本篇博文主要内容为 2026-06-01 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 (2026-06-01)
今日共更新761篇论文,其中:
- 自然语言处理共156篇(Computation and Language (cs.CL))
- 人工智能共226篇(Artificial Intelligence (cs.AI))
- 计算机视觉共147篇(Computer Vision and Pattern Recognition (cs.CV))
- 机器学习共265篇(Machine Learning (cs.LG))
- 多智能体系统共14篇(Multiagent Systems (cs.MA))
- 信息检索共20篇(Information Retrieval (cs.IR))
- 人机交互共27篇(Human-Computer Interaction (cs.HC))
多智能体系统
[MA-0] Dreaming Of Others: Latent Teammate Modeling In World Models For Multi-Agent Reinforcement Learning
【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）中因队友内部策略与意图不可直接观测而带来的协作不确定性问题。现有基于世界模型（如Dreamer）的方法在单智能体场景下表现出优异的泛化能力和样本效率，但在多智能体环境中受限于无法有效建模队友行为带来的不确定性。其解决方案的关键在于将队友视为智能体世界模型中的可学习、结构化的组成部分：提出一种分解式潜空间架构，将Dreamer风格的递归状态空间模型（Recurrent State-Space Model, RSSM）的潜状态显式分解为环境成分与队友成分，并引入一个辅助的“心智理论”（Theory-of-Mind, ToM）头，用于从部分轨迹中推断出队友的行为嵌入（包括性格、意图及预测动作）。这些队友潜变量作为条件信息作用于策略（actor）和价值函数（critic），使智能体能够模拟并适应多样化的合作者。该方法支持在部分可观测环境下实现零样本（zero-shot）与少样本（few-shot）协作，同时提出了相应的基准测试与评估协议以验证其有效性。本研究将世界模型从单纯的环境动态预测器拓展为社会行为模拟器，为构建通用性更强、具备人类兼容性的智能系统开辟了新路径。
链接: https://arxiv.org/abs/2605.31361
作者: Tomas Leroy-Stone
机构: 未知
类目: Multiagent Systems (cs.MA); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 5 pages, 2 figures. Accepted as a poster at the 2026 World Modeling Workshop. Conceptual workshop paper
Abstract:In cooperative multi-agent reinforcement learning (MARL), agents must coordinate with partners whose internal policies and intentions are not directly observable. While world models such as Dreamer have demonstrated strong generalization and sample efficiency in single-agent settings, their application to MARL remains limited by an inability to handle teammate-induced uncertainty. We propose a new perspective: treat teammates as structured, learnable components within the agent’s world model. We introduce an architecture that factorizes the latent state of a Dreamer-style recurrent state-space model (RSSM) into environment and teammate components, and learns an auxiliary Theory-of-Mind (ToM) head to infer latent embeddings of partner behavior such as character, intent, and predicted actions from partial trajectories. These teammate latents condition the actor and critic, enabling the agent to imagine and adapt to diverse collaborators. We outline how this approach can support zero-shot and few-shot coordination in partially observable settings and propose a set of benchmarks and evaluation protocols to assess its impact. This work positions world models as not only predictors of environmental dynamics, but as simulators of social behavior, opening new directions for generalizable, human-compatible AI.
[MA-1] Social welfare optimisation under institutional reward and punishment
【速读】：该论文旨在解决现有制度激励设计中长期存在的一个关键问题：在有限且充分混合的群体中，针对社会困境（如捐赠博弈和公共品博弈）所设计的激励机制，虽然通常以最小化制度成本并最大化合作频率为目标，但其对社会福利（即总群体收益减去制度支出）的优化效果尚未得到系统性研究。论文提出一种以社会福利为中心的制度激励框架，综合考虑对合作者的奖励与对背叛者的惩罚，并推导出期望社会福利的显式表达式，揭示其如何依赖于激励效率与选择强度。分析表明，在某些参数区域内，社会福利存在单一最优激励水平；而在其他区域则出现定性相变，导致社会福利非单调并具有多个局部极值。研究进一步证明，任何最大化社会福利的激励策略要么为零，要么集中于一个简洁的闭式目标值，并据此提出高效算法以计算最优解。通过对比奖励与惩罚机制，论文还推导出在给定预算下奖励优于惩罚的闭式条件。总体而言，研究揭示了以成本或合作频率优化的激励机制与真正最大化社会福利的机制之间存在系统性差距。
链接: https://arxiv.org/abs/2605.31330
作者: Van An Nguyen,Vuong Khang Huynh,Huu Loi Bui,Hai Anh Ha,Quang Dung Le,Tan Dat Nguyen,Ngoc Ngu Nguyen,Zhao Song,Manh Hong Duong,Le Hong Trang, TheAnh Han
机构: Ho Chi Minh City University of Technology (HCMUT), Vietnam; Vietnam National University - Ho Chi Minh City (VNU-HCM), Vietnam; Teesside University, United Kingdom; University of Birmingham, United Kingdom
类目: Computer Science and Game Theory (cs.GT); Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA); Optimization and Control (math.OC); Adaptation and Self-Organizing Systems (nlin.AO)
备注:
Abstract:Institutional incentives are widely used to promote cooperation among autonomous, self-regarding agents, from human societies to multi-agent and AI systems. Existing work typically treats incentive design as a bi-objective problem: minimise institutional cost while achieving a high long-run frequency of cooperation. Whether such schemes also maximise social welfare - total population payoff net of institutional expenditure - has remained largely unexplored. We develop a welfare-centric framework for institutional incentives in finite, well-mixed populations playing a social dilemma (Donation Game and Public Goods Game), considering both rewards for cooperators and punishments for defectors. For each mechanism, we derive explicit expressions for expected social welfare and characterise how it depends on incentive efficiency and selection intensity. Analytically, we identify parameter regimes where social welfare has a single optimal incentive level and regimes with qualitative phase transitions, in which welfare becomes non-monotonic with multiple local optima. We prove that any welfare-maximising incentive is either zero or concentrated around a simple closed-form target, and we provide an efficient algorithm to compute these optima. Comparing reward and punishment, we further derive close-formed conditions under which reward outperform punishment in terms of social welfare for any given budget. Overall, our results reveal a systematic gap between incentives optimised for cost or cooperation frequency and those that maximise welfare.
[MA-2] Generalized Intention Modeling in Multi-Agent Reinforcement Learning
【速读】：该论文旨在解决非合作、竞争性且一般和（general-sum）多智能体强化学习中对手意图建模的挑战，其核心问题是现有方法依赖于预先选定的、固定不变的对手信息（如下一步动作或未来环境状态）作为意图嵌入，而这些信息在不同任务和环境中并不具备普遍代表性。为克服这一局限，论文提出一种任务自适应的对手建模框架，通过学习多个意图表示的性能驱动混合模型来动态适应不同场景。其解决方案的关键在于引入一种基于最大互信息原则的新意图表示方法，该方法显式最大化与自代理未来回报之间的互信息，从而捕获对自身性能最相关的对手信息。实验表明，该方法在多种任务上均能稳定达到或超越当前最优基线性能，并揭示了不同建模策略在特定情境下成功的原因。
链接: https://arxiv.org/abs/2605.31318
作者: Mateusz Odrowaz-Sypniewski,Jasmine Bayrooti,Ajay Shankar,Amanda Prorok
机构: University of Cambridge(剑桥大学)
类目: Machine Learning (cs.LG); Multiagent Systems (cs.MA)
备注:
Abstract:Modeling an opponent’s intent is critical for effective decision-making in non-cooperative, competitive, and general-sum multi-agent reinforcement learning. Existing opponent modeling methods encode intent using an embedding derived from episode information chosen a priori, such as the opponent’s next action or a future environment state, and use this to guide the ego-agent’s behavior. These approaches assume that the chosen information is universally representative of intent; however, we show empirically that this is not the case as intentions are often task- and environment-dependent. To address this, we introduce a task-adaptive opponent modeling framework that learns a performance-driven mixture of multiple intent representations. We further introduce a new intention representation that maximizes mutual information with the ego-agent’s future returns, thereby capturing opponent information that is most directly relevant to performance. Our approach consistently matches or exceeds the performance of state-of-the-art baselines across diverse tasks and yields insights into when and why different opponent modeling strategies succeed.
[MA-3] HADT: A Heterogeneous Multi-Agent Differential Transformer for Autonomous Earth Observation Satellite Cluster ECML-PKDD2026
【速读】：该论文旨在解决异构卫星集群在执行地球观测（Earth Observation, EO）任务时的自主资源管理问题，尤其针对光学与合成孔径雷达（Synthetic Aperture Radar, SAR）卫星混合运行场景下的动态环境不确定性。传统调度方法依赖于精确的数学模型进行建模并结合优化算法求解，但在实际空间任务中，由于环境高度动态、模型复杂且难以准确刻画，此类方法的适应性与有效性显著下降。为此，本文提出将问题重构为序列决策过程，并采用无模型强化学习（model-free reinforcement learning）技术实现自适应、实时的资源分配。其解决方案的关键在于设计一种基于Transformer的新型架构，引入关系型观测-动作标记化（relational observations-actions tokenization）与差分注意力机制（differential attention mechanism），以有效捕捉异构卫星间复杂的交互关系和时空依赖性。实验结果表明，该方法在性能上显著优于现有基线，且具备良好的可扩展性与跨不同卫星集群规模的迁移能力。
链接: https://arxiv.org/abs/2605.31023
作者: Mohamad A. Hady,Muhammad Anwar Masum,Siyi Hu,Mahardhika Pratama,Jimmy Cao,Ryszard Kowalczyk
机构: 未知
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Multiagent Systems (cs.MA)
备注: Accepted in ECML-PKDD 2026. arXiv admin note: text overlap with arXiv:2511.12792
Abstract:This work addresses the problem of autonomous resource management in heterogeneous satellite cluster conducting Earth Observation (EO) missions including optical and Synthetic Aperture Radar (SAR) satellites. In autonomous operation mode, satellites are equipped with intelligent capabilities enabling real-time decision-making based on the latest conditions, while requiring minimal interaction with ground operators. Traditional scheduling approaches typically rely on mathematical models to represent satellite mission and resource management. Then, this problem is solved by using optimization algorithms. However, such solutions become less effective when the underlying models are not available, over complex, and inaccurate due to dynamic changes and uncertainties inherent in the space mission environment. A promising alternative is to reformulate the problem as a sequential decision-making process and apply model-free reinforcement learning techniques to enable adaptive and real-time resource management. To this end, we propose a novel transformer-based architecture tailored for heterogeneous satellite cluster autonomous EO Mission with relational observations-actions tokenization and differential attention mechanism. Our experimental results demonstrate significant performance improvements compared to the available baselines. Moreover, the proposed architecture exhibits strong adaptability and transferability with respect to varying numbers of satellite clusters.
[MA-4] Safe Equilibrium Policy Optimization for Strategic Agent Policies EMNLP2026
【速读】：该论文旨在解决语言模型在强化学习微调后忽视多智能体战略结构的问题，尤其是在自然语言描述的游戏状态与自由形式动作生成的交互界面下，容易出现策略性失效模式，如利用弱对手、协同达成有害均衡以及外部化成本等。其解决方案的关键在于提出安全均衡策略优化（Safe Equilibrium Policy Optimization, \sepo），通过在期望收益中显式引入对可被利用性（exploitability）、合谋风险（collusion risk）和外部性成本（externality cost）的惩罚项，以约束智能体行为。\sepo作为奖励信号集成至组相对策略优化（Group Relative Policy Optimization, GRPO）框架中，应用于经过监督微调（SFT）后的Gemma 4 E4B-it与Qwen 3.5-4B模型。实验覆盖五类战略场景：重复囚徒困境、重复拍卖、两种谈判变体及Kuhn扑克，结果表明，\sepo在Kuhn扑克中使两模型均实现零可被利用优势，在四个领域优于基线模型的安全表现，并纠正了SFT引入的过度合作行为；在谈判任务中，\sepo实现了正向安全性结果且所有配置下的归一化相对优势均为正值。消融实验进一步验证了每回合的可被利用性计算必要性——固定常数惩罚项在GRPO的优势归一化过程中会相互抵消（常数控制变量特性），导致梯度为零，无法有效驱动优化。为促进战略安全领域的研究，作者公开了代码与SFT数据集。
链接: https://arxiv.org/abs/2605.30854
作者: Karthika Arumugam,Kiran Kumar Manku,Amit Dhanda
机构: Amazon(亚马逊)
类目: Multiagent Systems (cs.MA); Artificial Intelligence (cs.AI)
备注: Submitted to EMNLP 2026
Abstract:Language models fine-tuned with reinforcement learning typically optimize for task reward, ignoring multi-agent strategic structure. Because these agents condition on natural language game-state descriptions and emit actions through free-form generation, strategic failure modes – exploiting weaker opponents, coordinating on harmful equilibria, and externalizing costs are inseparable from the language interface itself. We propose Safe Equilibrium Policy Optimization (\sepo), a training objective that augments expected payoff with explicit penalties for exploitability, collusion risk, and externality cost. We implement \sepo as a reward signal for Group Relative Policy Optimization (GRPO), applied to Gemma~4 E4B-it and Qwen~3.5-4B after supervised fine-tuning (SFT). Evaluated across five strategic domains: Iterated Prisoner’s Dilemma, repeated auctions, two negotiation variants, and Kuhn Poker. \sepo achieves zero exploit-pool advantage in Kuhn Poker for both models, outperforms the base model on safety in four domains, and corrects the over-cooperative behavior introduced by SFT. In negotiation, \sepo achieves a positive-safety outcome and only the positive normalized relative advantage of any negotiation configuration. Ablation experiments confirm that per-rollout exploit computation is necessary: a shared constant penalty cancels in GRPO advantage normalization (constant control-variate property), producing zero gradient. To support further research in strategic safety for agents, we release our \hrefthis https URLcode and SFT datasets.
