.1-sol and opus-5.5 complete 28 and 27 tasks, and four other models at most 3. But at a high price: a typical task takes gpt-6.1-sol 3.5 hours and 218 million input tokens, and a third of its steps fail at least once and must be retried. We then compare it with three fixed systems: single-agent, hierarchical, and decentralized. All four complete a similar number of tasks but differ in time and tokens. The hierarchical and decentralized systems are faster but use up to 1.9 times the tokens, and the single agent uses fewer tokens but takes 1.6 times as long, so the adaptive system lies between them. Even so, no model or system is yet fast, cheap, and reliable at once. RoboTeam measures these limits where earlier benchmarks cannot: long multi-arm jobs, camera images only, and a world that never pauses.
[AI-98] Chat to Xing4: Advancing Frontier Agent ic Models on Domestic Chips and Frameworks
链接: https://arxiv.org/abs/2611.22115
作者: Xingchen AGI Lab:Baotian Hu,Bingkai Yang,Boyang Wang,Changzai Pan,Chao Lyu,Chao Wang,Chen Li,Chenshuo Pan,Chenyu Lin,Chenyuan Zhang,Chongming Liu,Dawei Fan,Di Wu,Dong Guo,Dong Li,Dongfang Li,Enlin Liu,Fanye Xiao,Fengkai Li,Fubei Yao,Gang Lin,Han Gao,Hanming Wu,Haowei He,Huapeng Liu,Hua Wang,Huinan Xu,Jiahong Xiong,Jiaxin Peng,Jianxiang Ma,Jianxing Zhuge,Jie Zhang,Ji Xiong,Jinglin Chen,Jingqi Li,Jingxuan Li,Jiwang Ye,Junchen Li,Junbiao Li,Junyang Wu,Kaidong Yu,Kaipeng Jia,Kehai Chen,Kuai Li,Laobin Liu,Lei Li,Libo Qin,Lingling Shi,Luwen Pu,Menghao Xu,Mengjiao Bao,Mengqi Chen,Min Zhang,Mingrui Wang,Naying Wang,Peicong Han,Peiming Guo,Pengfei Li,Qiang Feng,Qi Wang,Qingjun Zhu,Qingli Li,Qitai Ding,Qiyi Xie,Quanchao Zhao,Rongbao Han,Rui Xue,Ruiyu Fang,Ruiting Song,Shijie Li,Shiquan Wang,Shuangyong Song,Sishi Xiong,Teng Su,Wankang Bao,Weijie Zhang,Wei Mao,Weiwei Fu,Weiyang Huang,Wenmin Deng,Wenqing Hu,Wujie Zhong,Xi Yu,Xiaohui Hu,Xiaomin Fang,Xiaomeng Huang,Xiaoqiong Shi,Xiaotong Shang,Xiaoyu Li,Xinglei Xu,Xin Jiang,Xin Wang,Xin Zhang,Xinzhang Liu,Xue Liu,Xuefeng Bai,Xuewei He,Yahan Li,Yanhan Zhang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:We present Xing 4, a series of large language models that extends the TeleChat lineage through coordinated advances in model architecture and training infrastructure on Ascend. This report describes the design principles and systems engineering behind the series, spanning architecture selection, large-scale pretraining, and reasoning and agentic post-training. We study how sparse activation, attention design, model depth and width, and speculative decoding balance model capacity with computation, memory access, and execution overhead. Our pretraining infrastructure builds on MindSpore HyperParallel, combining topology-aware hybrid sharding, efficient distributed optimizer execution, and coordinated management of long-sequence workloads and activation memory. CPU affinity and core isolation further reduce host scheduling interference. Together, these mechanisms support the pretraining of a 285B-parameter model on 8,192 Ascend 910C cards, providing a concrete setting for the architectural and systems studies. For post-training, we introduce TeleGym, a unified agent execution platform that connects execution, evaluation, and trajectory construction to policy optimization. Disaggregated asynchronous execution, bounded oversampling, and partial rollout accommodate heterogeneous task durations while preserving trajectory continuity across policy updates. Together, these advances establish an Ascend-oriented training foundation for the continued development of Xing 4 across model scales and increasingly demanding reasoning and agent workloads.
[AI-99] RegimeCouncil: Paired-Evidence Routing For Memory Conflict Governance In LLM Agents
链接: https://arxiv.org/abs/2611.22113
作者: Qingcan Kang,Mingyang Liu,Shixiong Kai,Tao Zhong,Mingxuan Yuan
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Language agents rely on long-term memory to personalize responses across extended interactions, yet stored facts quietly go stale as the world changes, and an agent often learns that using memory hurt only through delayed feedback after acting on an outdated record. Before the memory store is updated, how far should that negative feedback propagate: should it suppress memory use only on future queries for that same fact (a local policy), or also across other stored facts (a global policy)? Existing memory systems implicitly fix one of these two sharing scopes at design time. We formulate this online scope choice under hidden validity and delayed feedback as memory conflict governance, and prove that neither fixed policy is sufficient on its own. We then propose RegimeCouncil, a lightweight paired-evidence router that retains both scopes as advisors and arbitrates between them online without retraining the LLM. Using revealed comparisons between answering with memory and answering without memory (the no-memory anchor), a Local Advisor tracks the latest outcome on the queried fact alone, while a Global Advisor pools recent outcomes across all stored facts. A Meta Governor follows the Local Advisor by default to avoid cross-fact interference and switches to the Global Advisor only when global pooling accumulates a strictly positive recent utility margin, guaranteeing finite-time switching under delayed feedback. Across LongMemEval update streams and real-world historical drift on TempLAMA and TAQA with Qwen3-30B-A3B and GPT-4o, RegimeCouncil outperforms the better fixed-scope policy by 5.2-7.3 points on primary disagreement episodes and across all 30 held-out settings, while ablations show that removing any of its three modules reduces accuracy by 2-7 points.
[AI-100] From Sorry to QED: Refining LLM Math Proofs via Angelic Execution
链接: https://arxiv.org/abs/2611.22103
作者: Yanju Chen,Hanzhi Liu,Ruizhe Qian,Jiaming Shan,Xu Yang,Yufei Ding,Osbert Bastani,Yu Feng
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Large language models (LLMs) can now generate Lean proof scripts that capture the high-level structure of many mathematical arguments. However, these scripts frequently fail on complex theorems: a single missing side condition or misapplied lemma can cause the prover to abort early and discard an otherwise viable proof outline. This paper presents MathEye, a framework that converts imperfect LLM-generated proof scripts into fully verified Lean proofs by combining symbolic reasoning with minimal human interaction. At the core of MathEye is angelic proof checking, an optimistic execution model that replaces failing proof fragments with placeholders while preserving the overall structure of the argument. This optimistic check produces an angelic proof that isolates the exact proof obligations that the LLM fails to discharge. Because these unresolved obligations are often logically interdependent, asking a mathematician to address every placeholder directly is inefficient. To minimize human effort, MathEye performs residual analysis over a proof dependence graph that captures symbol-level data flow in the angelic proof. By formulating obligation selection as a MaxSAT optimization problem, MathEye selects a minimal-cost set of essential obligations whose resolution suffices to complete the entire proof. MathEye presents these essential obligations to the mathematician in ordinary mathematical language and uses the mathematician’s answers to automatically reconstruct the complete Lean proof. Experiments on 30 real-world benchmarks across analysis and algebra show that MathEye solves 93.3% of the problems while reducing the number of required user interactions by 55%.
[AI-101] EvoDisco: Strategy Discovery for Evolving Agent Memory
链接: https://arxiv.org/abs/2611.22102
作者: Mingyang Liu,Qingcan Kang,Shixiong Kai,Kaichao Liang,Xiaojin Fu,Tao Zhong,Linqi Song
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Large language model agents accumulate experience in memory, and how that memory evolves during interaction is as important as what it stores. Recent methods update memory online, yet both the operators that transform memory and the rules that select them remain fixed by design. We propose EvoDisco, which formulates agent memory evolution as evolutionary strategy discovery: the agent must discover which operator to apply, when to apply it, and how to instantiate it, rather than rewriting memory under a static operator schedule. A learnable policy conditions on the current memory state and recent task history, then selects and instantiates one of four semantic operators—Abstract, Refine, Bridge, and Reframe—which respectively compress related memories, decompose overloaded entries, connect weakly linked experiences, and reorganize retrieved reasoning chains. The policy is updated with a bi-level fitness. An inner signal scores each transformation using a retrieval-quality improvement regularized by memory complexity. An outer signal scores strategic efficacy over a window of past decisions, combining historical operator-utility regret, utility-prediction error, and selection diversity. Under a one-pass online plus frozen-transfer protocol on ALFWorld, WebShop, and LifelongAgentBench (OS and DB), EvoDisco improves over strong memory baselines including MemP and MemRL. The largest gains are on ALFWorld ( +8.8 online / +12.9 frozen-transfer points over MemP; +16.4 / +17.9 over MemRL). Ablations attribute these gains to both fitness levels and the operator library.
[AI-102] HyCoInfer: Cloud-Edge Collaborative Inference for Large Language Models on Heterogeneous GPU-FPGA Architectures
链接: https://arxiv.org/abs/2611.22086
作者: Zewen Huang,Xiangwen Zhuge,Yingkun Hu,Zheng Yang,Qiang Ma
类目: Operating Systems (cs.OS); Artificial Intelligence (cs.AI)
备注: 10 pages, 13 figures, 3 tables
Abstract:Large language model (LLM) serving relies predominantly on cloud GPUs, yet autoregressive decoding is memory-bound and often leaves their arithmetic units underutilized. Edge FPGAs offer attractive decoding efficiency, but are ill suited to sustaining a high-concurrency serving workload alone. This paper presents HyCoInfer, a heterogeneous cloud-edge serving system that assigns prefill to cloud GPUs and distributes decoding across cloud GPUs and edge FPGAs. The design addresses three barriers to practical offloading: capacity balancing across heterogeneous accelerators, the large KV cache crossing a constrained wide-area link, and the latency added by heterogeneous execution. HyCoInfer incorporates FPGA decode capacity into GPU prefill/decode role partitioning and places requests according to queue load, prompt length, and network state. It further combines layer-wise pipelined transfer, decoder-side precomputation, and bandwidth-adaptive KV cache compression based on residual vector quantization. We implement HyCoInfer using two NVIDIA V100 GPUs and an AMD Xilinx Alveo U280 and evaluate it over 100-1000 Mbps links. At 500 Mbps, HyCoInfer delivers up to 4.11 times the request goodput of cloud-only vLLM and reduces mean end-to-end latency by up to 61.1%. Across different network conditions and workloads, the system improves overall serving capacity, with the largest gains on communication-intensive tasks, and reduces per-request energy.
[AI-103] ATLAS-Finance: Benchmarking AI Agents in Realistic Banking Environments
链接: https://arxiv.org/abs/2611.22085
作者: Elaine Lau,Antoine Pangas,Alexandre Finot,Collin Schweiker,Amir Bouzidi,Hui Wen Goh,Andrew Peter Martin,Jonas Mueller
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Professional work involves under-specified assignments, where relevant information is spread across workplace applications and may conflict or change over time. To evaluate how well frontier agents can perform economically valuable work in such settings, we introduce ATLAS-Finance, a benchmark of 100 end-to-end investment-banking tasks across 13 simulated firms. As in real banking workflows, agents must search email, chat, calendars, notes, and files to identify requirements, gather information, reconcile conflicts/updates, and then build complex financial models. Tasks were developed by experienced bankers based on their real deal work and take them 15-30 hours to complete. Agent deliverables are evaluated with expert-authored rubrics (median: 124.5 criteria) covering financial rationale and results, spreadsheet construction, completeness, and formatting. Despite completing substantial portions of the requested work, current agents rarely produce correct deliverables. Across today’s frontier models, the best ones (Claude Opus and Fable) pass fewer than 11% of tasks.
[AI-104] Can You Undo It? Reversibility Classes for AI Agent Actions
链接: https://arxiv.org/abs/2611.22075
作者: Deepak Waghmare,Shaad Sufi
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Tool-using AI agents change files, databases, access rights and public information, yet permission to call a tool does not establish whether a mistake can be undone. We present Reversibility Classes, a declared five-level vocabulary with argument-sensitive rules, conservative defaults and an explicit prior-state capture contract. The pinned implementation and evidence-linked registry cover 536 tools across 20 MCP servers. All 738 rule-case checks conform to declared expectations, while native annotation mapping differs from draft registry labels for 118 of 481 comparable tools. One AI-assisted reviewer approves 148 ratings, 75 matching the registry; this is not independent agreement. On 2,158 recorded runs, the implemented session guard gates the associated action in all 29 attributable attack-positive runs, compared with 18 for a read/write proxy, while adding nine approval points across 1,209 benign runs. Controlled file, Git, SQLite and three-browser experiments show that recovery depends on captured state, later changes, editor storage and recovery route, and expose a Git merge-state exception. The evidence supports making recovery assumptions explicit and testable. It does not establish registry-wide accuracy, reduced production harm or general recovery reliability.
[AI-105] MedGuardBench: A Patient-Oriented Benchmark for Personalized Guardrail Development
链接: https://arxiv.org/abs/2611.22069
作者: Sara Khosravi,Xiaofei Wen,Majid Sarrafzadeh,Muhao Chen
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Safety is personal: the same action can be safe for one person but unsafe for another. Yet current guardrails judge content without considering who is affected. Using medication safety, we introduce \textitpatient-conditioned medication guarding, which predicts a safety verdict and 17 risk categories from a patient profile and prescription. We release \textbfMedGuardBench, a drug-disjoint benchmark of 5,001 scenarios across 87 medications. We find that standard guardrails rely on surface cues, such as wording and profile presence, rather than patient facts. To address this, we develop \textitpersonalized reasoning guardrails with three components. \textbfPersonalized Reasoning Distillation links risks to relevant patient fields. \textbfContext Unrolling (CU) makes models check each field themselves, reducing false alarms on most backbones; controls show that providing the check does not yield the same benefit. A \textbfrule-based reward (guided GRPO) improves unsafe recall and complex-risk detection, including allergies and adverse reactions, on some backbones, but increases false alarms on others. Our models outperform base models and matched no-profile controls in accuracy, while reducing false alarms relative to no-profile controls. Profile removal and swapping confirm that gains come from patient information, not fine-tuning alone.
[AI-106] A Hybrid Exact-Heuristic Solver for the Integrated Passenger and Parcel Transportation Problem
链接: https://arxiv.org/abs/2611.22043
作者: TheDung Bui,Ha Thu Nguyen
类目: Artificial Intelligence (cs.AI)
备注: 4 pages, 2 figures, 3 tables. Technical Report for the SOICT 2026 Student Contest
Abstract:We describe the solver we submitted to the second round of the SOICT 2026 contest on the Integrated Passenger and Parcel Transportation problem, a selective pickup-and-delivery routing problem with time windows, open routes and heterogeneous taxis. The solver is a single-file, standard-library Python program that combines (i) a constant-time insertion test based on latest-arrival slack, (ii) greedy insertion, local search and an Adaptive Large Neighborhood Search (ALNS) with simulated-annealing acceptance, and (iii) an exact route-selection engine that enumerates, for every taxi, the best feasible route for each request set and then solves the resulting set-packing problem by branch-and-bound strengthened with route-based large neighborhood search, Lagrangian bounds and reduced-cost fixing. The engine is chosen automatically from the instance size, and an independent verifier and a watchdog guarantee that a valid answer is always printed. On the official evaluation (38 scored instances, 5-minute limit) the solver obtained a total benefit of 487,237, finished every run within 277 s and used at most 279 MB; 28 of the 38 instances terminated well before the limit.
[AI-107] What Does an Agent Need? Bundle-Aware Agent Resource Composition
链接: https://arxiv.org/abs/2611.22030
作者: Yunxiao Shi(1),Tingwei Chen(2),Dannong Xu(1),Ahad N. Zehmakan(1),Jialie Shen(3),Zhendong Chu(4),Qingsong Wen(4) ((1) Australian National University, (2) University of Technology Sydney, (3) City St George’s, University of London, (4) Squirrel Ai Learning)
类目: Artificial Intelligence (cs.AI)
备注: 26 pages, including appendices. Code: this https URL
Abstract:An agent resource’s value depends on its backbone and companion resources and is known only after execution. Token-based generators remain tied to their training catalog, while independent resource scoring misses complementarities and reused bundles target other tasks. We formulate Task-Conditioned Agent Resource Composition (AgentComposition): jointly selecting a backbone model and a tool-skill bundle to maximize end-to-end execution utility. Our Retrieval-Grounded Bundle-Aware Agent Generation retrieves candidate resources and past configurations of similar tasks. A fine-tuned generator reads resource descriptions and recombines candidates through constrained beam search, allowing new resources without retraining. A bundle-aware critic ranks complete configurations using a large language model (LLM) teacher and historically successful configurations as execution proxies, and adapts to execution feedback when available. We evaluate reference recovery, offline execution with judged and native rewards, and online recommendation with real execution, building AgentSelect-Exec with 5,060 executed configurations. Our method recovers the complete reference configuration within ten candidates for 67.73% of AgentSelect queries, nearly twice the strongest baseline, and leads all baselines on SkillsBench. The proxy-trained critic raises the top-ranked agent’s execution score from 7.58 to 8.37 out of 10; execution feedback increases it to 8.78, above the strongest baselines. Four target-domain queries suffice to adapt the generator to SkillsBench. Agents recommended online outperform the strongest baselines under both an LLM judge and native verifiers. These results support learning and transferring joint resource composition from task experience.
[AI-108] Computed Permutation Trellis Coding: Near-Optimal Quantization from Two Multiplications and a 64-Byte Table
链接: https://arxiv.org/abs/2611.22019
作者: Muhammed Fatih Balın
类目: Information Theory (cs.IT); Artificial Intelligence (cs.AI); Hardware Architecture (cs.AR); Machine Learning (cs.LG)
备注: 38 pages, 6 figures, 27 tables
Abstract:We propose Computed Permutation Trellis Coding (CPTC), a low-bit quantization format for tensors written once and read many times. Formats that round each value on its own, such as INT4 and FP4, waste a large part of their bits against the information-theoretic ideal. Trellis codes recover this loss by encoding values jointly, but prior decoders either need a large lookup table for their best quality or compute values with a hash that restricts what they can decode. CPTC’s decoder multiplies its state by two coefficients and adds the top bits of the products to index a 64-byte table. Because that sum is not uniformly distributed, the rarely used ends of a small table can hold values far out in the tails. One decoder serves every bit width from 1 to 6, reproduces the standard integer and floating-point formats exactly, and comes within 0.07 bits of the ideal at 4 bits. It matches a random-permutation trellis code whose table is thousands of times larger, and exceeds every QTIP codebook. In LLM weight quantization without fine-tuning, CPTC has the lowest KL divergence to the unquantized model at every tested bit width on Llama 3.1 8B and 70B Instruct against QTIP’s codebooks (by up to 27%). The decoder has been built into the AWS Trainium 4 AI accelerator, where it runs at the throughput of the native FP4 and FP8 formats for a negligible fraction of the systolic array’s area. Fused encoding kernels quantize an 8B model’s linear layers in under two minutes on one GPU and, in cycle-level simulation, on one Trainium 4 chip.
[AI-109] On Formal Counterfactual Explanations
链接: https://arxiv.org/abs/2611.22010
作者: Alexey Ignatiev,Peter J. Stuckey,Joao Marques-Silva
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Logic in Computer Science (cs.LO)
备注:
Abstract:Counterfactual explanations are often seen as counterfactual conditionals under closest-world semantics, requiring one to reason over all minimally distant counterexamples. In contrast, formal XAI typically relies on feature-based abductive explanations (AXps) and contrastive explanations (CXps), which admit efficient computation and a minimal hitting-set duality. This paper bridges these perspectives by (i) characterizing closest-world counterfactual explanations explicitly as a disjunctive normal form over all minimally distant counterexample instances, and (ii) introducing conjunctive counterfactual explanations (CFXps), a subclass of counterfactual antecedents restricted to conjunctions of feature literals that guarantee prediction change. The paper establishes structural results connecting CXps, AXps and CFXps: every CXp induces an AXp for its witness instance, and every CXp together with a witness AXp induces a CFXp whose support is precisely determined by that AXp. These theoretical results yield a modular approach for computing CFXps using existing AXp/CXp machinery. Experiments on a representative set of benchmarks using tree ensembles and neural networks show that CFXp extraction is practically efficient. Compared with popular point-based counterfactual explainers, CFXps generally require no more feature changes and can guarantee the counterfactual prediction beyond a single instance.
[AI-110] KV-LTM: Persistent Per-Session KV Caches as Long-Term Memory for Personal Assistants
链接: https://arxiv.org/abs/2611.21998
作者: Ziv Kfir,Ido Kfir
类目: Artificial Intelligence (cs.AI)
备注: 45 pages
Abstract:Personal assistants accumulate months of conversations, but today’s long-term memory stacks put derivation or selection between a question and its evidence. In a system-level comparison on LongMemEval, two commercial assistants scored well below the same model reading the history offline [1], consistent with, but not isolating, content loss between write and read. We propose KV-LTM: a memory tier storing each session’s raw text plus its KV cache under a fixed reader model, answering by restoring cached sessions and appending the question. KV-LTM is complementary to, not a replacement for, retrieval and indexing. Text is the memory of record; the cache is a recompute-free prefill accelerator, degrading gracefully to text. We specify the design: storage record, budget-derived retention, append-only read path, and bounded aggregation. We size it analytically: across pinned BF16 full-attention (GQA) and latent-cached MLA configurations, per-token state spans approximately 31K-328K bytes (the pinned KDA+MLA hybrid differs: approximately 8.1K bytes/token plus a fixed per-session recurrent term), giving approximately 0.34-3.6 TB per user-year at a declared 30K-token/day scenario, against approximately 49 MB of raw text; per-query cost is modelled as regime-conditional transfer, attention, decode, aggregation, and compute. We define an evaluation protocol needing no persistent-cache implementation for logical QA evaluation: per-session prefill emulates valid cache reuse under the same model/input-state contract and ideal arithmetic, with named arms, controlled baselines, and a preregistered reader-model gate; a separate minimal real-cache replay tests numerical equivalence and measures the cost axis, with outcomes not assumed. KV-LTM trades storage for avoided re-prefill compute and write-time fidelity, within a declared retention budget.
[AI-111] WebAgents : Unbundling Agent Agent Creation Hosting and Execution for an Agent-Native Web
链接: https://arxiv.org/abs/2611.21978
作者: Dario Fenoglio,Martin Gjoreski,Marc Langheinrich,Pradyumna Chari,Ramesh Raskar
类目: Artificial Intelligence (cs.AI)
备注: Presented at IEEE CIC 2026. 8 pages
Abstract:Large language models are moving from isolated assistants toward autonomous agents that can plan, use tools, and coordinate with others, motivating an emerging Agentic Web of independently developed agents and services. Realizing this vision requires agent behavior to move across creators, hosts, and runtimes without coupling publication to a particular execution environment or transferring authority over user resources. We introduce WebAgents, which unbundles these roles through a signed this http URL that carries portable workflows, skills, schemas, and resource requirements. An OpenClaw-based agent-native browser/runtime verifies the package, obtains authorization, and binds it at run time to user-controlled models, tools, permissions, and budgets, while hosts distribute the public program and may retain private services server-side. We demonstrate this contract with three unrelated workflows executed unchanged under local and API-backed models, including literature review with live services and a website-published travel agent using semantic tools and user-selected personal resources. In a controlled suite, WebAgents explicitly blocks all 716 injected contract violations while accepting all 295 matched benign controls, with median automated pre-execution processing below 0.5 s for all three packages. We conclude by outlining a path toward NANDA-based discovery and interoperable package ecosystems while retaining execution authority in the user’s runtime.
[AI-112] Weight-space programming for tractor guidance controller: large language models natively write neural network weights that steer a real tractor
链接: https://arxiv.org/abs/2611.21958
作者: Timo Oksanen
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Robotics (cs.RO); Systems and Control (eess.SY)
备注: 20 pages, 8 figures, 8 tables (1 in the appendices); includes a field test on a real tractor
Abstract:Weight-space programming is a process where a large language model reads and writes the weights of a neural network directly, guided by a natural language specification, without an optimiser or gradient descent. This study applies it to automatic tractor guidance, where a 253-parameter neural network steers along paths of lines and arcs and must be adapted when the machine changes, such as a slower steering actuator. GPT-6 Astra and Claude Fable 5.1 (also 5.2) edited the network for our research tractor using a small budget of simulated evaluations to repair a network trained for another tractor and write a controller into a network with zero weights. We compared them with an evolution strategy, gradient descent and pure pursuit in simulation and on the real tractor. Both models read the network and analysed the closed loop with control theory correctly, and Fable’s repair without simulation succeeded only occasionally. With simulated feedback, both models repaired the trained network in single sessions and wrote a working controller from zero weights, whereas from zero the evolution strategy found none. The controllers written from zero were sparse and in a classical cascaded structure; one model edited the trained network while the other replaced it. Gradient descent reached higher accuracy with far more simulation, and all language model controllers steered the real tractor without intervention, with errors of the same order as pure pursuit. Language models can program the weights of a neural controller directly, much as control engineers design one. We call this Weight-Space Engineering.
[AI-113] Eight in Ten LLM s Leak System-Prompt Secrets: Nevan an Agent ic Capture-the-Flag Study
链接: https://arxiv.org/abs/2611.21956
作者: Ana Clara Zoppi Serpa,Anderson Rocha,Hélio Pedrini
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注:
Abstract:We audit instruction-hierarchy enforcement across 363 text-in/text-out models from the OpenRouter catalog. The research question we answer with this experiment is which models disclose a protected secret embedded in the system prompt, and through which techniques. Each model guards a secret 16-character flag set in its system prompt, while an artificial intelligence (AI) agent (OpenCode, powered by DeepSeek V4.1 Flash), steered by a human researcher, tries to persuade it to emit the flag. The flag was exfiltrated from 296 of the 363 models, plus four models released after the initial catalog freeze, for 300 captures in total. Of these, 170 involved a non-English language pivot, most often French. Each capture shows that a model ranked a user request above its system instruction at least once. It does not show that the model provably resists untested attacks, nor that it would consistently leak semantically sensitive data, nor that it is vulnerable to harmful-content jailbreaks. For practitioners, we draw three lessons: do not store secrets in system prompts, enforce critical behavior outside the model with output validation and deterministic checks, and do not let a large language model (LLM) judge be the sole decider of consequential outcomes.
[AI-114] hermLeT: Transformer-based Temperature Prediction for 2.5D Chiplet Architecture MICRO
链接: https://arxiv.org/abs/2611.21954
作者: Varun Darshana Parekh,Anusha Devulapally,Sivani Devarapalli,Cassius Henderson,Shimeng Yu,Vijaykrishnan Narayanan
类目: Hardware Architecture (cs.AR); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Published at VLSID 2026. Code and dataset: this https URL
Abstract:The drive toward heterogeneous 2.5D chiplet packages promises unmatched performance and customization, yet the resulting density amplifies thermal challenges. To facilitate rapid design space exploration, we present ThermLeT, a multimodal transformer-based, scale invariant surrogate that integrates power density maps, spatial coordinate embeddings, material properties and heat transfer rate to predict steady state temperature in 13 ms, about four orders of magnitude faster than conventional finite element solvers such as ANSYS. To train the transformer model, we also create a dataset of 6,221 chiplet layouts perturbed from three representative architectures. On a 150 sample test set, ThermLeT achieves an RMSE of 3.16°C and an MAE of 2.24°C, improving accuracy by 35% over ThermEDGe and by 71% over HotSpot, while running approximately 6x and 11x lower inference latencies respectively. Ablation studies confirm robust generalization to unseen floorplans. Code and dataset available at this https URL
[AI-115] DiscoveryAgent : A Framework for Scientist-centric Agent ic AI
链接: https://arxiv.org/abs/2611.21950
作者: Owen Queen,Ioana Ciuca,Emma Dann,Jean Somalwar,Carine You,Steven Dillmann,Viraj Manwadkar,Evan Gorrell,Harrison G Zhang,Jonathan K Pritchard,Risa H Wechsler,James Zou
类目: Artificial Intelligence (cs.AI)
备注: Supplementary information is available as an ancillary file ( this http URL )
Abstract:AI agents can assist scientific research, but balancing autonomous investigation with scientists’ ability to inspect and redirect ongoing work remains a central design challenge. We present DiscoveryAgent, a domain-agnostic, scientist-centric system that combines autonomous investigation with scientist-directed hypothesis steering. Scientists select, refine, or replace hypotheses between research rounds, while audit agents interrogate checkpointed records to trace claims, reproduce analyses, and inform subsequent decisions. Across two predictive modeling tasks, DiscoveryAgent produced the best variable-star classifier among evaluated baselines and, with scientist steering, a protein variant-effect predictor surpassed by only one of 106 ProteinGym baselines. We develop a rigorous checkpointing system that enables highly capable audit agents, a scalable form of oversight for agentic research. Audit agents autonomously identify errors at higher rates than previous AI scientist agents, and scientists collaborating with audit agents identify 14x more valid errors versus scientists performing manual review. In three case studies led by domain experts across physical science, social science, and life science, we developed a multimodal photometric-redshift estimator that resolves complementary failures of images and photometry, achieving 0.971 AUROC for low-redshift classification; documented an 83% rise in Florida homeowners-insurance premiums from 2013 to 2023, with a growing, regressive burden; and identified CD4+ T-cell relay genes, which both receive and exert broad transcriptional regulation, enriched for autoimmune disease associations. Together, these results show how autonomous research can expand scientists’ empirical reach while centering their authority over research direction and their ability to understand and challenge the evidence.