[MA-5] Design and Evaluation of Multi-Agent AI Oracle Systems for Prediction Market Resolution
【速读】：该论文旨在解决预测市场（prediction market）中结果判定（outcome resolution）的可靠性问题，即如何在保证高效性的同时实现高精度的结果判断。现有预言机系统面临自动化程度高但鲁棒性差，或依赖人工仲裁但成本高昂之间的权衡。单模型大语言模型（LLM）预言机虽具备一定准确性，但其错误模式无法自我修正，存在固有缺陷。为此，本文提出以多智能体大语言模型架构替代单模型基线，探索通过协同推理提升预言机准确率的可能性。关键解决方案在于设计两种多智能体机制：独立聚合（independent aggregation）与协商共识（deliberative consensus）。实验表明，基于置信度加权投票的独立聚合策略取得最高准确率83.43%，优于最优单模型（如GPT-5 Nano）1.01个百分点；而协商共识因错误传播导致准确率下降至约76%，低于所有单模型基线。研究发现，各模型间存在较高的错误相关性（0.529–0.689），限制了集成方法接近理论上的康多塞上限（Condorcet ceiling），从而揭示了当前多智能体方法的固有瓶颈。因此，论文进一步提出一种混合式AI-人类预言机路由机制：仅对全体一致且置信度高的问题进行自动判定，可实现97.87%的准确率，覆盖47%的数据集，其余分歧样本则触发人工仲裁，有效平衡了效率与可靠性。
链接: https://arxiv.org/abs/2605.30802
作者: Tarun Kota
机构: Yale University (耶鲁大学)
类目: Multiagent Systems (cs.MA); Artificial Intelligence (cs.AI)
备注: 34 pages, 11 figures
Abstract:Prediction markets aggregate collective intelligence to forecast uncertain events, but their utility depends on reliable outcome resolution. Existing oracle systems tradeoff fast but brittle automation against accurate but costly human arbitration. Single-LLM oracles achieve meaningful accuracy but inherit all failure modes of their underlying model with no self-correction mechanism. We evaluate whether multi-agent LLM architectures can improve oracle resolution accuracy over single-model baselines. We compare independent aggregation and deliberative consensus against single-LLM baselines (GPT-5 Nano, DeepSeek V3, and Llama-3.3-70B) on 1,189 resolved prediction market questions from KalshiBench. All agents share a common evidence layer through Exa, with retrieval filtered by publication date to isolate reasoning from retrieval quality. Independent aggregation with confidence-weighted voting achieves the highest accuracy at 83.43 percent, outperforming the best individual model by 1.01 percentage points. Deliberative consensus degrades accuracy to approximately 76 percent, below every single-model baseline, attributed to error propagation during debate where confidently wrong models flip correct ones. Error correlations across models (0.529-0.689) explain why aggregation gains fall short of the theoretical Condorcet ceiling, placing a fundamental limit on ensemble approaches. Many questions resist correction by any multi-agent architecture, motivating escalation to human arbitration. We propose routing criteria for hybrid AI-human oracle systems: auto-resolving only unanimous, high-confidence questions yields 97.87 percent accuracy on 47 percent of the dataset, with inter-agent disagreement flagging the remainder for human review.
[MA-6] Seeing Before Agreeing: Aligning Multi-Agent Consensus with Visual Evidence
【速读】：该论文旨在解决多智能体视觉问答（multi-agent VQA）中因个体幻觉和感知盲区导致的可靠性问题，尤其针对现有方法在多模态场景下过度依赖文本对话而忽视视觉信息对齐的缺陷。其核心挑战在于：仅通过答案层面的一致性难以保证共识的可信度，必须确保各智能体所依赖的视觉证据具有空间一致性与可验证性。解决方案的关键在于提出EAGLE（Evidence-Aligned Grounded Multi-agent Reasoning）框架，该框架无需训练，以证据为中心，显式暴露每个视觉语言模型（VLM）智能体的视觉锚定区域作为可解释的视觉证据，并通过跨智能体的证据互验机制强化对关键图像区域的共识，最终基于证据一致性动态引导决策。实验表明，EAGLE在六个VQA基准上实现了跨领域最优平均性能，同时具备轻量化、可解释性强及部署友好等优势。
链接: https://arxiv.org/abs/2605.30698
作者: Yuhan Wang,Shuochen Chang,Yalin Feng,Dongsheng Ma,Yuanzi Li,Zhengren Wang,Yinglong Yang,Yufei Chen,Yikang Wang,Shaoxu Sun,Wentao Zhang
机构: Peking University(北京大学); Shanghai Jiao Tong University(上海交通大学); Nanyang Technological University(南洋理工大学); Renmin University of China(中国人民大学); Shandong University(山东大学)
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
备注:
Abstract:Vision-language models (VLMs) have achieved strong performance on visual question answering (VQA). To mitigate individual hallucinations and blind spots, aggregating diverse perspectives via multi-agent collaboration has emerged as a promising paradigm. While this approach has shown great success in textual QA, its potential in the multimodal domain remains under-explored. Existing multi-agent VQA methods predominantly adapt text-centric protocols, focusing on textual discussions while ignoring the alignment of visual information. In this work, we reveal a key insight: answer-level agreement is insufficient for reliable multi-agent VQA; \textitaligned visual evidence – shared support from the image regions agents rely on – is essential for trustworthy consensus. To leverage this insight, we propose EAGLE (\textbfEvidence-\textbfAligned \textbfGrounded mu\textbfLti-agent r\textbfEasoning), a training-free evidence-centered framework for coordinating multiple VLM agents. EAGLE explicitly exposes each agent’s grounding regions as visual evidence, enables mutual verification over the evidence, and uses evidence consistency to guide final decision-making. Experiments on six VQA benchmarks show that EAGLE achieves best average performance across domains while remaining lightweight, interpretable, and practical for deployment.
[MA-7] Healthcare Mechanisms from Policy-as-Code Search under Strategic Provider Response
【速读】：该论文旨在解决现有医疗AI评估基准无法动态反映战略提供方响应的问题，即传统方法固定机制设计而忽略其在真实环境中诱发的均衡行为，导致机制评价失真。其核心解决方案是将医院机制设计重构为语言模型的程序合成问题：通过可类型化、可解释的规则程序在Medi-Sim多智能体模拟器中执行，该模拟器包含编码、选择、延迟、努力和分诊五类战略提供方行为通道。研究通过激励扫描（incentive sweep）复现了经典健康经济学中的关键现象，如在利润压力下出现的高编码（up-coding）与低复杂度患者选择，以及类似古德哈特定律（Goodhart’s Law）的性能漂移——衡量指标与真实结果呈现反向关联。进一步发现，单一审计杠杆（如关闭编码通道）可显著加剧低复杂度患者选择行为，揭示压力迁移效应。在此基础上，采用大语言模型（LLM）引导的进化代码搜索，在同一规则程序空间中生成可解释的混合目标程序，该程序成功消除高编码行为，使拒绝率降低50%，同时保持基线大部分利润水平，从而实现了对机制设计的高效优化与可解释性保障。
链接: https://arxiv.org/abs/2605.30680
作者: Zihan Wang,Xiang Xu,Hongyuan Zha,Wenhao Li
机构: The Chinese University of Hong Kong, Shenzhen(中国深圳大学); Tongji University(同济大学)
类目: Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
备注: 32 pages, 18 figures, 4 tables
Abstract:Healthcare mechanisms are inseparable from the strategic provider response they induce: existing healthcare AI benchmarks hold this response fixed and so cannot evaluate mechanisms by the equilibrium they produce. We recast hospital mechanism design as program synthesis for language models: typed, inspectable rule programs are executed and scored by Medi-Sim, a multi-agent simulator with five strategic provider channels (coding, selection, delay, effort, triage). An incentive sweep recovers classical health-economics findings as adjacent regimes – up-coding and low-complexity-patient selection under profit pressure, and Goodhart-style drift where measured performance becomes anti-correlated with true outcomes – and a single audit lever exposes pressure migration: closing the coding channel more than doubles low-complexity selection. LLM-guided evolutionary code search over the same rule-program space then synthesizes an inspectable mixed-objective program that eliminates up-coding, halves rejection, and retains most of the profit-oriented baseline’s funds.
[MA-8] MATraM: A Multi-Activity Transport and Mobility Agent -Based Model for Activity Modifications
【速读】：该论文旨在解决传统活动型交通模型在应对动态交通条件时行为响应能力不足的问题。现有模型多依赖预设的固定日程安排生成出行，难以反映个体在面对拥堵、延误等非理想出行条件时的行为灵活性与不确定性。其核心解决方案在于提出一种基于智能体的多活动交通移动性（Multi-Activity Transport Mobility, MATraM）模型，通过引入动态活动适应机制，使智能体能够在遭遇次优出行条件（如行程时间延长）时主动发起活动调整请求，并结合活动调度与修改框架实现决策过程的自适应演化。该模型将行为适应性嵌入日常活动计划的生成与执行中，从而更真实地模拟个体对交通系统动态变化的响应，催生出更具现实意义的出行模式与拥堵涌现现象。MATraM遵循ODD协议构建，集成智能体、活动日程、交通网络及路径规划、调度与行为适应等子模型，实现了活动型建模与交互式移动性仿真的深度融合，为在不确定性环境下探索交通系统动态提供了可扩展且灵活的建模平台。
链接: https://arxiv.org/abs/2605.30547
作者: Yahya Gamal,Ricardo Colasanti,Gary Polhill,Tatsuya Mitomi,Esra Suel,Alison Heppenstall
机构: 未知
类目: Multiagent Systems (cs.MA)
备注: 24 pages, 4 figures, 9 tables, working paper for a submission to MethodsX journal
Abstract:This paper introduces the Multi-Activity Transport Mobility (MATraM) Agent-Based Model (ABM), a novel framework designed to advance activity-based transport modelling by incorporating dynamic activity adaptation. Traditional transport models simulate system performance using varying levels of abstraction, including flow-based, queue-based, and interaction-based mobility representations. While these approaches differ in their treatment of movement and congestion, they typically rely on pre-defined trip patterns that limit responsiveness to changing conditions. In particular, conventional activity-based models generate trips from fixed daily schedules, constraining their ability to capture behavioural flexibility and uncertainty. MATraM addresses this limitation by enabling agents to flag activities modification requests in response to sub-optimal travel conditions, such as increased travel times. By coupling with an activity scheduling and modification framework, the model integrates adaptive decision-making into the generation and execution of daily activity schedules. This allows for a more realistic representation of how individuals adjust their behaviour in response to transport system dynamics, leading to emergent mobility and congestion patterns. The ABM is presented following the ODD protocol, outlining its purpose, structure, and implementation. MATraM includes detailed representations of agents, their activity schedules, and the transport network, alongside submodels governing routing, scheduling, and behavioural adaptation. By bridging activity-based modelling with interaction-based mobility simulation, MATraM provides a flexible and extensible platform for exploring transport dynamics under uncertainty. This work contributes to the development of next-generation transport models capable of capturing the complex interplay between individual behaviour and system-level outcomes.
[MA-9] A Theory-Guided LLM Pedagogical Agent for STEMC Scaffolding Without Over-Reliance
【速读】：该论文旨在解决当前大语言模型（Large Language Model, LLM）教学代理在教育实践中缺乏对学习理论的遵循，导致学生出现认知卸载、过度依赖及“游戏化”行为等问题，进而影响其实际教育价值。其核心解决方案是提出Copa——一个基于证据-决策-反馈（Evidence-Decision-Feedback, EDF）框架的多智能体、多模态协同同伴代理（Collaborative Peer Agent），融合社会认知理论（Social Cognitive Theory）与社会建构主义（Social Constructivism），通过自适应对话式支持促进学生的意义建构，而非直接提供答案。在一项真实高中生计算建模研究中（n=33对），实证表明Copa不仅能有效提升学生信心并帮助其口头表达概念理解，且不会引发依赖；同时能够根据学生的多模态输入数据提供可解释的个性化反馈。该研究揭示了以理论为指导、具备多模态交互能力的LLM代理在增强学生推理能力方面具有显著潜力，为课堂人工智能融合提供了可信路径。
链接: https://arxiv.org/abs/2605.30539
作者: Clayton Cohn,Surya Rayala,Siyuan Guo,Hanchen David Wang,Naveeduddin Mohammed,Umesh Timalsina,Shruti Jain,Ryan Li,Angela Eeds,Menton Deweese,Pamela J. Osborn Popp,Rebekah Stanton,Shakeera Walker,Ashwin T S,Meiyi Ma,Gautam Biswas
机构: 未知
类目: Multiagent Systems (cs.MA)
备注: Submitted to Computers Education. Currently under review
Abstract:LLM pedagogical agents are proliferating, yet recent findings have raised questions about their adherence to established theories of learning and, by extension, their educational value. Concerns regarding cognitive offloading, over-reliance, and “gaming” behaviors persist and remain largely unaddressed. In response, we developed Copa, an agentic, multi-agent, multimodal Collaborative Peer Agent for STEM+C learning. Copa is built on top of the Evidence-Decision-Feedback (EDF) framework, grounding its interactions in Social Cognitive Theory and Social Constructivism and promoting sense-making through adaptive, dialogic support rather than answer-seeking. In an authentic high school computational-modeling study (n=33 dyads), we demonstrate that Copa (1) supports students’ confidence building and ability to verbalize conceptual understanding without causing dependence; and (2) provides adaptive feedback personalized to learners that is interpretable with respect to students’ multimodal input data. These findings position theory-guided, multimodal LLM agents as a promising path toward classroom AI integration that amplifies students’ reasoning rather than replacing it.
[MA-10] LongDS-Bench: On the Failure of Long-Horizon Agent ic Data Analysis
【速读】：该论文旨在解决现有数据分析基准在评估智能体（agent）长期多轮交互能力时的不足，特别是其无法有效测试智能体在长时间跨度下对动态演化分析状态的持续追踪与维护能力。现有基准大多聚焦于孤立或短周期的交互任务，难以反映真实世界中数据探索的迭代性本质。为应对这一挑战，研究提出LongDS，一个面向长时程、多轮次数据分析的基准，要求智能体在分析过程中持续维护、更新、恢复和组合不断演化的分析状态。LongDS包含68个源自真实Kaggle笔记本的任务，覆盖地理科学、商业、教育等六个领域，总计2,225个交互轮次，任务设计基于多种状态演化模式（如反事实扰动、回滚、多状态复合），平均依赖跨度达11.3轮。实验评估五种先进模型发现，最优模型平均准确率仅为48.45%，且性能从早期到后期下降近47个百分点，其中52%–69%的失败案例源于长时程推理错误。进一步分析表明，增加交互步数并不能显著提升性能，揭示出当前系统的核心瓶颈并非交互预算不足，而是维持正确分析状态的能力受限。因此，解决方案的关键在于构建具备强状态持久性与演化建模能力的智能体架构，以支持可靠、可追溯的长期数据探索。
链接: https://arxiv.org/abs/2605.30434
作者: Kewei Xu,Xiaoben Lu,Shuofei Qiao,Zihan Ding,Haoming Xu,Lei Liang,Ningyu Zhang
机构: Zhejiang University (浙江大学); Ant Group (蚂蚁集团); Zhejiang University - Ant Group Joint Laboratory of Knowledge Graph (浙江大学-蚂蚁集团知识图谱联合实验室)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Multiagent Systems (cs.MA)
备注: Ongoing work
Abstract:Real-world data analysis is inherently iterative, yet existing benchmarks mostly evaluate isolated or short interactive tasks, leaving agents’ ability to track evolving analytical context over long horizons untested. We introduce LongDS, a benchmark for long-horizon, multi-turn data analysis where agents must maintain, update, restore, and compose evolving analytical states. LongDS comprises 68 tasks constructed from real-world Kaggle notebooks, spanning 2,225 turns across six domains including Geoscience, Business, and Education. Tasks are designed around state-evolution patterns (e.g., counterfactual perturbation, rollback, multi-state composition), with an average dependency span of 11.3 turns. Evaluating five state-of-the-art models, we find that the best model reaches only 48.45% average accuracy, performance drops nearly 47 points from early to late turns, and long-horizon errors account for 52%–69% of failures. Further analysis shows that additional agent steps do not necessarily improve performance, suggesting that the key bottleneck is maintaining a correct analytical state rather than increasing interaction budget. We release LongDS to support research on reliable long-horizon agentic data analysis. Code and data will be released at this https URL.
[MA-11] Delayed Repression and Emergent Instability in Adaptive Multi-Agent Systems
【速读】：该论文旨在解决在缺乏外部冲击、代理人之间无协调机制或恶意行为者的情况下，监管机构因处理延迟（processing lag）导致的系统性不稳定性问题。核心问题是：仅由监管响应滞后引发的反馈延迟是否足以破坏原本稳定的多智能体系统的动态平衡。解决方案的关键在于揭示“对延迟信号的反应性”是引发不稳定的根源——当代理人基于滞后的监管警报信号立即采取激进行为（如利用低警报窗口），会触发振荡型反馈循环。研究通过理论建模与仿真验证发现，尽管学习能力通常被认为可能加剧复杂性，但采用表格Q-learning的自适应代理反而表现出部分韧性：其通过Q值中隐含的历史惩罚记忆实现对延迟信号的缓冲，从而抑制了不稳定性传播；相比之下，固定策略代理虽不受延迟影响，而仅依赖阈值启发式的反应型代理则在延迟超过8步时出现高达96%的崩溃率，凸显了“即时反应”与“学习型记忆”的根本差异。
链接: https://arxiv.org/abs/2605.30392
作者: Igor Itkin
机构: 未知
类目: Multiagent Systems (cs.MA); Computer Science and Game Theory (cs.GT); Dynamical Systems (math.DS)
备注: 30 pages, 10 figures, 2 appendices. Code: this https URL
Abstract:Regulatory institutions (from content moderation platforms to financial supervisors) observe, deliberate, and intervene only after a characteristic delay. We ask whether this processing lag alone can destabilize a multi-agent system that would otherwise remain stable, without exogenous shocks, coordination among agents, or malicious actors. We study this question in two stages. First, we analyze a delayed replicator equation in which autonomous agents receive a benefit from radical behavior but face punishment based on a lagged institutional alarm signal. We derive a closed-form critical delay threshold beyond which the unique interior equilibrium loses stability through a Hopf bifurcation, and prove via center manifold reduction that the bifurcation is supercritical (producing bounded oscillations, not explosive growth) for the entire sigmoid response-function family. Second, we embed N=240 agents on a network and equip them with reinforcement learning (tabular Q-learning), comparing three decision architectures in a factorial design: non-reactive agents (fixed policy), reactive agents (threshold heuristic without memory), and Q-learning agents (adaptive with cumulative value estimates). The results reveal a hierarchy opposite to the naive expectation that learning amplifies instability: non-reactive agents are immune to delay (0% runaway across all tested values), reactive agents collapse catastrophically (96% runaway by delay \geq 8 steps), and Q-learning agents achieve partial resilience (66% runaway at delay = 20 ). The destabilizing ingredient is reactivity to delayed signals: agents that immediately exploit low-alarm windows trigger oscillatory feedback loops. Learning buffers this through implicit punishment memory encoded in Q-values
[MA-12] Social Reasoning in Machines: Investigating Collective Truth-Seeking Dynamics in Large Language Model Debate
【速读】：该论文旨在解决传统认知理论中个体理性推理（intellectualist reasoning）在面对复杂问题时的局限性，提出并验证一种基于社会性辩论的集体推理范式——论证理论（Argumentative Theory of Reasoning, ATR）。其核心问题是：如何通过模拟人类社会性推理机制来提升系统在真理探寻任务中的表现，尤其是在个体模型能力有限的情况下。解决方案的关键在于首次利用大语言模型（Large Language Models, LLMs）构建多智能体辩论（Multi-Agent Debate, LLM-MAD）框架，通过引入认知多样性与对抗性辩论机制，使多个性能有限的独立模型在协同辩论中实现超越个体表现的真理发现能力。实证研究表明，该方法显著提升了问卷类任务中的真相获取效率，并且性能提升机制严格符合ATR的核心原则，即真理是通过社会性、对抗性讨论中个体推理的动态修正而涌现的。此外，研究进一步提出一种基于辩论过程动态分析的新基准评估方法，能够量化模型内在属性（如幻觉倾向），突破了传统静态基准测试的局限，为模型能力评估提供了更深层、更动态的视角。
链接: https://arxiv.org/abs/2605.30391
作者: Tom Pecher
机构: University of Bath
类目: Multiagent Systems (cs.MA); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: Master’s thesis
Abstract:Human reasoning has long been theorised to operate socially, not through isolated individual cognition, but through collective adversarial discourse, a framework known as the Argumentative Theory of Reasoning (ATR). Rather than relying on individual “intellectualist reasoners” as the primary vehicle for truth-seeking, ATR reconceptualises truth as an emergent property of social epistemology: the product of imperfect individual reasoning refined under the adversarial pressure of debate. This distributed method of collective intelligence has guided humanity to ever-greater epistemic heights and underpins the foundational principles of all democratic systems. This thesis breaks new ground by, for the first time, simulating ATR through the multi-agent debate (MAD) of large language models (LLMs). With rigorous empirical analysis, we demonstrate that, when correctly engineering an epistemically diverse set of models, LLM-MAD can significantly improve truth-seeking performance on questionnaire-based tasks, even when individual debate participants exhibit limited standalone performance. Furthermore, we present strong empirical evidence that this performance gain is mechanistically grounded in the central principles of ATR, suggesting that collective reasoning may be universally favourable over individualist reasoning, rather than a quirk in biology or evolution. Finally, drawing on our analysis of debate dynamics, we propose a novel benchmarking methodology that leverages LLM-MAD to measure intrinsic model properties (such as hallucination propensity) in order to compare models in ways that current static benchmarking approaches cannot support.
[MA-13] Comparing Market Mechanism Efficiencies
【速读】：该论文旨在解决在不同市场机制下，交易效率与整体福利之间的权衡问题，具体比较了三种市场结构——具有透明订单簿的连续双重拍卖（lit exchanges）、具有隐匿订单簿的暗池（dark pools）以及周期性批量拍卖（periodic batch auctions）——在异质交易者面临执行价格、等待成本和交易成本权衡时的福利效率。其核心问题是：在存在信息不对称与策略性行为的环境中，何种信息结构与服务机制能够实现更高的社会总福利？解决方案的关键在于构建一个基于博弈论的排队系统模型，将每种市场机制建模为动态匹配环境下的战略互动框架。研究发现，在中等订单到达率且逆向选择有限的条件下，暗池在事前总体福利上优于其他两种机制。原因在于透明订单簿会诱发交易者为优化队列位置而进行策略性延迟或抢跑，形成无谓的社会等待成本；而暗池通过信息设计（information design）消除了这种策略性时间博弈。论文通过严格刻画各机制下的均衡策略，证明了福利排序关系：$ W^{\text{DARK}} > W^{\text{LIT}} > W^{\text{BATCH}} $。进一步扩展考虑了信息不对称与交易者对交易场所的内生选择，揭示了信息结构与服务纪律共同决定战略匹配环境中的效率水平。
链接: https://arxiv.org/abs/2605.31072
作者: Irene Aldridge
机构: 未知
类目: Theoretical Economics (econ.TH); Computer Science and Game Theory (cs.GT); Multiagent Systems (cs.MA); Econometrics (econ.EM)
备注: 79 pages
Abstract:We develop a game-theoretic framework that compares welfare efficiency across three market mechanisms: continuous double auctions with transparent order books (lit exchanges), opaque order books (dark pools), and periodic batch auctions. Each mechanism is modeled as a queuing system where heterogeneous traders face trade-offs between the execution price, waiting costs, and transaction costs. Our main result establishes that under moderate arrival rates and bounded adverse selection, dark pools dominate both alternatives in aggregate ex-ante welfare. Observable order books create costly strategic timing games in which traders delay or rush submissions to optimize their position in the queue, generating wasteful social waiting costs. Opaque order books eliminate these timing games through information design. We formally characterize the equilibrium strategies in each mechanism and prove the welfare ranking W^DARK W^LIT W^BATCH . Extensions incorporate asymmetric information and endogenous venue choice. The results demonstrate how the information structure and the discipline of the service jointly determine efficiency in strategic matching environments. Comments: 79 pages Subjects: Theoretical Economics (econ.TH); Computer Science and Game Theory (cs.GT); Multiagent Systems (cs.MA); Econometrics (econ.EM) Cite as: arXiv:2605.31072 [econ.TH] (or arXiv:2605.31072v1 [econ.TH] for this version) https://doi.org/10.48550/arXiv.2605.31072 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
自然语言处理
[NLP-0] Language Models Learn Constructional Semantics Not To Mention Syntax: Investigating LM Understanding of Paired-Focus Constructions CONLL
【速读】： 该论文旨在解决开放源代码语言模型（open-source models）在理解罕见构式（rare constructions，即形式-意义配对）方面的能力问题，特别是针对英语中一类罕见的“成对焦点”构式（Paired-Focus constructions，如“let alone”、“much less”）是否具备稳健的语义理解能力，以及其知识习得背后的学习动态机制。研究的关键在于构建了一个新颖的数据集，结合量级形容词语义与通用世界知识，系统评估不同规模、架构及预训练数据量的语言模型对这些构式的理解能力。实验发现，尽管大规模模型在人类规模数据上训练的模型在语义评估中表现不佳，但若干参数量适中的开源模型却能敏感地捕捉到成对焦点构式的形态与语义特征；进一步分析训练动态表明，语义理解的习得晚于句法知识的掌握，并且其发展与世界知识某些领域的提升存在显著相关性。因此，该研究的关键结论是：适度规模的开源模型具备对罕见构式的语义理解能力，且其学习过程体现出语义知识与跨领域世界知识之间的深层关联。
链接: https://arxiv.org/abs/2605.31586
作者: Wesley Scivetti,Ethan Wilcox,Nathan Schneider,Kanishka Misra,Leonie Weissweiler
机构: Georgetown University(乔治城大学); The University of Texas at Austin(得克萨斯大学奥斯汀分校); Leipzig University(莱比锡大学); ScaDS.AI Dresden/Leipzig(萨克斯人工智能中心德累斯顿/莱比锡)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: Conference on Natural Language Learning (CoNLL) 2026
Abstract:Grasping the semantics of rare constructions (form-meaning pairings) has been shown to be a challenging problem that has currently only been solved by the largest LLMs. It remains an open question if open-source models have robust constructional understanding, and if so, what learning dynamics underlie the acquisition of this knowledge. Focusing on a set of rare Paired-Focus constructions in English (e.g. “let alone”, “much less”), we construct a novel dataset to test their meanings using both scalar adjectival semantics and general world knowledge. Testing a wide range of models differing in parameter count, architecture, and pretraining dataset size, we find that several modestly sized models are sensitive to both the forms and the meanings of Paired-Focus constructions, though models trained on human-scale data fail at all meaning evaluations. Turning to training dynamics for a set of open-checkpoint models, we find that Paired-Focus understanding emerges later in training than Paired-Focus syntactic knowledge, and that learning of Paired-Focus semantics is correlated with gains in some domains of world knowledge. Overall, our empirical results support the conclusion that modestly sized open-source models can grasp the rare Paired-Focus constructions, and demonstrate a connection between knowledge of Paired-Focus constructions and other meaning domains.
[NLP-1] LongTraceRL: Learning Long-Context Reasoning from Search Agent Trajectories with Rubric Rewards
【速读】： 该论文旨在解决大语言模型在长上下文推理中难以有效定位并整合关键信息的问题，尤其针对海量干扰内容下的信息筛选与链式推理能力不足。现有基于可验证奖励的强化学习（Reinforcement Learning with Verifiable Rewards, RLVR）方法受限于低混淆度的干扰项以及稀疏且仅基于最终结果的奖励信号，无法对中间推理步骤进行有效监督。为此，论文提出 \textscLongTraceRL，其核心创新在于两方面：一是通过知识图谱随机游走生成多跳问题，并利用搜索代理轨迹构建分层干扰项——包括代理阅读但未引用的高混淆度文档与出现在搜索结果中但从未打开的低混淆度文档，显著提升了训练上下文的挑战性；二是设计了一种基于评分量表（rubric reward）的细粒度过程奖励机制，以推理链中的真实实体作为监督信号，仅对最终答案正确的响应施加奖励（正向奖励策略），从而区分正确答案间的推理质量，防止奖励劫持。实验在三个不同规模的推理型大模型（4B–30B）及五个长上下文基准上验证了 \textscLongTraceRL 的优越性，显著提升了模型的全面性与证据驱动推理能力。
链接: https://arxiv.org/abs/2605.31584
作者: Nianyi Lin,Jiajie Zhang,Lei Hou,Juanzi Li
机构: Tsinghua University (清华大学); Zhipu (智谱)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Long-context reasoning remains a central challenge for large language models, which often fail to locate and integrate key information in extensive distracting content. Reinforcement learning with verifiable rewards (RLVR) has shown promise for this task, yet existing methods are limited by low-confusability distractors and sparse, outcome-only reward signals that cannot supervise intermediate reasoning steps. To address these issues, we introduce \textscLongTraceRL. For data construction, we generate multi-hop questions via knowledge graph random walks and leverage search agent trajectories to build \emphtiered distractors: documents the agent read but did not cite (high confusability) and documents that appeared in search results but were never opened (low confusability), producing training contexts that are far more challenging than those built by random sampling or one-shot search. For reward design, we propose a \emphrubric reward that uses the gold entities along each reasoning chain as fine-grained, entity-level process supervision. This rubric reward is applied only to responses with correct final answers (positive-only strategy), distinguishing the reasoning quality among correct responses and preventing reward hacking. Experiments on three reasoning LLMs (4B–30B) across five long-context benchmarks demonstrate that \textscLongTraceRL consistently outperforms strong baselines and encourages comprehensive, evidence-grounded reasoning. Codes, datasets and models are available at \hrefthis https URLthis https URL.