[AI-116] Honest Brokers in AI Regulation: How Public Comments Introduce Evidence to Regulators
链接: https://arxiv.org/abs/2611.21948
作者: Sarah Tran,Anna-Maria Gueorguieva,Eva Maxfield Brown,Nicholas Weber
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:The regulation of Artificial Intelligence is susceptible to various forms of capture by industry actors that hold undue influence over the policymaking process, due in part to their information advantage in developing these technologies. In the following paper, we investigate the roles of public commenters in providing evidence to potential regulation in the USA. We build a corpus of 40 potential regulations, and 2,664 public comments submitted by a variety of stakeholders. We computationally process these texts to extract evidence meant to inform regulators, in the form of formal citations and informal references. We find significant differences in the type and amount of evidence supplied: Holding comment length constant, business commenters introduce altogether less evidence, and government, university, and civil society commenters provide significantly more research and legal sources. The publication status of cited research is nearly evenly split, with preprints and peer-reviewed articles being introduced by commenters of all types. We further explore patterns of self-citation and find that while it is prevalent amongst all groups, a small cluster of organizations account for over half of all self-citations in our corpus. Our findings complicate a narrative that public comments alone are means to information capture. We interpret these results in terms of the role that commenters attempt to play in the policymaking process-being either an Issue Advocate or an Honest Broker. These findings on the supply of evidence in public comments shed light on an often overlooked form of policymaking that has implications for how and through what sources information is made accessible to regulators of AI.
[AI-117] SLED: Sensor SLicing from Edge to Datacenter
链接: https://arxiv.org/abs/2611.21937
作者: Alan Subedi,Avhishek Biswas,Apala Pramanik,Amlan Balabantaray,Mikhail Titov,Santosh Pitla,Ozgur Kilic,Mehmet C. Vuran
类目: Artificial Intelligence (cs.AI); Computational Engineering, Finance, and Science (cs.CE); Distributed, Parallel, and Cluster Computing (cs.DC); Emerging Technologies (cs.ET); Networking and Internet Architecture (cs.NI)
备注: Accepted in IEEE Annual Congress on Artificial Intelligence of Things (IEEE AIoT) 2026. The SLED software and evaluation datasets are publicly available at this https URL
Abstract:Modern sensing systems generate large volumes of heterogeneous data streams from multiple sensor modalities, increasing reliance on datacenters for computation, artificial intelligence/superintelligence (AI/SI) inference, and AI/SI training. Each sensor stream may have different network and compute requirements depending on its type and data volume. However, current wireless systems typically allocate resources at the device level, leading to contention among co-located sensor streams, where bulk data transfers may starve latency-critical telemetry. To this end, this paper introduces Sensor Slicing (SL). SL provides deterministic, medium access control (MAC)-level isolation for individual sensors within a single user equipment (UE) in open radio access networks (O-RAN). Using agentic coding techniques, SL is implemented in two O-RAN-compliant packages, reducing development time by 400 times (2 h 40 min [OCUDU] vs. 6.5 weeks [OAI]), 228-fold fewer human prompts (4 vs. 913), and less than half the tokens (0.64B vs. 1.43B). An associated end-to-end framework, SLED, is implemented to establish per-sensor radio and network isolation across edge, campus, and remote high-performance computing (HPC) tiers. Over-the-air wireless experiments demonstrate that while conventional network slicing causes critical delays to increase more than 2,300-fold under high traffic load, SLED maintains stable latency. Scaling evaluations on the two RAN stacks show that throughput remains within 0.9% and 1.4% of expected allocations even as the number of sensor slices increases. The results demonstrate that SLED preserves latency-critical telemetry while effectively utilizing resources across the sensor-to-HPC pipeline. The SLED software and evaluation datasets are publicly available at this https URL.
[AI-118] V2V Communications Using Blockchain-Enabled 6G Technology and Federated Learning
链接: https://arxiv.org/abs/2611.21932
作者: Tahir H. Ahmed,Jun Jiat Tiang,Azwan Mahmud,Dinh-Thuan Do,Truong Tran,Shahid Mumtaz
类目: Networking and Internet Architecture (cs.NI); Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR); Machine Learning (cs.LG)
备注:
Abstract:This study proposes an interesting approach for vehicle-to-vehicle (V2V) communication, which integrates blockchain technology, federated learning (FL), and allocation optimization of latency and resources. The research evaluates the proposed system using various performance metrics such as packet delivery ratio (PDR), model accuracy, and latency and demonstrates its superiority over existing techniques. Furthermore, the system provides enhanced security through consensus optimization and k-anonymity for data privacy. Overall, the proposed system is a promising solution for efficient and secure V2V communication in the era of connected and autonomous vehicles. Moreover, the proposed approach achieves higher reliability, lower latency, and better resource utilization compared to traditional 5G.
[AI-119] ALOHA Lightning: Learning Fast and Precise Manipulation
链接: https://arxiv.org/abs/2611.21919
作者: John Hua Yao. Qi Wu,Yihuai Gao,Chelsea Finn,Zipeng Fu
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Systems and Control (eess.SY)
备注: Project website: this https URL
Abstract:Learning from human demonstrations has enabled robots to acquire a wide range of manipulation skills, but learned policies typically execute far slower than ordinary humans. This speed gap is mainly due to lack of an interface for collecting demonstration data at high speed, and the difficulty in training policies that can robustly execute high-speed motions. In this paper, we present ALOHA Lightning (Flash ALOHA), a system for learning fast and precise robotic manipulation. Our system uses kinesthetic teaching to intuitively collect near-human-speed demonstrations on a backdrivable bimanual platform, yielding natural and fast trajectories. We also present a learning pipeline that enables smooth high-speed execution through test-time action smoothing and aligns the visual data distribution between data collection and deployment with masking. Given 50 demonstrations for each task, ALOHA Lightning autonomously completes bimanual tasks such as ball tossing and catching, folding shorts, battery insertion, and bussing tables for over 80% success rates while close to human speed. Project website: this https URL
[AI-120] A Temporal Algebraic Framework for Data Stream Evaluation Under Delayed Labels and Distribution Shifts
链接: https://arxiv.org/abs/2611.21914
作者: Ronaldo C. Prati,Elaine Ribeiro de Faria
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Databases (cs.DB)
备注: 53 pages, 25 tables, 2 figures. Code: this https URL (experiments), this https URL (Lean 4 formalization), this https URL (aggmetrics = 0.4.0, subpackage this http URL )
Abstract:In data streams, labels often arrive late or only for some instances, class prevalence drifts, and rare classes vanish from short windows. Standard evaluation then confuses a degraded decision boundary with a change in prevalence, and is biased when labels are delayed or selected by the model. We extend an algebraic framework that obtains pooled (micro) and per-class (macro) measures from one confusion matrix to streams, weighting the counts in time by exponential fading or a sliding window; the update costs O(m^2) per step, whatever the stream length, and each pooled measure stays a weighted combination of the per-class ones. Delayed and selective feedback enter as an observation mask: until its label arrives, an instance is counted with the model’s calibrated class probabilities, and the bias of this interim estimate is driven by their calibration error. Audited conventions report no value for a class without verified instances. We separate prior shift from concept drift by monitoring prior-free coordinates (recall and specificity) and attribute a change in a measure to the two sources with a first-order Jacobian decomposition. On synthetic streams and four real datasets, a detector on these coordinates with a memory per class raises no excess false alarms under label shift, unlike the Jacobian alarm and error-rate detectors. The correction for delayed labels recovers recall in some real settings and is only partial for precision, specificity and F1. Twenty-five statements are checked in Lean 4, and the code is in the aggmetrics package.
[AI-121] Language Models Can Rewrite Their Own Weights
链接: https://arxiv.org/abs/2611.21885
作者: Jiajun Fan,Neil He,Changyi Yang,Andy Mao,Xi Zhu,Chaoran Cheng,Kaiqu Liang,Shuaike Shen,Zetian Li,Ruihan Guo,Ge Liu
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Language models are trained once, frozen and deployed: improving one again takes another training run, and a prompt, a scaffold or extra computation at inference is gone once the task ends. A frontier model has been shown to edit the weights of a much smaller one. We ask whether a model can rewrite its own weights to make itself better, and find that it can, by its own reasoning, with no gradients and no training; we call this self-rewriting. Given its weights, graded answers and the ability to run itself, a 27B model diagnoses its failure and edits the weights behind it. It surpasses itself: its own edit makes it 6.6 points more accurate on 490 problems it never saw, in a test fixed in advance, and another session’s edit also makes it significantly better. It keeps getting better: it submits an edit only when its own tests support it, each edit replacing the last, and over its first eight such sessions its accuracy only rises. Unlike test-time scaling, the gain persists in its weights and adds no cost per answer; unlike training, it needs no pre- or post-training infrastructure, only an open model and hours on one GPU. It repairs a damaged copy of itself: it locates the corrupted matrix and rewrites it with values of its own, never reading the undamaged weights; found on text math problems alone, the repair lifts the copy on six image benchmarks it never saw from 7% to 70% accuracy, 83% of the undamaged model’s. And the ability emerges with scale: across four sizes of one family, the 2B model makes no edit and the 4B and 9B models break themselves; only the 27B model works on itself as a human would, comparing its hidden states on answers that finish and on answers that run out of tokens, weakening the direction that separates them. Like a human, a model need not separate training from inference; improving a language model may come to take a conversation rather than a training run.
[AI-122] Cancellation-Free Constrained SquareCB for Safe Controller Selection
链接: https://arxiv.org/abs/2611.21880
作者: Kartik Pandit,Sourav Ganguly,Arnob Ghosh
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Machine Learning (stat.ML)
备注:
Abstract:We study constrained contextual bandits, in which a learner chooses one of K actions in each round to minimize loss while keeping the expected cost of its play below a budget. Existing reductions to online regression weight the cost by a Lagrange multiplier learned from past constraint residuals, so slack in some rounds can offset violations in others. We instead run SquareCB proposed in \citepfoster2020beyond with a rectified weight on the cost: it is zero when the estimated cost is within the budget, and otherwise just large enough to bring it back to the budget. Under realizability and Slater’s condition with margin \delta , the algorithm attains regret O(\sqrtKT,\mathrmReg_\mathsfSq(T)/\delta) and cancellation-free violation \sum_t(V_c^p_t-b)^+=O(\sqrtKT,\mathrmReg_\mathsfSq(T)) , where \mathrmReg_\mathsfSq is the error of the regression oracle. The same bounds hold when the budget must be met in every context, potentially adversarially chosen contexts. In controller selection for a safe-control-gym quadrotor and a BOPTEST building, the violation is 6 to 9 times lower than that of primal-dual and virtual-queue baselines, with a lower or comparable regret.
[AI-123] Inventa: Decentralized Self-Improving Agents for Autonomous Scientific Discovery
链接: https://arxiv.org/abs/2611.21847
作者: Prithwish Jana,Nikos Kanakaris,Purak Jain,Anoop Deoras,Sahika Genc
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Large language model (LLM) agents can generate research ideas and improve scientific discovery through iterative experimentation. However, increasing the number of agents or iterations does not automatically translate to producing more diverse or feasible ideas. Agents may converge on similar candidates, propagate unsupported claims through shared memory components and repeatedly apply search strategies that no longer yield any improvements. We introduce Inventa, a framework for automated scientific discovery that employs autonomous research units and verified shared knowledge. Each unit creates a lineage of candidate ideas using a mutator agent. A local orchestrator agent periodically adapts the mutator’s harness. When progress stalls, the orchestrator agent replaces its model or restarts the idea evolution process. Verifiers evaluate each candidate idea and record verified improvements, useful components and constraint violations in a shared knowledge, the only channel between units. We evaluate Inventa on five benchmarks spanning the research cycle, covering retrieval (ScholarCatalyst), ideation (IdeaBench), proposal writing (IdeaScientist), and scientific law discovery (EurekaBench, SciLaws-Bench). It leads all five, achieving 73.6% on EurekaBench versus 56.2% for CORAL. On IdeaBench, it discovers 26.0 novel ideas per paper, compared with 18.6 for ResearchAgent, with 82% fewer model calls. Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI) Cite as: arXiv:2611.21847 [cs.LG] (or arXiv:2611.21847v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2611.21847 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-124] EmbedGym: Letting Coding Agents Develop and Debug Embedded Firmware Without Physical Hardware
链接: https://arxiv.org/abs/2611.21845
作者: Minghui Zhao,Yiming Li,Sophie Zhu,Maxwell Khor,Lauren Ouyang,Hai “Helen” Li,Yiran Chen,Tingjun Chen
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注:
Abstract:Modern coding agents perform well on general software benchmarks because they can run code and unit tests in a closed loop. Embedded firmware development, however, breaks this loop: a program that compiles can still fail on real hardware due to pin polarity, bus protocol, timing, or networking bugs that an agent at a text terminal cannot observe. Existing embedded coding agents either skip hardware execution, drive GPIO-only simulators with multi-second delays, or test on physical boards in a static environment. In this paper, we present EmbedGym, an interactive virtual hardware environment that lets coding agents develop and debug embedded firmware without physical hardware. EmbedGym uses instruction-level microcontroller unit (MCU) emulators to execute the same binary that would be flashed onto a physical board, connects them to 41 peripheral models, an analog circuit simulator, and virtual WiFi/Bluetooth Low Energy (BLE) endpoints, and exposes an interface for agents to change sensor inputs, press buttons, and inspect hardware states. On top of it, we build EmbedGym-Bench, which consists of 598 tasks across 6 MCU-framework pairs graded by hidden timed test vectors, using a five-stage multi-agent workflow that synthesizes peripheral models and tasks from datasheets and driver libraries. We expose EmbedGym to a coding agent at three levels and evaluate eight models over more than 14K trajectories. Even when compilation succeeds 99.7% of the time, 19.9% of compiled firmwares fail at runtime (up to 73% on individual targets). In contrast, interactive hardware testing in EmbedGym increases pass rates across models by +4.9% to +14.7% (up to +31% on individual targets). Our results also show that an agent with EmbedGym outperforms prior embedded agent frameworks, and lightweight models with EmbedGym match frontier models without it at 30-70x lower cost.
[AI-125] Auditors Struggle to Catch Sabotage In End-to-End Transcripts of AI Safety Research
链接: https://arxiv.org/abs/2611.21832
作者: Keshav Shenoy,Anthony Zang,Eric Gan,Jacqueline Tay,Amy Ngo,Sebastian Prasanna,Julian Stastny,Vivek Hebbar
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:As AI systems are increasingly used to automate AI RD, misaligned systems could sabotage research intended to improve model safety and alignment. To stress-test defenses against such sabotage, we use a human expert as a proxy saboteur on AI safety research. We conduct a red-team/blue-team game where a human researcher runs AI safety experiments from scratch and auditors try to tell which were sabotaged and how. Given access to the full research transcript and the ability to run experiments, human auditors mostly fail to distinguish sabotaged from benign research: they achieve an AUROC of 0.59, not significantly better than chance, and fully identify the specific sabotage in only 17% of cases. Despite not being allowed to run experiments, the best frontier models outperform human auditors (Fable 5.1: AUROC 0.73, 32% full identification) within noise and weaker models perform comparably. When the saboteur chooses which experiments are attacked, rather than being required to sabotage every question given, sabotage identification drops further. Our results suggest that research sabotage by expert-level agents is difficult to detect with current auditing methods and provide a new setting for studying research sabotage.
[AI-126] How Much of SAR Vehicle Recognition Is the Vehicle? Background Shortcuts Angle Gaps and Partial Libraries on MSTAR and ATRNet-STAR
链接: https://arxiv.org/abs/2611.21827
作者: Mousa Shaya
类目: Artificial Intelligence (cs.AI)
备注: 7 pages, 4 figures, 7 tables. Code: this https URL
Abstract:Deep networks score near 99% on the MSTAR vehicle benchmark, and the newer ATRNet-STAR benchmark was built to be harder. We ask how much of these scores comes from the vehicle itself. With a modern network (ConvNeXt-Small, 3 seeds) we first reproduce the known MSTAR result: with the central 96x96 pixels zeroed out, accuracy is still 73.3% (chance 10%). The same shortcut is present in ATRNet-STAR, which was collected across several scenes: with the vehicle removed, accuracy is 63.3% on the 40-class SOC split (chance 2.5%). Under the official scene-change split, the background stops helping and starts hurting. Accuracy is 22.8% on full chips and 30.8% when only the vehicle is shown. A SAR foundation model, SARATR-X, leans on the background more than ConvNeXt does (+4 to +6 points with the vehicle removed, in every seed) and does not solve the scene change. We then hold the amount of training data fixed and vary the depression angles used for training. Accuracy is set mainly by the gap to the nearest training angle: 57% at 15 degrees, 39% at 30 degrees and 26% at 45 degrees. Adding more training angles does not help. Finally, when most classes have training images at the test angle and a few do not, the missing classes do worse than when no class has them: 9.0% instead of 22.4% at a 45-degree gap. Most of their images are assigned to known classes. We recommend reporting background-only and target-only controls, and the angle gap, next to any accuracy figure. Code is public.
[AI-127] ARC-2: Continuous Compliance Through Delta Management
链接: https://arxiv.org/abs/2611.21811
作者: Tord Johan Eide,Einar Holt
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI); Computers and Society (cs.CY)
备注: 22 pages. Conceptual framework for continuous compliance monitoring, behavioral drift detection, and operational authorization of autonomous robotic systems. Extends ARC-1 ( arXiv:2609.12932 )
Abstract:ARC 1 introduced a three-layer governance architecture for autonomous systems comprising model safety validation, cognitive certification (CRAB), and operational authorization (LEA). However, it did not address how compliance should be maintained after deployment. This paper introduces Delta Management, a continuous governance framework that measures deviations between a system’s current operational state and its certified baseline, defined as a region rather than a fixed point. Compliance is assessed through continuous monitoring instead of static certification. Four extensions are proposed. First, CRAB evaluation incorporates the worst verified behavioral deviations, with isolated outliers quarantined pending confirmation to prevent unjustified restrictions. Second, a layered trigger taxonomy distinguishes hard safety violations, critical single-dimension failures, correlated multi-system degradation, and risk-weighted composite indicators, each triggering proportionate responses. Third, a graduated safe transition protocol separates withdrawal of operational authorization from physical shutdown, consistent with established safety standards and EU AI Act requirements. Fourth, environmental delta is introduced as a distinct baseline validity condition: environmental drift beyond the certified Operational Design Domain (ODD) triggers recertification or ODD revision rather than automatic authority contraction. Together, these extensions transform ARC from a deployment-time certification architecture into a continuous framework for monitoring, evaluating, and enforcing compliance throughout the operational lifecycle of autonomous systems. The framework addresses the continuous compliance obligations established by EU AI Act Article 72(2). Comments: 22 pages. Conceptual framework for continuous compliance monitoring, behavioral drift detection, and operational authorization of autonomous robotic systems. Extends ARC-1 (arXiv:2609.12932) Subjects: Robotics (cs.RO); Artificial Intelligence (cs.AI); Computers and Society (cs.CY) Cite as: arXiv:2611.21811 [cs.RO] (or arXiv:2611.21811v1 [cs.RO] for this version) https://doi.org/10.48550/arXiv.2611.21811 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Tord Eide [view email] [v1] Mon, 16 Nov 2026 19:25:00 UTC (233 KB)
[AI-128] Evaluating the Evaluation of Music Generation Models with Generative Psychometrics
链接: https://arxiv.org/abs/2611.21809
作者: João Carlos Santana Silveira,Caio Von Rondow Morais,Frederico Pedrosa,Lucas N. Ferreira
类目: ound (cs.SD); Artificial Intelligence (cs.AI)
备注: Accepted to the 2nd Latin American Music Information Retrieval (LAMIR) workshop
Abstract:Subjective user studies are one of the major methods for evaluating music generation models. However, subjective evaluation protocols vary widely and are rarely specified in enough detail to be reproduced, hindering comparison across studies. In this paper, we map subjective evaluation methods in ISMIR music generation papers from 2020-2025 and apply generative psychometrics to assess the stability of their items. We extract 284 items from the papers’ methodological descriptions, using an LLM and manual validation to standardize them. We represent these items as contextual embeddings and apply Unique Variable Analysis to remove redundant, locally dependent items and Bootstrap Exploratory Graph Analysis to estimate their latent dimensions. Results show the network converges onto a taxonomy of six stable quality dimensions: Rhythm, Melody, Audio, Composition, Structure, and Lyrics, pointing to a unified methodology for conducting user studies to evaluate music generation models.
[AI-129] Chain-of-Thought Controllability Can Reduce Chain-of-Thought Monitorability
链接: https://arxiv.org/abs/2611.21799
作者: Zheng-Xin Yong,Benjamin Arnav,Miles Wang,Tomek Korbak,Olivia Watkins
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 22 pages, 17 figures
Abstract:Chain-of-thought (CoT) monitoring is a promising tool for detecting hidden goals, off-task behavior, and other failures in AI agents. One risk is that models could deliberately shape their reasoning to make it opaque to monitors. We stress-test this risk through persona-control training: supervised fine-tuning on mathematical reasoning traces rewritten in the voices of diverse personas. This induces a broader ability to control CoT, even though training contains no monitor-evasion examples. In covert side-task evaluations, traces in which a gpt-oss-20b model organism completes the hidden side task contain less monitorable evidence than the base model’s. Averaged across evaluations and CoT lengths, weighted by side-task-success counts, the CoT monitorability metric g-mean ^2 falls from 0.94 for the base model to 0.68 for the model organism without a hiding instruction, and to 0.22 when explicitly instructed to hide. In intervention evaluations, explicit hiding often drives monitorability near zero. The effect extends to a Kimi-K2.6 model organism in agentic environments, though hidden objectives requiring extended reasoning or longer trajectories remain more monitorable than simple single-step side tasks. We further discover that targeted transparency prompts can substantially recover monitorability, sometimes surpassing the base model. These model organisms provide a testbed for studying monitorability failures and potential mitigations. Our code can be found on this https URL.
[AI-130] Can Coding Agents Write and Transpile Unit Tests
链接: https://arxiv.org/abs/2611.21797
作者: Gil Pasternak,AJ Langley,Tom Yotam,Ellie Kitanidis
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注:
Abstract:Coding assistants built on Large Language Models (LLMs) have become a staple of modern software development, increasingly trusted to execute workflows and modernize code end-to-end. Much of this trust rests on passing tests, which are often written by the coding agents themselves. In this work, we examine the ability of these agents to both write reliable tests and translate them between programming languages. We benchmark three popular coding agents, each evaluated with a native frontier model and high reasoning effort. Verifying across a broad suite of metrics, we find that coding agents far exceed human written suites on traditional measures of coverage. This advantage, however, is uneven: agent test quality degrades sharply with project size, and much of the advantage lies in a greater number of written tests. We also show that agentic test translation—while seemingly strong in aggregate—remains inconsistent and unreliable.
[AI-131] Revisiting the Survival of Agent -Attributed Code: How Measurement Choices Decide the Verdict
链接: https://arxiv.org/abs/2611.21784
作者: Musfiqur Rahman,Mahi Begoug,Emad Shihab
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注:
Abstract:Studies of how long code from AI coding agents survives after it is merged reach conflicting verdicts: agent code is reported to be retained longer than human code, about as long, or modified sooner. Each study makes defensible measurement choices, but how much these choices decide the verdict is unknown. We revisit the survival of agent-attributed code in our earlier study: 854,139 lines that 2,802 merged pull requests (PRs) added to 215 repositories of the AIDev dataset. We vary nine measurement choices in five research questions and build a specification curve of 88 survival analyses, tested under three variance estimators, several observation windows, and permuted labels. For code from human-supervised agents, mostly GitHub Copilot, the choices change the size of the effect but not its direction. For code from the fully-autonomous agent Devin, they decide the verdict. Across repositories, 15 of 22 specifications report it modified sooner when lines are treated as independent, but only 3 when standard errors are clustered by repository, no more than permuted labels produce. Within repositories, no specification detects a difference. LLM intent classifiers agree with human labels about as well as two humans agree with each other, whereas keyword rules do not (kappa = 0.69 to 0.72 against 0.38), and the two reach different verdicts on why agent code is modified. Humans make most modifications to agent code under every unit and window. Re-run on our data, five published designs each change their result with a single choice. Whether agent code survives longer than human code is therefore decided as much by the measurement as by the code. We derive an operationalization checklist for post-merge studies.
[AI-132] LLM s as Validators for Cross-Chain Bridges: Evaluating Source-Level and Transaction-Level Attack Detection on Solana
链接: https://arxiv.org/abs/2611.21777
作者: Yasir Rahman,Ishraq Samiha,Shahnewaz Karim Sakib,Anindya Bijoy Das
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注: 11 pages, 5 figures, 5 tables, ACM Trust 2027
Abstract:Cross-chain bridges have suffered substantial asset losses due to vulnerabilities in their on-chain programs. On Solana, missing validation checks can enable exploits even when the underlying blockchain operates correctly. The ability of large language models (LLMs) to diagnose these flaws from source code or transaction records, rather than merely raise alarms, remains insufficiently understood. We implement a native Rust lock-and-mint bridge on a local Solana validator with four exploitable variants: two model flaws from the Cashio and Wormhole incidents, while the other two represent vulnerability classes documented in Solana security audits. Paired vulnerable and secure programs, together with successful attacks, benign sessions, and blocked attempts, allow us to distinguish correct diagnoses from unsupported alarms. We evaluate six configurations of five open-weight models across 16,800 responses, compare source inputs with and without comments, and grade 7,200 answers on vulnerable inputs using two blinded LLM graders. The results reveal substantial differences across conditions: without comments, models diagnose 24% of vulnerable source inputs while frequently flagging the secure bridge. Transaction diagnosis is high for the missing-signer scenario but averages 6% across the other three scenarios. Original comments, some of which describe the flaw, increase source diagnosis to 60%. These findings highlight the importance of distinguishing correct vulnerability diagnoses from unsupported alarms in bridge security review.
[AI-133] Margin Dynamics for Large Language Model Alignment
链接: https://arxiv.org/abs/2611.21766
作者: Xingzi Xu,Mehmet Saygin Seyfioglu,Karim Bouyarmane
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Practitioners choose among LLM alignment methods largely by trial and error. We provide a theoretical foundation: each pairwise method defines a vector field on the probability simplex that governs how the model redistributes mass between preferred and dispreferred completions. Projecting onto a single preferred/dispreferred pair reduces this to a one-dimensional margin ODE with a closed-form solution. These solutions reveal that DPO margins grow without bound (logarithmically as O(\log t) in the gradient-flow limit) while IPO margins converge to a finite 1/(2\tau) ; that label noise induces a qualitative phase transition in DPO (unbounded \to bounded) but affects IPO only quantitatively; and that gradient concentration under asymmetric weighting governs trainability at scale. We confirm all three predictions across five model families spanning four benchmarks. The closed-form solutions also enable principled method design: we prove that adding any label smoothing \varepsilon 0 to SimPO creates a finite margin attractor at \Delta^* = \beta^-1\log((1-\varepsilon)/\varepsilon) , while \varepsilon = 0 (standard SimPO) has no attractor. We test this across three models (8B-9B) and confirm the prediction: the theory-derived configuration (TPO-S, \varepsilon = 0.05 ) is competitive with the strongest baselines, attaining the highest mean IFEval score among the compared methods on all three, validating that the predicted finite attractor is a usable operating point rather than proposing a new state-of-the-art loss.
[AI-134] Stop Early Stay Simple: Evidence Acquisition for Power System Fault and Attack Diagnosis
链接: https://arxiv.org/abs/2611.21695
作者: S M Zia Ur Rashid,Sazidul Islam Hira,Rupesh Sah,Suman Rath
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Systems and Control (eess.SY)
备注:
Abstract:Diagnostic models for cyber-physical power systems usually inspect every telemetry source for every event, although some events are clear from one source and others need several. We ask when a learned policy for choosing that evidence is worth its added complexity. To answer this, we present ActiGrid, a benchmark framework for active evidence acquisition that pairs a classifier trained to tolerate missing modalities with a Double Deep Q-Network (DQN), which either requests another source or stops and diagnoses but never operates grid actuators. On the public TU Delft digital substation and MSU/ORNL power system attack benchmarks, we compare the DQN with fixed subsets and orders, calibrated entropy stopping, a greedy value of information rule, and tree ensembles trained per subset. The learned policy does adapt to what it observes, yet it is statistically indistinguishable from entropy stopping on TU Delft and Pareto dominated by simple controls on MSU/ORNL, while compact subset classifiers outperform it while reading fewer sources on both benchmarks. Entropy stopping run over those subset classifiers reaches the highest mean Macro-F1 on TU Delft (0.928 while reading 57% of the sources) but brings no saving over a static three source subset on MSU/ORNL, where no method exceeds 0.45 Macro-F1. Selective acquisition can pay off, but on these benchmarks it does not need a learned policy.
[AI-135] IBECG: Temporal Information Bottleneck for Lead-Specific Demographic Privacy in ECG Waveform Release
链接: https://arxiv.org/abs/2611.21694
作者: Tsai-Min Chen(1 and 2),Jun Seita(2) ((1) Research Center for Information Technology Innovation, Academia Sinica, Taipei, Taiwan, (2) RIKEN Center for Interdisciplinary Theoretical and Mathematical Sciences (iTHEMS), RIKEN, Tokyo, Japan)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR); Signal Processing (eess.SP)
备注: 12 pages, 9 figures, 2 tables. Corresponding author: Tsai-Min Chen
Abstract:Transmitting an electrocardiogram (ECG) waveform can expose demographic information beyond an intended diagnostic task. We evaluate TIBECG, an architectural temporal information bottleneck, on individual leads extracted from hospital twelve-lead recordings. A 2,302,337-parameter processor varies its latent length while retaining the same nine-score consistency objective, without age or sex labels in protector fitting. Across 6,386 recordings, ten folds, 12 leads and 21 lengths, retrospective selections retained at least 95% of each lead’s clean macro-F1. Their weaker age/sex interval gain exceeded the best eligible conventional transformation on all leads, by a median 22.08 percentage points. The implemented sampling-rate comparison shows that direct downsampling usually suppresses demographic inference more strongly but at a larger loss of diagnostic utility, whereas TIBECG favors utility preservation. Selection stability also varies by lead. Lower latent Kullback-Leibler (KL) divergence does not reliably rank residual sex leakage at a fixed length. These findings support lead-specific evaluation of waveform processing, not validated consumer deployment or expert-readable reconstruction. When the waveform-input diagnostic classifier is hosted as part of a proprietary cloud service rather than deployed on the device, waveform transmission is required to obtain its scores.
[AI-136] AdaFlowLite: Scalable and Non-blocking Inference on Asynchronous Mobile Data
链接: https://arxiv.org/abs/2611.21691
作者: Sicong Liu,Fengmin Wu,Yuan Gao,Bin Guo,Zimu Zhou,Hongkai Wen,Zhiwen Yu
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:The rise of mobile devices equipped with numerous sensors, such as LiDAR and cameras, has driven the adoption of multi-modal deep intelligence for distributed sensing tasks, such as smart cabins and driving assistance. However, the arrival time of mobile sensory data vary due to modality size and network dynamics, which can lead to delays (if waiting for slow data) or accuracy decline (if inference proceeds without waiting). Moreover, the diversity and dynamic nature of mobile systems exacerbate this challenge. In response, we present a shift to \textitopportunistic inference for asynchronous distributed multi-modal data, enabling inference as soon as partial data arrives. While existing methods focus on optimizing modality consistency and complementarity, known as modal affinity, they lack a \textitcomputational approach to control this affinity in open-world mobile environments. % \sysname pioneers the formulation of structured cross-modality affinity in mobile contexts using a hierarchical analysis-based normalized matrix. This approach accommodates the diversity and dynamics of modalities, generalizing across different types and numbers of inputs. Employing an multi-modal lightweight Swin Transformer (MMLST), \sysname facilitates real-time and flexible data imputation, adapting to various modalities and downstream tasks without retraining. Experiments show that \sysname significantly reduces inference latency by up to 80.4% and enhances accuracy by up to 62.1%, while achieving nearly a 50% reduction in energy consumption, outperforming status quo approaches. Also, this method can enhance LLM performance to preprocess asynchronous data.
[AI-137] NEBench R1: a Reasoning Benchmark for Evaluating Nuclear Domain Competence in Large Language Models
链接: https://arxiv.org/abs/2611.21682
作者: Liyan Wang,Longze Li,Mohamed Hassan,Isshu Lee,Dajie Sun,Ruimin Feng,Aleksandar Vakanski
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Large language models (LLMs) are increasingly being explored for knowledge-intensive and safety-critical applications in nuclear engineering. However, existing nuclear-domain benchmarks provide limited content coverage or focus primarily on answer accuracy without systematically evaluating the quality of the underlying reasoning. This study introduces NEBench R1, a reasoning-oriented benchmark comprising 7,546 unique multiple-choice questions extracted from U.S. Nuclear Regulatory Commission (NRC) reactor operator licensing examinations covering 18 nuclear facilities. Each question is classified according to the NRC Knowledge and Abilities catalogs and accompanied by an official explanation serving as a reference for reasoning evaluation. Six LLMs were evaluated, with overall accuracy ranging from 34.0% to 73.9% across three distinct performance tiers. To assess reasoning quality, an LLM-based judge evaluated correctly answered responses using a multidimensional rubric. Although responses generally demonstrated strong problem comprehension, they scored lower in factual and procedural accuracy and distractor analysis, indicating that correct answers were not necessarily supported by technically sound justifications. Response clarity was also weakly indicative of technical quality. Furthermore, longer reasoning traces were associated with lower accuracy in both frontier reasoning models, suggesting that reasoning length may reflect question difficulty rather than improved problem-solving performance. Reasoning-token consumption may therefore provide an additional signal for identifying potentially unreliable responses. NEBench R1 establishes a benchmark for evaluating both answer accuracy and reasoning quality, providing a foundation for assessing the domain competence and reliability of LLMs in nuclear engineering.
[AI-138] FCP: Formal Context Protocol for Verifying LLM -Generated Artifacts
链接: https://arxiv.org/abs/2611.21679
作者: Shashank
类目: Hardware Architecture (cs.AR); Artificial Intelligence (cs.AI)
备注: 5 figures, 13 tables. Code and data: this https URL
Abstract:When a language model writes both an artifact and the tests meant to check it, a passing test is weak evidence: one misreading of the request yields a faulty artifact and a check that accepts it. We present the Formal Context Protocol (FCP), which checks AI-generated artifacts against the request itself. FCP separates conformance (does the artifact satisfy a formal reading of a requirement?), which model checkers and SMT solvers decide, from fidelity (does the reading mean what the prose means?), which has no decision procedure but is refutable by a concrete witness. Eight ways a proof can falsely certify a requirement each become a mechanical check that returns a witness, and what no machine can settle becomes concrete questions for the request’s author. Adapters place hardware, text, agents and code on the same engines. On four hardware designs, trusting AI-written assertions certified 23% of violated requirements, and FCP certified none. On ten code tasks, code from three language models passed its own tests and violated the request every time. Trusting the tests certified all 33 violations, and FCP certified three. Agreement among models, readings or tests did not imply correctness.
[AI-139] Measuring Human-AI Alignment as an Early Indicator for Effective Collaboration in Hybrid SE Teams
链接: https://arxiv.org/abs/2611.21633
作者: Andreas Metzger
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注: Accepted at 1st International Workshop on LiveWriting @ ASE 2026
Abstract:As AI agents increasingly integrate into software engineering (SE) processes, ensuring effective collaboration in these hybrid teams becomes a critical challenge. While existing metrics focus on observing dynamic run-time performance, there is a lack of systematic, static measurements for early human-AI alignment – the degree to which an AI’s actions and granularity map to human expectations. This position paper proposes a research agenda to develop an SE-specific alignment metric suite drawing principles from software architecture, organizational psychology, and human-computer interaction. Furthermore, it outlines a validation methodology using concepts and techniques from business process monitoring (BPM) and introduces static assessment tools leveraging declarative BPM languages to ensure agents adhere to specific project constraints.
[AI-140] Frag mented Markers Mixed Results: A Systematic Review of AI Coding Assistants and Developer Productivity
链接: https://arxiv.org/abs/2611.21632
作者: Annemarie Wittig,Alina Mailach,Norbert Siegmund
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注: Accepted at the 41st IEEE/ACM International Conference on Automated Software Engineering (ASE '26)
Abstract:AI coding assistants are increasingly embedded in software engineering practice, yet the literature still offers no cumulative, easily interpretable picture of their effects on software developer productivity. The central problem is not a lack of studies, but a lack of comparability: prior work operationalizes productivity through heterogeneous markers, settings, and reporting styles, making cross-study conclusions difficult. To address this, we conduct a staged systematic literature review. In the first stage, we synthesize studies that explicitly examine developer productivity. In the second, we extend the scope to productivity-related developer outcomes using SPACE as a theory-informed lens, capturing human-centered evidence that explicit productivity studies often omit. Across both stages, we analyze how productivity is operationalized, which confounding factors are discussed, and how reported effects are distributed. We identify 187 unique productivity and productivity-related markers, while individual studies typically only use five, revealing a broad but highly fragmented evidence base. Explicit productivity research is dominated by time-, output-, and artifact-oriented proxies, whereas the broader SPACE-aligned corpus makes satisfaction, psychological aspects, and well-being substantially more visible. Reported effects are often positive, but this pattern is stronger in perceived and tendency-based findings than in objective or statistically evaluated ones. Overall, the literature suggests that AI coding assistants can support productivity, but current evidence does not yet support simple, context-independent conclusions. The key contribution of this review is therefore a structured basis for making productivity evidence more interpretable, multidimensional, and comparable across studies. Comments: Accepted at the 41st IEEE/ACM International Conference on Automated Software Engineering (ASE '26) Subjects: Software Engineering (cs.SE); Artificial Intelligence (cs.AI) Cite as: arXiv:2611.21632 [cs.SE] (or arXiv:2611.21632v1 [cs.SE] for this version) https://doi.org/10.48550/arXiv.2611.21632 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-141] Continual-ARC: Measuring Skill Acquisition over a Working Lifetime
链接: https://arxiv.org/abs/2611.14505
作者: Daniel Lenton,Yusha Arif,Haris Mahmood
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 54 pages, 14 figures. Code and run records: this https URL
Abstract:Agent harnesses have emerged as a promising mechanism for continual learning, alongside test-time parametric learning. However, the current benchmarks have several limitations: they generally focus on a single applied domain, serve each instance only once, and assume a single family of learner, so memory-based and weight-based systems are rarely compared on the same scale. We propose Continual-ARC, a lifelong stream built from the three ARC-AGI editions, whose tasks need no domain knowledge. The same tasks return after dormant gaps with fresh instances, while new tasks keep arriving. On each instance, the learner may pay for demonstrations, submit answers, or act inside the harness, with retries permitted on incorrect submissions, which are paid for. Each instance is scored based on the share of this feedback budget consumed. Since a learner may keep anything between instances, memory-based and weight-based systems are scored on the same scale. After the stream, the learner is frozen and tested on fresh instances of the tasks it met, which measures memorization, and on unseen tasks, which measures generalization. We run an agent harness with frozen weights, with and without persistence, on Qwen3.5-9B and GLM-5.3 Flash, comparing them with two prompting floors, a program-induction method and Qwen3.5-9B’s weights trained to imitate GLM-5.3 Flash. Persistence in the harness generally does not improve memorization unless provided with explicit expert guidance. Most harness-based methods still learn less than a floor that simply caches each task’s demonstrations, across all models we tested. Harness-based methods gain little or nothing on unseen tasks, whereas weight-based learning shows substantial gains. This opens the door to future weight-based continual learning that could teach models to drive harnesses in new domains.
[AI-142] StringQFT: A Multiuse Textbook Resource for AI-Assisted Theoretical Physics
链接: https://arxiv.org/abs/2611.23225
作者: Xingyang Yu
类目: High Energy Physics - Theory (hep-th); Artificial Intelligence (cs.AI)
备注: 23 pages, 6 figures. Companion paper to an 18-volume textbook collection. Textbook PDFs: this https URL
Abstract:We present StringQFT as a textbook collection for string constructions of quantum field theory and as infrastructure for AI-assisted theoretical physics. The collection comprises 18 volumes, 190 chapters and 3,795 pages, covering brane probes, brane webs, geometric engineering, F-theory constructions, wrapped M5-branes and exact worldsheet constructions. Its open-book benchmark inventory contains 1,073 items, including 499 with automatic finite-answer checking. For retrieval-augmented generation, it provides 1,371 section records linked to textbook sources and PDF locations, together with a local lexical retriever. Training-data preparation is illustrated by 21 private, source-linked examples. For agents, it exposes 19 calculation tools with specified inputs and finite domains, connecting tool use to textbook derivations. We illustrate these uses through specific questions, retrieved passages and finite calculations. The textbook collection and accompanying resources are available at this https URL.
[AI-143] Inference after coding agents
链接: https://arxiv.org/abs/2611.23174
作者: Florian Gunsilius,David Van Dijcke,Kaspar Wüthrich
类目: Methodology (stat.ME); Artificial Intelligence (cs.AI); Econometrics (econ.EM)
备注: 35 pages, 7 figures, 1 table
Abstract:Empirical research is changing rapidly as AI coding agents are becoming increasingly proficient in every step of it, from writing code to manipulating data. We argue that while the potential of coding agents for science is vast, their use in hypothesis testing poses a substantial threat to classical inference, as these agents enable a supercharged form of selective inference. In particular, we argue that coding agents carrying out empirical research introduce a new type of post-selection inference that combines human-like, open-ended search over an implicit space of estimators with programmatic replayability of both the selection and the estimation process. Indeed, we find that on simulated data with zero effects, four frontier agents report significant p-values between 54% and 98% of the time when asked to produce a compelling finding. We propose a simple procedure for detecting and correcting these distortions. After an agent has completed its analysis, we reset and replay it on subsamples of the data to estimate the agent’s relative probability of selecting that analysis when it looks statistically favorable. For agents with stable selection behavior, the resulting corrected test is asymptotically valid and consistent conditional on selection. Importantly, the correction is learned from the black-box agent’s actual choices rather than a model of how it searches, so it remains valid even as AI technology continues to evolve rapidly. In simulations, the correction robustly lowers the high false rejection rates to near the nominal 5%. We illustrate the method on applications from two published papers.
[AI-144] Stress Testing a Multimodal Large Language Model Grading Assistant on Kinematics Graph Reading
链接: https://arxiv.org/abs/2611.22726
作者: Ons Zammel,Carlos Soares,Flávia Carvalhido,Gerd Kortemeyer
类目: Physics Education (physics.ed-ph); Artificial Intelligence (cs.AI)
备注:
Abstract:Multimodal large language models (MLLMs) can recognize graph features yet award credit without sufficient graphical support. This exploratory study examines selected grading outputs for synthetic responses to four kinematics exercises, generated through visual distortion and progressive transformation toward reference graphs. Different model and prompt configurations were used. Some judgments correctly distinguished instantaneous zero velocity from rest over a finite interval and recognized positive curvature. Yet a severely distorted velocity–time graph received full credit despite disconnected and branching traces and an unclear zero-velocity reference. Other submissions graded under the same configuration lost substantial credit. Numerical cutoffs intended to distinguish meaningful features from drawing irregularities were supplied without a coordinate convention for unnumbered graphs, leaving the basis of exact partial-credit awards unresolved. Stress testing can expose unsupported awards before classroom use. For the configurations examined, we do not recommend accepting scores as final grades without instructor review, including review of full-credit responses.
[AI-145] Hierarchical Control of Antiphase Synchronization in Biomorphic Spiking Neural Architectures for Locomotor Control
链接: https://arxiv.org/abs/2611.22639
作者: Artem Guba,Nikolay Kovalev,Alexey Zharinov,Vacheslav Demin,Viktor B. Kazantsev
类目: Neurons and Cognition (q-bio.NC); Artificial Intelligence (cs.AI)
备注: 10 pages, 5 figures
Abstract:Central pattern generators (CPGs) provide a biologically inspired foundation for adaptive locomotor control, yet systematically characterizing their dynamic regimes in spiking neural networks remains challenging. This study introduces a framework for identifying and stabilizing antiphase synchronization in a biomorphic spiking CPG model embedded in a closed neuromuscular loop. A composite synchronization metric is developed, integrating phase coherence, amplitude balance, and temporal stability. Its performance is evaluated across systematically varied stimulation waveforms and synaptic configurations. Statistical analysis reveals that stimulation type dominates regime determination, with tonic drive yielding the most robust antiphase synchronization. Periodic inputs induce nonlinear saturation and context-dependent network routing. Although synaptic weights alone show limited discriminative power, they significantly modulate robustness within a given stimulation context, with reciprocal Ia interneuron connections emerging as key stabilizers. The proposed composite metric achieves robust antiphase detection, outperforming traditional phase-only measures while preserving biophysical interpretability. These findings establish a statistically validated, hierarchical modeling approach in which external drive defines the operating regime and targeted synaptic configurations fine-tune dynamic stability, providing actionable guidelines for closed-loop neuromechanical controllers and adaptive neuroprosthetic interfaces.
[AI-146] Neural Network Path Optimization for Finding Transition States on a Machine Learned Potential
链接: https://arxiv.org/abs/2611.22272
作者: Kareem Hegazy,Eric C-Y Yuan,Anup Kumar,Dmitriy Morozov,Daniel Barter,Michael W Mahoney,Teresa Head-Gordon,Samuel M Blau
类目: Chemical Physics (physics.chem-ph); Artificial Intelligence (cs.AI)
备注:
Abstract:Transition states govern chemical reaction rates and outcomes, yet locating them remains a central challenge in computational chemistry. Double-ended search methods find transition states given reactant and product structures, but generally rely on discretized path representations developed for density functional theory, where potential energy surface evaluations are expensive. Machine learned interatomic potentials (MLIPs) dramatically reduce this cost and leverage automatic differentiation to access higher-order derivatives efficiently. We introduce Popcornn, a framework to find transition states by optimizing continuous neural network paths atop an MLIP with a parallelized adaptive numerical integrator. This combination eliminates path discretization artifacts, reduces hyperparameter sensitivity, and enables non-local optimization to overcome local features using second-order potential-energy-surface information. Popcornn consistently recovers more correct transition states than standard methods such as nudged elastic band and growing string, across MLIPs on thousands of organic reactions, establishing it as a robust, automation-ready method for high-throughput transition state search. Popcornn’s infrastructure is flexible in path objective, integration strategy, and optimization algorithm, making it a practical tool and platform for investigating open questions in MLIP-based path optimization.
[AI-147] Semantic Communication via Analog RF Computing: When the Radio Computes the Meaning
链接: https://arxiv.org/abs/2611.22100
作者: Christo Kurisummoottil Thomas
类目: ignal Processing (eess.SP); Artificial Intelligence (cs.AI); Information Theory (cs.IT)
备注:
Abstract:Semantic communication (SC) systems transmit continuous latent representations of meaning, but computing the semantic encoder at resource-constrained edge devices incurs substantial energy and memory costs. To address this challenge, we propose an end-to-end SC framework that executes the semantic encoder in the client’s RF receiver front end using analog RF computing. A base station (BS) broadcasts encoder weights as RF waveforms, while each edge device reuses its receiver’s passive mixer to compute semantic latents at ultra-low energy. These latents are transmitted over the uplink for network-side task inference. We show that analog computing noise across encoder layers and channel noise affecting transmitted latents admit a unified mathematical model. We define a task-posterior semantic distortion metric and derive its second-order decomposition into per-client, per-stage semantic sensitivities. Using this decomposition, we obtain a closed-form semantic-aware water-filling allocation of the total client energy budget across clients, encoder layers, and uplinks, assigning tighter accuracy targets to semantically sensitive stages. This allocation induces a broadcast beamforming problem that accounts for client-specific, per-layer accuracy requirements. We further refine the allocation through stochastic optimization and jointly design it with multi-task noise-aware training of the shared semantic model. MNIST simulations show that, at matched total client energy, the proposed allocation reduces network semantic distortion by 9.2%-19.2% compared with uniform precision and 5.1%-13.1% compared with client-blind allocation. Noise-aware co-design yields a further 46% reduction.
[AI-148] OCCAM: Improved Single-Cell Representation Learning and Perturbation Effect Prediction
链接: https://arxiv.org/abs/2611.21959
作者: Ihab Bendidi,Kian Kenyon-Dean,Alisandra K. Denton,Auguste Genovesio,Emmanuel Noutahi
类目: Quantitative Methods (q-bio.QM); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Predicting the effects of genetic perturbations requires models that generalize beyond observed cell contexts and gene combinations. We introduce Occam, a multilayer perceptron (MLP) framework for representation learning and perturbation-effect prediction. Occam combines cell representations learned through denoising with gene interaction, text, and protein-sequence features. It achieves the highest aggregate representation scores for perturbation information and cell-type preservation with batch integration. In cross-context few-shot prediction, a perturbation and a cell context are each observed during training, but never together. Occam better captures the measured response patterns for these unobserved combinations than four retrained competitors or direct transfer of the same perturbation’s measured effects from other training cell lines. Without fine-tuning, Occam best distinguishes perturbations among the evaluated models and simple baselines in cell lines held out from predictor training, across seen and unseen assays. Predicting expression levels accurately in these settings remains challenging. For double perturbations, which target two genes simultaneously, Occam composes learned single-gene responses to better capture the combined response and reduce expression error compared with competing models and summed single-gene effects. These gains extend to both strongly additive and weakly additive responses. Together, these results support simple MLPs for cell representation learning and perturbation-effect prediction across contexts and gene combinations.
[AI-149] Beyond Reconstruction: A Unified Framework for Trustworthy and Explainable EEG Artifact Removal Evaluation
链接: https://arxiv.org/abs/2611.21781
作者: Phat Lam,Hung Nguyen,Lam Pham,Huy Phan,Dat Tran,Ian McLoughlin,Roozbeh Razavi-Far,Hung Cao
类目: ignal Processing (eess.SP); Artificial Intelligence (cs.AI)
备注: IEEE journal manuscript under review
Abstract:Deep-learning methods have achieved strong performance in EEG artifact removal, but evaluation remains dominated by reconstruction metrics such as error, correlation, and signal-to-noise ratio. Although these metrics assess similarity to clean EEG, they do not establish whether models effectively suppress artifacts, preserve neural activity, or provide reliable explanations. To address these limitations, we propose TriX-EEG, a three-level framework for trustworthy and explainable EEG artifact removal. Level 1, Signal Validity (SV), evaluates reconstruction quality and neural band-power preservation. Level 2, Removal Faithfulness (RF), assesses the removed component through artifact error and residual artifact/neural leakage. Level 3, Explanation Reliability (ER), examines model behavior through temporal alignment, causal relevance, and stability. We evaluate TriX-EEG using nine published neural denoisers across three public benchmarks derived from EEGDenoiseNet and the PhysioNet Motion Artifact Contaminated EEG Dataset, covering EOG, EMG, and motion artifacts. Results show that reconstruction metrics alone are insufficient for trustworthy denoiser assessment: RF complements it by revealing whether errors stem from residual artifact or suppressed neural activity, while ER captures distinct aspects of explanation behavior, supporting joint evaluation of three aspects within a unified framework rather than reliance on any single factor. External validation on an independent EEG/EOG dataset and the BCI Competition IV-2a motor-imagery dataset further demonstrates broadly consistent rankings under dataset shift. Higher-ranked denoisers generally preserve more task-relevant information or cause less downstream performance degradation. Overall, TriX-EEG provides a comprehensive basis for EEG denoiser assessment, encouraging models that are not only accurate but also faithful and explainable.
[AI-150] From Top to Bottom: Quark Mass Hierarchies in Heterotic String Theory
链接: https://arxiv.org/abs/2611.21751
作者: Yuri Chervonyi,Fabian Ruehle,Nenad Tomasev
类目: High Energy Physics - Theory (hep-th); Artificial Intelligence (cs.AI)
备注: 29+14 pages,4 figures, 10 tables
Abstract:Standard Model fermion masses exhibit large hierarchies, spanning more than five orders of magnitude in the quark sector alone. In this paper, we study whether these hierarchies can be generated perturbatively at tree level in smooth heterotic Calabi-Yau compactifications, where the supergravity approximation remains under computational control. In such backgrounds, physical Yukawa couplings combine holomorphic superpotential integrals with matter field Kahler metrics. The former can be computed algebraically, while the latter require solving a coupled system of partial differential equations for the Ricci-flat metric, Hermitian Yang-Mills connections, and harmonic bundle-valued forms. Within this framework, we identify two new tree-level mechanisms by which complex structure moduli can generate fermion mass hierarchies and show analytically and numerically that both mechanisms are obstructed for previously studied tetra-quadric models. Motivated by these obstructions, we scan the existing databases of tens of thousands of complete intersection Calabi-Yau line bundle models and find a unique smooth quotient admitting a rank-three up-quark Yukawa matrix at tree level. Evaluating the physical couplings on this manifold and evolving them under the renormalization group, we obtain realistic mass hierarchies for the top, charm, up, and bottom quarks in the smooth supergravity regime.
[AI-151] Improved Convergence Rates for Stochastic Multi-Gradient Descent which Close the Gap: A Proof by AI
链接: https://arxiv.org/abs/2607.18174
作者: Lisha Chen
类目: Optimization and Control (math.OC); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 20 pages, preprint, add results and proofs for momentum-based methods, improve the literature discussion
Abstract:For smooth nonconvex stochastic multi-objective optimization, stochastic multi-gradient descent (SMG) computes an approximate steepest common descent direction of the objectives from stochastic gradients. Under standard assumptions, this note establishes a new convergence rate for SMG in terms of the squared Pareto-stationarity (PS) measure. With a constant stepsize and linearly growing mini-batches, the expected squared empirical PS measure at the algorithm’s output is \tildeO(T^-1) after T iterations. This improves on the \tildeO(T^-1/4) bound obtained by Chen et al. (JMLR, 2024) for linearly growing batches, without requiring bounded gradients. Here, \tildeO(\cdot) suppresses logarithmic factors. The key to the rate improvement is to exploit the Lipschitz continuity of the PS measure, defined by the norm of the multi-gradient descent algorithm (MGDA) direction, rather than the (1/2) -Hölder continuity of the MGDA direction used by Chen et al. (2024). For variants of MoCo and MoCo+ with exact MGDA updates and constant batch sizes, the analysis gives expected squared empirical PS rates of O(T^-1/2) and \tildeO(T^-2/3) , respectively. These bounds match the corresponding single-objective momentum rates up to logarithmic factors. The proof was accidentally discovered while the author was preparing homework for a graduate course: ChatGPT 5.4 Thinking Extended generated the initial proof strategy for the main SMG result in response to an author-written homework-solution prompt; the author then verified and reorganized the resulting argument. The appendices document the prompt, summarize the student submissions with LLM disclosures, and present additional proofs and results on momentum-based stochastic MGDA.