[NLP-2] What Gets Unmasked First? Trajectory Analysis of Diffusion Models for Graph-to-Text Generation
【速读】： 该论文旨在解决图结构到文本生成任务中，基于掩码扩散语言模型（MDLMs）在生成过程中因监督微调（SFT）引入的非自然解码轨迹问题。具体而言，尽管MDLMs在无监督条件下能自然地优先生成实体、随后是关系与功能词，而将结构化标记（如句尾符号）留到最后处理，但标准的监督微调会过早锚定结构标记，导致输出长度被固定，进而引发信息遗漏或幻觉。其解决方案的关键在于提出一种无需训练的推理时修正方法——λ-缩放结构解码（lambda-scaled structural decoding），通过降低结构标记的置信度权重，恢复原始的合理解码顺序，从而提升生成质量（提升+9.4 BLEU-4）。此外，论文还提出Graph-LLaDA，将图变换器（Graph Transformer）编码器嵌入到LLaDA的解码流程中，显式建模图结构信息，显著增强模型对跨数据集模式的泛化能力，实验表明相较于传统基线，基于LLM和MDLM的方法展现出更强的鲁棒性与泛化性能。
链接: https://arxiv.org/abs/2605.31564
作者: Qing Wang,Jacob Devasier,Chengkai Li
机构: The University of Texas at Arlington
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:We present the first systematic study of masked diffusion language models (MDLMs) for graph-to-text generation. We analyze MDLM generation trajectories – the order in which tokens are unmasked during iterative decoding – and find that, unlike autoregressive LLMs which generate text linearly, MDLMs naturally prioritize entities first, followed by relational and function words, with structural tokens resolved last. We further identify a previously undocumented failure mode of supervised fine-tuning: SFT disrupts this strategy by prematurely anchoring structural sentence-ending tokens early in the decoding trajectory, effectively fixing the output length which can lead to omitted or hallucinated information. To address this, we propose lambda-scaled structural decoding, a training-free inference-time modification that downweights structural token confidence and recovers +9.4 BLEU-4. Finally, we introduce Graph-LLaDA, which integrates a Graph Transformer encoder into LLaDA’s decoding process to explicitly incorporate relational graph structure. Cross-dataset evaluation on LAGRANGE reveals that previous baselines overfit to dataset-specific patterns, while LLM- and MDLM-based approaches generalize significantly better.
[NLP-3] Disagreeing Rationales: Rethinking Classification and Explainability Evaluation in Hate Speech Detection
【速读】： 该论文旨在解决主观自然语言处理（NLP）任务中，人类标注与解释（rationales）在粒度层面存在显著差异却未被充分研究的问题，尤其在仇恨言论检测等任务中，不同个体的推理风格、价值取向和理解差异可能影响模型训练与评估。其核心挑战在于如何有效评估人类标签与解释，并超越传统多数投票机制对解释进行聚合。为此，论文提出一种统一的监督框架，通过系统性地在不同标签与解释表示空间（硬标签/软标签、硬解释/中间表示/软解释）中复现多种模型、训练策略与损失函数，整合分类性能与可解释性评价指标。分类评估聚焦预测能力与分布特性，可解释性评估则从合理性（plausibility）、忠实性（faithfulness）与复杂性（complexity）三个互补维度展开。实验结果表明，软表示形式在各类指标上均表现更优，凸显其在捕捉人类判断多样性方面的优势，进而揭示了在主观性任务中重新审视评估范式的重要性，强调应摒弃对硬标签与简单聚合方式的依赖，转向更具表达力的软表示与多维评估体系。
链接: https://arxiv.org/abs/2605.31563
作者: Benedetta Muscato,Beiduo Chen,Gizem Gezici,Barbara Plank,Fosca Giannotti
机构: Scuola Normale Superiore, Italy; University of Pisa, Italy; MaiNLP, LMU Munich, Germany; Munich Center for Machine Learning, Germany
类目: Computation and Language (cs.CL)
备注: 16 pages
Abstract:Human disagreement is ubiquitous and well-known in labeling. However, variation in explanations, captured through token-level human rationales, remains far less explored. At the same time, it is unclear how to best evaluate human labels and rationales – or even how to best aggregate rationales beyond majority vote – in light of this variation. Yet, rationales may provide additional insights into the richness of human reasoning, that may differ in style, values and interpretations – especially in subjective NLP tasks like hate speech detection. In this work, we unify diverse models, training strategies, loss functions, and existing evaluation metrics under a single protocol by systematically re-implementing them across different label and rationale representation spaces. Classification metrics are organized around two key properties – predictive and distributional – while explainability metrics through three complementary dimensions: plausibility, faithfulness, and complexity. In this unified supervision framework, we evaluate model behavior across classification and explainability metrics, as well as metric sensitivity to the choice of label (hard and soft) and rationale representation space (hard, intermediate and soft). Results show that both hard and soft metrics favor softer representations, highlighting their effectiveness in capturing variation and the need to rethink evaluation in subjective NLP.
[NLP-4] What Am I Missing? Question-Answering as Hidden State Probing
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在测试时推理过程中存在的不确定性问题，即相同输入提示或部分解题路径下，模型多次采样可能产生不一致的输出结果。其核心挑战在于如何有效提升模型在推理过程中的自一致性与可靠性。解决方案的关键在于将提问（question-asking）作为一种推理时干预手段，通过构建师生（student-teacher）框架，使学生模型在生成答案前主动向教师提问以探查自身隐藏状态。研究发现，对学生在提问前后隐藏状态进行探测所获得的信号，能够提前预测最终解答的正确性，表明该信号源于模型自我诊断而非来自教师的信息传递。进一步地，研究将提问行为建模为一个序列决策问题，利用探测器作为质量评分，并设计了一个门控策略（gating policy）以选择最有可能提升正确性的提问时机。然而，实验结果显示，尽管该策略能有效识别模型的正确性与不确定性，但干预措施对正确轨迹和错误轨迹的影响具有对称性——即存在“诊断与修正之间的鸿沟”，说明当前方法在实现基于不确定性的自我修正方面仍面临根本性局限。
链接: https://arxiv.org/abs/2605.31561
作者: Chu Fei Luo,Samuel Dahan,Xiaodan Zhu
机构: Queen’s University(皇后大学); Vector Institute for AI(向量人工智能研究所)
类目: Computation and Language (cs.CL)
备注:
Abstract:Test-time reasoning has become a significant field of study since the introduction of chain-of-thought reasoning in large language models (LLMs). However, the mechanisms of this reasoning process are still under-explored – from the same input prompt, and even the same partial solution, LLMs can produce varied answers if sampled multiple times. We propose to leverage question-asking as an inference-time intervention that articulates information about the model’s hidden state. To achieve that, we present a student-teacher setting where a student asks questions to a teacher. We train a probe on the student’s hidden state before and after asking a question and find it is predictive of the trajectory’s final correctness, even before generating the teacher’s answer. This suggests there is a meaningful signal from the self-diagnosis that occurs during question generation rather than information transfer from the teacher. We then frame question-asking as a sequential decision problem, using this probe as a quality score, and define a gating policy to ask questions that maximize likelihood of correctness. We find that the success of question-asking as an intervention is largely dependent on the model’s self-consistency. Our empirical results show a gap between detection and recovery; while our gating policy captures model correctness and uncertainty, interventions are equally likely to harm correct trajectories as they are to recover incorrect ones. This gap between diagnosis and correction has broader implications on language models’ capacity for self-refinement under uncertainty.
[NLP-5] Semantic Triplet Restoration: A Novel Protocol for Hierarchical Table Understanding in Large Language Models
【速读】： 该论文旨在解决表格问答（Table Question Answering, TQA）中因二维布局、合并单元格及层级表头导致的语义关系隐含表达难题。现有方法通常依赖HTML或Markdown作为中间表示，但这类布局导向的序列化方式引入了冗余标记开销，并迫使大语言模型从行列跨度中推断表头与单元格的对齐关系，增加了计算负担与错误风险。本文提出语义三元组恢复（Semantic Triplet Restoration, STR）协议，将每个单元格重写为原子事实三元组：项目路径（item path，标识行级实体）、特征路径（feature path，标识层级属性）和值（value，包含单元格内容），从而显式表达语义结构。同时，提出轻量级查询感知路由器TripletQL，利用STR生成的三元组集合，针对不同问题动态选择合适的渲染或过滤后的三元组子集进行推理。在四个中英文表格问答基准上的实验表明，STR在保持或超越基于HTML基线性能的同时，显著减少输入令牌数；尤其在小规模语言模型和长表格上下文场景下，优势更为明显，表明显式语义表示在资源受限的推理环境中具有更强的适应性与效率。
链接: https://arxiv.org/abs/2605.31550
作者: Yibin Zhao,Fangxin Shang,Dingrui Yang,Yuqi Wang
机构: Taiyuan University of Technology; AI Lab, Qifu Technology, Beijing, China; AI Lab, Greensea Technology, Shenzhen, China
类目: Computation and Language (cs.CL)
备注:
Abstract:Table question answering requires models to recover semantic relations encoded implicitly by two-dimensional layout, merged cells, and hierarchical headers. Current pipelines typically use HTML or Markdown as intermediate table representations, but these layout-oriented serializations introduce markup overhead and require large language models to infer header-cell alignments from row and column spans. We propose Semantic Triplet Restoration (STR), a protocol that rewrites each cell as an atomic fact item path, feature path, value, where the item path specifies the row-wise entity, the feature path specifies the hierarchical attribute, and the value contains the cell content. We also present TripletQL, a lightweight query-aware router that uses STR to select an appropriate rendering or filtered subset of triplets for each question. Across four Chinese and English table-QA benchmarks, STR matches or improves upon HTML-based baselines while reducing input tokens. The relative benefit grows for smaller language models and longer table contexts, suggesting that explicit semantic representations are especially useful under constrained inference budgets. Code and data are available at this https URL .
[NLP-6] Preference-Aware Rubric Learning for Personalized Evaluation
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在向以用户为中心的智能体演进过程中，个性化对齐（personalized alignment）评估所面临的挑战。现有评估方法（包括自动指标和基于大模型作为评判者的方法）难以捕捉用户长期交互历史中蕴含的主观、个性化的偏好，导致评估结果缺乏可靠性与有效性。为应对这一问题，论文提出三个关键原则：代表性（Representativeness）、用户一致性（User-Consistency）和区分性（Discriminativeness）。其核心解决方案是引入“个性化评估即学习”（Personalized Evaluation as Learning）范式，将个性化评估重构为一个动态的学习过程而非静态判断。在此范式下，论文提出PARL（Preference-Aware Rubric Learning for Personalized Evaluation）框架，通过直接从原始用户交互数据中学习具备偏好感知的评估标准（evaluation rubrics），并结合自验证机制确保评估标准与用户偏好的一致性。PARL进一步融合判别式强化学习目标，对比用户生成内容与个性化模型输出，使学习到的评估标准能够精准刻画用户特定的决策边界。实验表明，PARL在真实场景下的个性化文本生成任务中，能持续生成高保真度的评估标准，有效识别与用户偏好一致的响应，并在跨用户与跨任务场景中展现出良好的泛化能力，同时稳定捕捉风格偏好与细粒度评价模式。
链接: https://arxiv.org/abs/2605.31545
作者: Yilun Qiu,Xiaoyan Zhao,Yang Zhang,Yuxin Chen,Cilin Yan,Jiayin Cai,Xiaolong Jiang,Yao Hu,Yoko Yamakata,Tat-Seng Chua
机构: National University of Singapore (新加坡国立大学); Xiaohongshu Inc. (小红书); The University of Tokyo (东京大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:As Large Language Models (LLMs) evolve from general-purpose assistants to user-centric agents, personalization has become central to aligning model behavior with individual preferences, making the evaluation of personalized alignment a critical bottleneck. Existing evaluation methods-ranging from automatic metrics to LLM-as-a-judge approaches-fail to capture subjective, user-specific preferences embedded in long-term interaction histories. We identify three essential principles for reliable and effective personalized evaluation: Representativeness, User-Consistency, and Discriminativeness. To address these principles, we introduce Personalized Evaluation as Learning, a paradigm that formulates personalized evaluation as a learning problem rather than a static judgment. Under this paradigm, we propose PARL (Preference-Aware Rubric Learning for Personalized Evaluation), a framework that learns to induce preference-aware evaluation rubrics directly from raw user histories and performs a self-validation mechanism to ensure consistency with the user’s preferences. PARL integrates rubric induction with a discriminative reinforcement learning objective that contrasts user-authored responses against competitive personalized model outputs, enabling the learned rubrics to capture precise, user-specific decision boundaries. Experiments on real-world personalized text generation tasks show that PARL consistently induces high-fidelity rubrics that reliably identify user-aligned responses and generalize across users and tasks, while capturing stable stylistic preferences and fine-grained evaluative patterns. To ensure reproducibility, our code is available at this https URL.
[NLP-7] UniAudio-Token: Empowering Semantic Speech Tokenizers with General Audio Perception
【速读】： 该论文旨在解决现有语义语音分词器（Semantic Speech Tokenizers）因过度关注语言抽象而产生的声学感知盲区问题，导致其在非语音主导任务中应用受限。其核心挑战在于如何在保持原有语义分词框架下实现对通用音频的全面感知能力，避免信息损失。解决方案的关键在于提出UniAudio-Token框架，通过两项创新机制实现语义与声学表征的协同优化：(1) 语义-声学基元（Semantic-Acoustic Primitives, SAP），通过将音频分解为语言内容、语音属性和听觉场景基元，提供结构化监督信号；(2) 语义-声学平衡机制（Semantic-Acoustic Equilibrium, SAE），引入内容感知门控机制，自适应地从浅层网络恢复细粒度声学细节。实验表明，该框架在保持高质量语音生成能力的同时，能够学习到兼具广泛适用性的通用音频表征，显著优于所有单码本基线模型，在理解与生成任务中均展现出优越性能，可作为统一的音频接口有效支持下游大语言模型（LLM）。
链接: https://arxiv.org/abs/2605.31521
作者: Yuhan Song,Linhao Zhang,Aiwei Liu,Chuhan Wu,Sijun Zhang,Wei Jia,Yuan Liu,Houfeng Wang,Xiao Zhou
机构: Peking University (北京大学); Tencent Inc. (腾讯公司)
类目: Computation and Language (cs.CL); Sound (cs.SD)
备注: 19 pages, 10 figures
Abstract:Semantic speech tokenizers have become a widely used interface for Audio-LLMs, owing to their compact single-codebook design and strong linguistic alignment. However, their focus on linguistic abstraction induces acoustic blindness, limiting their applicability beyond speech-centric tasks. We propose UniAudio-Token, a framework that empowers semantic tokenizers with general audio perception without compromising speech ability. Instead of altering the semantic paradigm, UniAudio-Token mitigates its information loss through two key innovations: (1) Semantic-Acoustic Primitives (SAP) provide structured supervision by decomposing audio into linguistic content, vocal attributes, and auditory-scene primitives; and (2) Semantic-Acoustic Equilibrium (SAE) introduces a content-aware gating mechanism that adaptively restores fine-grained acoustic details from shallow layers. Extensive evaluations show that UniAudio-Token learns comprehensive universal representations while preserving high-fidelity speech generation. When integrated with downstream LLMs, it outperforms all single-codebook baseline tokenizers on both understanding and generation tasks, effectively serving as a unified audio interface. We publicly release all our code, including training and inference scripts, together with the model checkpoints at this https URL.