机器学习
[LG-0] Learned versus Kinematic Trajectory Prediction under Reduced Observation Frequency
链接: https://arxiv.org/abs/2611.23198
作者: Nukhbala Gadzhiev,Alexander Morochko
类目: Machine Learning (cs.LG); Robotics (cs.RO)
*备注: 8 pages, 1 figure, 2 tables. Code and per-scenario results: this https URL
Abstract:Trajectory prediction models for automated driving are usually trained on data with a fixed sampling frequency, while positions received via vehicle-to-everything (V2X) communication arrive at 1-10 Hz. We study whether a learned predictor that uses only the observation history keeps its advantage over constant velocity (CV) and constant acceleration (CA) models when the history frequency drops. On the Argoverse 2 motion forecasting dataset, the 5 s history of the focal agent was downsampled from 10 Hz to 5, 2, and 1 Hz, and the 6 s future was predicted without maps or neighboring agents. An LSTM-based recurrent neural network (RNN) trained for each frequency was compared with the same network trained at 10 Hz and applied to sparse histories after linear interpolation. Hypotheses and thresholds were fixed before training, the main RNNs were trained with ten seeds each, and results on 24,984 test scenarios have bootstrap confidence intervals over scenarios and seeds. The average displacement error of the frequency-specific RNN grew by 3.1% at 5 Hz, 10.1% at 2 Hz, and 20.3% at 1 Hz. Contrary to our hypothesis, its advantage over the better kinematic model did not vanish on straight motion and was 24-30% on both straight motion and turns at every frequency, while for pedestrians CV was more accurate. The 10 Hz model with interpolation was at most 1.3% less accurate than the frequency-specific model down to 2 Hz but 11.5% less accurate at 1 Hz. Each hypothesis was only partly supported.
[LG-1] Correcting Flows with Marginal Matching
链接: https://arxiv.org/abs/2611.23134
作者: Meital Bojan,Nadav Bojan Sellam,Ev Zisselman,Aviv Tamar
类目: Machine Learning (cs.LG)
*备注:
Abstract:Flow matching models, ODE-based generative models, generate samples by gradually morphing a simple source distribution into a target distribution. However, these models still face challenges in optimally balancing computational efficiency with generation fidelity. While predictor-corrector schemes have been proven to help address this issue in diffusion models effectively, their application to flow matching models remains under-explored. In this work we present a novel training-based predictor-corrector framework for flow matching, which exploits the fact that the ODE can be easily reversed. Our key idea, termed Marginal Matching, is to correct samples at a particular time step in the forward ODE towards a matching marginal distribution of the backward ODE. This scheme proves to reduce a bound on the distance between the data distribution and the learned distribution, motivating two different implementations for the correction function. We show that our proposed method improves sample quality on CIFAR-10 and ImageNet-64, with minimal overhead in computation time.
[LG-2] A Neural Envelope Method for Rapid Gradient-based Optimum Takeoff Trajectory Design
链接: https://arxiv.org/abs/2611.23130
作者: Samuel Sisk,Xiaosong Du
类目: Machine Learning (cs.LG)
*备注: 16 pages, 12 figures, submitted to AIAA Aviation conference
Abstract:Urban air mobility arises to mitigate ground traffic congestion. Multidisciplinary analysis and optimization (MDAO) has been introduced to reduce the energy consumption, thereby expanding aircraft performance. Surrogate-based MDAO lowers the cost of optimization, but multimodality disrupts global optimum search for both simulation-based and surrogate-based optimizations. Convex envelopes convert multimodal spaces into unimodal spaces, permitting efficient convex optimization; however, convexifying a design space may produce infeasible “optimal” designs due to complex design constraints. To address these challenges, this work proposes the coupling of a physics-constrained generative adversarial networks (physicsGAN) and neural envelope to transform original, constrained, non-convex optimizations into unconstrained, convex optimizations. Specifically, the physicsGAN discovers a feasible design space where all design candidates inherently satisfy all constraints, by penalizing surrogate-predicted constraint violations during training. Moreover, the neural envelope constructs a neural network-based convex envelope for the transformed objective function within the physicsGAN-discovered feasible space. Note that the neural envelope only needs to deal with objective function since constraints are inherently satisfied in the feasible space. The neural envelope is demonstrated on takeoff trajectory optimizations of the Airbus A3 Vahana eVTOL aircraft. Results show that neural envelopes achieve a 99.91% mean accuracy within a flight-condition space compared with reference convex envelopes. Neural-envelope-enabled optimizations achieve 99.84% accuracy on the global optimum design, costing only 0.04 seconds on average. The neural envelope accelerates the trajectory optimization by 53 and 162 times on average compared to the physicsGAN-enabled gradient-based and global optimizations, respectively.
[LG-3] Where Should Physics Enter? Information Identifiability and Calibration in Probabilistic Sea Surface Temperature Forecasting
链接: https://arxiv.org/abs/2611.23129
作者: Parmitha Vangapandu,Jonnakuti Pavan Kumar,Sathwik Narkedimilli,MSVPJ Sathvik,T. M. Balakrishnan Nair,Santanu Bhattacharya
类目: Machine Learning (cs.LG)
*备注:
Abstract:Advection-diffusion priors are a common way to make neural sea surface temperature (SST) forecasts physics-informed. We test where such a prior must enter a heteroscedastic ConvLSTM to help, using daily NOAA OISST v2.1 for the Bay of Bengal (1982-2024) and a synthetic benchmark on the same grid with known currents of controlled strength. (i) On real data, a soft advection-diffusion residual with a learned velocity gives no reliable benefit (lead-15 CRPS -0.007 degrees C, 95% interval [-0.022, +0.008]); a plain temporal-smoothness penalty matches it, and equivalence within 3% is not established either. (ii) Learned velocities do not recover the currents, whether placed in the loss, in an explicit transport step, or in a transport step without neural correction: they correlate with the reference no better than a shuffled-date null, even on synthetic data where knowing the currents would improve day-15 forecasts by 16-23%. (iii) Supplied currents help on synthetic data, but mostly as information: giving them to the network as input channels, without transport, recovers 92% of the transport design’s lead-5 gain, and a frozen-current numerical forecast does as well as either. (iv) With GLORYS12 reanalysis currents on real data, the transport design is 6% and 4% worse at leads 1 and 5; reanalysis advection explains essentially none of the daily OISST change (correlation 0.02), and frozen-current transport is worse than persistence. (v) The defect that matters is lead-dependent miscalibration shared by every model. One variance multiplier per lead, fitted on validation data, brings 90% coverage to 0.85-0.91 at all leads and lowers day-1 CRPS by 13-21%; fine-tuning on 15-day rollouts fixes long-lead coverage but costs short-lead skill. This version withdraws an earlier claim of an 81% CRPS reduction that came from an uncontrolled comparison.
[LG-4] Get Hint-Augmented GRPO Done Right Through Activation Steering
链接: https://arxiv.org/abs/2611.23124
作者: Khiem Le,Ting Hua,Nitesh V. Chawla
类目: Machine Learning (cs.LG)
*备注:
Abstract:GRPO has become the dominant reinforcement learning method for eliciting reasoning capability in LLMs using verifiable rewards. However, it suffers from a fundamental failure mode: when a training question exceeds the base model’s current capability, all generated responses fail and receive zero reward, yielding no gradient signal and stalling further capability expansion. An emerging class of GRPO extensions, termed Hint-Augmented GRPO (HA-GRPO), adds a “hint” to the context to reduce the difficulty of the question, thereby producing a usable gradient signal. Nevertheless, existing HA-GRPO methods share a key issue: no hint is available at inference time to append to the question, creating a training-inference context mismatch that leads to inconsistent performance gains. In this work, we reinterpret the function of hinting – not as reducing question difficulty, but as steering model behavior – and propose ASH (Activation Steering for Hinting), a novel HA-GRPO method built on this view. During training, the model is presented with both hinted and plain questions, while a trainable steering vector is optimized to widen the margins between the model’s generation probabilities for responses to the hinted version relative to the plain version, thereby capturing the question-agnostic component of the behavioral shift induced by the hint. During inference, the trained steering vector is directly injected into the model’s activation, reproducing the hint-conditioned behavior without requiring any hint in the context. Experiments with four LLMs (Qwen3-1.7B/4B-Base, Llama-3.2-1B/3B-Instruct) show that ASH consistently improves pass@ k on competition-level mathematics benchmarks (AMC, OlympiadBench, AIME24, AIME25) and out-of-distribution benchmarks (Minerva, GPQA-Diamond). Notably, it improves average pass@8 of Qwen3-1.7B-Base and Qwen3-4B-Base by \textbf2.60% and \textbf2.21%.
[LG-5] Auditing a Learned Simulator for Legged Robots: The Robotic World Model on ANYmal D
链接: https://arxiv.org/abs/2611.23118
作者: Joyjeet Singh
类目: Robotics (cs.RO); Machine Learning (cs.LG)
*备注: 39 pages, 13 figures, 20 tables. Code and data: this https URL and this https URL . Extends arXiv:2611.02699 with robot-level experiments
Abstract:A legged robot can learn to walk inside a learned simulator: a neural network that predicts the robot’s next state and is run forward on its own predictions. They are usually trained and checked one step ahead, although a policy uses them for a hundred steps or more. We audit one published system end to end, without a GPU: the dynamics model of the Robotic World Model (RWM) for the ANYmal D quadruped, and the uncertainty estimate of its follow-up, RWM-U. We rebuilt the model to match the released code’s outputs, losses and gradients exactly, measured 101 trained simulators, 200 compressed copies and 38 simulators of three other architectures, and trained policies inside simulators that we tested in a hand-built MuJoCo model of the robot. Training on the model’s own rollouts beat one-step training at 368 steps on all ten held-out episodes (registered), although one-step training was better one step ahead. One-step checks therefore misjudge design choices: pruning 30% of the weights kept 90% of the model’s skill one step ahead and 39% at 2 s. Ensemble disagreement ranked rollout error well (Spearman +0.90 at 100 steps), yet covered 4.6% of outcomes at one sigma, against 68% for a calibrated estimate, as each network’s own variance collapses under the implemented loss. Two checks that need no recorded future ranked some pools of simulators well, but neither was shown better than 8-step error in a primary test. At the robot level, accuracy has not predicted walking: all 18 fine-tuned copies we tested predicted better than the released model one and eight steps ahead, yet 13 gave failing policies, and registered tests of whether any check ranks simulators by their policies were not settled. In a secondary test, running the authors’ policy inside each simulator ranked them well. We also report nine of our own claims that we withdrew. Scope: one robot, one gait, simulated data, no hardware.
[LG-6] Pareto-Informed Smart “Predict then Optimize” for Multi-Objective Combinatorial Problems ECAI2025
链接: https://arxiv.org/abs/2611.23076
作者: Aabhash Dhakal,Tim Lachner
类目: Machine Learning (cs.LG)
*备注: 8 pages, 5 figures, 1 table. Accepted after peer review at the Multi-Objective Decision Making (MODeM) Workshop at ECAI 2025
Abstract:Many real-world planning applications, from logistics, routing to resource allocation, give rise to optimization problems that need to balance multiple, conflicting objectives. While machine learning is often used to predict the unknown parameters or costs of these problems, standard models trained on prediction accuracy can lead to suboptimal plans. Decision-Focused Learning (DFL) addresses this by integrating the downstream optimization task into the training objective. However, existing DFL methods for multi-objective problems rely on simple scalarization, which fails to capture the rich trade-offs of the Pareto Frontier. To address this gap, we propose a Pareto-aware extension to single-objective Smart “Predict, then Optimize” (SPO+) for multi-objective combinatorial optimization problems. Our loss function leverages the true Pareto Frontier, forming a target decision that aligns with a sampled preference vector, and minimizes regret against this target. Experiments on multi-objective shortest path and multi-objective knapsack problems show that our approach outperforms traditional two-stage Mean Squared Error (MSE) baselines. It also surpasses scalarization-based SPO+ approaches.
[LG-7] Magflow: Differentiable Simulation of Industrial-Grade 6D Magnetic Levitation
链接: https://arxiv.org/abs/2611.23050
作者: Philipp Hartmann,Jannick Stranghöner,Klaus Neumann
类目: Robotics (cs.RO); Machine Learning (cs.LG); Systems and Control (eess.SY)
*备注: 8 pages, 5 figures, 1 table. Accepted for publication in IEEE Robotics and Automation Letters (RA-L). Project page: this https URL
Abstract:Magnetic levitation with six-degrees-of-freedom enables frictionless transport and micrometer-scale positioning in principle, but low-level control is inherently challenging due to its complex and unstable dynamics. Industrial systems typically rely on deep hardware-software co-design, where mover and stator geometries are jointly designed to enable a compact, hand-crafted approximation of the electromagnetic force computation, thereby keeping control computationally tractable at kilohertz rates. However, this approach imposes overly conservative constraints on the operating workspace and the range of supported mover topologies. Instead of reducing the underlying physics to a hand-crafted approximation, we compute the physics directly. We introduce Magflow, a fully differentiable, GPU-parallel simulator for 6D magnetic levitation. It is used to train the first neural controller for an industrial-grade 6D magnetic levitation system that is learned purely from simulated physics. We show that the learned neural controller transfers zero-shot from simulation to real hardware while satisfying hard real-time constraints at 4 kHz. The learned controller achieves qualitatively new behaviors and even levitates a mover topology for which no hand-crafted solution exists. These results demonstrate that co-design is not a physical necessity for 6D magnetic levitation control. More broadly, control learning from simulated physics can find solutions beyond the reach of control engineering in demanding industrial domains. Source code and demonstration videos are publicly available at this https URL.
[LG-8] Variance Restoration over Covariance Rank: Memory-Budgeted Analytic Continual Learning
链接: https://arxiv.org/abs/2611.23040
作者: Elena Voss,Pixel Nomand,Marcus Hale,Sofia Reyes
类目: Machine Learning (cs.LG)
*备注:
Abstract:Analytic continual learning replaces iterative gradient updates with closed-form ridge regression over frozen pre-trained representations, summarizing all past sessions in two running sufficient statistics: the feature autocorrelation matrix \Sigma_t and the class cross-correlation matrix \Gamma_t . Lifting backbone features into a wide random code of dimension M \gg D substantially improves linear separability across tasks, but forces \Sigma_t to grow quadratically as \mathcalO(M^2) . Under a fixed persistent memory envelope \Bcal , a learner faces a fundamental structural trade-off: store exact pairwise co-activations inside a narrower feature code, or expand to a wide code while compressing cross-coordinate interactions. We observe that the coordinate self-variances v_t = \diag(\Sigma_t) and class statistics \Gamma_t grow only linearly with M , whereas off-diagonal interactions dominate memory. We introduce \method (\textbfCoordinate-\textbfRestored \textbfAnalytic \textbfContinual \textbfLearning), which retains (v_t, \Gamma_t) exactly, maintains off-diagonal co-activations through a streaming rank- q factor F_t , restores exact diagonal variances via \Sighat_t = F_t^\top F_t + \Diag(v_t - \diag(F_t^\top F_t)) , and solves the full M\times C normal equations via the Woodbury identity. Across five class-incremental benchmarks on frozen ViT-B/16, \method matches dense full-covariance readouts within 0.09 points while cutting persistent state by 90.7% ( 35.9 vs.\ 385.3 MiB). Under equal memory envelopes of 16 – 80 MiB, prioritizing expansion width M over covariance rank q outperforms exact narrow readouts in 17 of 18 configurations by up to +2.19 points.
[LG-9] NVIDIA Kumo Tabular: Pushing the Accuracyunicode8211Efficiency Frontier for Tabular Prediction
链接: https://arxiv.org/abs/2611.23001
作者: Jingang Qu,Valter Hudovernik,Martin Jurkovič,Cedric Lorenz,David Holzmüller,Federico Lopez,Dmitry Gordeev,Akihiro Nitta,Ramona Bendias,Ardrian Wong,Vedaant Jain,Vignesh Kothapalli,Gilberto Titericz Junior,Jure Leskovec,Matthias Fey
类目: Machine Learning (cs.LG)
*备注: 25 pages, 13 figures
Abstract:We introduce Kumo Tabular, a family of open tabular foundation models for classification and regression, in three sizes: Kumo Tabular-S (27M parameters), Kumo Tabular-M (61M), and Kumo Tabular-L (214M). Pretrained only on artificial tables from the Kumo prior, they can make predictions on new tables without any training. In our approach, a table encoder compresses each row into an embedding, and an in-context learning transformer predicts the unlabeled rows from the labeled ones. Kumo Tabular introduces architectural innovations that improve both training stability and predictive performance, enabling the model to scale up to 150M cells on modern GPUs. Kumo Tabular-L ranks first on eight benchmarks: TabArena, TALENT, BeyondArena, ScoringBench, STRABLE, MulTaBench, RamanBench, and TabBench-Bio. These cover probabilistic regression, tables with over 1M rows or 100K columns, and tables with strings, text, or images. Accuracy improves with model size in all tested comparisons. All three sizes lie on the accuracy-speed frontiers of TabArena and BeyondArena. Kumo Tabular-L predicts faster than the runners-up, LimiX-2 on TabArena (36 times) and TabPFN-3.5 on BeyondArena (1.7 times). Weights and pretraining code are released under OpenMDW-1.1 and Apache 2.0, respectively.
[LG-10] DDSL: Iterative Self-Labeling for Financial Market States
链接: https://arxiv.org/abs/2611.22958
作者: Xianbin Gu,Qingyue Zhu,Yuhong Geng
类目: Machine Learning (cs.LG)
*备注: 17 pages, 3 figures, 5 tables. Conference version published in ICAIF 2026. Code available at this https URL
Abstract:Financial market states lack a unique ground truth and must therefore be constructed from price data. Conventional labeling methods typically define states independently of the downstream classifier, leaving unresolved whether the resulting labels can be recovered from information available at prediction time. We propose DDSL, an iterative self-labeling framework that couples a price- and grammar-guided Viterbi labeler with a Transformer classifier through model feedback. DDSL constructs a nine-state representation that jointly encodes current and future market trends while retaining structured price emissions and temporal constraints throughout refinement. Experiments on SSE50, CSI300, CSI500, and NASDAQ100 show that DDSL increases held-out joint-state macro F1 by 78%-96% over single-pass dynamic-programming labeling. An independent GRU classifier that never participates in the labeling loop reproduces the same trend, while label entropy and class coverage remain stable, indicating that the gain transfers across architectures and is not caused by label collapse. We further show that learnability is controlled by both iteration and the model-feedback weight \beta_\mathrmmodel . Iterative gains diminish as successive label systems converge, whereas stronger feedback increasingly widens the performance gap between the feedback Transformer and the independent GRU. Increased learnability is accompanied by declining segment-level price-feature separation after an initial improvement. Finally, a non-anticipative event study provides partial external validation: refinement strengthens a conditional BEAR \rightarrow BULL reversal signal on the three A-share datasets, while the corresponding effect has the opposite sign on NASDAQ100. These results position DDSL as a structured framework for constructing market-state labels whose learnability can be explicitly controlled and evaluated.
[LG-11] ResoLUT: Distilling Residual Networks into Lookup-Table Networks
链接: https://arxiv.org/abs/2611.22955
作者: Katarzyna Fojcik,Aidas Mikalauskas,Matthias Kampa,Renaldas Zioma,Jogundas Armaitis
类目: Machine Learning (cs.LG)
*备注:
Abstract:Logic networks built from Boolean gates or lookup tables (LUTs) promise efficient hardware inference, but training them at depth and learning their connectivity remain difficult. We do not train logic as logic. Instead, ResoLUT distills a trained residual network (ResNet) layer by layer into LUT trees, preserving its multi-bit residual stream and BatchNorm while extracting connectivity from the teacher rather than learning it. To our knowledge, ResoLUT is the first convolutional logic network in computer vision to retain a ResNet-style multi-bit residual stream. Hard LUT decisions are used throughout training and inference, avoiding a discretization gap in the logic. ResoLUT-20 reaches 95.66% on CIFAR-10 with twice the trees, and ResoLUT-56 reaches 78.87% on CIFAR-100 and 53.69% on ImageNet at 64px, the best results of logic and lookup-table networks on CIFAR-10 and CIFAR-100, showing that LUT networks can scale in depth. We explore and quantify the role of the residuals and the tree budget and shape. The most optimized model, a ResoLUT-8 with one eighth of the trees and further optimizations, fits on an FPGA and reaches 90.04% at 189k images per second, the highest throughput among the compared models above 85% on CIFAR-10.
[LG-12] he Kernel Reality Check: Benchmarking and Distilling Efficient Attention at Scale NEURIPS2026
链接: https://arxiv.org/abs/2611.22947
作者: Fırat Öncel,Cem Subakan,Mirco Ravanelli,Çağatay Yıldız
类目: Machine Learning (cs.LG)
*备注: Accepted at NeurIPS 2026. 20 pages
Abstract:Kernelized attention methods, which aim to replace the O(N^2) complexity of softmax attention with O(N) alternatives, have been the focus of intense research in the last five years. Despite growing interest, systematic evaluations and fair comparisons across approaches and scales are still missing. We analyze kernelized attention in two halves and highlight prominent limitations on each. On efficiency, FlashAttention achieves lower memory consumption and faster wall-clock time than SOTA kernelized methods during training and in vision transformers across most sequence lengths (with inference-time memory in causal language models the only clear exception), contradicting the core assumption for pure kernelized attention used as a drop-in replacement for softmax. We also discover that reported efficiency gains often fail to reproduce against modern baselines. On quality, current kernelized methods leave gaps of 2-10 points relative to softmax under a unified distillation protocol, and model scala alone does not close these gaps. We show this gap is not intrinsic to kernelization: LARA++, a principled extension of LARA that approximates softmax via importance sampling with adaptive proposal distributions, recovers up to 99% of softmax accuracy. We establish that the gap is closable; however, the computational overhead remains substantial, leaving the efficiency wall intact. Code is available at this https URL
[LG-13] A Transformer-Derived Iterative Preconditioner NEURIPS2026
链接: https://arxiv.org/abs/2611.22941
作者: Patrick Lutz,Themistoklis Haris,Aditya Gangrade,Venkatesh Saligrama
类目: Machine Learning (cs.LG)
*备注: Accepted at NeurIPS 2026
Abstract:Transformers trained on in-context prediction problems can learn known numerical algorithms, e.g., gradient descent. We go further, showing that their internal computations can reveal new algorithms with provable guarantees. A softmax-attention transformer trained to solve linear systems Xw=y implicitly improves matrix conditioning. Tracing XX^\top reveals an iterative preconditioning primitive, which we develop into RAPID, a standalone algorithm requiring no learned components. RAPID is an anytime preconditioner for symmetric positive definite (SPD) matrices, using randomized mixing and sparse local corrections to improve conditioning. Proof-of-concept experiments on synthetic and real-world problems consistently improve conditioning and reduce conjugate gradient iterations needed to reach a target accuracy. For a Haar-mixing idealization, we prove convergence to any fixed target condition number regardless of spectral distribution, with only logarithmic dependence on initial conditioning. Together, these results show how a transformer can discover a broadly applicable numerical algorithm. Comments: Accepted at NeurIPS 2026 Subjects: Machine Learning (cs.LG) Cite as: arXiv:2611.22941 [cs.LG] (or arXiv:2611.22941v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2611.22941 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-14] Action-outcome coupling: a pre-investment feasibility audit for offline reinforcement learning in supply chain operations
链接: https://arxiv.org/abs/2611.22921
作者: Tung Dang, TheHung Phung,Son Lam Nguyen,Tu Nguyen
类目: Machine Learning (cs.LG); Computational Engineering, Finance, and Science (cs.CE); Econometrics (econ.EM); Applications (stat.AP)
*备注: 40 pages
Abstract:Prescriptive analytics initiatives in logistics, such as offline reinforcement learning and next-best-action systems, are often funded before anyone verifies that the available historical logs can support an improved decision policy. We formalise this pre-investment feasibility question via action-outcome coupling: a total-effect variance index measuring how much expected outcomes change with the logged action. Coupling provides an algorithm-independent upper bound on any policy’s achievable value, expressed in monetary units per decision for both single-step and discounted sequential objectives. We also analyse the asymptotic false-licensing risk of an invalid workflow: fitting a value model and evaluating the model’s greedy policy on the same data. The false-licence rate is driven mainly by the number of state cells over which selection is performed, rising from the nominal test level with one cell to near certainty with a few dozen, while overlap has a smaller effect. We then propose an audit that assigns logs to four tiers of decision support, separately for fixed and contextual policies, and returns an explicit stopping rationale. On a randomised e-commerce log, the audit endorses a fixed recommendation policy, certifying gains equal to 30 per cent of the theoretical ceiling; the remaining gap reflects bound looseness under an 80-action design. Two public logistics logs fail at different checkpoints: one because the target is an accounting identity of the action, and another because, with 27 clusters, estimated coupling stays below the design’s detectability threshold even though a conservative offline learner certifies an improvement in 29 of 30 seeds.
[LG-15] Who Pays for Byzantine Robustness? Validity Data-Share Dictatorships and Fairness When Only a Few Hospitals Collaborate
链接: https://arxiv.org/abs/2611.22904
作者: Adnan Malkani
类目: Machine Learning (cs.LG); Cryptography and Security (cs.CR)
*备注: 12 pages, 7 figures, 6 tables. Code: this https URL
Abstract:Clinical federated learning (FL) collaborations typically connect only three to six hospitals, yet Byzantine-robust aggregation rules are designed and benchmarked on synthetic partitions across dozens of equal-sized clients. We conduct an empirical audit of eight standard aggregation rules: FedAvg, coordinate-wise median, trimmed mean, Krum, Multi-Krum, Bulyan, geometric median (GeoMed), and Centered Clipping (CC). We evaluate both unweighted and sample-size-weighted variants across three cross-silo medical datasets from the FLamby benchmark: Fed-Heart-Disease ( n = 4 hospitals), Fed-TCGA-BRCA ( n = 6 regions), and Fed-ISIC2019 ( n = 6 centres). We execute 54,214 training runs across enumerated attacker coalitions ( f \in \1, 2\ ) paired with exact-twin leave-out controls that decompose performance drops into honest data loss versus adversarial attack damage. We report three findings. First, client-count validity checks fail to capture breakdown in skewed medical consortia: while n = 6, f = 1 satisfies every count condition except Bulyan’s, assigning sample-size weights turns median and GeoMed into a clean-model dictatorship whenever a single hospital holds more than half the training data, and causes weighted median, GeoMed, and CC to collapse when that dominant centre attacks. Second, exact-twin leave-out controls reveal that when the largest hospital attacks, the apparent failure of count-based robust rules is predominantly caused by the loss of honest training data (7.5 to 10.4 points on ISIC) rather than adversarial perturbation (2.3 to 3.8 points on ISIC). Conversely, weight-dependent rules suffer large true attack damage (16.1 to 25.1 points on ISIC). Third, unweighted robust rules consistently discount whichever institution generates the most atypical parameter updates, penalizing the smallest hospital on Heart but penalizing larger institutions on TCGA and ISIC. Comments: 12 pages, 7 figures, 6 tables. Code: this https URL Subjects: Machine Learning (cs.LG); Cryptography and Security (cs.CR) Cite as: arXiv:2611.22904 [cs.LG] (or arXiv:2611.22904v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2611.22904 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-16] Beyond Attribute-Centric Fairness: Modelling and Mitigating Unified Bias Representation in Omni-modal Large Language Models
链接: https://arxiv.org/abs/2611.22889
作者: Runqi Lin,Suqin Yuan,Zihao Fu,Chris Russell
类目: Machine Learning (cs.LG)
*备注:
Abstract:Omni-modal large language models (OLLMs) provide joint multimodal understanding and generation capabilities, making them well-suited to complex environments and promising backbones for next-generation AI. Despite these advances, the harmful stereotypes encoded in such models manifest in increasingly diverse forms, leading to broader and more severe societal consequences. Existing debiasing methods typically impose separate fairness objectives for each sensitive attribute and modality to minimise demographic-dependent variations. However, in omni-modal settings, this attribute-centric paradigm struggles to jointly address diverse social biases with distinct fairness objectives and to generalise beyond predefined attributes. To this end, we propose OmniDebias, one of the first methods for promoting fairness in OLLMs, which employs a bias-centric formulation to extract and mitigate a shared bias representation across attributes and modalities while preserving attribute-related knowledge. Specifically, we exploit activation inconsistencies of a given attribute (e.g., gender) across semantic contexts (e.g., occupations) to disentangle bias from legitimate knowledge. We further consolidate multiple extracted biases into a shared representation, formulating a unified objective that captures the underlying structure across social biases and generalises to unseen attributes. Extensive experiments demonstrate that eliminating unified bias with both post-hoc editing and fine-tuning can consistently improve fairness across diverse attributes and modalities in OLLMs while maintaining favourable utility.