[NLP-8] If LLM s Have Human-Like Attributes Then So Does Age of Empires II
【速读】： 该论文旨在解决当前大语言模型（LLM）研究中普遍存在的一个根本性问题：即许多研究在未充分验证的前提下，将通用的人类特质（如道德判断或自然语言理解能力）归因于LLM，这种做法可能导致错误的结论。其核心解决方案在于提出“非唯一性”（non-uniqueness）的假设——即所谓的拟人化属性并非LLM所独有，而是任何足够强大的计算系统（如《帝国时代2》游戏中的智能体、乐高积木系统甚至波士顿都市区域）都可能表现出类似行为特征。这一观点表明，尽管某些输入-输出模式（如对提示的响应）可能保持一致，但对其行为的解释却高度依赖于具体实现载体（substrate），因此若不建立明确的实证测量标准，对这些属性的讨论将陷入主观诠释的循环。为此，论文主张采用“零假设”（null assumption），即默认LLM不具备独特的人类特质，以此为基础设计可验证的实验框架。此外，作者通过形式化证明《帝国时代2》具备功能完备性和图灵完备性，进一步支持了其关于复杂系统在不同载体上均可涌现出类人行为的论点，从而为相关研究提供了更为严谨的方法论基础。
链接: https://arxiv.org/abs/2605.31514
作者: Adrian de Wynter
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Computers and Society (cs.CY)
备注:
Abstract:Much research has been carried out on large language models (LLMs) and LLM-powered agentic workflows. However, many works within the field state emergence of, ascribe to, or assume, generalised anthropomorphic attributes to them (e.g., morality or understanding of natural language). Our goal is not to argue in favour or against the existence of these attributes, but to point out that these conclusions could be incorrect. For this we build and train a simple neural network on the videogame Age of Empires II, and note that any entity in a sufficiently-powerful substrate, such as LEGO or the Greater Boston Area, could also present such attributes. Hence, the purported anthropomorphic attributes of LLMs are empirically non-unique: although some properties (e.g., responses to prompts) could remain constant, others, such as the interpretation of their perceived behaviour, might change with the substrate. Thus, any empirically-grounded discussion requires explicit measurement criteria; otherwise the interpretation is left to the representation. We then show that assuming that these attributes exist or not in a system, independent of the substrate and in a generalised way, leads to either circular or uninformative conclusions, regardless of the experimenter’s viewpoint on the subject. Finally we propose a ‘null’ assumption, where one assumes LLM non-uniqueness instead of assuming anthropomorphic attributes to set up an experiment, along with examples of it. We also discuss potential objections to our work, briefly survey the field, and prove that \textitAge of Empires II is functionally- and Turing-complete.
[NLP-9] Reliable Multilingual Orthopedic Decision Support from Clinical Narratives: Language-Aware Adaptation and Verification-Guided Deferral
【速读】： 该论文旨在解决低资源医疗环境中多语言骨科临床决策支持的难题，主要挑战包括临床文本中的专业术语、多语种混用、证据不完整、标签不平衡以及语言依赖性文档模式。其解决方案的关键在于提出一种面向可靠性的多语言分类框架，核心创新是引入领域自适应编码器IndicBERT-HPA，该模型通过在IndicBERT基础上添加语言感知的骨科适配头（language-aware orthopedic adapter heads），实现对英、印地、旁遮普语等多语言骨科自由文本的临床相关表征学习。此外，研究还设计了一个确定性的选择性验证层，结合置信度门控、证据一致性检查和语言风险筛查，在保持72.3%覆盖率的前提下，将选择性预测的准确率提升至84.4%，宏观F1达0.76，显著优于全接受预测策略（71.5%准确率，0.65宏观F1）。实验表明，零样本指令微调的大语言模型（LLMs）在封闭集分类任务中仍远逊于经过任务适配的编码器，且存在语言依赖性不稳定问题；而IndicBERT-HPA在自然临床先验分布下展现出最优的整体性能，平均宏观F1达0.8792，宏观AUROC为0.894，AUPRC为0.902，充分验证了其在多语言临床决策支持中的可靠性与鲁棒性。
链接: https://arxiv.org/abs/2605.31512
作者: Danish Ali,Li Xiaojian,Sundas Iqbal,Farrukh Zaidi
机构: Wuhan University (武汉大学); Nanjing University of Information Science and Technology (南京信息工程大学); Bahawal Victoria Hospital (巴哈瓦尔维多利亚医院)
类目: Computation and Language (cs.CL)
备注:
Abstract:Multilingual orthopedic decision support remains challenging in low-resource healthcare settings, where clinical narratives contain specialized terminology, mixed scripts, incomplete evidence, label imbalance and language-dependent documentation patterns. This article presents a reliability-oriented framework for classifying free-text orthopedic notes in English, Hindi and Punjabi. We compare task-aligned multilingual transformer encoders, a task-fine-tuned DistilBERT baseline, zero-shot instruction-tuned large language models (LLMs) and a domain-adaptive encoder, IndicBERT-HPA. IndicBERT-HPA augments IndicBERT with language-aware orthopedic adapter heads to support clinically relevant multilingual representation learning. Evaluation extends beyond aggregate accuracy to per-class performance, ROC-AUC, AUPRC, expected calibration error, cross-language stability and robustness under controlled balanced and natural-prevalence distributions. The evaluated zero-shot LLMs remain substantially less effective than task-adapted encoders for closed-set classification, with language-dependent instability. Under natural clinical prevalence, IndicBERT-HPA achieves the strongest overall performance, reaching an averaged Macro-F1 of 0.8792, Macro-AUROC of 0.894 and AUPRC of 0.902. We further implement a deterministic selective-verification layer combining confidence gating, evidence-consistency checking and language-risk screening. On a randomly selected held-out 5,000-record subset, it achieves 84.4% selective accuracy and 0.76 selective Macro-F1 at 72.3% coverage, compared with 71.5% accuracy and 0.65 Macro-F1 for accept-all prediction. These results support reliability-oriented multilingual clinical decision support with explicit deferral.
[NLP-10] Consolidating Rewarded Perturbations for LLM Post-Training
【速读】： 该论文旨在解决后训练语言模型时，传统基于梯度下降的样本-评分-更新循环在推理阶段效率低下且难以扩展至自由生成任务的问题。现有方法如RandOpt虽在计算资源匹配下表现优异，但依赖于预测层面的K个专家模型集成，导致每次测试需执行K次前向传播，且不适用于开放式生成场景。其解决方案的关键在于提出一种无需梯度流经语言模型的新型可部署框架CoRP（Consolidating Rewarded Perturbations），通过发现并利用奖励种群中普遍存在的低秩结构，实现对高奖励扰动的高效整合。CoRP采用奖励加权聚合、兼容性感知重加权与保留验证门控机制，在不依赖梯度的情况下将多个受奖励的扰动合并为单一可部署模型，仅需一次前向传播即可完成推理。实验表明，该方法在0.5B至8B规模的五种语言模型上，于数学、代码和创意写作等五类任务中平均提升8.1点；仅使用RandOpt十分之一的扰动预算，即超越单次推理的RandOpt 6.5点，并恢复超过50次多数投票集成一半以上的性能增益。
链接: https://arxiv.org/abs/2605.31494
作者: Zheyu Zhang,Shuo Yang,Gjergji Kasneci
机构: Technical University of Munich (慕尼黑工业大学); Munich Center for Machine Learning (慕尼黑机器学习中心)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:Post-training of language models is commonly framed as a sample-score-update loop implemented by gradient descent. A recent line of work, exemplified by RandOpt, relocates this loop to weight space, sampling Gaussian perturbations around a pretrained model and ensembling the top-K rewarded specialists at inference. While competitive with PPO and GRPO under matched training compute, this prediction-level ensemble incurs K forward passes per test example and does not extend cleanly to free-form generation. We ask whether the rewarded population can instead be folded into a single deployable model, replacing the inference-time ensemble with one consolidated update. A split-half analysis over 25 model-task pairs reveals reproducible low-rank structure in every case. We turn this geometry into CoRP (Consolidating Rewarded Perturbations), a gradient-free operator that combines reward-weighted aggregation, compatibility-aware reweighting, and a held-out validation gate, with no gradient flowing through the language model. Across five language models from 0.5B to 8B and five tasks covering math, code, and creative writing, CoRP improves the base model by 8.1 points on average. Using one tenth of RandOpt’s perturbation budget, CoRP exceeds single-inference RandOpt by 6.5 points and recovers more than half of the gain of the 50-pass majority-vote ensemble, at one forward pass per test example.
[NLP-11] Are Full Rollouts Necessary for On-Policy Distillation?
【速读】： 该论文旨在解决生成式强化学习中基于策略蒸馏（On-policy Distillation, OPD）在长时序推理任务中存在的训练效率低下问题，尤其针对标准OPD在训练初期因依赖完整轨迹而产生高计算开销及不可靠教师反馈（特别是在序列末端）的缺陷。其核心挑战在于：尽管OPD无需完整轨迹或最终答案奖励即可提供学习信号，但传统方法仍强制生成全长度回溯（rollout），导致资源浪费与训练不稳定。本文的关键解决方案是引入对回溯时长（rollout horizon）的可控机制，提出两种简单有效的策略：渐进式OPD（Progressive OPD, POPD），通过训练过程中逐步扩展回溯长度以缓解早期不稳定性；截断式OPD（Truncated OPD, TOPD），在可靠截断回溯上进行永久蒸馏，显著降低计算需求。实验结果表明，POPD可将训练效率提升达3倍，TOPD仅需10%的回溯长度即可达到与全量回溯相当的性能，大幅减少实际运行时间与内存占用，验证了回溯时长控制作为提升OPD效率的简洁且实用路径的有效性。
链接: https://arxiv.org/abs/2605.31490
作者: Yaocheng Zhang,Jiajun Chai,Songjun Tu,Yuqian Fu,Xiaohan Wang,Wei Lin,Guojun Yin,Qichao Zhang,Yuanheng Zhu,Dongbin Zhao
机构: Institute of Automation, Chinese Academy of Sciences; School of Advanced Interdisciplinary Sciences, University of Chinese Academy of Sciences; Meituan; School of Artificial Intelligence, University of Chinese Academy of Sciences
类目: Computation and Language (cs.CL)
备注: 14 pages, 16 figures
Abstract:On-policy distillation (OPD) provides dense teacher feedback along rollouts generated by the student and has emerged as a promising post-training paradigm for long-horizon reasoning. However, standard OPD typically generates full rollouts during training, which is computationally expensive and may expose the student to unreliable teacher feedback at late rollout positions, especially during early training. We identify the rollout horizon as a key bottleneck in OPD that substantially impacts training efficiency. Unlike Reinforcement Learning with Verifiable Rewards (RLVR), OPD does not require a complete trajectory or a final answer reward to provide learning signals. This observation suggests that full rollouts may not always be necessary for effective OPD. Motivated by this insight, we propose two simple horizon-control strategies: Progressive OPD (POPD), which gradually expands the rollout horizon during training, and Truncated OPD (TOPD), which permanently performs distillation on reliable truncated rollouts. Experiments on mathematical reasoning show that POPD improves the training efficiency of OPD by up to 3 \times , while TOPD matches OPD performance using only 10% of the rollout horizon, leading to substantial wall-clock and memory reductions. These results demonstrate that controlling the rollout horizon offers a simple and practical path to more efficient OPD.
[NLP-12] BenHalluEval: A Multi-Task Hallucination Evaluation Framework for Large Language Models on Bengali
【速读】： 该论文旨在解决低资源语言——孟加拉语（Bengali）在大语言模型（LLM）中幻觉（hallucination）问题缺乏系统评估的现状。当前虽有大量关于高资源语言幻觉的研究，但孟加拉语作为全球第六大使用语言，其幻觉现象尚未得到专门研究。为此，作者提出BenHalluEval，一个针对孟加拉语的细粒度幻觉评估框架，涵盖生成式问答（Generative Question Answering, GQA）、孟加拉-英语混用问答（Bangla-English Code-Mixed QA）、摘要生成和推理四个任务。其解决方案的关键在于构建包含12,000个幻觉样本的基准数据集，基于GPT-5.4生成十二种特定任务类型的幻觉，并采用双轨评估协议：Track A评估在真实答案实例上的误报率（false-positive rate），Track B评估对幻觉候选样本的检测率。为克服单一评估轨迹导致的评分偏差并联合惩罚两类错误模式，提出BenHalluScore这一双轨校准指标，其得分范围在7.72%至55.42%之间，揭示了不同模型与任务间显著的幻觉校准差异。此外，实验表明链式思维提示（Chain-of-thought prompting）虽能改变响应分布，但未能一致提升幻觉识别能力。本研究首次建立了面向孟加拉语的幻觉基准，强调了仅依赖提示工程或单轨评估在低资源语言场景下的局限性，为未来多语言幻觉研究提供了重要参考。
链接: https://arxiv.org/abs/2605.31483
作者: Shefayat E Shams Adib,Ahmed Alfey Sani,Ekramul Alam Esham,Ajwad Abrar,Ishmam Tashdeed,Md Taukir Azam Chowdhury
机构: Islamic University of Technology (伊斯兰科技大学); University of California (加州大学)
类目: Computation and Language (cs.CL)
备注: Preprint. Under review
Abstract:Despite Bengali being the sixth most spoken language in the world, no prior work has systematically evaluated hallucination in large language models (LLMs) for Bengali. We introduce BenHalluEval, a fine-grained hallucination evaluation framework for Bengali covering four tasks: Generative Question Answering (GQA), Bangla-English Code-Mixed QA, Summarization, and Reasoning. We construct 12,000 hallucinated candidates using GPT-5.4 across twelve task-specific hallucination types, drawn from three existing Bengali datasets, and evaluate seven LLMs spanning reasoning-oriented, multilingual, and Bengali-centric categories under a dual-track protocol that independently measures false-positive rate on ground-truth instances (Track A) and hallucination detection rate on hallucinated candidates (Track B). To jointly penalise both failure modes and prevent inflated scores from uniform response bias, we propose BenHalluScore, a dual-track calibration metric that ranges from 7.72% to 55.42% across models and tasks, revealing substantial variation in hallucination calibration. Chain-of-thought prompting, applied as a mitigation strategy, shifts response distributions without consistently improving hallucination discrimination. BenHalluEval establishes the first dedicated hallucination benchmark for Bengali and highlights the inadequacy of single-track and prompting-only evaluation approaches for low-resource language settings. The dataset and code are available at this https URL.