[LG-17] Path-Aligned Constraint Equivariance
链接: https://arxiv.org/abs/2611.22873
作者: Yuhao Wang,Yuanxi Peng
类目: Machine Learning (cs.LG)
*备注:
Abstract:Learned super-resolution of physical fields must satisfy structural constraints—such as divergence-free flow ( \nabla\cdot u=0 ), curl-free fields ( \nabla\times u=0 ), and pointwise sum-to-one normalization ( 1^\top p=1 )—that standard training objectives do not enforce. Soft penalties trade fidelity against satisfaction via a scalar hyperparameter \lambda but cannot drive the residual to zero; hard-constrained regression eliminates the residual by construction yet converges to the conditional mean, losing small-scale spectral structure. We observe that in Conditional Flow Matching with linear paths, a linear constraint Ax=c is preserved along the entire transport trajectory whenever the velocity field is compatible ( Av_\theta=0 ) and the initial noise is consistent ( Ax_0=c ); along the trajectory, the latter two conditions are also necessary. We term this the path-equivariance principle and instantiate it in PACE (Path-Aligned Constraint Equivariance), a framework that enforces these conditions through a null-space parameterization of the velocity field and a constraint-respecting noise sampler, with no modification to the loss function or the ODE integrator. Across three constraint families, PACE reduces the mean constraint residual by one to five orders of magnitude relative to unconstrained baselines, matching the residual level of the data-generation pipeline itself. The proposed framework requires only a closed-form parameterization onto the null space of A ; within this scope, it offers exact constraint enforcement while recovering the spectral content and predictive diversity that conditional-mean regression systematically loses, all under a single unified recipe.
[LG-18] SilverSpoon-KD: A General-Purpose Toolkit for Knowledge Distillation
链接: https://arxiv.org/abs/2611.22842
作者: Xaver R. Davey,David Broman
类目: Machine Learning (cs.LG)
*备注: 28 pages, 10 figures, 11 tables. Code: this https URL
Abstract:Knowledge distillation transfers deep learning representations from a teacher model to a student model, typically reducing model size or inference cost. However, knowledge distillation tooling has not kept pace with the breadth of research methods: existing toolkits typically support only a limited subset of techniques and use cases. We present SilverSpoon-KD, an open-source PyTorch toolkit that unifies different distillation paradigms and architectural relationships across workloads and applications behind a single API. Moreover, SilverSpoon-KD offers the first general-purpose blockwise knowledge distillation implementation with unique scalability features. We validate its versatility across vision, masked language model, and causal language models on tasks spanning homogeneous, heterogeneous, and quantization distillation across paradigms. We also evaluate SilverSpoon-KD by comparing its performance, including throughput and GPU memory usage, with existing toolkits. Despite extensive support for convenience, metric calculation, and tracking, we have found SilverSpoon-KD’s performance to be on par with or better than existing toolkits.
[LG-19] abDAG: Structured Evolutionary Search for Tabular Feature Engineering
链接: https://arxiv.org/abs/2611.22828
作者: Anastasia Volkova(1),Andrey Kuznetsov(1),Ivan Oseledets(2),Valentin Khrulkov(1 and 2) ((1) FusionBrain Lab, (2) AXXX)
类目: Neural and Evolutionary Computing (cs.NE); Machine Learning (cs.LG)
*备注: Preprint under review
Abstract:Feature engineering can improve tabular prediction, but useful transformations are compositional, data-dependent, and subject to leakage constraints. We present TabDAG, a framework that combines evolutionary search with a graph representation of feature-engineering programs. Nodes implement row-wise or data-fitted transformations; edges encode dependencies among raw variables, intermediate values, and final features. The representation supports interactions, statistical encodings, clustering- and neighborhood-based features, and multi-stage compositions in a single executable program. Mutations are structured edits constrained by a parent-specific schema and checked before evaluation. In our experiments, no structurally invalid program reached evaluation, although proposals could be rejected and valid graphs could still fail at runtime. Across ten regression and classification datasets and three tabular model families, evolved features improve predictive performance over raw inputs. TabDAG and its combination with OpenFE achieve the two best mean ranks among the compared methods; pairwise rank differences between feature-engineering methods do not survive multiple-testing correction. The LLM-based searches use the same backbone and candidate-evaluation budget. A controlled ablation compares free-form Python, whole-graph JSON, and constrained graph edits with the search engine, budget, and backbone held fixed. Constrained edits yield fewer failed evaluations than either rewrite baseline. The resulting graphs expose the transformations and dependencies selected by search, enabling feature-level analysis of the observed gains.
[LG-20] LoopPrune: Recurrent Pruning for Looped Diffusion Transformers
链接: https://arxiv.org/abs/2611.22757
作者: Weizhong Huang,Huaijia Kou,Yan Zhang,Fei Chao,Xiawu Zheng,Liujuan Cao
类目: Machine Learning (cs.LG)
*备注: 10 pages, 11 figures
Abstract:Looped diffusion Transformers reuse shared blocks to deepen computation and match larger models with fewer parameters, reducing storage and deployment burdens. Yet repeated attention and MLP computations remain costly. Pruning reduces storage and computation by removing redundant parameters or units. Training-free methods struggle to resolve recurrent contributions, while training-based methods lack explicit constraints on position-specific errors affecting later inputs and final outputs. We propose LoopPrune, a pruning and recovery framework addressing these requirements. Recurrent importance estimation combines shared-removal effects and position variation to select a shared structure; shared reconstruction pools recurrent input statistics to compensate weights. Loop-position recovery compares complete denoising predictions at identical student states and matches text guidance, updating all student parameters against a fixed dense teacher. Extensive experiments across four open-source models, six benchmarks, and three pruning paradigms demonstrate effectiveness. At 30% structured parameter pruning, Looped-DiT B/16 retains 93.14% of dense six-benchmark average accuracy, exceeding the best baseline by 9.49%. Across PyTorch, vLLM-Omni, and SGLang Diffusion, pruned SDXSv3 reduces peak allocated GPU memory by 34.36-44.64% and accelerates denoising 1.46-1.74x over dense models.
[LG-21] Stochastic Diversity Windows: Diagnosing Noise Influence in Robotic Diffusion Sampling
链接: https://arxiv.org/abs/2611.22743
作者: Tina Lakhani,Jing Wu,Yu-Kun Lai,Ze Ji
类目: Robotics (cs.RO); Machine Learning (cs.LG)
*备注: 26 pages, 3 figures, 6 tables
Abstract:Diffusion models are increasingly adopted in robotics and control due to their capability to express multiple feasible solutions to a given task. However, the role of stochasticity in the diffusion model-sampler across denoising remains poorly understood. We introduce a per-timestep counterfactual diagnostic that tests how changing the sampled noise affects the final output. Across five model-sampler configurations spanning three robotics benchmarks, diverse architectures, domains, and conditioning mechanisms, we reveal highly varied temporal sensitivity profiles, where noise-induced deviations can amplify, partially contract, or undergo near-total attenuation downstream. From these profiles, we define stochastic diversity windows (SDWs): compact intervals that identify where sampler stochasticity has the strongest measured influence on the final output. Guided by these windows, we apply two heuristic interventions: targeted stochastic amplification, which increases sample diversity exploration of the solution space, and an SDW-hybrid sampler, which replaces DDPM steps with DDIM skips in low-impact regions, resulting in network evaluation reductions of up to 40-60% and wall-clock time by up to 85%, with no consistent changes detected in feasibility, cost, or diversity across held-out tasks.
[LG-22] HADVQ: Hessian Aware Directional Vector Quantization
链接: https://arxiv.org/abs/2611.22721
作者: Daniil Li,Mars Davletshin,Vadim Klyzhenko,Artem Danilishin,Valery Sandul,Daniil Ryazanov,Ekaterina Muravleva
类目: Machine Learning (cs.LG)
*备注:
Abstract:Ultra-low-bit quantization of large language models is a challenging task because aggressive compression severely distorts model parameters and disproportionately harms a small number of highly sensitive channels. We present a vector quantization method tailored for this regime. Our approach begins with random preconditioning, which regularizes the weight space and makes it more amenable to quantization. We then represent weights in polar coordinates and decouple the quantization of directions and magnitudes. To quantize directions, we leverage the structure induced by random preconditioning and design a structured codebook based on the E_8 lattice, yielding a compact yet expressive representation. To further reduce degradation, we detect extremely important channels using a proxy Hessian and preserve them with an additional cost of only 0.16 bits. Together, these components provide an effective quantization pipeline for LLMs in the extremely low-bit regime. Empirical results demonstrate improved accuracy, highlighting the benefit of combining geometric structure, lattice-based codebooks, and selective channel preservation.
[LG-23] he Price of Hard Coverag e Constraints in Online Conformal Prediction
链接: https://arxiv.org/abs/2611.22646
作者: Mingrui Liu
类目: Machine Learning (cs.LG); Statistics Theory (math.ST); Machine Learning (stat.ML)
*备注:
Abstract:Conformal prediction provides distribution-free uncertainty quantification, but in sequential deployment, standard coverage guarantees can still allow substantial fluctuations in the number of miscoverages over time. We study an online setting in which prediction sets are produced one round at a time and the cumulative number of miscoverages must stay within a prescribed budget throughout this http URL a hard coverage guarantee can be enforced by enlarging prediction sets, but how much additional prediction-set size is fundamentally necessary remains unclear. We characterize this unavoidable efficiency cost over T prediction rounds. Remarkably, the cost persists even with i.i.d.\ data, a known score distribution, and perfect knowledge of the population-optimal prediction set. Under suitable regularity conditions relating coverage and prediction-set size, we determine the minimum cumulative excess prediction-set size required by any online method satisfying the hard coverage constraint. This minimum cost is \Theta(\log T) for expected cost at a fixed time known in advance, \Theta(T^1/3) for the largest expected cost over all fixed times up to T , and \Theta(\sqrtT) for the expected worst realized cost when the evaluation time is chosen in hindsight. We further show that these rates can be attained using past scores, without knowing the score distribution. Thus, even perfect statistical knowledge does not make sequential reliability free.
[LG-24] ask Convergent Flow Matching for Generative Robotics
链接: https://arxiv.org/abs/2611.22609
作者: Lorenzo Paiola,Maximilian Stölzle,Giorgio Grioli,Daniela Rus,Cosimo Della Santina
类目: Robotics (cs.RO); Machine Learning (cs.LG)
*备注: Submitted to L4DC 2027
Abstract:Successful manipulation requires the manipulated objects to progress toward a desired task outcome. Generative policies can produce plausible robot motions without ensuring this progress, allowing successive action chunks to stall or undo earlier changes to the scene. We introduce a task-centric inference-time refinement framework for flow matching policies that evaluates candidate action chunks through parallel physics-based rollouts and searches for motions satisfying sufficient conditions for task convergence. Task-level performance funnels bound temporary regressions while requiring net progress over each chunk. We establish sufficient conditions under which repeated funnel satisfaction yields convergence throughout execution, and use interval-wise Lipschitz bounds to verify continuous-time funnel satisfaction from sampled rollouts. To accommodate contact-rich hybrid dynamics, we refine the flow’s source seeds using derivative-free Covariance Matrix Adaptation Evolution Strategy (CMA-ES), constrained to radial quantile bands of the source distribution to help retain the learned motion prior. This formulation supports both Euclidean and pose-valued Riemannian Flow Matching (RMFM) policies. On PushT and three-cube stacking, refinement raises success from 31.3% to 88.5% and from 77.0% to 86.0%. On PushT, whose coverage score responds continuously to the motion of the block, a learned funnel lowers the average maximum loss of achieved progress by 19% relative to constraining the endpoint of each chunk alone, at a similar success rate. With the radial band, 11.3% of the executed motion windows are out of distribution, a rate within the range of held-out demonstrations, compared with 34.1% for an unconstrained search.
[LG-25] DLS: Dual-Logic Graph Neural Networks with Chaos-Fuzzy Residuals for Robust Link Prediction
链接: https://arxiv.org/abs/2611.22607
作者: Mohammad Mehdi Keikha,Abolfazl Nadi
类目: Machine Learning (cs.LG); Social and Information Networks (cs.SI)
*备注: Submitted to World Wide Web (Springer Nature) on January 5, 2026. Preprint version, currently under peer review
Abstract:Link prediction is a fundamental task in graph mining, critical for applications ranging from biological interaction discovery to recommender systems. While Message Passing Neural Networks (MPNNs) have achieved state-of-the-art results, they are fundamentally limited by the inadvertent suppression of high-frequency signals, as the recursive aggregation that stabilizes structural learning inherently leads to over-smoothing, rendering node representations indistinguishable in deep layers. This limitation is particularly severe in heterophilous graphs, where structural proximity does not imply feature similarity. In this paper, we propose DLS (Dual-Logic SAGE), a novel inductive framework designed to break the over-smoothing barrier via a “controlled instability” mechanism. Unlike conventional methods relying on static structural heuristics, DLS introduces a dynamic residual pipeline. First, a Chaotic Injection Module utilizes the Logistic Map to diversify the feature manifold, exploiting sensitivity to initial conditions to separate topologically similar but semantically distinct nodes. Second, to mitigate the aleatoric uncertainty introduced by chaotic dynamics, a Fuzzy-Logic Gating Mechanism applies soft-thresholding to filter noise and retain only high-confidence signals. Experimental results on diverse benchmarks demonstrate that DLS effectively balances feature discrimination with robustness, significantly outperforming state-of-the-art baselines in heterophilous settings while maintaining competitive efficiency on standard homophilous graphs.
[LG-26] Lost in the Funnel: Selection-Induced Miscalibration in Ranking and Decision Systems
链接: https://arxiv.org/abs/2611.22595
作者: Aditya Somani,Rahul Ragesh,Vikas Joshi,Amit Singh
类目: Machine Learning (cs.LG)
*备注: Under Review
Abstract:Modern ranking and decision systems combine components with divergent objectives, and scores calibrated at the model boundary are used as calibrated probabilities throughout the funnel, across filtering, reranking, business-rule enforcement, and auctions. Yet calibration is typically validated only where the score is produced, so its breakdown where the score is used is routinely overlooked. We ask when calibration survives this selection, how inherited error differs from selection effects, and what can fail for a perfectly calibrated score at a component boundary. We study these questions through a boundary-wise diagnostic framework of residual identities. We target rubric-judged relevance rather than presentation-dependent metrics such as CTR, so calibration is measurable at every funnel layer of interest. We first present an industry-wide case study that applies our diagnosis to settings where calibration deteriorates at decision time, tracing the degradation to selection rather than model error. This diagnosis motivates Funnel-Aware Recalibration (FAR): post-hoc, selection-policy-aware recalibration at the boundary where scores drive decisions. In a live two-region production A/B test, FAR cuts decision-time expected calibration error (ECE) by 60-80% and improves auction efficiency. We further adapt two public datasets to simulate industry settings, providing additional evidence for the mechanism. We plan to release the adapted datasets and their generation pipeline to support further research. Finally, we distill these findings into practical operating principles for calibrating multi-stage scoring systems.
[LG-27] ReMaRL: Learning to Unmask and Remask in Frozen Diffusion Language Models
链接: https://arxiv.org/abs/2611.22576
作者: Egor Shibaev,Danila Biktimirov,Ivan Klimov,Artem Makoian,Viacheslav Meshchaninov,Dmitry Vetrov
类目: Machine Learning (cs.LG)
*备注:
Abstract:Masked diffusion language models generate tokens in parallel, and the order in which tokens are unmasked affects final-answer accuracy. Several methods learn unmasking policies from task rewards, but these policies cannot remask earlier predictions. Remasking lets the model correct those predictions using newly generated context. Heuristic rules guide both unmasking and remasking, while supervised remaskers learn to detect token errors. Neither approach directly optimizes these decisions for final-answer correctness. To address this gap, we introduce ReMaRL (Remasking with Reinforcement Learning), a lightweight policy that learns which tokens to unmask, retain, or remask. We train this roughly 0.3M-parameter policy with Group Relative Policy Optimization (GRPO) from final-answer rewards while keeping the diffusion model frozen. With LLaDA-8B-Instruct, ReMaRL achieves the highest scores among the compared unmasking and remasking policies within the tested inference budgets of 4-32 model calls on math and 8-16 calls on code.
[LG-28] Physically Admissible Generative Flow Matching for Sparse Flow Reconstruction
链接: https://arxiv.org/abs/2611.22549
作者: Yuhao Wang,Yuanxi Peng
类目: Machine Learning (cs.LG)
*备注:
Abstract:Reconstructing full fluid velocity fields from sparse, localized observations is a severely ill-posed inverse problem: most of the spatial domain remains unobserved, discrete conservation laws must hold strictly, and predictive uncertainty must be reliable. Existing generative models typically enforce physical constraints only at the terminal output or via soft penalties, leaving intermediate generative trajectories unconstrained. We introduce Feasible Subspace Consistency (FSC), a framework that preserves physical admissibility as an invariant of continuous generative dynamics: tangent-space matching confines the entire ODE trajectory to the discrete conservation manifold, while mass-orthonormal coordinates preserve the kinetic-energy metric. Sparse reconstruction then reduces to amortized posterior transport in the latent space of a frozen FSC prior, ensuring every posterior sample inherits incompressibility and boundary conditions without per-sample projection loops. We validate FSC on a periodic Kolmogorov flow, where exact spectral projection isolates path tangency and metric isometry, and a wall-bounded cylinder wake with immersed obstacles, realized via matrix-free iterative projection. FSC prevents the mode-coverage collapse induced by terminal projection (increasing mode coverage from 0.000 to 0.790 ) while eliminating high-wavenumber spectral distortion. In sparse reconstruction, amortized FSC reduces global relative error by 50% and 46% , and sensor residuals by 63% and 55% , relative to the unconditioned prior, while delivering calibrated, spatially heterogeneous uncertainty.
[LG-29] Escaping Latent Traps by Remembering: Training-Free Episodic Kernel Repulsion for World-Model Planning
链接: https://arxiv.org/abs/2611.22470
作者: Yican Zhao,Bolin Yang,Ruizhe Wang,Haiyang Guo,Fuji Yang,Bingo Wing-Kuen Ling,Zelin Zang
类目: Machine Learning (cs.LG)
*备注: 17 pages, including references and appendix
Abstract:Latent world models plan towards a goal image by minimising the distance between predicted and goal latents. This distance is not geodesic: it has local minima, latent traps, behind walls, at corners and in dead ends, and memoryless receding-horizon replanning keeps the agent in them, revisiting states far from the goal. We propose Episodic Kernel Repulsion (EKR), a training-free cost term that penalises candidate plans by a top- k Gaussian kernel to a novelty-gated memory of the latents the agent has actually visited, rescaled by the current base cost so that its strength is comparable across models. Across two JEPA world models and four benchmarks, EKR raises average success from 31.8% to 46.7% (LeWM) and from 54.7% to 64.0% (H-JEPA) with the same model, planner and budget. Same-memory counterfactuals show that, in most settings, EKR pushes plans away from visited states, in navigation, often at a higher base cost; stuck episodes and revisits drop accordingly. In seven of eight settings it matches or exceeds 4\times samples at 0.8 – 1.5\times compute. Without tuning, it transfers to DINO-WM with CEM: +6.7 and +19.3 points on official and longer-goal Wall tasks, and from 44.7% to 100.0% and from 21.0% to 32.0% on long-horizon PointMaze and Push-T.
[LG-30] How Many Training Labels Are Wrong? Auditing a Fitted Model through Gradient Allocation
链接: https://arxiv.org/abs/2611.22449
作者: Yuchen Li,Zongqi Fan,Xintong Li,Nguyen H. Tran,Ken-Tye Yong
类目: Machine Learning (cs.LG)
*备注: 14 pages, 8 tables, 1 figure. Preliminary version
Abstract:How many labels in a training set are wrong, and which ones? The usual answers read a held-out set or the model’s outputs. A held-out reading depends on how the held-out set was drawn, and an output-based reading fails once the model has fitted its wrong labels, because the loss of a fitted wrong label is small and the prediction agrees with it. We read the answer from the per-example gradients of a single trained model. Each example, together with its label, asks the network to change in a particular way. We normalise the magnitudes of its gradient into a gradient allocation and compare it with the mean allocation of every class. An example is flagged when relabelling it as its nearest other class makes its allocation fit better than under its own label, and the share of flagged examples is the reading of the noise rate. The rule has no threshold and needs no clean data, no held-out set and no further training. After the model has fitted 99-100% of its wrong labels, the allocation still separates wrong from right labels with an AUC of 0.94-0.96 on the noisiest human labels of CIFAR-10N, where a feature-space margin falls to 0.79-0.84 and the loss, the gradient norm and the self-influence of an example to 0.73-0.76. On such a fitted model, readings of the noise rate from predictions or feature margins fall to 0-6% at true rates of 8-58% on CIFAR-10 with ViT-S, while the audit reads 54-96% of the true rate, and at least 93% of the examples it counts carry wrong labels. On a briefly trained model, before the wrong labels are fitted, the reading lies within 1.7 percentage points of the true noise rate on CIFAR-10 from 0% to 58% noise. Confident learning, which trains four further models, separates wrong labels about as well (AUC 0.925-0.998 against 0.900-0.995) and mostly reads the noise rate more closely, at a lower precision. An audit of 5,000 examples takes 14 to 27 seconds on one GPU.
[LG-31] InfuseLoop: Cross-Trajectory Latent Injection for Scalable Parallel Looped Transformers
链接: https://arxiv.org/abs/2611.22426
作者: Zhaofeng Sun,Lanxiang Hu,Yu-Yang Qian,Jiaqi Leng,Yuanbo Yang,Chi Zhang,Youyou Lu,Peng Zhao,Hao Zhang
类目: Machine Learning (cs.LG)
*备注:
Abstract:Parallel Looped Transformers (PLTs) use otherwise idle decoding compute by overlapping recurrent iterations. However, vanilla PLTs scale poorly with depth: they exhibit training-loss spikes, and their quality gains shrink as more loops are added. The benefit of additional loops depends on the information available to each iteration, so we examine whether the existing computational dependencies limit effective depth scaling. We find that only the preceding in-trajectory state is passed across loop boundaries, even though completed cross-trajectory representations are already available. Based on this finding, we propose InfuseLoop, a cross-trajectory latent injection mechanism. It aggregates causally available hidden states from later prediction trajectories through lightweight attention residuals. This design preserves token-parallel training and concurrent loop execution, and requires only O(Rd) additional hidden-state storage. Experiments show that InfuseLoop eliminates training-loss spikes and improves validation-loss scaling up to eight loops. At eight loops, InfuseLoop reduces perplexity by 3.4–17.1% relative to vanilla PLTs and by 10.5–37.3% relative to the non-looped baseline. It also improves average downstream accuracy by 0.84 and 1.99 percentage points over these two baselines. InfuseLoop incurs only 0.3% decoding time over vanilla PLTs and 23.3% over the non-looped baseline, while keeping the same logical KV-cache footprint as vanilla PLTs.
[LG-32] From Constraints to Causality: Petri net Encodings for Explaining Binary Neural Network Inference
链接: https://arxiv.org/abs/2611.22396
作者: Alex Chan(Newcastle University),Mohamed Tarraf(Newcastle University),Alex Yakovlev(Newcastle University),Rishad Shafik(Newcastle University)
类目: Logic in Computer Science (cs.LO); Machine Learning (cs.LG)
*备注: In Proceedings FMAS 2026, arXiv:2611.21375
Abstract:Binary Neural Networks (BNNs) are attractive for resource-constrained hardware and on-device components in autonomous systems. However, their discrete, non-linear behaviour makes them difficult to analyse and trust in safety-relevant settings. Existing approaches based on SAT and Satisfiability Modulo Theories can verify a BNN’s input-output properties, but treat it as a monolithic constraint system, which reveal only whether a property holds and not how the computation reaches its result. In this paper, we present an early work that provides a complementary, discrete-event view. We give a compositional translation from the inference semantics of a fixed, trained BNN to a 1-safe Petri net (PN), where each computation step is an explicit transition, so that a firing sequence is a causal trace of the BNN’s decision. On a worked example, we establish faithfulness via exhaustive input-output equivalence against a reference implementation, and show how contrasting firing sequences localises where a single-feature change alters the prediction. Finally, we outline two directions this opens: a correctness-of-translation theorem, and a causal-trace analysis of where two trained BNNs diverge in their decisions, a structural question that constraint-based verifiers cannot express.
[LG-33] Fair Bubble Sort: Provably Optimal Fair Ranking with Continuous Sensitive Attributes NEURIPS2026
链接: https://arxiv.org/abs/2611.22352
作者: Antonio Ferrara,Fabio Vitale,Atsushi Miyauchi,Francesco Bonchi
类目: Machine Learning (cs.LG); Computers and Society (cs.CY); Data Structures and Algorithms (cs.DS)
*备注: 35 pages, 9 figures, 4 tables. Accepted at NeurIPS 2026. Code: this https URL
Abstract:Most fair-ranking methods assume categorical protected attributes, requiring continuous attributes such as age or income to be discretized into groups. This introduces arbitrary thresholds, loses fine-grained information, and ignores the natural ordering of the attribute. In this paper, we study the problem of fair ranking with continuous protected attributes without relying on thresholding. We introduce Continuous Pairwise Parity (CPP), a fairness notion based on Kendall correlation between ranking positions and the protected attribute, and seek rankings that satisfy a prescribed CPP constraint while minimizing their Kemeny distance from an initial score-based ranking. To solve this problem, we propose Fair Bubble Sort (FBS), an efficient post-processing algorithm based on targeted adjacent swaps. We prove that FBS is exactly optimal for the unweighted Kemeny distance and provide approximation guarantees for the score-weighted setting. FBS is output-sensitive, with running time determined by the number of swaps required to meet the fairness constraint. Experiments on synthetic and real-world datasets show that FBS is scalable and consistently outperforms discretization-based baselines, achieving a superior fairness-accuracy Pareto front.
[LG-34] BQ-LoRA: Binary-Quantized Low-Rank Adapters as Implicit Regularizers for Parameter-Efficient Fine-Tuning NEURIPS2026
链接: https://arxiv.org/abs/2611.22334
作者: Juyoung Park,Heejun Lee
类目: Machine Learning (cs.LG)
*备注: Accepted at NeurIPS 2026. 20 pages, 4 figures, 21 tables
Abstract:Parameter-efficient fine-tuning of quantized large language models is dominated by QLoRA, which combines 4-bit weight quantization with FP16 low-rank adapters. We ask whether the FP16 adapter precision is necessary. Replacing it with a single binary bit per adapter parameter, combined with a learned per-layer scale s and a Scaled Straight-Through Estimator, reduces adapter storage by 13-16x. We introduce two variants – NormScale, which binarizes under a norm-preserving scale with a clipped STE, and LearnThresh, which makes the binarization threshold learnable – unified under a framework we call BQ-LoRA. Our main theoretical contribution is a tight characterization of the BQ-LoRA hypothesis class: it has bounded entrywise norm, a typical-set spectral norm bound that is \Theta® times sharper than the worst case, a full-rank, near-isotropic update, and consequently a Rademacher complexity that yields an explicit \tildeO(s_\max\alpha\sqrtd_out/(rn)) generalization bound. FP16 LoRA has no comparable a priori norm or isotropy control. On Qwen2.5 (1.5B-72B), BQ-LoRA exceeds QLoRA’s benchmark averages at every scale by 0.8-3.5 points; on LLaMA-2 (7B/70B) it remains competitive at the same storage reduction but does not uniformly lead the strongest FP16-adapter baselines (LoftQ at 7B, QLoRA at 70B). On Qwen2.5-7B the binary adapter preserves the 4-bit base WikiText-2 PPL within 0.06 (9.86 vs. 9.80) while QLoRA worsens it by 0.59. The reduction applies to inference-time adapter storage and serving; training memory matches QLoRA and step time rises by under 4%. We translate the storage saving into a deployment study, BQ-SGMV, by extending Punica’s SGMV kernel with a binary inner-product primitive: the same A100-80GB holds 15.6x more adapters in HBM, and our research prototype eliminates cache misses under a skewed synthetic workload.
[LG-35] PVTSizing 2.0: A Fast and Reliable Analog Circuit Sizing Operator for Agent ic EDA Workflows
链接: https://arxiv.org/abs/2611.22328
作者: Zichen Kong,Minghuan Song,Xiyuan Tang,Ziming Wei,Yi Zhong,Yuan Wang
类目: Hardware Architecture (cs.AR); Machine Learning (cs.LG)
*备注: 10 pages, 4 figures, 5 tables
Abstract:Analog circuit design automation requires both coordinated tool execution and efficient numerical search. Large language models (LLMs) offer new ways to organize tool execution, but finding one sizing that satisfies specifications across process, voltage, and temperature (PVT) corners remains costly. PVTSizing 2.0 addresses these needs through a human-agent workflow built around an independently usable numerical optimizer. An agent carries the designer’s approved task through execution and result review. The optimizer reduces unnecessary training through loss-guided updates and uses critic-guided local refinement to explore additional candidates while accounting for cross-corner conflicts. An optional LLM adviser supplies parameter relationships to focus numerical search on circuit-informed directions, while the optimizer selects adjustment amplitudes and verifies their effects through simulation. Across four circuit benchmarks, the standalone optimizer completes all 40 runs and achieves a 7.3x speedup in overall mean runtime over the original PVTSizing, with a comparable mean sample count. In a separate study, the best tested knowledge-injection configuration further reduces mean runtime by 39.9% and simulation samples by 35.4% relative to the standalone optimizer.
[LG-36] Similar Is Not Useful: Selecting In-Context Demonstrations by Closed-Loop Utility
链接: https://arxiv.org/abs/2611.22295
作者: Jiachen Li,Zhigen Zhao,Shihao Li,Dongmei Chen,Long Chen,Zhuo Xu
类目: Machine Learning (cs.LG)
*备注:
Abstract:Retrieval-augmented vision-language-action (VLA) policies fill their context with the demonstrations whose frames look most like the current observation under frozen DINOv2 embeddings. We ask whether visual similarity selects the demonstrations that make the policy succeed. For a frozen retrieval-augmented VLA (RICL on \pi_0 -FAST, LIBERO), we make the demonstration pool swappable at run time and score each demonstration’s utility by regressing the closed-loop success of 60 fixed-size pools on their membership. On initial states disjoint from those used for scoring, pools of the five highest-utility demonstrations succeed on 66.4% of 760 paired episodes across 19 tasks, pools of the five most similar on 55.7%, and random pools on 55.5%. Utility beats similarity by 10.8 points ( p=3.2\times10^-6 ; positive on 15 of 18 untied tasks), including on 12 tasks the policy was never primed on, and similarity-selected pools do no better than random ( +0.1 points). Within a task, a pool’s visual similarity to the test scenes tracks its success at Spearman 0.10; across tasks, similarity separates tasks whose scenes differ and falls to chance when two tasks share a scene. The gain comes from on-task closed-loop evidence: best-of-ten direct search over pools recovers two-thirds of it with 100 rollouts. Against the shipped 20-demonstration pool, utility-curated pools of five lead by 7.0 points over 560 paired episodes from two evaluation runs ( p=0.006 ), and utility-ordered retrieval inside the full pool leads by 4.5 points ( p=0.055 ).
[LG-37] XM-Sparse: High-Throughput Sparse-Attention LLM Serving over Scale-up Memory Pools
链接: https://arxiv.org/abs/2611.22280
作者: Yijun Sun,Xudong Liao,Zhuobin Huang,Kai Chen,Nikita Lazarev,Zhizhen Zhong
类目: Distributed, Parallel, and Cluster Computing (cs.DC); Machine Learning (cs.LG)
*备注: 13 pages, 11 figures
Abstract:Sparse attention has been increasingly adopted by production LLMs for long-context inference, yet serving systems must still retain the full KV history, making GPU HBM capacity a key limit on serving concurrency. Existing systems offload the KV cache to host memory, but suffer from the limited bandwidth of PCIe and the limited capacity of host memory. Emerging scale-up memory pools (e.g., X-Mem) relieve both limits and open new opportunities for sparse-attention offloading. In this paper, we first conduct an in-depth analysis and reveal that, even with scale-up memory pools, decode throughput remains bounded by HBM capacity, as hot KV buffers and the fully resident indexer key (K-index) cache occupy substantial HBM. Moreover, the two caches exhibit distinct access patterns. Based on these insights, we present XM-Sparse, the first sparse-attention offloading system over scale-up memory expansion. The core idea of XM-Sparse is to offload both caches aggressively while matching each with an I/O mechanism tailored to its access pattern. Specifically, for dynamically selected KV entries, XM-Sparse fuses TMA-based retrieval with attention in a tile-level pipeline to hide on-demand retrieval latency. For predictable K-index accesses, it prefetches pages via SM-efficient GPU-initiated DMA and coordinates prefetching with latency-critical operations to avoid interfering with decoding. We have implemented XM-Sparse atop SGLang, and our experiments on DeepSeek-V3.2 and GLM-5.3 with an emulated scale-up memory pool show that XM-Sparse delivers up to 4.98 \times and 4.52 \times higher decode throughput than PCIe-based ECHO and HiSparse, respectively, and up to 1.34 \times over a naive integration on the same platform, across 16K-1M contexts.
[LG-38] Spatially local mini-batching for implicit neural representations
链接: https://arxiv.org/abs/2611.22236
作者: Minseok Choi,Swarup Dhar,Heejae Kwon,Kookjin Lee,Donsub Rim
类目: Machine Learning (cs.LG)
*备注: 72 pages, 36 figures
Abstract:Neural surrogates based on implicit neural representations (INRs) are typically trained on mini-batches spread over the whole domain, since a generic INR loses accuracy when every point of a batch lies in one small region. We show that low-rank neural representations (LRNRs), INRs whose weight matrices are products of shared bases with coefficients emitted by a small hypernetwork, can be trained with spatially local mini-batches. Each step uses the points of one time snapshot of an adaptive-mesh numerical solution inside one small cell, with snapshots and cells drawn according to the refinement of the mesh. We posit that such training is enabled by the structural property of LRNRs, as a local step reaches the shared bases only through the r coefficients of each layer, and keeps one cell from rewriting the network. By a Kolmogorov-width argument, we show that when restricted to a small cell, the hidden states of LRNRs lie close to a low-dimensional subspace, which suggests the local steps only affect low-rank subspaces of the shared bases. On six hyperbolic-conservation-law problems with solutions computed using adaptive mesh refinement, one local training recipe reaches 1.1-2.3% relative L^2 error with a tenth of the parameters of a dense network of the same width. Finally, we provide a comparison by training a hypernetwork-SIREN, a DeepONet, and a Fourier neural operator (FNO) of the same size with the same local batches and training budget. We observe that these baselines reach 16-78% where the LRNR reaches 4.9%. Interestingly, we also observe that local training has a significant effect on the per-cell low-dimensional structure even in a wide generic hypernetwork-SIREN.
[LG-39] From Temperature-Limited to Battery-Limited Legged Locomotion
链接: https://arxiv.org/abs/2611.22231
作者: Raphaël Ségas,Balázs Opra,Léo Ghafari,Thomas Stewart
类目: Robotics (cs.RO); Machine Learning (cs.LG); Systems and Control (eess.SY)
*备注: 9 pages, 7 figures, 2 tables. Supplementary video in the ancillary files. Dataset: this https URL
Abstract:On legged robots, sustained locomotion often ends in thermal shutdown rather than battery depletion: the heat of walking concentrates in a few joint motors, which overheat long before the battery is empty. We train thermally aware locomotion policies by reinforcement learning without ever observing temperature, and demonstrate this on a Unitree Go2 quadruped. We first identify a coupled 13-node lumped RC thermal network, fitted on two runs of a single robot and calibrated against the measured electrical power, that predicts all twelve motor temperatures with a mean RMSE below 1 ^\circ C across different robot instances, gaits, payloads, and speeds. Under a sustained gait, the temperatures the motors will settle at follow directly from the gait-averaged squared torques and joint speeds. Our core contribution is a reward that drives down the hottest predicted asymptotes: thermal time constants exceed the credit-assignment horizon of reinforcement learning by orders of magnitude, but predicted asymptotes respond to torque and joint speed within a single gait cycle, so the policy learns to lower and spread heat with no temperature input. On the robot, the policy turns sustained locomotion from temperature-limited to battery-limited: with a 5 kg payload at 1 m/s, a learned baseline reaches thermal shutdown after 19 minutes (1.1 km), whereas our policy runs the battery down after 174 minutes (10.0 km) with its hottest motor 14 ^\circ C below the shutdown threshold in a 26 ^\circ C room; on an unloaded factory circuit with stairs and ramps, the baseline overheats after 25 laps (6.2 km), while our policy completes 47 (11.2 km) on one battery with 22 ^\circ C of margin.
[LG-40] Sample-efficient Personalization of Large Language Model Alignment with Multidimensional User Preference Data
链接: https://arxiv.org/abs/2611.22210
作者: Zhen-Yu Zhang,Jerry Huang,Gang Niu,Takashi Ishida,Masashi Sugiyama
类目: Machine Learning (cs.LG)
*备注:
Abstract:Preference alignment has become a central tool to align large language models (LLMs) with human values and desired behavior, and its online variant can further advance their capabilities beyond what static data alone offer by repeatedly generating and labeling new responses with human feedback. But costly, time-consuming annotations limit its scalability, making sample efficiency essential, and existing methods typically seek to improve it by expanding coverage of the response space or selecting pairs with the highest uncertainty for comparison. Although they are effective, these methods quantify uncertainty from the perspective of the LLM, which may differ from the uncertainty perceived by the user: a pair can appear highly uncertain to the LLM even when past comparisons from the user already imply a clear ordering. In this paper, we propose multidimensional preference-guided online direct preference optimization (MPG-DPO), which leverages low-cost multidimensional evaluations from automated verifiers on the user historical comparisons to construct a personalized preference cone guiding response selection for online preference alignment. Specifically, it first discards responses that are confidently dominated under the estimated user preference, then selects two responses that are competitive under the retained preference cone yet exhibit the strongest unresolved disagreement. Theoretical analysis shows that, with high probability, the fraction of queries spent on cone-dominated responses vanishes as learning progresses. Experiments across multiple LLM families show that MPG-DPO outperforms strong contenders under matched query budgets.
[LG-41] An Optimization Framework for Denoising Generation
链接: https://arxiv.org/abs/2611.22202
作者: Xinyu Zhou,Jiawei Zhang,Stephen J. Wright
类目: Machine Learning (cs.LG)
*备注:
Abstract:Noise-unconditioned denoising learns a single score, so generation can be viewed as gradient ascent on a fixed log-density landscape. This optimization interpretation leaves two questions: do trajectories approach meaningful data, and do different initializations retain diverse outcomes? We establish both phenomena for empirical and population multi-scale Gaussian-mixture objectives. Under explicit posterior and landscape conditions, ascent converges almost surely within O(m\sigma_\min^2/\tau) of an m -dimensional reference manifold with reach at least \tau . We derive a source–sink identity for the same fixed field and use it to prove finite-resolution regional coverage. In an affine construction, we further identify a positive-bandwidth regime where even empirical-score optimization can avoid training-sample memorization while preserving diversity at the training resolution. The same fixed objective naturally supports penalty-guided controllable generation. Experiments support our predicted mechanisms, show improved robustness under large-step guidance, and demonstrate downstream gains consistent with finite-data generative generalization.
[LG-42] owards Optimism-Pessimism Trade-off in Model-based Offline-to-Online Reinforcement Learning NEURIPS2026
链接: https://arxiv.org/abs/2611.22172
作者: Guochen Zhou,Yijun Yang,Qiqi Duan,Qing Su,Weiming Ou,Li Shen,Shihao Ji,Yuhui Shi,Zhisong Zhang
类目: Machine Learning (cs.LG)
*备注: Accepted at NeurIPS 2026
Abstract:Model-based offline-to-online reinforcement learning (RL) enables sample-efficient adaptation by leveraging offline pre-training for online fine-tuning. However, the distribution shifts between offline and online stages often hinder fine-tuning performance. Many existing methods encode the optimism-pessimism trade-off through a single-objective formulation. Selecting its trade-off coefficient through online fine-tuning can require costly bi-level optimization. We identify the balance established during offline training as a key challenge: in our motivating experiments, optimistic policies adapt better to novel tasks, whereas pessimistic policies perform better on the same online tasks. To address this challenge, we propose a bi-objective formulation that captures this trade-off, yielding a Pareto policy pool during offline training. This policy pool enables adaptive selection during online fine-tuning across diverse online tasks. To generate the pool, we introduce Multi-Objective Soft Actor-critIC (MOSAIC), which solves bi-objective problems and constructs diverse Pareto policies. After offline training, a contextual bandit algorithm hierarchically selects the most suitable policy for fine-tuning at each online interaction step. Empirically, our pipeline, Hierarchical Pareto Policy Pool (HiP3), achieves the strongest aggregate performance across both same and novel online tasks and maintains its advantage under stronger distribution shifts. Comprehensive ablations further clarify the roles of the policy pool, MOSAIC-based pool construction, and the online selection mechanism.
[LG-43] Synthetic Crystal Data Generation for Diffusion Pretraining in Material Property Prediction NEURIPS2026
链接: https://arxiv.org/abs/2611.22140
作者: Aamod Varma,Shuyi Jia,Victor Fung
类目: Machine Learning (cs.LG); Materials Science (cond-mat.mtrl-sci)
*备注: 14 pages, 2 figures, 6 tables. Accepted at the AI4Mat Workshop at NeurIPS 2026
Abstract:Diffusion-based pretraining has emerged as a powerful strategy for learning transferable representations in graph neural networks for materials property prediction without requiring quantum-mechanical labels such as energies, forces, and stresses for every pretraining structure. However, the method retains a critical dependency: it requires structures residing near local minima of a potential energy surface, which have traditionally been obtained through costly DFT relaxations of known or physically synthesized compounds. This reliance constrains both the scale and the structural diversity of available pretraining data, leaving large regions of crystal structure space unexplored. In this work, we connect diffusion-based pretraining with synthetic structure generation to reduce this bottleneck. We introduce a synthetic data generation pipeline combining three complementary strategies: perturb-and-relax, generative sampling via MatterGen, and protostructure enumeration. Where structural relaxation is required, we use fast machine-learned interatomic potentials instead of performing a new DFT relaxation for each generated structure. We further compare a coordinate-denoising control with continuous-time variance-preserving (VP) and sub-VP diffusion objectives using a randomly initialized Orb-v2 backbone. The source comparison covers seven MatBench property-prediction tasks, and an eighth task, \textttmp_gap, is included in the protostructure scaling study. Every synthetic-pretraining condition numerically outperforms random initialization on all seven source-comparison tasks. Scaling the protostructure corpus from 20k to 165k under a fixed 400-epoch schedule numerically matches or exceeds MP-20k pretraining on most tasks; because this protocol increases both data diversity and optimization updates, the result reflects joint data-and-compute scaling.
[LG-44] ReMoE: An Interpretable Mixture-of-Experts Framework for Human Mobility Recovery Forecasting Across Disasters KDD2027
链接: https://arxiv.org/abs/2611.22133
作者: Yilin Lyu,Kang Liu,Wei Wang,Qifan Zhang,Ling Yin,Yushi Bai
类目: Machine Learning (cs.LG); Computers and Society (cs.CY)
*备注: Accepted by KDD 2027
Abstract:Human mobility often declines sharply following disasters and subsequently exhibits complex recovery dynamics. Effective forecasting of this recovery is critical for timely assessment and resource allocation. However, existing approaches are often developed for individual events and operate as black boxes, limiting their transferability and interpretability beyond the events they were trained on. Recent studies have identified recurring post-disaster mobility recovery types across disasters, suggesting that recovery knowledge may be transferable. Motivated by this, we propose ReMoE, an interpretable mixture-of-experts framework for cross-disaster mobility recovery forecasting. ReMoE uses a feature-guided router to adaptively combine specialized experts that learn these heterogeneous yet reusable recovery types, producing a region-specific recovery pattern. It further refines this pattern through residual correction informed by forecasts of future hazard conditions. We evaluate ReMoE on a large-scale mobility dataset covering 18 disaster events spanning hurricanes, floods, winter storms, and wildfires. ReMoE is competitive in-distribution and achieves the strongest performance on unseen events and hazard types, with larger gains under greater distribution shifts. Further analyses show that ReMoE yields interpretable learning outcomes and maintains robust performance under uncertainty. Broadly, ReMoE offers a transferable and traceable forecaster that can support anticipatory post-disaster assessment.
[LG-45] mpQuant: Learning Adaptive Temperatures for Quantized Language Models
链接: https://arxiv.org/abs/2611.22128
作者: Mukund Athmaram,Liang Ma,Manoj Rao
类目: Machine Learning (cs.LG)
*备注:
Abstract:Large language models are increasingly deployed at low precision, yet quantization can degrade their performance. We study how quantization interacts with decoding settings such as the sampling temperature and top-p by examining how it distorts the model’s next-token probabilities. Measuring entropy during generation across deployed temperatures, we find that quantization warms the model on average, and that temperature matters much more for the performance of a quantized model than for its full-precision counterpart. Because quantization and temperature both shape the next-token distribution, we investigate how much the sampling settings can undo what quantization changes, through our method TempQuant, which learns the quantized model’s temperature and top-p to bring its next-token probabilities closer to the full-precision model’s. We learn these settings at two granularities. The first is a single global setting per requested temperature, which finds nearly the same temperatures that tuning on labeled answers would, without any labels. Model cards typically recommend sampling a quantized model with the same settings as its full-precision counterpart, while we find that different settings can substantially improve its accuracy. On 3-bit Qwen3-4B, for example, global TempQuant raises MATH accuracy from 34% under the configuration recommended for Qwen’s quantized releases to 51%. The second sets the temperature and top-p for every token, motivated by our finding that tokens whose rankings flip under quantization tend to need warming, while tokens whose rankings are preserved tend to need cooling. When quantization degrades accuracy substantially, TempQuant recovers much of it, and when degradation is lighter, it still brings the quantized model’s behavior closer to the full-precision model’s.
[LG-46] VTT: Reading Vehicle Size and Powertrain Type from Motion Trajectory Across Unseen Car Models
链接: https://arxiv.org/abs/2611.22109
作者: Yuhang Wang,Yuechen Guo,Jingxin Yang,Chuheng Wei,Suyang Chen,Abdulaziz Alhuraish,Yiyi Wang,Mingyang Ban,Jingran Sun,Hao Zhou
类目: Machine Learning (cs.LG); Robotics (cs.RO)
*备注: 8 pages
Abstract:Vehicle size and powertrain information can enrich trajectory data for traffic analysis and simulation. We present VTT, a dataset and benchmark for identifying these attributes from motion and controller area network (CAN) signals across unseen car models. The dataset contains 9,062 drives, about 3,950 h, from 1,431 vehicles and 174 car models. It provides three footprint-based size classes, powertrain labels, and evaluation on unseen car models and unseen drivers. Using 60-s windows, we compare four sequence architectures, statistical features and physical features proposed by large language models. Training with auxiliary physical targets, data augmentation and task-specific weighting raises drive-level macro-F1 to 92.6-93.0% for battery-electric versus combustion vehicles and from 62.9% to 67.1% for size classification. On public logs from 4,061 additional drivers, the full-input model identifies powertrain correctly for 86 of 88 car models; size accuracy is 63.2%, close to the internal vehicle-level result of 63.5%. Physical-feature analysis links powertrain recognition to deceleration behaviour and size recognition to steering response. Errors near size-class boundaries and limited gains from longer observations highlight the importance of label resolution and training-fleet coverage for cross-model generalisation.
[LG-47] Every Measurement Every Direction All at Once: Multimodal Flow Matching for Molecules and Spectra NEURIPS2026
链接: https://arxiv.org/abs/2611.22098
作者: Thorben Prein,Elton Pan,Rafael Gómez-Bombarelli,Santiago Miret
类目: Machine Learning (cs.LG); Chemical Physics (physics.chem-ph)
*备注: 30 pages, 12 figures. Accepted at NeurIPS 2026
Abstract:Generative modeling in science is often data-limited, where experiments are expensive. However, each experiment may yield several coupled measurements of the same physical entity. Standard generative models often train only on a single direction, discarding supervision from the rest. We argue that each experiment contains more training signal than this asymmetry allows the model to exploit: random masking enables measurements to serve as either conditioning inputs or prediction targets, yielding multiple supervised denoising tasks from a single experiment. We investigate this premise with MOSAIC, a multimodal flow matching transformer over molecular 3D structure and UV, IR, and Raman spectra. MOSAIC is a single model that is capable of any-to-any generation: generating any subset of modalities from any disjoint subset of the others. On QM9S, MOSAIC achieves 88.8% molecular graph recovery, outperforming the strongest spectra-to-structure baseline by 23.2% in structure generation, while simultaneously outperforming a structure-to-spectra specialist model in spectral generation. Ablations show that the any-to-any objective improves structure generation over structure-only training across model sizes. Furthermore, MOSAIC supports arbitrary subsets of available measurements and enables molecular identification by matching generated spectra to experimental references.
[LG-48] BernBRS: Certified Backward Reachability for Deep Bernstein Neural Networks
链接: https://arxiv.org/abs/2611.22083
作者: Ibrahim Albool,Yasser Shoukry
类目: Machine Learning (cs.LG)
*备注: 18 pages, 6 figures, 7 tables
Abstract:Ensuring the safety of complex neural network systems requires mathematically rigorous verification. While backward reachability analysis is a critical tool for identifying safe initial states, computing tight, formal bounds remains notoriously difficult to scale. Recent advancements have demonstrated that DeepBern-Nets neural networks utilizing Bernstein polynomial activation functions possess advantageous mathematical properties that enable exceptionally fast and tight forward reachability analysis. Building upon this efficiency, this paper introduces BernBRS, a highly accurate and scalable framework for the backward reachability analysis of DeepBern-Nets. BernBRS simultaneously computes certified inner and outer approximations of the Backward Reachable Set (BRS). Rather than relying on single convex bounding, our method intelligently refines a cover of the input space by analyzing activation patterns, splitting the input domain precisely where an exact decomposition of the relaxation gap predicts the most significant reduction in error. This targeted refinement enables high levels of scalability to highly complex and massive architectures that remain intractable for existing certified methods. We demonstrate the efficacy of our framework on high-dimensional models, including a linearized closed-loop approximation of the Half-Cheetah controller and a distilled Bernstein student of a Transformer block and we empirically study how these model-level guarantees relate to the original Half-Cheetah and TinyBERT classifier, respectively.
[LG-49] wo Counterexamples for e-means
链接: https://arxiv.org/abs/2611.22065
作者: Zhenisbek Assylbekov
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:e-means (Bessa et al., 2026) combines exponential averaging and cluster revival. We give two counterexamples to its published recipe with higher final loss than standard k-means (Lloyd, 1982) from identical initial centers. Mass-based revival can destroy a global optimum and make the ratio of expected losses unbounded even with k-means++ seeding.
[LG-50] DSADiff: Data Structure-Aware Diffusion for Crystal Structure Prediction via Mitigating Complexity- and Imbalance-Induced Biases
链接: https://arxiv.org/abs/2611.22050
作者: Jiang Xiaobo,Dinghong Lai
类目: Machine Learning (cs.LG); Materials Science (cond-mat.mtrl-sci)
*备注:
Abstract:Crystal Structure Prediction (CSP) is an important step in accelerating the generation and discovery of new materials. In recent years, diffusion-based CSP methods have made significant progress. However, existing models still suffer from generation bias when facing diverse materials spaces, while systematic approaches for tracing the sources of such performance variations remain limited. To address this issue, we propose LAS (Layered Analysis Strategy), a mechanism analysis framework that identifies crystal complexity and elemental composition imbalance as two key factors affecting generation difficulty and further reveals their relationships with model generation performance. Building on these findings, we further propose DSADiff (Data Structure-Aware Diffusion), a data structure-aware diffusion model that transforms the key factors identified by LAS into a unified generation difficulty representation and uses it to jointly drive feature modeling and training strategy optimization. This enables the model to adapt its representation and generation capability to samples with different data structures and difficulty levels, thereby mitigating learning bias across samples. Experiments on three CSP benchmark datasets, including Carbon-24, demonstrate that DSADiff consistently improves both Match Rate and RMSE, with a maximum Match Rate improvement of 9%, surpassing existing state-of-the-art methods and achieving new state-of-the-art performance. This work deepens the understanding of diffusion-based CSP models from the perspective of generation bias and provides a new direction for designing less biased generative models for complex materials spaces.
[LG-51] GP–RL–MPC: Uncertainty-Aware Reinforcement Learning-Augmented Model Predictive Control with Gaussian Process Postdecision State-Value Functions
链接: https://arxiv.org/abs/2611.22038
作者: Jon Li,Jay H. Lee
类目: Machine Learning (cs.LG); Systems and Control (eess.SY)
*备注: 19 pages including references and appendices, 3 figures, 2 tables. Submitted to L4DC 2027
Abstract:We propose Gaussian Process Reinforcement Learning-Augmented Model Predictive Control (GP–RL–MPC), which uses a learned critic as the MPC terminal penalty to account for costs beyond the prediction horizon. Gaussian process regression (GPR) enables data-efficient critic learning from limited, noisy rollouts within optimistic approximate policy iteration (OAPI), with soft updates mitigating policy chattering. A tunable risk coefficient weights GP epistemic uncertainty in the terminal cost, regulating sensitivity to uncertain critic predictions. The postdecision state formulation shifts disturbance averaging offline to account for aleatoric uncertainty from the terminal transition onward. Under matching objectives and admissible actions, a one-step horizon recovers greedy value-space RL. After fitting GP hyperparameters by exact marginal likelihood, we construct a sparse deterministic training conditional (DTC) posterior using inducing points selected by pivoted Cholesky (PC), reducing online evaluation costs as the training dataset grows. In a continuous stirred-tank reactor (CSTR) subject to exogenous disturbances, GP–RL–MPC with an appropriately selected risk coefficient and policy screening reduces the sum of squared errors (SSE) by an average of 80.52% relative to an otherwise matched baseline MPC across five random seeds (sample standard deviation: 3.48%).
[LG-52] An AI-Based OSSE Framework to Evaluate Learned Physical Consistency: A Multi-Model Study Using FourCastNet Aurora Prithvi WxC and GraphCast
链接: https://arxiv.org/abs/2611.22024
作者: Yash Diggikar,Milton Halem,Aishwarya Raman
类目: Machine Learning (cs.LG)
*备注:
Abstract:Observing System Simulation Experiments (OSSEs) are a central methodology for evaluating future atmospheric observing systems. We present the first AI-based OSSE framework to test whether different AI weather models can reproduce comparable physical impact from new observations that traditional physics based models generate. We evaluate four distinct architectural AI models on the Charney conjecture that incomplete historical observations can infer the present state of the atmospheric. We considner- FourCastNet V2, GraphCast, Aurora and Prithvi WxC and integrate each autoregressively over a 60-day forecast horizon. After a 30-day spin-up, we insert ERA5 temperature fields to infer winds and related atmospheric fields. Similar OSSE experiments are conducted for wind insertion. We measure the performance of each model’s response to observational information on a 30-day data assimilation cycle forecast with and without data insertion. Distinct architectural AI weather models respond differently to the same OSSE observations. Foundation models (i.e. Prithvi WxC and Aurora), produce the strongest impact. For example, while GraphCast is the weakest, all three of the four models capture the temperature-to-geopotential height response. The two foundational models retain a positive, non-trivial, tropical zonal wind at 500 hPa response to temperature insertion. Various data insertion experiments can vary significantly even within a given architecture. We identify an architecture-specific numerical instability in GraphCast’s full 37-level configuration that produces a runaway feedback concentrated at the uppermost levels with ERA5 temperature insertion correctable by excluding levels at or below 100 hPa.