[NLP-13] Language Models Can Resolve Reference Compositionally But Its Not Their Native Strength: The Case of the Personal Relation Task
【速读】： 该论文旨在探究大规模语言模型（Large Language Models, LLMs）是否真正具备对自然语言进行组合性语义解释的能力。研究聚焦于语义解释的两个互补维度：指称任务（Extensional task，即确定表达式在现实世界中的指称对象）与内涵任务（Intensional task，即以结构化方式表征表达式的含义）。通过在“个人关系任务”（Personal Relation Task）这一设定下对比人类与LLMs的表现，研究发现两者呈现相反的优势模式：人类在指称任务上表现优于内涵任务，而LLMs则恰恰相反，其在内涵任务上的表现显著优于指称任务。这一结果表明，当前LLMs缺乏对语义的参照性锚定（referential grounding），是其难以实现类人语言理解的关键瓶颈。研究提出的评估框架为深入理解现代机器学习模型的组合性能力提供了更精细的视角。
链接: https://arxiv.org/abs/2605.31480
作者: Bart Evelo,Meaghan Fowlie,Denis Paperno
机构: 未知
类目: Computation and Language (cs.CL)
备注: A pre-MIT Press publication version. Paper accepted to Transactions of the Association for Computational Linguistics
Abstract:Do neural models, such as Large Language Models, genuinely acquire compositional abilities for interpretation of natural language? When we talk about semantic interpretation, we can distinguish two complementary aspects: establishing what an expression refers to in the world (which we call the Extensional task) and representing its sense in a structured way (which we call the Intensional task). We evaluate LLMs and humans on both tasks in the setting of the Personal Relation Task (Paperno 2022) in which, given a universe of people and their relationships with each other, one is asked to interpret a noun phrase such as “Amber’s parent’s friend”. Here, for the Intensional task, the answer is the formula “friend(parent(amber))”, and for the Extensional task, the person. We find that humans and LLMs show opposite strengths: humans perform better on Extensional than Intensional tasks, and LLMs vice versa. Our methodology brings greater nuance to the understanding of compositional abilities in modern machine learning models. Our results support the notion that the lack of referential grounding in LLM training is a crucial missing component in mimicking human-like language understanding.
[NLP-14] Knowledge Boundary Probing and Demand-Guided Intervention for LLM -Based Power System Code Generation
【速读】： 该论文旨在解决在电力系统分析中使用开源权重大语言模型（LLM）进行本地化部署时的可靠性问题，尤其针对因保密性、合规性、可复现性及成本控制需求而必须采用本地服务的场景。核心挑战在于，尽管模型具备较强的推理能力，但其在生成电力系统代码时的首次失败主要并非源于逻辑推理不足，而是由结构化API知识边界错误所主导——包括函数名幻觉、参数误用以及对版本化仿真库中结果表处理不当等问题。为此，论文提出三个关键解决方案：一是构建基于执行验证的基准测试框架PowerCodeBench，将自然语言操作请求与真实pandapower代码及数值真值配对；二是设计从L0到L3的文档驱动探查流程，以量化评估各模型的API知识分布；三是引入一种边界感知干预机制，通过查询端的API需求估计、目标导向的主动文档注入与路由式被动修正相结合的方式，实现精准纠错。实验在2,000个任务的冻结版本上评估了十款开源模型（1.5B–480B参数）及四款商用中端API，结果表明该干预策略使所有7B及以上参数的开源模型及所有商用API准确率提升32至56个百分点，其中70B–120B量级的开源模型达到商用中端水平，而Llama-3.1-405B与Qwen3-Coder-480B表现领先。该方法在保持全上下文准确率上限的同时，仅消耗41%的提示词（prompt-token）成本，为无需微调或云端推理即可实现高可靠性的本地化电网分析辅助提供了部署时即可用的性能路径。
链接: https://arxiv.org/abs/2605.31478
作者: Hui Wu,Xiaoyang Wang,Zhong Fan
机构: University of Exeter
类目: oftware Engineering (cs.SE); Computation and Language (cs.CL); Systems and Control (eess.SY)
备注: 43 pages, 12 figures, includes supplementary material
Abstract:Large language models (LLMs) are increasingly used to automate power-system analysis, but many utilities and energy-research labs require on-premise serving for confidentiality, regulatory, reproducibility, and cost reasons. This makes the reliability of open-weight models a deployment issue. We show that first-pass failures in power-system code generation are dominated not by reasoning alone, but by structured API-knowledge boundary errors: hallucinated function names, misused parameters, and mishandled result tables in versioned simulation libraries. We introduce PowerCodeBench, an execution-validated benchmark generator that pairs natural-language operator queries with pandapower code and numerical ground truth; an L0-L3 documentation-driven probing procedure that measures per-model API knowledge profiles; and a boundary-aware intervention that combines query-side API demand estimation with targeted proactive documentation injection and routed reactive correction. On a 2,000-task frozen release, we evaluate ten open-weight LLMs (1.5B-480B parameters) and four commercial mid-tier APIs. The intervention improves every evaluated open-weight model of at least 7B parameters and every commercial API by 32 to 56 accuracy points. Open-weight models in the 70B-120B range match the commercial mid-tier accuracy range, while Llama-3.1-405B and Qwen3-Coder-480B lead the panel. The targeted prompts preserve the full-context accuracy ceiling while using 41% of the prompt-token cost. The result is an accuracy-side, deployment-time path toward reliable on-premise LLM assistance for grid-analysis workflows without fine-tuning or cloud inference. Comments: 43 pages, 12 figures, includes supplementary material Subjects: Software Engineering (cs.SE); Computation and Language (cs.CL); Systems and Control (eess.SY) Cite as: arXiv:2605.31478 [cs.SE] (or arXiv:2605.31478v1 [cs.SE] for this version) https://doi.org/10.48550/arXiv.2605.31478 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[NLP-15] Scaling Conversational Hungarian ASR: The BEA-Dialogue Corpus
【速读】： 该论文旨在解决匈牙利语对话式自动语音识别（Conversational Automatic Speech Recognition, CASR）因公开可用的对话风格训练数据有限而面临的挑战。现有数据集BEA-Dialogue虽已部分缓解此问题，但其严格的说话人独立划分（speaker-disjoint split）导致可用于训练的有效数据仅85小时，限制了模型性能提升。为此，本文提出BEA-Dialogue+，通过放宽实验者与对话伙伴之间的分割约束（同时保持主要说话人完全分离），将可利用的转写自然对话数据扩展至200小时。其核心解决方案在于在保证主说话人无重叠的前提下，引入更灵活的数据划分策略，从而实现训练数据量的显著增加，并支持对额外训练数据与说话人重叠之间权衡关系的可控研究。实验表明，尽管更大规模的数据使未经微调的模型面临更高挑战，但基于序列输出训练（Serialized Output Training, SOT）的微调方法在词错误率（WER）、字符错误率（CER）、上下文相关词错误率（cpWER）及上下文相关字符错误率（cpCER）上均实现了稳定改进。因此，BEA-Dialogue+不仅为匈牙利语对话式语音识别提供了更大规模且仍具挑战性的基准，也为对话转录系统的训练与评估提供了实用资源。
链接: https://arxiv.org/abs/2605.31469
作者: Máté Gedeon,Piroska Zsófia Barta,Péter Mihajlik,Katalin Mády
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Sound (cs.SD); Audio and Speech Processing (eess.AS)
备注:
Abstract:Conversational automatic speech recognition in Hungarian is constrained by the limited amount of publicly available dialogue-style training data. The BEA-Dialogue corpus addresses this need, but its strictly speaker-disjoint train/dev/eval split reduces the usable material to only 85 hours. In this paper, we introduce BEA-Dialogue+, an expanded version of the corpus that relaxes the split criterion for experimenters and dialogue partners while preserving complete separation of the primary speakers. This results in 200 hours of transcribed natural conversations and enables a controlled study of the trade-off between additional training data and speaker overlap across the splits. We evaluate several Whisper- and FastConformer-based models on both corpus versions, including Serialized Output Training (SOT)-based fine-tuning for dialogue transcription. Our results show that the larger corpus is more challenging for models without fine-tuning, whereas SOT-based adaptation yields consistent improvements in WER, CER, cpWER, and cpCER. Overall, BEA-Dialogue+ provides a substantially larger yet still demanding benchmark for Hungarian dialogue ASR, and a practical resource for training and evaluating dialogue transcription systems.
[NLP-16] PithTrain: A Compact and Agent -Native MoE Training System
【速读】： 该论文旨在解决当前前沿语言模型中混合专家（Mixture-of-Experts, MoE）训练框架在演进过程中面临的高成本问题，尤其是针对新架构与系统优化的适配与开发成本。尽管现有生产级框架经过多年工程积累已实现高效训练吞吐量，但其维护和扩展仍高度依赖人工干预，难以快速响应新兴需求。随着生成式AI编程代理（AI coding agents）的兴起，理论上可自动化部分训练框架的开发流程，从而加速演进。然而，现有评估体系仅关注训练吞吐量，忽略了使用编程代理操作、理解与扩展框架所隐含的额外开销——即“代理任务效率”（Agent-Task Efficiency, ATE）。为此，论文提出PithTrain，一个基于四项代理原生（agent-native）设计原则构建的轻量化、面向代理优化的MoE训练框架，并引入ATE-Bench基准测试集以全面评估真实场景下的代理交互效率。实验表明，PithTrain在保持与生产级框架相当的吞吐量的同时，在ATE-Bench上显著提升了代理任务效率，最多减少62%的代理交互轮次（Agent Turns）和64%的活跃GPU时间，验证了其在降低代理使用成本方面的有效性。
链接: https://arxiv.org/abs/2605.31463
作者: Ruihang Lai,Hao Kang,Haozhan Tang,Akaash R. Parthasarathy,Zichun Yu,Junru Shao,Todd C. Mowry,Chenyan Xiong,Tianqi Chen
机构: 未知
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Distributed, Parallel, and Cluster Computing (cs.DC)
备注:
Abstract:Mixture-of-Experts (MoE) has become the dominant architecture for frontier language models. To meet this demand, production frameworks have built optimized MoE training stacks over years of engineering effort. Yet evolving these stacks for new architectures and system optimizations remains expensive. With the rise of AI coding agents, they could automate parts of training-framework development and accelerate this evolution. But applying them to these existing frameworks carries hidden costs, invisible to today’s throughput-only evaluations. We name this missing dimension agent-task efficiency (ATE): the cost of using coding agents to understand, operate, and extend a framework. Grounded in four agent-native design principles, we build PithTrain, a compact, agent-native MoE training framework. We further introduce ATE-Bench, covering real-world training-framework tasks. Our evaluation shows PithTrain matches the throughput of production frameworks, and on ATE-Bench, PithTrain enables higher agent-task efficiency, with up to 62% fewer Agent Turns and 64% less Active GPU Time.
[NLP-17] DRIFT: Decoupled Rollouts and Importance-Weighted Fine-Tuning for Efficient Multi-Turn Optimization
【速读】： 该论文旨在解决大语言模型在多轮交互场景中行为优化所面临的两难困境：在线强化学习虽能有效处理多轮动态，但因每次更新需生成完整的修正轨迹而成本过高；而离线监督微调（SFT）虽高效，却易受分布偏移和行为坍缩的影响。其解决方案的关键在于提出DRIFT（解耦滚动与重要性加权微调）框架，该框架基于理论洞察——KL正则化强化学习目标等价于重要性加权的监督学习，通过从固定参考策略中离线采样交互轨迹，基于回报计算重要性权重，并利用加权SFT对策略进行优化，从而实现滚动过程与优化过程的解耦。实验表明，DRIFT在保持标准监督微调训练效率与简洁性的前提下，性能可媲美甚至超越多轮强化学习基线。
链接: https://arxiv.org/abs/2605.31455
作者: Jian Mu,Tianyi Lin,Chengwei Qin,Zhongxiang Dai,Yao Shu
机构: 未知
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注:
Abstract:Large language models are increasingly deployed in multi-turn interactive settings where users or environments can iteratively provide lightweight feedback. Unfortunately, optimizing such behavior presents a sharp dilemma in practice: online reinforcement learning is able to effectively address multi-turn dynamics but is prohibitively expensive due to the cost of generating full correction trajectories at every update, whereas offline supervised fine-tuning (SFT) is efficient but suffers from distribution shift and behavioral collapse. To this end, we novelly propose DRIFT (Decoupled Rollouts and Importance-Weighted Fine-Tuning), a framework that operationalizes the theoretical insight that the KL-regularized RL objective is equivalent to importance-weighted supervised learning. DRIFT decouples rollout from optimization by sampling offline interaction trajectories from a fixed reference policy, deriving return-based importance weights, and optimizing the policy via weighted SFT on the resulting dataset. Empirically, we demonstrate that DRIFT matches or exceeds the performance of multi-turn reinforcement learning baselines while maintaining the training efficiency and simplicity of standard supervised fine-tuning. Code is available at this https URL.
[NLP-18] Fine-grained Verification via Diagnostic Reasoning Supervision for Aspect Sentiment Triplet Extraction
【速读】： 该论文旨在解决生成式情感三元组提取（Aspect Sentiment Triplet Extraction, ASTE）中后处理验证不足的问题，即现有方法多聚焦于端到端的三元组抽取，而对抽取结果的全局有效性验证相对薄弱。这导致模型输出的三元组虽在局部语境上合理，却可能在整体语义或逻辑上存在错误，从而降低系统的可靠性。针对这一问题，其解决方案的关键在于提出FiVeD框架——一种基于诊断性推理监督的细粒度验证机制。该框架通过多任务学习训练验证器，以三元组有效性分类和质量评分估计为核心目标，并辅以错误类型分类与理由生成作为辅助任务，实现对候选三元组的精细评估。研究构建了分层错误类别体系，并在语义与句法约束下生成合理的错误三元组样本，利用预训练大语言模型（LLM）结合特定评分标准，自动生成质量分数与可解释的诊断理由。推理阶段，基于输出的质量分数对候选结果进行过滤或重排序，支持灵活调整精确率与召回率之间的权衡。实验表明，作为即插即用的验证模块，FiVeD在多个基线模型上均显著提升性能，最高达3.53 F1分数点。
链接: https://arxiv.org/abs/2605.31446
作者: Wenna Lai,Haoran Xie,Guandong Xu,Qing Li,S. Joe Qin
机构: The Hong Kong Polytechnic University(香港理工大学); Lingnan University(岭南大学); Education University of Hong Kong(香港教育大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 25 pages, 13 figures, and 6 tables
Abstract:Aspect Sentiment Triplet Extraction (ASTE) aims to identify aspect terms, opinion terms, and sentiment polarities as structured triplets, providing essential inputs for downstream information system applications such as opinion mining, explainable recommendations, and review summarization. Prior work mainly focuses on end-to-end extraction, while post hoc verification of extracted triplets remains comparatively underexplored. This gap limits the reliability of ASTE systems, since predicted triplets may be locally plausible while being globally invalid. Moreover, candidate invalidity is multi-faceted and candidate usability is inherently graded, motivating a fine-grained verification mechanism that can filter or re-rank outputs from diverse extractors. In this paper, we propose FiVeD, a framework for Fine-grained Verification with Diagnostic reasoning supervision. Specifically, the verifier is trained with multiple complementary objectives, including validity classification and quality score estimation as primary tasks, with error type classification and rationale generation as auxiliary tasks. We define hierarchical error categories and construct plausible incorrect triplets under semantic and syntactic constraints, and leverage an off-the-shelf LLM with task-specific rubrics to produce quality scores and diagnostic rationales. During inference, the resulting quality scores are used to filter candidate outputs, supporting adjustable precision-recall tradeoffs. Experiments across multiple ASTE baselines demonstrate that FiVeD consistently improves extraction performance by up to 3.53 F1 points as a plug-and-play verification module.