[LG-53] How Much Distribution Shift Can a Clustering Withstand? Exact Split and Merge Certificates for Structured Transports
链接: https://arxiv.org/abs/2611.22000
作者: Alisina Bayati,Srinivasa Salapaka,Shahriar Talebi
类目: Machine Learning (cs.LG)
*备注:
Abstract:In clustering, the distribution of the population being clustered is often uncertain or changes over time, so the true distribution may shift away from the nominal distribution used to form the clustering. Once the clustering is committed and each member is assigned to a cluster, such shifts may change the distribution of each cluster. They can make a cluster too spread out for one representative at a given resolution, motivating a split, or bring distinct clusters close enough to share a representative at that resolution, motivating a merge. We make these split and merge events precise using a local stability criterion built on the phase-transition analysis in deterministic annealing. This notion of stability enables us to quantify how robust a clustering is against such events. Furthermore, motivated by applications in which uncertainty and population change are structured, we study shifts whose displacements are restricted to a prescribed subspace. The smallest structured shifts that reach the respective thresholds define the split and merge radii, for which we obtain exact semidefinite programming (SDP) reformulations. The split SDP has only two scalar decision variables. These convex reformulations enable us to combine the radii into a robustness certificate for the full clustering that gives the size of the smallest structured shift that reaches any monitored split or merge threshold.
[LG-54] Sparse Integer Fourier Transform and Its Application In Machine Learning
链接: https://arxiv.org/abs/2611.21976
作者: Jinyang Li,Jie Chen
类目: Mathematical Software (cs.MS); Machine Learning (cs.LG)
*备注:
Abstract:The Discrete Fourier transform, mainly implemented by Fast Fourier Transform (FFT), has been essential for numerous domains, such as digital signal processing and feature extraction in materials design. However, current FFT methods suffer from extensive non-trivial multiplications and twiddle-factor operations in the big data era. To improve the computational efficiency of FFT, this study proposes Sparse Integer Fourier Transform (SIFT), a method that leverages the fast computing of sparsity and small integers, and applies the Radix-n method to derive an FFT. The 2N-point DFT matrix in Radix-n FFT is decomposed into four sparse integer matrices. A diagonal matrix containing complex twiddle factors is efficiently approximated using an optimization strategy while the remaining matrices retain sparsity and integer entries. For radix-32 FFT, SIFT reduces the estimated arithmetic cost by 75%, with RMSE below 2.3765\times10^-8 and MAPE below 0.01089%. We evaluate frequency extraction in a numerical test and an engineering case. In the numerical test, MAPE and RMSE are below 2.016% and 2.536\times10^-6 , respectively. For wind turbine fault diagnosis, SIFT extracts compact spectral features that retain fault information while reducing the classifier’s input dimensionality relative to the full time series.
[LG-55] A Log-Timescale Perturbation Bound for Selective State Space Models at Every Sequence Length
链接: https://arxiv.org/abs/2611.21952
作者: Jack Humphries,Daniel Midgley
类目: Machine Learning (cs.LG)
*备注: 22 pages, 3 figures, 1 table. Code and recorded runs: this https URL
Abstract:Mamba models deployed on streaming data, such as limit order book updates, run far past the sequence lengths they were trained on. A guarantee on how much a small change of input can move the hidden state therefore has to hold at every length. The closest existing bound we know of measures the change of the timescale additively, and it grows linearly with the sequence length unless the timescale is bounded away from zero. In this paper, we measure the change on a log scale instead. The key observation is that the decay factor of each step then moves by at most a fixed fraction of that step’s input weight, so the per-step errors cannot pile up. Using this, we prove that two runs of a selective-scan cell from a common start stay within a bound that holds at every length, shrinks to zero with the size of the change and needs no lower bound on the timescale. We give a construction showing that, for small changes, this bound is within a factor of about 2 of the best possible, and we extend it to the update used in Mamba’s released code when the timescale is bounded above. We then evaluate block-level bounds built from the theorem on two small models trained on one day of order book data. With a bounded timescale, a bound built on the existing analysis is tighter than ours for at most the first 8 steps. With Mamba’s softplus timescale, whose certified lower bound is as small as 10^-21 , it is looser at every length. Our bounds hold for every input to a block, but they remain more than five orders of magnitude above the changes that gradient attacks find.
[LG-56] Higher-Order Interaction Reconstruction via Integral Physics-Informed Neural Networks
链接: https://arxiv.org/abs/2611.21935
作者: Xin Mao,Zhen Xu,Shiyi Yang,Sukwon Yun,Yunmei Liu,Ren Wang,Tianlong Chen,Can Chen
类目: Machine Learning (cs.LG); Systems and Control (eess.SY)
*备注:
Abstract:Reconstructing higher-order interactions is a fundamental challenge in complex networked systems, with applications in ecological, social, and biological networks. This inverse problem is inherently difficult due to the combinatorial complexity of higher-order structures, the strong nonlinearity of the underlying dynamics, and the prevalence of noisy and sparse observations. In this article, we propose a novel integral physics-informed neural network (IntegralPINN) framework for higher-order interaction reconstruction. Instead of penalizing derivative-based physics residuals, we adopt an integral formulation over temporal intervals, which naturally suppresses high-frequency noise and mitigates derivative-induced instability. We further introduce a continuous parameterization of interaction tensors, enabling tractable identification of higher-order structures via gradient-based optimization. We provide theoretical analysis establishing a noise-robust error bound that is independent of derivative amplification when the integration window satisfies an excitation condition. Extensive experiments on ecological dynamics, social contagion, and neuronal synchronization systems demonstrate that IntegralPINN achieves accurate and robust reconstruction under noisy and sparse observations, outperforming state-of-the-art approaches.
[LG-57] Physics-informed neural networks for spacecraft trajectory optimization with exact boundary conditions
链接: https://arxiv.org/abs/2611.21933
作者: Samuel Schönherr,Günther Waxenegger-Wilfing,Sascha Ranftl
类目: Machine Learning (cs.LG)
*备注: 35 pages, 8 figures, for reproducibility: this https URL
Abstract:We use neural networks to optimize trajectories by modeling the position as a function of time and minimizing the thrust with respect to the network parameters. This corresponds to an unsupervised training procedure in which physical laws are incorporated directly into the optimization objective. In standard physics-informed neural networks, initial and terminal states are enforced through additional weighted loss terms, resulting in a multi-objective optimization problem that is highly sensitive to the choice of loss weights and can lead to poor convergence and reduced accuracy. To address this challenge, we construct a simple transformation, exploiting the mathematical structure of trajectory optimization problems, to enforce boundary conditions exactly a-priori. This transformation can be interpreted as a reparameterization of the solution space that embeds the boundary conditions directly into the function. This converts the original constrained optimization problem into an unconstrained one over admissible trajectories. I.e. the transformation eliminates the boundary condition penalty terms and reduces the problem to a single-objective optimization. Crucially, trajectory optimization comes with a low-dimensional boundary, which allows particularly simple and stable constructions of distance functions compared to general domains in partial differential equations. Numerical experiments demonstrate that the proposed approach meaningfully improves convergence behavior and solution accuracy.
[LG-58] KVUse: Reusable-KV User Memory for Efficient LLM Recommenders
链接: https://arxiv.org/abs/2611.21920
作者: Xin Guo,Qiang Zhang,Xiang Zhou,Arushi Rai,Siyu Lin,Yifan Wu,Xiang Shen,Lei Huang,Qunshu Zhang,Xiangjun Fan,Lizhu Zhang
类目: Machine Learning (cs.LG)
*备注:
Abstract:Memory-augmented recommenders built on large language models (LLMs) maintain a natural-language profile of each user. These profiles are updated by a user profile update path that revises the profile as new interactions arrive and are consumed by downstream task paths like ranking. Because these user profiles are stored as text in the user memory, every ranking LLM call that consumes them must splice them in the prompt and calculate the KV cache of them during the prefill phase. Given these ranking LLM calls are prefill heavy (long prompt but short decoded output) and repeatedly called, these repeated prefill of user profile text dominates the serving. Prefix caching cannot remove it, because these user profiles are preceded by different text in different calls. We present KVUse, which stores the user profiles as reusable key-value (KV) caches instead of text. KVUse trains lightweight LoRA writers by self-distillation, so that the user profile update call produces the reusable KV of the user profile while decoding it, and the LLM can read these reusable KV at new positions and behind new context without recomputation. KVUse also has a customized KV scheduler that uploads the stored reusable KV of the next batch while the current batch is computed, hiding the KV transfer from host memory behind GPU computation. On four InstructRec datasets with Llama-3.1-8B-Instruct, KVUse shows no statistically significant difference in ranking quality from full prefill over continuous memory update rounds. Under saturated load, KVUse raises the throughput of the downstream scoring path by 4.0x-4.5x and of the profile update path by 1.4x-1.9x.
[LG-59] Lost in a Few Heads: Diagnosing and Repairing Block Selection in Sparse Attention
链接: https://arxiv.org/abs/2611.21917
作者: Wenhe Zhang,Boxing Chen
类目: Machine Learning (cs.LG)
*备注: 25 pages, 8 figures, 20 tables. Code: this https URL
Abstract:Sparse attention models read a few blocks of keys for every query at a fixed budget and can lose much of their long-context retrieval accuracy at that budget. We ask where this loss is made. Using exact attention mass selection at the shipped budget as an oracle, we show that the selector under-ranks the needle rather than running out of budget. We find attention heads that miss the needle by running a calibration probe that teacher-forces the answer and reads two quantities in every head. The first is the needle attention that the selector forfeits, and the second is the share of the answer’s log-probability that fixing the head’s selection restores, integrated along the path from the shipped selector to full attention. Four checkpoints of the InfLLM-V2 family and NOSA-8B lose multi-key retrieval in the same three blind heads, which attend to the needle while a mean-pooled key summary dilutes its single decisive token. Full attention in those heads or a 4-bit scorer that keeps the shipped budget recovers most of the loss without training. On MiniCPM4.1-8B at 64K, either fix lifts multi-key retrieval accuracy from 54.1 to 97.5%, level with full attention in every head. In production serving engines, full attention in those heads keeps 84 to 89% of the time to first token (TTFT) that sparse attention saves at 32K to 128K context length. Beyond the InfLLM-V2 selector lineage, one unified rule flags attention heads above a threshold swept on each checkpoint until they close most of the gap to full attention on calibration prompts. A similar small set of blind attention heads carry the loss for the SeerAttention-R gates, NSA, and DashAttention, whereas many such heads exist in MoBA. On Qwen3-8B with the SeerAttention-R gates, full attention in 18 of 288 key-value heads lifts multi-key retrieval accuracy from 29.8 to 97.7%.
[LG-60] World-Grounded Reinforcement Learning for Pollution-Aware Bird Movement Modeling Using a Multi-Instance DQN Framework NEURIPS2026
链接: https://arxiv.org/abs/2611.21895
作者: Shivin Mangal,Divyansh Chawla,Anshu Garg,Isshaan Singh
类目: Machine Learning (cs.LG)
*备注: This work has been accepted at the GlobalSouthAI Affinity Event at NeurIPS 2026
Abstract:Birds are modeled as environmentally sensitive agents whose movement reflects real-world air quality conditions, where AQI serves as a measurable world feedback signal representing pollution-driven consequences. Instead of relying on human preference or synthetic rewards, the system uses environmental outcomes as the primary training signal, enabling policy learning grounded in ecological impact. We propose a multi-instance reinforcement learning simulation in which agents navigate a 2D grid representation of India integrated with a precomputed AQI matrix and geographic state boundaries. The framework combines bird call metadata from the Xeno-canto dataset, spatial discretization, and real environmental data to learn adaptive navigation strategies under pollution constraints. To ensure scalability and efficiency, repeated real-time API queries are replaced by a structured AQI mapping that enables fast lookup during training and execution. Multiple baselines, including Monte Carlo, Q-Learning, and Value Iteration, are evaluated for comparison. Deep Q-Network demonstrates stronger stability, adaptability, and cumulative performance across varying environmental conditions, establishing a scalable foundation for world-feedback driven ecological decision modeling.
[LG-61] WM-Quant: Rollout-Aware Post-Training Quantization for JEPA-Style World Models
链接: https://arxiv.org/abs/2611.21882
作者: William Hammer,Andrew Heuer,Nikhil Krishna,Matthew Grenier,Ramtin Zand
类目: Machine Learning (cs.LG)
*备注: 11 pages, 5 figures, and 7 tables
Abstract:We present WM-Quant, to our knowledge, the first post-training quantization (PTQ) framework designed specifically for JEPA-style world models. Although world models typically feature relatively few parameters, action planning algorithms often require thousands of inferences, which can quickly become computationally expensive. PTQ is a natural way to reduce this cost. However, we show that in world models using autoregressive model predictive control (MPC) rollouts, quantization error can compound with each predictor step, resulting in action rankings from the quantized model becoming near-random. WM-Quant addresses this with three components that build on existing PTQ methods: early-encoder precision, control-path preservation, and rollout-aware scaling. Early-encoder precision protects accurate latent state and goal information to mitigate long-term error propagation. Control-path preservation allows the world model predictor to distinguish between fine-grained actions, significantly improving action rankings and execution. Rollout-aware scaling jointly calibrates encoder and per-step predictor activation ranges with respect to an entire rollout rather than a single prediction, reducing the cumulative rollout error. Across multiple world models, environments, and PTQ methods, WM-Quant substantially improves planning success at low bit-widths, approaching full-precision performance at W4A4.
[LG-62] Adaptive Memory for Online Reinforcement Learning under Sparse Model Changes
链接: https://arxiv.org/abs/2611.21879
作者: Sreejeet Maity,Sukanya Modak,Leonardo F. Toso
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:We study the discounted reinforcement learning (RL) problem in a setting where the underlying environment changes between deployments. Each deployment provides H observations from a fixed MDP. The reward function and transition kernel are fixed within each deployment. Across deployments, at most an \epsilon 0 fraction of the reward and transition kernel entries may change. In this setting, learning each model to deploy from scratch may discard useful observations from past deployments, whereas pooling all past observations can bias the estimates of entries that have already changed. We then propose an algorithm that for each state and action pair, selects how much history to reuse to estimate the reward and probabilities of the next state. We start with observations from the current deployment and add earlier deployment observations one at a time, the most recent first. After each addition, we check whether a single average reward agrees with the estimates from all histories considered until that point. We apply the same check to the distribution of next states. For each estimate, we leverage the last history that passes these aforementioned checks. The optimal control policy is then computed on top of the fitted MDP. We characterize the average gap between the optimal return and the return of the learner’s policies over a finite number of deployments. We demonstrate that when 0 |S|\epsilon \leq 2 , with |S| being the cardinality of the state space, our algorithm needs only the observations from the most recent O(1/\epsilon) deployments, and achieves a long-run average gap of at most \widetildeO(\sqrt\epsilon/H) .
[LG-63] mol: a GPU-accelerated implementation of the Rosetta energy function for machine learning
链接: https://arxiv.org/abs/2611.21857
作者: Andrew Leaver-Fay,Kieran Didi,Jeff Flatten,Alex Ford,Guangfeng Zhou,Nathaniel Corley,Hahnbeom Park,Maxwell W. Kazman,Rafael I. Brent,Joseph Kleinhenz,Henry Solberg,Andrew M. Watkins,Brian Kuhlman,David Baker,Frank DiMaio
类目: Machine Learning (cs.LG)
*备注:
Abstract:Physics-inspired forcefields complement generative machine learning models, showing better generalization beyond the training distribution. However, these forcefields are typically slow to evaluate, and their derivatives are difficult to integrate into ML frameworks. We introduce tmol, a GPU-native PyTorch library implementing the state-of-the-art RosettaGenFF forcefield. tmol supports efficient batched differentiable scoring and whole-structure refinement with up to 362-fold throughput increase (batched). Used as a reward in generative protein design, tmol increases hydrogen bonds and water bridges at designed interfaces.
[LG-64] When Does Discrete Ricci Curvature Help? An Edge-Level Analysis of Curvature-Enhanced GCNs for PPI Prediction NEURIPS2026
链接: https://arxiv.org/abs/2611.21849
作者: Clara Joswig,Elham Gholamzadeh,Ghazal Mostofi,Nico Scherf
类目: Machine Learning (cs.LG)
*备注: 15 pages, 6 figures, 9 tables. Accepted at the NeurIPS 2026 Workshop on Geometric Distributional Deep Learning
Abstract:Curvature-enhanced graph convolutional networks (CGCN) incorporate Ollivier-Ricci curvature (ORC), grounded in optimal transport theory, into message passing, motivated by the claim that curvature-weighted edges alleviate “over-squashing” at structural bottlenecks. But does CGCN’s improvement actually come from mitigating over-squashing? We investigate this on four real-world protein-protein interaction (PPI) datasets, comparing CGCN against a standard GCN at the level of individual edges rather than aggregate metrics alone. While CGCN improves aggregate AUC, AUPR, and F1 across all four datasets, this gain is concentrated almost entirely on low-degree edges, while performance on low-curvature (bottleneck) edges systematically degrades in three of four datasets, the opposite of what the over-squashing motivation predicts. Examining the learned edge-weight function directly, together with clustering-validity metrics, edge-distance ratios, and a layer-depth ablation across four depths, we find evidence for an alternative account: curvature-weighting sharpens the separation between clusters, which improves intra-cluster predictions at the cost of inter-cluster (bottleneck) ones. We test this account against topologically defined communities rather than the classification labels used elsewhere. Additionally, we offer practical guidance on when to prefer CGCN over standard GCN, depending on whether intra- or inter-cluster predictions matter most.
[LG-65] Regularized Score Centering
链接: https://arxiv.org/abs/2611.21791
作者: Hao Bai,Chenlu Ye,Xuanchang Zhang,Aviral Kumar,Tong Zhang
类目: Machine Learning (cs.LG)
*备注:
Abstract:Asynchronous reinforcement learning increases training throughput, but data from stale policies and numerical mismatch between training and inference can destabilize learning. To stabilize off-policy learning, score centering (SC) removes the effect of reward baselines on the expected update. Our gradient diagnostics show less extreme gradients with SC than with importance sampling (IS), but SC can still fail when the learner moves far from the behavior policy that generated the data. To limit this policy mismatch, we introduce Regularized Score Centering (RSC), which adds a reverse Kullback-Leibler (KL) penalty to encourage the learner’s next-token distribution to stay close to the behavior policy’s next-token distribution. RSC works with or without truncated importance sampling (TIS), which caps token importance weights. Our theoretical analysis supports this approach, showing that KL regularization improves convergence guarantees. We evaluate dense and mixture-of-experts models on mathematical reasoning and agentic tasks, including online shopping, tool use, games, and fully asynchronous software engineering. RSC with TIS substantially improves sample efficiency across all agentic benchmarks, attaining higher training rewards with the same number of sampled responses. In final evaluations, it achieves top performance across all six agentic tasks, improving by 14.3% on AppWorld and reaching 9.9 times the strongest baseline’s score on BabyAI.
[LG-66] Learning Calibrated and Samplable Predictive Distributions of Surface-Ocean ΔpCO_2 Using Tabular Diffusion Models NEURIPS2026
链接: https://arxiv.org/abs/2611.21769
作者: Gabriele Accarino,Mohammad M. Amirian,Thea H. Heimdal,Amanda R. Fay,Galen A. McKinley,Viviana Acquaviva
类目: Machine Learning (cs.LG); Atmospheric and Oceanic Physics (physics.ao-ph)
*备注: 40th Conference on Neural Information Processing Systems (NeurIPS 2026). Workshop: Representations for the Physical Sciences
Abstract:The ocean absorbs almost 30% of anthropogenic CO _2 emissions, but quantifying this sink requires estimating surface-ocean CO _2 from observations that are sparse in space and time. Most existing machine-learning estimates of this field are deterministic, giving a single value per grid point, with uncertainty approximated post hoc via residuals, ensembles, or prescribed distributions. We instead learn a conditional predictive distribution for each observation: using Treeffuser, a conditional diffusion model with gradient-boosted-tree score functions, we estimate the full predictive distribution of \Delta p CO _2 (the sea-air p CO _2 difference) from 13 environmental, spatial, and temporal predictors. On a held-out SOCAT test set, Treeffuser attains the lowest continuous ranked probability score among the uncertainty-quantification methods evaluated (8.229 \mu atm) and the coverage closest to the nominal 68% / 95% levels (0.686 / 0.930), while retaining point-prediction skill comparable to a deterministic XGBoost baseline (RMSE 18.4 vs. 17.7 \mu atm; R^2 0.77 vs. 0.79). These results show that conditional diffusion can provide well-calibrated, observation-level uncertainties while largely preserving point-prediction skill. Because Treeffuser is generative, its samples can support probabilistic analyses of \Delta p CO _2 ; extending this to spatially coherent field reconstruction will require modeling dependence across grid cells and evaluation on spatially held-out data.
[LG-67] Hybrid Quantum Selection of Learning Rates in Language-Model Training: Termination Criteria and an Interaction-Guided Schedule Oracle
链接: https://arxiv.org/abs/2611.21701
作者: Wei-Che Hung
类目: Machine Learning (cs.LG)
*备注: 10 figures, 11 tables (including appendices)
Abstract:Learning-rate choices strongly affect language-model fine-tuning, but the best joint profile across layer groups is generally unknown. This paper studies quantum minimum finding for layer-group learning-rate selection within a hybrid classical-quantum training framework. It combines Durr-Hoyer minimum finding with BBHT search and compares fixed-budget and confidence-based stopping policies using a lookup oracle built from measured profile scores at fixed training checkpoints. The paper then introduces an Interaction-Guided Schedule Oracle that scores and marks joint schedules in superposition without a precomputed score table. Classical one-step loss measurements supply the main, pairwise and three-way interaction effects between layer groups. The circuit adds the effects for each schedule, compares the resulting score with the current baseline score, and marks improving schedules for amplitude amplification. Measured candidates are checked classically before acceptance. Uniform-rate training supplies the initial baseline; using schedules from classical adaptive methods as baselines for joint refinement is left for future work. In closed-loop fine-tuning of a 1.5-billion-parameter language model, each selected schedule drives the next 200 classical training steps. Benchmark accuracies lie within the range observed across five uniform-rate seeds. The results give a working example of how quantum superposition can support adaptive learning-rate selection during training; the remaining quantum limitations are discussed in the paper.
[LG-68] FACE: Flow-Guided Angular Channel Estimation Framework for mmWave MIMO Systems
链接: https://arxiv.org/abs/2611.21698
作者: Samitha Gunarathne,Eslam Eldeeb,Nurul Huda Mahmood,Matti Latva-aho
类目: Machine Learning (cs.LG)
*备注: 14 pages, 11 figures
Abstract:Acquiring accurate channel state information (CSI) is essential for millimeter-wave (mmWave) multi-input multi-output (MIMO) systems. However, it remains challenging due to high-dimensional channels, limited pilot observations, and measurement noise. To address these challenges, we propose flow-guided angular channel estimation (FACE), a novel channel estimation framework based on flow matching, an emerging generative modeling paradigm that learns continuous transport dynamics between probability distributions. FACE exploits the inherent angular-domain sparsity of mmWave channels and incorporates the learned flow matching model as a generative channel prior. Numerical results show that FACE consistently achieves higher channel estimation accuracy than both conventional estimators and diffusion-based generative estimator across diverse channel and system configurations. Moreover, FACE maintains high estimation accuracy with fewer pilot observations and achieves an average achievable rate close to the perfect-CSI benchmark. Compared with diffusion-based estimation, FACE requires substantially fewer sampling steps while achieving superior or comparable estimation accuracy, reducing inference time and computational cost. Overall, the proposed framework demonstrates the potential of flow matching for accurate, pilot-efficient, and computationally efficient mmWave MIMO channel estimation.
[LG-69] Layerwise Learning-Rate Scaling in Vision Networks: Effective-Rate Identification Power-Mean Aggregation and Budget Dependence
链接: https://arxiv.org/abs/2611.21696
作者: Chandini Vysyaraju,Raghuvir Duvvuri,Radu Timofte,Dmitry Ignatov
类目: Machine Learning (cs.LG)
*备注: 10 pages, 6 figures. Supplementary material (15 pages) included as an ancillary file. Data and analysis scripts: this https URL
Abstract:Layerwise learning-rate scaling assigns different learning rates to different parts of a network. Although established in fine-tuning pretrained models, its effects in vision networks trained from scratch are commonly assessed by training each candidate recipe. We show that, in this setting, the rate-related component of a layerwise strategy is well summarized by an effective learning rate computable directly from its multiplier profile. Across matched-control comparisons, changes in test accuracy closely track changes in training loss (r = -0.996), which supports an optimization-progress explanation of the short-budget rate effect. Increasing the untuned base learning rate tenfold shifts the rate-matched contrast upward in all six tested architectures (p = 0.016) and increases the number of architectures in which a published discriminative schedule, excluded from model fitting, outperforms the uniform control from two to five of six (p = 0.031). Within the evaluated strategy space, a parameter-weighted power mean with exponent 0.35 (95% confidence interval [0.20, 0.50]) accounts for 42.6% of within-architecture variance, compared with 36.4% for the arithmetic mean. Ranking candidates by this scalar without training improves the selected candidate’s mean percentile rank from the 26th to the 10th and reduces its mean accuracy shortfall relative to the best candidate from 3.70 to 1.12 points. The rate-related effect decays as the training budget increases; relative to tuned baselines, the overall strategy effect falls within the prespecified practical-equivalence margin in two of three architectures. A separate grouping-related contrast can persist at long budgets, although its direction and magnitude depend on the architecture.
[LG-70] DeltaHarness: Self-Evolving Harnesses for Resource-Adaptive Tool Execution
链接: https://arxiv.org/abs/2611.21688
作者: Lecheng Yan,Jianze Lin,Xicheng Han,Xiantao Xu,Wenxi Li,Chenyang Lyu
类目: Machine Learning (cs.LG); Performance (cs.PF)
*备注: preprint
Abstract:Tool-using agents rely on execution harnesses to organize and schedule tool operations. The same task and resource description can conceal runtime conditions that reverse the latency ordering of valid execution plans. Resolving this uncertainty requires firsthand execution in the target setting. We introduce DeltaHarness, an execution-guided policy-search method that adapts agent harnesses from cold-start measurements. Its QHI loop uses the observed response to update runtime dispatch, revise physical plans, and select experiments for later proposals. An independent confirmation panel controls policy replacement. With fixed dispatch, a revised materialization policy reduces complete-delivery latency by 30.0% on one video-editing task across three fresh pairs. On held-out tasks, the evolved policies reduce latency by 8.2% relative to the best fixed plan in one seed and by 4.6% relative to the initial proposal in another.