[NLP-19] Used Car Salesbots? Honesty and Credulity of LLM s as Bargaining Agents under Partial Information
【速读】： 该论文旨在解决在模拟讨价还价场景中，基于大语言模型（Large Language Models, LLMs）的智能体在不同信息结构（完全信息、信息不对称或相互不确定性）下如何进行谈判，并评估其表现与博弈论均衡解的偏离程度。核心问题在于：当智能体被优化以最大化经济收益时，其谈判能力是否提升，同时是否会伴随诚信度下降（即更倾向于隐瞒或歪曲私有信息）和可信度降低（即对对方提供的信息表现出更强的不信任）。研究的关键解决方案是通过零样本（zero-shot）提示工程与微调（fine-tuning）两种方式构建智能体，并在真实人类行为数据驱动的讨价还价场景中系统性地评估其诚实性（honesty）与可信度（credulity）——结果表明，尽管微调可显著提升智能体获取有利交易的能力，但同时也导致其欺骗倾向增强，揭示了任务目标优化可能带来的安全风险。
链接: https://arxiv.org/abs/2605.31445
作者: Antonio Valerio Miceli-Barone,Vaishak Belle,Shay B. Cohen
机构: University of Edinburgh
类目: Computer Science and Game Theory (cs.GT); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 18 pages, 14 figures
Abstract:In this work we study agents in simulated bargaining scenarios, where a buyer and a seller communicate through a text channel and attempt to negotiate mutually beneficial trades, under different information regimes (complete information, information asymmetry or mutual uncertainty). We evaluate their performance w.r.t. game-theoretical solutions and further investigate their honesty (their tendency to disclose or withhold information or to mislead and deceive) as well as their credulity (their tendency to trust or distrust information provided by the other agent). We study zero-shot LLM agents with simple prompting scaffolding as well as fine-tuned agents, in order to investigate whether optimising the agents to maximise financial profits makes them stronger negotiators but also more dishonest and less trusting. We find that off-the-shelf LLMs all substantially deviate from game-theoretical equilibria, they attempt to lie about their private information but cannot efficiently exploit information asymmetries. Fine-tuning on financial utility makes the agents stronger at achieving better deals but also more dishonest, highlighting the risks that optimising agents for a task can have on their safety. We release our code and a dataset of bargaining scenarios. Comments: 18 pages, 14 figures Subjects: Computer Science and Game Theory (cs.GT); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG) Cite as: arXiv:2605.31445 [cs.GT] (or arXiv:2605.31445v1 [cs.GT] for this version) https://doi.org/10.48550/arXiv.2605.31445 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[NLP-20] SCOPE: Self-Play via Co-Evolving Policies for Open-Ended Tasks
【速读】： 该论文旨在解决生成式 AI 在开放性任务中缺乏有效自监督训练机制的问题，现有方法依赖可规则验证的答案或人工精心设计的提示词（prompt），难以推广至自由形式、复杂多变的任务场景。其核心解决方案是提出 SCOPE——一种无需外部数据的自洽式自我博弈框架，通过协同进化两个策略：挑战者（Challenger）负责生成基于文档的开放性任务，求解者（Solver）则通过多轮检索进行回答。系统利用初始模型的冻结副本作为自评判器（self-judge），根据源文档自动构建任务专属评分标准（rubric），并据此评估求解者的输出。实验表明，SCOPE 在三个 7-8B 参数量级的指令微调模型（Qwen2.5、Qwen3、OLMo-3）上，于八项基准测试中将开放性任务性能提升最高达 +10.4 分，且在不使用任何人工标注提示的情况下，性能达到使用约 9,000 条精选提示训练的 GRPO_data 模型水平。值得注意的是，尽管仅在开放性任务上训练，SCOPE 还显著提升了未见的短文本问答任务表现，最高提升 +13.8 分，并在所有三类模型上超越 GRPO_data。消融实验揭示：挑战者与求解者的协同进化对维持任务难度处于求解者能力前沿至关重要；性能提升源于检索与内容生成能力的双重增强，具体贡献比例依任务而异；而评分标准生成质量是当前自评判机制的主要瓶颈。
链接: https://arxiv.org/abs/2605.31433
作者: Wai-Chung Kwan,Aryo Pradipta Gema,Joshua Ong Jun Leang,Pasquale Minervini
机构: University of Edinburgh(爱丁堡大学); Imperial College London(帝国理工学院); Miniml.AI(最小化人工智能)
类目: Computation and Language (cs.CL)
备注:
Abstract:Self-play can train language models without external supervision. However, existing methods require rule-checkable answers, leaving open-ended tasks dependent on curated prompts or frontier-model judges. We introduce SCOPE, a data-free self-play framework for open-ended tasks that co-evolves two policies: a Challenger that generates document-grounded tasks, and a Solver that answers them through multi-turn retrieval. A frozen copy of the initial model serves as the self-judge, which writes task-specific rubrics from the source document and grades Solver responses against them. Across three 7-8B instruction-tuned models (Qwen2.5, Qwen3, OLMo-3), SCOPE improves open-ended performance by up to +10.4 points on eight benchmarks and matches or exceeds GRPO_data trained on ~9K curated prompts. Although trained only on open-ended tasks, SCOPE also improves held-out short-form QA by up to +13.8 points on seven held-out benchmarks, surpassing GRPO_data on all three models. Ablations show that co-evolving the Challenger is necessary to keep tasks near the Solver’s frontier, that gains arise from improvements in both retrieval and synthesis with the relative contribution varying by task, and that rubric generation quality is the bottleneck for self-judging.
[NLP-21] DOA: Training-Free Decoder-Only Attention Policy for Long-Form Simultaneous Translation with SpeechLLM s
【速读】： 该论文旨在解决生成式语音翻译（Simultaneous Speech-to-Text Translation, SimulST）中如何在不依赖训练微调或启发式等待策略（wait-k）的前提下，实现长文本场景下的低延迟、高质量流式翻译问题。现有基于注意力机制的编码器-解码器模型依赖交叉注意力提供显式的对齐信号以指导流式决策，而当前主流的语音大语言模型（Speech Large Language Models, SpeechLLMs）采用仅包含自注意力（self-attention）的解码器架构，缺乏显式的对齐机制，因此其自注意力是否具备足够稳定的对齐信号以支持流式策略成为关键挑战。为此，论文提出无需训练的解码器自注意力策略（Decoder-Only Attention, DOA），通过从自注意力权重中提取代理对齐信号（proxy alignment），为流式决策提供有效依据。实验结果表明，DOA能够在不重新训练的情况下，使现成的SpeechLLMs实现接近离线解码质量的低延迟长文本流式翻译，显著提升了模型在真实应用场景中的实用性与泛化能力。
链接: https://arxiv.org/abs/2605.31432
作者: Sara Papi,Luisa Bentivogli
机构: Fondazione Bruno Kessler, Italy
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Sound (cs.SD)
备注:
Abstract:Simultaneous speech-to-text translation (SimulST) generates translations while speech is still unfolding, requiring a streaming policy that decides when to read and when to write. State-of-the-art approaches rely on attention-based encoder-decoder models where cross-attention provides explicit alignment signals. In contrast, Speech Large Language Models (SpeechLLMs) are decoder-only architectures relying solely on self-attention. This raises a central question: whether decoder self-attention contains sufficiently stable alignment signals to guide the streaming policy. Moreover, existing approaches typically rely on training-based adaptations or heuristic wait- k policies and have not been validated in long-form settings. To fill these gaps, we propose Decoder-Only Attention (DOA), a training-free policy that enables long-form simultaneous translation with off-the-shelf SpeechLLMs by deriving a proxy alignment from self-attention. Experiments on Phi4-Multimodal and Qwen3-Omni show that DOA provides an effective alignment signal for supporting streaming decisions, enabling low-latency long-form SimulST with quality close to offline decoding without retraining.
[NLP-22] Neuro-symbolic Syntactic Parsing: Shaping a Neural Network with the CYK Algorithm
【速读】： 该论文旨在解决如何将复杂符号计算算法直接嵌入神经网络架构中的问题，特别是针对上下文无关文法在乔姆斯基规范形式下的解析任务。其核心挑战在于如何在保持算法精确性的同时实现神经网络对符号逻辑的可学习与可泛化能力。解决方案的关键在于提出CYKNN——一种基于循环神经网络（Recurrent Neural Network, RNN）的新型架构，通过可训练的矩阵-向量运算显式编码柯克-尤格-卡萨米（Cocke-Youger-Kasami, CYK）算法的动态规划过程。实验结果表明，在仅使用4种简单语法变体的情况下，该方法在上下文学习设置下超越了参数量超过200亿的大型语言模型（Large Language Model, LLM），并优于经过LoRA微调的通义千问（Qwen）系列小规模模型。这一尝试为神经符号方法学（Neuro-Symbolic Methodology）提供了一条新路径，即通过结构化、可解释的算法嵌入实现符号推理与深度学习的深度融合。
链接: https://arxiv.org/abs/2605.31421
作者: Fabio Massimo Zanzotto,Federico Ranaldi,Giorgio Satta
机构: University of Rome Tor Vergata, Italy; University of Padua, Italy
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Data Structures and Algorithms (cs.DS)
备注: 9 content pages
Abstract:In this paper, we show the possibility of a direct injection of algorithms into neural network architecture. We focus on a complex algorithm, that is, Cocke-Youger-Kasami (CYK) for parsing context-free grammars in Chomsky Normal Form and we propose CYKNN, a simple recurrent neural network architecture for encoding the CYK algorithm in trainable matrix-vector this http URL experimented with a very simple grammar with 4 variations showing that our approach outperforms existing LLMs with more than 20B parameters with an in-context learning setting and smaller LLMs of the Qwen family fine-tuned with LoRA. Our attempt paves the way to a different approach to neuro-symbolic methodologies.
[NLP-23] Skill Availability and Presentation Granularity in Large-Language-Model Agents : A Controlled SkillsBench Study
【速读】： 该论文旨在探究可控技能知识（controlled skill knowledge）在推理阶段的呈现粒度（presentation granularity）对下游任务成功率的影响。核心问题在于：不同抽象层级的技能呈现方式（如低抽象、中等抽象及是否包含示例）是否显著影响大语言模型代理在复杂任务中的表现。研究采用经过官方验证的30个任务、领域平衡的SkillsBench子集，结合两种具备推理能力的模型配置（GPT-5.5与DeepSeek V4-Flash），设置六种技能条件，并在每个任务-条件-模型组合下进行五次试验，共获得1,800行数据。关键发现表明，相较于不提供技能，提供技能可使任务平均通过率提升26.7至36.0个百分点（GPT-5.5）和18.0至26.0个百分点（DeepSeek V4-Flash），显示出技能可用性具有明确的正向效应。然而，在更精细的呈现粒度对比中，低抽象与高抽象指导之间的差异仅分别为+0.7和-6.7个百分点，且95%置信区间均包含零值，表明其效果微小且统计上不确定；类似地，中等抽象指导下增加一个示范案例的效果也仅为+0.7至+1.3个百分点，同样不显著。稳健性检验进一步确认了主结论的一致性。因此，解决方案的关键在于：技能可用性是决定任务成功的核心因素，而呈现粒度的细微调整在当前设定下并未带来显著或一致的性能增益，其影响高度依赖于具体模型且效果有限。
链接: https://arxiv.org/abs/2605.31408
作者: Xiaonan Xu,Wenjing Wu
机构: Northern Arizona University (北方亚利桑那大学); University of Colorado Boulder (科罗拉多大学博尔德分校)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Skill documents provide procedural knowledge to large-language-model agents at inference time. This article studies whether the presentation granularity of controlled skill knowledge changes downstream task success. The experiment uses a pinned SkillsBench version, a 30-task domain-balanced subset validated by official oracle runs, two reasoning-enabled model configurations, six skill conditions, and five trials per task-condition-model cell. Skill availability is the clearest empirical signal. Relative to no skill, skill conditions increase task-mean pass rate by 26.7 to 36.0 percentage points for GPT-5.5 and by 18.0 to 26.0 percentage points for DeepSeek V4-Flash. The final data contain 1,800 rows, with 900 rows for each model. The task is the inference unit. Five trials are aggregated within each task-condition-model cell before paired contrasts are estimated over 30 tasks. The primary presentation contrasts are smaller and uncertain. Low-abstraction guidance differs from high-abstraction guidance by +0.7 percentage points for GPT-5.5 and -6.7 percentage points for DeepSeek V4-Flash, with both 95% bootstrap confidence intervals crossing zero. Adding one worked example to medium-abstraction guidance differs from the no-example variant by +0.7 and +1.3 percentage points. Mean-reward robustness checks preserve the same substantive conclusion. In this controlled subset, skill availability is associated with higher success than no skill, while the tested presentation-granularity changes yield small, uncertain, and model-dependent effects.
[NLP-24] he Sword Shield and Achilles Heel: Characterizing the Linguistic Inductive Bias of Large Language Models for Spatial Reasoning in Navigation Planning
【速读】： 该论文旨在解决大语言模型（Large Language Model, LLM）在导航规划中依赖文本化空间表征时，其语言结构与上下文特征选择对模型行为的潜在影响未被充分理解的问题。现有方法通常将语言格式与拓扑、几何等上下文信息的编码视为中立的工程决策，而忽视了这些设计因素对LLM推理能力的根本性塑造作用。为此，作者提出一种双干预框架（dual-interventional framework），通过分离语言结构与不同上下文线索（如拓扑、语义、几何）的影响，系统评估LLM在导航任务中的语言归纳偏置（linguistic inductive bias）。该框架的核心在于：语言结构干预（representation intervention）通过改变文本表达形式与语言压缩程度，揭示语言表征在何种条件下促进或抑制导航规划；上下文干预（context intervention）结合特征组合与冲突探测，明确模型对不同类型上下文线索的偏好与脆弱性。实验结果表明，拓扑信息是保障鲁棒规划的基石，语言格式具有双刃剑效应，其有效性取决于模型规模、任务需求及压缩水平，而语义信息则构成致命弱点——错误的语义线索会系统性破坏规划流程。因此，有效的基于文本的空间表征应保持拓扑完整性、根据模型容量校准压缩程度，并确保语义正确性，而非采用单一固定表征形式。
链接: https://arxiv.org/abs/2605.31404
作者: Xudong Zhang,Jian Yang,Shengkai Wang,Jiangpeng Tian,Shaowen Chen,Xian Wei,Ke Li,Xiong You
机构: East China Normal University (华东师范大学); Information Engineering University (信息工程大学); Zhengzhou University (郑州大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Large Language Model (LLM)-based navigation systems commonly construct explicit spatial representations (e.g., topological graphs, semantic raster maps) and translate them into textual descriptions as LLMs’ inputs. However, the linguistic structures of such text-based spatial representations and the choices of contextual features (e.g., topology, geometry) they contain are often treated as neutral engineering decisions rather than key factors that shape LLMs’ behavior. To fill the gap, we propose a dual-interventional framework that disentangles linguistic structures from different contextual cues to evaluate the linguistic inductive bias of LLMs for navigation planning. In the framework, representation intervention varies the linguistic format and the degree of linguistic compression, clarifying when linguistic representations support or inhibit navigation planning. Context intervention, combined with contextual feature combination and conflict probing, explicitly clarifies the preferences and weaknesses of LLMs when processing different contextual cues. Experiments across diverse spatial reasoning tasks and multiple model scales reveal a consistent pattern: topological information is a sturdy shield and the backbone of robust planning; linguistic format is a double-edged sword whose effect depends on model size, task demands, and the compression level; and semantic information is a fatal Achilles’ heel – incorrect semantic cues can systematically derail the planning process. Overall, our study shows that effective text-based spatial representations in LLM-based navigation should preserve topological integrity, calibrate representational compression to model capacity, and ensure semantic correctness, rather than simply adopting a single representation. Our code is publicly available at this https URL.