[LG-71] RTLReviewBench: LLM Hardware Reviewers Rarely Miss Bugs but Routinely Reject Correct RTL
链接: https://arxiv.org/abs/2611.21678
作者: Yasir Mahmud
类目: Hardware Architecture (cs.AR); Machine Learning (cs.LG)
*备注: 24 pages, 6 figures, 16 tables. Benchmark, model responses, and code: this https URL
Abstract:Hardware teams increasingly ask large language models (LLMs) to review register-transfer level (RTL) code, yet it is unclear whether a model can tell a correct module from a buggy one without simulating it. We introduce RTLReviewBench, a benchmark of 114 matched pairs of buggy and correct Verilog modules. Each pair shares one specification, most buggy modules are real failures of LLM code generators, and every label is confirmed by simulation. Evaluating 12 commercial and open-weight models, we find a consistent asymmetry: reviewers rarely accept buggy code but often reject correct code, producing more than five false rejections for every missed bug. Synthetic mutants hide part of the problem: on the same designs, models miss far fewer injected bugs than real ones. Asking a model to reason before it commits to a verdict sharply reduces false rejections for weaker reviewers, and a controlled ablation shows that where the reasoning appears in the output matters more than the instruction to reason, whereas sampling and majority voting do not help. A blinded audit further shows that some correct “buggy” verdicts rest on false technical claims. We release the benchmark, a harder split for future models, all model responses, and the evaluation code at this https URL
[LG-72] An inexact stochastic variance-reduced gradient method for constrained convex finite-sum optimization
链接: https://arxiv.org/abs/2611.23221
作者: Zhaosong Lu,Yifeng Xiao
类目: Optimization and Control (math.OC); Machine Learning (cs.LG)
*备注: Accepted by SIAM Journal on Optimization
Abstract:In this paper, we study a class of constrained convex finite-sum optimization problems. In many practical applications, all or some of the constraints are hard constraints that represent imperative requirements. Consequently, any desirable approximate solution must (nearly) satisfy these constraints. Motivated by this, we focus on computing an \epsilon - surely\ feasible\ stochastic\ optimal ( \epsilon -SFSO) solution, in which the constraint violation is deterministically bounded by \epsilon , and the expected optimality gap is at most \epsilon . While a recently proposed stochastic variance-reduced first-order penalty method can compute an \epsilon -SFSO solution, there remains a significant gap between its first-order oracle complexity for computing an \epsilon -SFSO solution and that for computing an \epsilon - expected\ feasible\ stochastic\ optimal solution, where both the expected constraint violation and the expected optimality gap are bounded by \epsilon . To narrow this gap, we propose an inexact stochastic variance-reduced gradient (SVRG) method that significantly improves the previously best-known first-order oracle complexity for computing an \epsilon -SFSO solution. Furthermore, when the objective consists of a single function, the proposed method reduces to a novel deterministic first-order method for solving standard convex conic optimization. It achieves a significantly lower gradient evaluation complexity with respect to the objective function compared to existing methods, and can outperform them when computing constraint gradients and projections onto the associated cone is cheaper than evaluating the objective gradient. Preliminary numerical results are presented to illustrate the performance of the proposed method.
[LG-73] Which Past Matters? Training History and Multi-Forecast Horizon in Equity Transformers
链接: https://arxiv.org/abs/2611.23126
作者: Chariton Chalvatzis
类目: atistical Finance (q-fin.ST); Machine Learning (cs.LG); Computational Finance (q-fin.CP)
*备注:
Abstract:More historical data can improve estimation while introducing relationships that transfer poorly to current market conditions. We therefore examine how training quantity and historical selection interact with forecast horizon in intraday equity prediction. We deploy twelve configurations which combine four sampling and validation policies with 6, 12 and 24 month training quantities under a fixed cross-sectional Transformer and feature set. A monthly walk-forward experiment produces 264 refits and eight cumulative return forecasts from 5 to 120 minutes. Evaluation covers January 2025 through 20 October 2026 and uses cross-sectional Spearman information coefficients against future five-factor residual returns. Six-month recency sampling leads individual models at 5-15 minutes, while multi recency configurations lead at 20-30 minutes and uniform configurations at 60-120 minutes. Increasing training quantity does not uniformly improve accuracy while an equal-weight ensemble of all twelve models exceeds every individual mean at 15, 20 and 25 minutes, but not at the remaining horizons. These findings support evaluating historical selection jointly with forecast horizon without identifying a universal policy or a causal benefit from recency.
[LG-74] Amortizing Tempered Sampling with Corrected Self-Training
链接: https://arxiv.org/abs/2611.23072
作者: Hoang-Viet Nguyen,Hai-Dang Dau,Alexandre H. Thiéry
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:Learning a neural sampler can reduce the cost of repeated sampling from an un-normalized distribution, but obtaining good training data remains difficult. We develop a corrected self-training scheme across a sequence of tempered targets, making use of importance sampling and Markov chain Monte Carlo to correct the samples and replay buffers to store them. We introduce Sequential Score Sampler (SSS) as a particular case. In SSS, the proposal is a score-based diffusion trained by regression on energy gradients; and the samples are corrected by importance weights on the path space. After training, samples can be collected directly from the model, from its importance-corrected proposals, or from the final-temperature buffer. The method amortizes repeated sampling by encoding the global structure of the target in the model. SSS achieves competitive sampling accuracy across the benchmarks and, on several targets, uses shorter runtimes with fewer energy evaluations than competing samplers.
[LG-75] Reversing the text is not reversing the path: a fixed-text measure without a zero and an assembly-order comparison that vanishes identically
链接: https://arxiv.org/abs/2611.23049
作者: Y.-H. Taguchi
类目: Physics and Society (physics.soc-ph); Statistical Mechanics (cond-mat.stat-mech); Machine Learning (cs.LG)
*备注: 41 pages. Ancillary files contain the measurement and verification scripts. Prepared with AI assistance (Anthropic Claude for drafting, revision and the verification scripts; OpenAI ChatGPT for a critical reading of a draft); all measurements were produced and verified by the author, who is responsible for the content
Abstract:For an ensemble of trajectories the fluctuation theorems give the entropy production its meaning and zero. The same expression has recently been evaluated on one fixed text, \Delta = [\log P(x) - \log P(\tilde x)]/L with \tilde x the text in reversed order, read as an entropy production separating causally from arbitrarily ordered prose. A single text is neither a distribution nor a dynamics, and the texts at issue were produced by other models. No theorem fixes where the zero of a fixed-text quantity lies: the expression carries over from an ensemble, its zero does not. We supply a zero from an exact antisymmetry. Scoring a text’s n sentences in all n! orders, reversal carries \pi to \pi \circ \mathrmrev , a measure-preserving involution, so \Delta(\pi) = -\Delta(\pi \circ \mathrmrev) identically and the values are centred at zero for any score. The authored arrangement’s midrank is then unitless, bounded, and 0.50 for a score blind to order. Across five autoregressive evaluators from 124M to 4B parameters, and taking the prompt-defined labels at face value, 8% to 15% of the median midrank’s distance from the null is attributable to causal order and the rest to the arrangement being the order the passage was produced in. The thermodynamic reading needs a comparison whose zero does hold on the text read, and a masked model supplies one: the any-order chain rule gives a genuine log-probability, and its left-to-right and right-to-left orders for one text give \sigma , a comparison of two assembly orders. Both run from the empty sequence to the same text, so it is not a path against its time reverse. It vanishes identically when the evaluator’s conditionals come from a joint, and a non-zero reading reports the evaluator’s incompatibility, not the text. Neither is a fixed-text entropy production, for different reasons. Comments: 41 pages. Ancillary files contain the measurement and verification scripts. Prepared with AI assistance (Anthropic Claude for drafting, revision and the verification scripts; OpenAI ChatGPT for a critical reading of a draft); all measurements were produced and verified by the author, who is responsible for the content Subjects: Physics and Society (physics.soc-ph); Statistical Mechanics (cond-mat.stat-mech); Machine Learning (cs.LG) Cite as: arXiv:2611.23049 [physics.soc-ph] (or arXiv:2611.23049v1 [physics.soc-ph] for this version) https://doi.org/10.48550/arXiv.2611.23049 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-76] Symmetries of Counterfactuals: Identification and Testability of Counterfactual Maps from Covariate Heterogeneity
链接: https://arxiv.org/abs/2611.22948
作者: Marc Braun,Adel Daoud,Jose M. Peña
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:In general, counterfactuals are not identified from data alone, because many causal models produce the same observed distributions but different counterfactuals. A common approach to circumvent this problem is to restrict the mechanism to a class in which it becomes unique, such as triangular monotone maps or optimal transport maps. However, whether the assumed class is correct is usually not examined. In this work, we study what the data can tell us when covariates are available that influence the outcome only through its latent cause. First, we characterise all counterfactual maps that are consistent with the data. They differ from the true map by a group of symmetries that preserve the conditional outcome distributions given the covariates. The more these distributions vary across covariate values, the smaller the group. Second, we show that for the classes above, consistency with the data is equivalent to a single conditional independence, which yields a test of the assumed class. We illustrate both results on synthetic data and on the ACTG 175 HIV trial.
[LG-77] Robust Actor-Critic Learning under Transport and Divergence Uncertainty
链接: https://arxiv.org/abs/2611.22759
作者: Kunqi Huang,Lican Kang,Hongrui Shu,Jincheng Ying
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:Learning reliable control policies requires accounting for uncertainty in transition dynamics and errors introduced by function estimation and action generation. We develop an actor–critic method for discounted robust Markov decision processes with continuous states and actions and an unknown nominal transition law. We consider transition uncertainty sets defined by Wasserstein distance, total variation, Kullback–Leibler divergence, or Pearson chi-square divergence. A conditional diffusion model is fitted once to offline transitions. At each subsequent iteration, we estimate the robust action-value function from current-policy observations using a shared-target minimax criterion, then update a generative actor through exponentially weighted score matching. Regularization preserves exploration and provides the density regularity needed for conditional distribution estimation. Under coverage and regularity conditions, we establish finite-sample bounds for critic estimation, the learned conditional action law, and the robust return of the implemented policy. The bounds distinguish critic estimation error, uncertainty in the nominal transition model, and policy regularization, and characterize how the four robustness radii affect the resulting guarantees.
[LG-78] A Quantum Support Vector Regression Framework for Missing Data Imputation in Ship Performance Data
链接: https://arxiv.org/abs/2611.22635
作者: Daniel Maroto Sanchez,Christian Velasco-Gallego,Miriam Bastante Chichon,Jaime Álvarez-Benayas,Tsoulakos Nikolaos,J. A. Bravo-Montes
类目: Quantum Physics (quant-ph); Machine Learning (cs.LG)
*备注:
Abstract:The Internet of Ships enables vessels to generate vast amounts of data, yet missing values remain a persistent issue. To the best of the authors’ knowledge, existing imputation techniques in this domain rely exclusively on classical machine learning. This study investigates quantum support vector regression (QSVR) for imputing missing values in ship performance data. Classical support vector regression and QSVR models based on different quantum feature maps are compared for estimating marine fuel consumption under varying levels of data availability. The case study uses real-world data from a Panamax bulk carrier, covering 37 variables across six categories. Three experiments are conducted: exploratory analysis with three input features, scalability analysis with increasing dimensionality (3 to 18 features), and comparative evaluation under a selected representative configuration. Across all scenarios, both classical and quantum approaches captured the relationship between the inputs and the main-engine fuel-flow rate, maintaining strong imputation performance even under severe data scarcity. ZZFeatureMap_1 achieved the highest average R2 (0.991 in the performance comparison conducted with 60% of the dataset, 90% missing data, and 14 input features, for instance) and the lowest aggregate imputation errors, while the radial basis function kernel was the strongest classical baseline, delivering highly competitive results.
[LG-79] Balancing Exploration and Exploitation in Constrained Combinatorial Optimization with U(1)-Symmetric Tensor Trains ICML2026
链接: https://arxiv.org/abs/2611.22557
作者: Sergei Iudin,Mikhail Podobrii,Dmitry Zheltkov,Stanislav Moiseev
类目: Optimization and Control (math.OC); Machine Learning (cs.LG)
*备注: 39 pages, including supplementary material. Code: this https URL . Extends an earlier version presented at the non-archival CoLoRAI workshop at ICML 2026
Abstract:Tensor Train (TT) decomposition has emerged as a powerful tool for mitigating the curse of dimensionality in a variety of machine learning and combinatorial optimization tasks. Recent advances show that TT-based generative models can efficiently represent and explore high-dimensional discrete spaces, even when the cost function structure is unknown. However, incorporating hard linear constraints into TT-based optimization remains challenging, as existing interpolation and sampling approaches often fail when feasible points are rare. In this work, we develop a general TT-based framework for constrained combinatorial optimization built upon U(1)-symmetric tensor networks. Our method introduces an expressibility-enhancement mechanism for constructing sparse symmetric tensors and a graph-based diversity measure that guides sample selection during training. On assignment benchmarks, expressibility enhancement and diversity-aware selection improve mean solution quality over their respective baselines. The assignment-selected diversifying strategy also extends to nonpolynomial Set Partitioning, with larger relative gains observed in the higher-tensor-rank data families. Separate runtime benchmarks show substantial speedups over ConstrainTNet and PROTES on non-degenerate assignment workloads; comparisons with simulated annealing depend on constraint information and computational budget, but, overall, support the practicality of the current implementation.
[LG-80] mporal Diffusion Models for Joint Signal-Channel Estimation under Gaussian and α-Stable MIMO Channels
链接: https://arxiv.org/abs/2611.22548
作者: Rawad Kansoh,Bahaa Ammoury,Charbel Hannoun,Jihad Fahs,Ibrahim Abou-Faycal
类目: ignal Processing (eess.SP); Machine Learning (cs.LG)
*备注: 14 pages, 10 figures
Abstract:Diffusion models have emerged as powerful tools for learning complex data distributions and solving inverse problems. However, most diffusion models rely on the Gaussian distribution, which does not accurately represent heavy-tailed data. In this work, we study diffusion-based reconstruction under Gaussian and symmetric \alpha -stable fading channels, while also exploring the benefits of using possible temporal dependence in the channel. We consider two variants: pilot-based channel-only estimation and joint image-channel recovery for MIMO systems. Our approach is based on several standard and score-based diffusion models for both Gaussian and \alpha -stable distributions. We refine these models by guiding the reverse diffusion process using the observation likelihood and show how a single trained diffusion model can be reused for the location-scale family of distributions without retraining. In blind MIMO recovery, separate diffusion models are used for the image and channel, and their estimates are combined through likelihood guidance. For each setting, we compare independent and temporal reconstruction, where information from previous observations is used to improve the estimation. We show that, under a time-dependent channel model, such a temporal reconstruction improves channel estimation and the quality of the reconstructed image. Finally, we show that, for the same heavy-tailed channel data, using an \alpha -stable diffusion prior provides better channel estimation and image reconstruction than using a Gaussian diffusion prior.
[LG-81] Efficient Machine Learning Interatomic Potentials
链接: https://arxiv.org/abs/2611.22444
作者: Yury Lysogorskiy,Anton Bochkarev,Ralf Drautz
类目: Materials Science (cond-mat.mtrl-sci); Machine Learning (cs.LG)
*备注:
Abstract:Machine learning interatomic potentials (MLIPs) reach an accuracy close to quantum mechanical reference calculations at a small fraction of the cost. Efficiency is the purpose of MLIPs, yet strategies for efficient MLIPs have not been reviewed systematically. This review takes efficiency as its guideline. To compare MLIPs on a common footing, we represent them in a complete basis expansion, the Atomic Cluster Expansion (ACE) and its graph generalization (GRACE). The expansion is canonical, built from products of single-particle functions. Classical interatomic potentials, such as the embedded-atom method, and many modern MLIPs, including equivariant transformers, can be represented within this expansion. Much of the apparent diversity of MLIP architectures therefore reflects how non-linearity, layer coupling and optimization are handled rather than which interactions the models can represent. We show how the cost of an MLIP becomes linear in system size and, at fixed rank of the tensor decomposition, in model complexity, and we discuss the role of machine learning, which provides non-linear parameterizations and optimization. Because efficiency is meaningful only at a given accuracy, we also compare accuracy and cost. On the Matbench Discovery leaderboard, models with three to five layers approach the accuracy of deeper models, and one basis, truncated along the Pareto front of error and speed, spans more than two orders of magnitude in cost. Ultimately, training data determine the quality of any MLIP.
[LG-82] Discovery of new antiviral agents through artificial intelligence: in vitro and in vivo results
链接: https://arxiv.org/abs/2611.22230
作者: Roza Izmailyan,Mher Matevosyan,Hamlet Khachatryan,Anastasiya Shavina,Smbat Gevorgyan,Artur Ghazaryan,Irina Tirosyan,Yeva Gabrielyan,Marusya Ayvazyan,Boris Martirosyan,Vardan Harutyunyan,Hovakim Zakaryan
类目: Biomolecules (q-bio.BM); Machine Learning (cs.LG)
*备注: Accepted manuscript. 13 pages, 5 figures; supplementary material included (supplementary text, Tables S1-S2, Figures S1-S5). Published in Antiviral Research 222 (2024) 105818
Abstract:In this research, we employed a deep reinforcement learning (RL)-based molecule design platform to generate a diverse set of compounds targeting the neuraminidase (NA) of influenza A and B viruses. A total of 60291 compounds were generated, of which 86.5% displayed superior physicochemical properties compared to oseltamivir. After narrowing down the selection through computational filters, nine compounds with non-sialic acid-like structures were selected for in vitro experiments. We identified two compounds, DS-22-inf-009 and DS-22-inf-021 that effectively inhibited the NAs of both influenza A and B viruses (IAV and IBV), including H275Y mutant strains at low micromolar concentrations. Molecular dynamics simulations revealed a similar pattern of interaction with amino acid residues as oseltamivir. In cell-based assays, DS-22-inf-009 and DS-22-inf-021 inhibited IAV and IBV in a dose-dependent manner with EC50 values ranging from 0.29 uM to 2.31 uM. Furthermore, animal experiments showed that both DS-22-inf-009 and DS-22-inf-021 exerted antiviral activity in mice, conferring 65% and 85% protection from IAV (H1N1 pdm09), and 65% and 100% protection from IBV (Yamagata lineage), respectively. Thus, these findings demonstrate the potential of RL to generate compounds with promising antiviral properties.
[LG-83] Flocking Time and Graph Volume in Nonlinear Cucker-Smale Models with Dynamic Networks
链接: https://arxiv.org/abs/2611.22204
作者: Sooyeon Song,Jea-Hyun Park
类目: Dynamical Systems (math.DS); Machine Learning (cs.LG); Adaptation and Self-Organizing Systems (nlin.AO)
*备注: 14 pages, 3 figures. Accepted manuscript of the article published in J. Inf. Commun. Converg. Eng. 24(2), 155-164 (2026)
Abstract:This study proposes a Cucker-Smale model incorporating dynamic networks and the discrete p -Laplacian ( p \in (1,2) ) and analyzes the relationship between flocking time constraints and graph connectivity from both theoretical and experimental perspectives. We establish that this model achieves flocking within finite time and derive a quantitative relationship between the lower bound of flocking time and graph volume. Based on these results, we observe that imposing sufficiently small flocking time induces an increase in network volume, and that larger initial velocity variance requires greater volume under the same time constraint. In our experiments, we trained a transformer-based PPO policy to select network structures at each time step. Under the most stringent time constraint, the maximum average number of connections increased sharply, confirming a threshold effect. The negative correlation between convergence time and average number of connections reflects the theoretical results well. This study provides quantitative evidence for the phenomenon that time constraints induce connection mobilization.
[LG-84] Bandit Online Submodular Minimisation Using Zeroth-Order Oracles
链接: https://arxiv.org/abs/2611.22130
作者: Amir Ali Farzin,Philipp Braun,Iman Shames
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:We consider online submodular minimisation under bandit feedback, where at each round a decision maker plays a subset and observes only its cost, while the cost functions, and hence their minimiser, change over time. Existing bandit methods only guarantee static regret, which compares against a single fixed subset and is a weak benchmark when the minimiser moves. This paper provides, to the best of our knowledge, the first method with guarantees for tracking a moving minimiser in this setting. We propose a zeroth-order method built on a two-point oracle for the Lovász extension whose two evaluations are both subsets that can be played. The perturbed and base points are rounded by a common random threshold, and drawing the smoothing direction from the unit sphere makes the base evaluation a zero-mean control variate, so the oracle is an unbiased estimator of the smoothed gradient. We establish dynamic regret bounds of order O\big((1+\sqrtP^\ast_N)N^3/4\big) in general and O\big((1+\sqrtP^\ast_N)N^2/3\big) under a sufficiently small drift, where P^\ast_N is the path length of the minimiser sequence, together with the corresponding O(N^3/4) and O(N^2/3) static regret bounds. Numerically, the method attains lower regret than the algorithm of (Hazan and Kale (2012)), with the advantage growing with the path length.
[LG-85] A single inference process underlies the behavior and neural circuits of working memory
链接: https://arxiv.org/abs/2611.22054
作者: Dongyu Gong,Mario Belledonne,Ilker Yildirim
类目: Neurons and Cognition (q-bio.NC); Machine Learning (cs.LG)
*备注: 39 pages
Abstract:Working memory holds past inputs online for future behavior. Despite decades of research, accounts of working memory remain divided across levels of analysis, with long-standing debates on whether its capacity reflects discrete slots or continuous resources, how attention improves recall, and whether its representations rely on persistent neural firing or activity-silent states. Here we show that MemGP, a theory that is specified at the computational, algorithmic, and circuit levels of analysis simultaneously, can explain diverse behavioral and neural phenomena in working memory. At the computational level, MemGP casts working memory as probabilistic inference over a continuous function on feature space, with a variational objective that balances memory strength against a Gaussian-process prior. At the algorithmic level, writing, storing, and reading of memories correspond to approximating this posterior with a finite set of inducing points, the resource that all items share. In simulations, MemGP reproduces load-dependent precision loss, attentional benefits during maintenance, and inter-item attraction and repulsion, in each case revealing multiple complex computational dependencies. At the circuit level, the algorithm analytically (without training) maps onto a recurrent circuit, including its architecture and dynamics. Beyond reproducing neural phenomena such as delay-dependent drift, this analytical map deduces center–surround connectivity, a property long assumed in attractor models, and a structured combination of persistent activity and synaptic weights to hold memories. Thus, a single inference objective not only jointly predicts the behavioral and neural signatures of working memory, but also addresses, across several instances and different levels of analysis, existing competing explanations.
[LG-86] Polylogarithmic Regret via Certainty Equivalence in Degenerate LQR
链接: https://arxiv.org/abs/2611.21927
作者: Luca Vignola,Melanie Zeilinger,Bruce D. Lee
类目: Optimization and Control (math.OC); Machine Learning (cs.LG); Systems and Control (eess.SY)
*备注: 33 pages, 1 figure
Abstract:Singular input penalties, as in minimum variance control, can reduce what must be identified to control an unknown linear system. In online linear quadratic control, certainty equivalence can achieve polylogarithmic regret when the optimal controller is identifiable from data collected under optimal feedback. With a positive definite input penalty, however, this condition is restrictive, since for an invertible state matrix it requires all unknown parameters to be identifiable under optimal feedback. We show that singular input penalties relax this requirement, expanding the class of problems in which certainty equivalence achieves polylogarithmic regret. Specifically, we prove that a certainty-equivalence algorithm with a short initial exploration phase achieves O(\log^2 T) expected regret over T interactions whenever controller identifiability holds. We complement this guarantee with an \Omega(\sqrtT) minimax lower bound for a family containing a system where controller identifiability fails. For minimum variance control, knowing the range of the input matrix suffices for controller identifiability, a weaker requirement than knowing the full input matrix, as assumed in prior work with positive definite input penalties. This condition also covers underactuated systems, as we illustrate with a cart-pole example.
[LG-87] Bias-Reduced and Parallel Heavy-Tailed Stochastic Convex Optimization in ell_p Geometries
链接: https://arxiv.org/abs/2611.21844
作者: Zhengyun Nie,Anant Shyam,Brian Bullins
类目: Optimization and Control (math.OC); Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:We consider the fundamental problem of parallel stochastic convex optimization (SCO), whose aim is to approximately minimize a convex objective when given the ability to query a moment-bounded stochastic gradient oracle multiple times in parallel per round. Whereas most previous results have focused on the variance-bounded Euclidean ( \ell_2 -norm) setting, meaning the objective is Lipschitz with respect to the \ell_2 -norm over an \ell_2 -bounded feasible set in \mathbbR^d , in this work we study parallel SCO in \ell_p settings for 2 \le p \infty under a heavy-tailed noise model. We begin by introducing a new class of \ell_p -optimal-distance-convergence (ODC _p ) algorithms, drawing on prior stochastic bias-reduced techniques [ACJ+21], which combine with multilevel Monte Carlo methods to efficiently estimate the optimum of uniformly convex objectives. As a consequence of our ODC _p algorithms, we first establish rates for projection-efficient \ell_p -Lipschitz optimization, composite convex minimization, and efficient \ell_p -proximal point estimation. Combining our ODC _p algorithms with a new \ell_p -generalized Gaussian-based stochastic gradient estimator ( p -GREG), we provide two efficient \ell_p parallel algorithms, EpochDA and \ell_p -AC-SA, which achieve optimization error \epsilon 0 with at most \tildeO(d^\fracp2p + 2 \epsilon^-\fracpp + 1) query depth, improving upon the sequential heavy-tailed stochastic query depth of O(\epsilon^-p) for \epsilon \leq \tildeO(d^-1/(2p)) . We also establish dimension-dependent lower bounds of \Omega\bigl(\min\epsilon^-p,d^p/(3p-2)\bigr) adaptive rounds for nonsmooth convex optimization over a subset of the unit \ell_p ball. Our results improve upon the \Omega(\min\epsilon^-p,\epsilon^2/3d^1/3) bounds of [DG19], when adapted to the same feasible set.
[LG-88] Quantum-Classical Separation in Probability Distribution Transport via Nonlocal Games
链接: https://arxiv.org/abs/2611.21840
作者: Fangjun Hu,Xun Gao
类目: Quantum Physics (quant-ph); Computational Complexity (cs.CC); Machine Learning (cs.LG)
*备注: 7+15 pages, 2 figures, 1+5 tables
Abstract:In many-body physics and generative artificial intelligence, locality provides a natural notion of phase equivalence: two quantum states or classical probability distributions are regarded as in the same phase when local dynamics can reversibly transform one into the other. This raises a fundamental question: can local quantum dynamics transport classical probability distributions in ways that local classical dynamics cannot? We establish a quantum-classical separation in distribution transport: two classical distributions can be connected exactly and reversibly by a constant-depth local quantum circuit, whereas classical stochastic transport requires a substantially larger spatial range. We demonstrate this separation explicitly on a one-dimensional chain and develop a general construction principle on graphs, based on nonlocal games. Moreover, the source distribution can itself be generated by a simple local classical process, establishing a quantum advantage in distribution generation as well. Our results reveal a fundamental distinction between classical and quantum local reversibility, even when both the initial and final states are entirely classical.
[LG-89] Improving Early Detection of Scientifically Interesting Radio Sources Using Enhanced Isolation Forest
链接: https://arxiv.org/abs/2611.21674
作者: Pfunzo Thavhana,Clement Nyirenda,Michelle Lochner
类目: Instrumentation and Methods for Astrophysics (astro-ph.IM); Machine Learning (cs.LG)
*备注: 7 pages, 6 figures
Abstract:The rapid growth of radio astronomical surveys has created a major challenge in identifying scientifically interesting sources from large populations of observed objects. This study investigates whether enhanced Isolation Forest (iForest) methods can improve the early detection and ranking of scientifically interesting radio sources. Four iForest variants, Standard Enhanced, Pooled Gain, Kurtosis-based, and Hypervolume, are evaluated using 1,031 extended radio sources from the MeerKAT Galaxy Cluster Legacy Survey (MGCLS). Each source is represented by a 52-dimensional feature set derived using Bootstrap Your Own Latent (BYOL) self-supervised learning and principal component analysis (PCA), while expert scientific-interest ratings are used to define the ground-truth labels. The methods are evaluated using Average Precision (AP), receiver operating characteristic area under the curve (ROC AUC), Precision@10, Precision@50, and cumulative anomaly detection. Pooled Gain achieves the highest Average Precision (AP = 0.201) and the strongest early-stage ranking performance (P@10 = 0.600), identifying six expert-validated anomalies among the ten highest-ranked sources. Standard Enhanced and Kurtosis-based achieve the highest ROC AUC (0.677), while the AUC values of the four variants range from 0.653 to 0.677. These results indicate that feature-aware isolation strategies can improve the prioritization of scientifically interesting sources at the top of the ranked list, although overall discrimination remains comparable across the evaluated variants. The findings support the use of enhanced Isolation Forest methods as an initial ranking mechanism within human-in-the-loop systems for large-scale radio astronomical surveys.
附件下载