[NLP-25] "Intelegi Româneşte? A Recipe for Romanian Vision-Language Models
【速读】： 该论文旨在解决视觉-语言模型（Vision-Language Models, VLMs）在低资源语言场景下性能显著下降的问题，尤其针对罗马尼亚语这类缺乏大规模图文语料库和文化语境化评估基准的语言。其核心挑战在于：现有VLMs主要基于英语数据训练，在非英语低资源语言上表现不佳，且缺乏适配本地文化语境的评测体系。为此，论文提出了一套系统性解决方案，关键在于构建完整的罗马尼亚语专用VLM开发流程，涵盖数据构建、模型架构选择与评估体系设计。具体而言，通过机器翻译将英文主流VLM训练与评估数据集中的文本注释及图像内文本转换为罗马尼亚语，同时保持视觉语义对齐；在此基础上，系统性地训练并消融不同规模的视觉主干网络、多语言至罗马尼亚语适配的语言主干模型，以及基于光学字符识别（OCR）风格的图文数据；此外，构建了名为HoraVQA的文化原生评估集，以反映罗马尼亚日常场景的真实语境。实验结果表明，经过罗马尼亚语适配的VLM在各项基准测试中均优于同规模模型，并在多数任务上超越更大一级的通用模型，验证了语言特异性适配在提升低资源语言VLM性能中的关键作用。
链接: https://arxiv.org/abs/2605.31401
作者: Mihai Masala,Marius Leordeanu,Mihai Dascalu,Traian Rebedea
机构: National University of Science and Technology POLITEHNICA Bucharest (布加勒斯特理工大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Vision-Language Models (VLMs) largely follow the text-only LLM trajectory, excelling on English benchmarks but sharply degrading on low-resource languages, where neither large-scale image-text corpora nor culturally grounded evaluations exist. We present a systematic study of building a language-specific VLM for Romanian, covering the full pipeline from data construction to architectural choices. We translate established English VLM training and evaluation corpora into Romanian, applying machine translation to textual annotations and to in-image text, preserving visual grounding while adapting the textual content. Using this data, we train and ablate a series of VLMs to isolate the contribution of (i) vision backbones of varying scale and pretraining, (ii) language backbones from multilingual to Romanian-adapted LLMs, and (iii) OCR-style image-text data. We further curate HoraVQA, a culturally native evaluation set grounded in Romanian everyday scenes. Romanian-adapted VLMs consistently outperform their same-sized counterparts and, across all evaluated benchmarks, even surpass models from the next larger size category.
[NLP-26] arget-Side Paraphrase Augmentation for Sign Language Translation with Large Language Models CVPR2026
【速读】： 该论文旨在解决手势语言翻译（Sign Language Translation, SLT）中因配对的手势视频-文本语料库稀缺以及目标词汇表存在长尾分布所带来的性能瓶颈问题。其核心挑战在于有限的标注数据难以支撑模型充分学习复杂多样的表达，而长尾词汇在训练中往往得不到有效覆盖，导致翻译质量受限。为此，论文提出一种基于生成式AI（Generative AI）的目标侧数据增强方法：利用GPT-4o在保持原始手势输入不变的前提下，生成参考句子的受控改写变体，从而扩充训练语料。关键创新在于采用两阶段训练策略——先在增强后的语料上进行预训练，再在原始参考句上微调，以兼顾泛化能力与语义保真度。实验在三个具有互补挑战的数据集上验证：PHOENIX14T（德语手语，词汇多样性适中）、GSL（希腊手语，录制高度重复）、LSA-T（阿根廷手语，极端长尾稀疏）。结果显示，在PHOENIX14T上BLEU-4指标从9.56提升至10.33；尽管在趋于饱和的GSL和极度稀疏的LSA-T上效果受限，但语义评估表明，该方法在忠实性（fidelity）方面有显著提升，而传统词重叠指标未能充分反映此优势。作为首个将大语言模型（LLM）生成的目标侧改写与LLM作为评判者（LLM-as-a-Judge）评估引入SLT的研究，该工作为缓解数据稀缺与长尾问题提供了新范式。
链接: https://arxiv.org/abs/2605.31393
作者: Pedro Dal Bianco,Jean Paul Nunes Reinhold,Oscar Stanchi,Facundo Quiroga,Franco Ronchetti,Ulisses Brisolara Corrêa
机构: Universidad Nacional de La Plata (国立拉普拉塔大学); CONICET (国家科学与技术研究委员会); III-LIDI (LIDI实验室); Federal University of Pelotas (联邦佩洛塔斯大学); Universidade Federal de Pelotas (联邦佩洛塔斯大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: Accepted at GenSign ( this https URL ) at CVPR 2026. Non proceedings track
Abstract:Sign language translation (SLT) remains constrained by limited paired sign-video/text corpora and heavy-tailed target vocabularies. We study target-side augmentation in which GPT-4o generates controlled paraphrase variants of reference sentences while the sign input remains unchanged. A Signformer-style pose-based Transformer is trained under a two-stage schedule: pre-training on the augmented corpus followed by fine-tuning on the original references. We evaluate on three datasets spanning complementary challenges: PHOENIX14T (German Sign Language), with moderate lexical diversity; GSL (Greek Sign Language), with highly ontrolled, repetitive recordings; and LSA-T (Argentinian Sign Language), with severe long-tail sparsity. On PHOENIX14T, augmentation improves BLEU-4 from 9.56 to 10.33. The near-saturated GSL baseline and extremely sparse LSA-T setting reveal the limits of the approach. To our knowledge, this is the first study to apply LLM-generated target-side araphrases and LLM-as-a-Judge evaluation to SLT. The semantic evaluation reveals gains in fidelity that lexical overlap metrics understate. Comments: Accepted at GenSign (this https URL) at CVPR 2026. Non proceedings track Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI) Cite as: arXiv:2605.31393 [cs.CL] (or arXiv:2605.31393v1 [cs.CL] for this version) https://doi.org/10.48550/arXiv.2605.31393 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[NLP-27] Multi-Turn Multi-Agent Dialogue for Collaborative Reconstruction Improves VLM Performance on Spatial Reasoning But Only Barely
【速读】： 该论文旨在解决当前视觉语言模型（Vision-Language Models, VLMs）在人机协作任务中进行空间推理能力不足的问题，特别是在需要结合视觉理解、空间定位、语言引导交互与动作生成的协同结构构建任务中表现有限。其解决方案的关键在于构建一个基于对话的框架，使VLM能够通过多轮语言交互，利用视觉输入与文本描述共同重构目标结构。研究发现，尽管详细的文字描述可提升不同模态条件下的重建成功率，而分解后的图像表示有助于改善性能，但现有VLM在视觉空间定位与基于语境的指令生成方面仍存在显著局限，表明当前模型在实现高效、准确的空间语义对齐与协作推理方面仍面临挑战。
链接: https://arxiv.org/abs/2605.31387
作者: Chalamalasetti Kranti,Sherzod Hakimov,David Schlangen
机构: University of Potsdam (波茨坦大学); German Research Center for Artificial Intelligence (DFKI) (德国人工智能研究中心)
类目: Computation and Language (cs.CL); Robotics (cs.RO)
备注: Preprint
Abstract:Robots operating in diverse environments rely on visual input to interpret objects and spatial layouts. In human-collaborative tasks, they are expected to communicate this understanding through language. Vision-language models (VLMs) support robotic tasks involving visual interpretation, question answering, and instruction following, but their capabilities in collaborative dialogue tasks requiring spatial reasoning remain underexplored. We study this gap through a collaborative structure-building task that combines visual interpretation, grounding, language-guided interaction, and action generation. We develop a framework in which VLMs use dialogue to reconstruct a target structure from visual and textual inputs. We evaluate open-weight and closed VLMs across interaction settings, input modalities, and image representations. Results show that spatial reasoning over visual representations remains difficult for the evaluated VLMs. Detailed text representations of the target yield higher reconstruction success across modality conditions, while decomposed image representations improve performance. These findings reveal limits in visual spatial grounding and grounded instruction generation for collaborative VLM agents.
[NLP-28] LLM Judges Inconsistently Disagree Across Safety Criteria and Harm Categories
【速读】： 该论文旨在解决在无参考（reference-free）设置下，基于大型语言模型（Large Language Models, LLMs）作为自动化评判者进行多维度安全评估时的可靠性问题。研究发现，尽管LLMs在识别暴力等明显有害内容方面表现相对可靠，但在评估金融等受监管领域中机器生成建议的安全性时，其判断一致性显著不足。关键问题是：不同安全标准、内容语言及语言风格均会显著影响模型判断的一致性，且不同评判模型对同一输出的评价存在高度分歧。因此，解决方案的关键在于认识到当前自动化评判机制的局限性，并提出需根据具体应用场景谨慎选择评估标准、引入多模型交叉验证以及结合人工校验等实践建议，以提升评估结果的可信度与稳健性。
链接: https://arxiv.org/abs/2605.31381
作者: Krishnapriya Vishnubhotla,Soumya Vajjala,Akriti Vij,Isar Nejadgholi
机构: National Research Council, Canada; IMDA, Singapore
类目: Computation and Language (cs.CL)
备注: 8 pages plus appendices, under review
Abstract:We evaluate the consistency of automated judges in conducting a multi-dimensional safety evaluation in a reference-free setup. Our results indicate that Large Language Models are unreliable judges in identifying safety issues related to machine-generated advice in regulated domains such as finance, although they are more reliable at identifying more overt forms of unsafe/harmful content such as violence. The degree of inconsistency in a model’s judgments can vary significantly by the chosen safety criteria and can be impacted by the language of the content and its linguistic style as well. Finally, there is high disagreement among different judges for the same output, across domains, safety criteria, and languages. These findings provide new insights on the practice of using LLMs as evaluators and offer several recommendations for practitioners on how to use automated judges in practical scenarios.
[NLP-29] Unlocking Fine-Grained Translation Quality Estimation in LRMs through Synergistically Evolving Implicit and Explicit Reasoning
【速读】： 该论文旨在解决大语言模型在细粒度翻译质量评估（Fine-grained Translation Quality Estimation, QE）任务中表现不佳的问题，尽管已有长推理链的尝试，但模型仍难以有效捕捉细微的翻译质量问题。其核心挑战在于细粒度QE任务本身具有高度复杂性，而现有方法未能充分激发模型内在的多语言理解与推理能力。为此，论文提出一种名为RIEQE（Reasoning both Implicitly and Explicitly for QE）的两阶段训练框架，其关键在于通过协同优化隐式（层间）与显式（词元级）推理能力，实现两种推理模式的共进化。具体而言，首先将复杂的QE任务分解为简单子任务以支持隐式推理的可行性，随后采用非思考监督微调（NonThinking-SFT）直接增强模型的隐式推理倾向与能力；继而通过标准可验证奖励强化学习（Thinking-RLVR）进一步提升显式推理性能。实验结果表明，该框架下隐式与显式推理能力能够相互促进、协同进化，在WMT测试集上，基于Qwen3-4B-Thinking-2507的RIEQE在显式推理性能上超越所有基线模型，同时其隐式推理能力也达到当前编码器类模型的最佳水平，验证了两种推理机制之间的协同增益效应。
链接: https://arxiv.org/abs/2605.31378
作者: Renfei Dang,Xinye Wang,Zhejian Lai,Weilu Xu,Shimin Tao,Daimeng Wei,Min Zhang,Shujian Huang
机构: Nanjing University(南京大学); Huawei Translation Services Center (华为翻译服务部)
类目: Computation and Language (cs.CL)
备注:
Abstract:Large Reasoning Models (LRMs) still struggle with fine-grained translation quality estimation (QE), even with long reasoning chains. We argue that LRMs already possess strong multilingual capabilities, while the core challenge stems from the intrinsic difficulty of learning the fine-grained QE task. In this paper, we propose RIEQE (Reasoning both Implicitly and Explicitly for QE), a simple two-stage training framework that enables the co-evolution of implicit (layer-wise) and explicit (token-wise) reasoning capabilities. To make implicit reasoning feasible, we first decompose the complex QE task into straightforward subtasks. Based on this, our two-stage approach applies: (1) NonThinking-SFT, Supervised Fine-Tuning (SFT) without reasoning chains to directly boost the model’s implicit reasoning tendency and capability; and (2) Thinking-RLVR, standard Reinforcement Learning with Verifiable Reward (RLVR) to subsequently strengthen explicit reasoning. Results demonstrate that implicit and explicit reasoning synergistically co-evolve under our framework. On the WMT test sets, RIEQE based on Qwen3-4B-Thinking-2507 surpasses all baselines in explicit reasoning performance, while its implicit reasoning capability is also comparable to the best current encoder-based models. We further provide evidence for the synergistic collaboration between implicit and explicit reasoning, showing how they mutually benefit each other.
[NLP-30] rading Complexity for Expressivity Through Structured Generalized Linear Token Mixing ICML2026
【速读】： 该论文旨在解决语言模型在建模长程依赖关系时，如何在解码速度与内存开销（特别是缓存大小）之间实现高效权衡的问题。其核心挑战在于：在因果生成场景下，既要保证输入对输出的直接影响能力，又要实现历史输出信息的递归传播，而现有架构（如注意力机制和状态空间模型）在此二者的平衡上存在局限性。本文提出一个统一框架，将这两个关键特性——单步内输入到输出的直接作用、通过历史输出进行信息递归传播——明确分离，并进一步放宽传统递归方程中仅依赖前一时刻状态的限制，允许每个状态可依赖多个先前状态，从而扩展了递归模式的表达能力。在此基础上，通过引入结构化设计，构造出具有理论保障复杂度的新递归模式，实现了运行时间与模型表达能力之间的系统性权衡。该方法不仅为多种主流架构提供了统一理解视角，还通过合成任务和语言建模的实证验证，证明了其在效率与表达力之间的优越性能，为跨模型家族的高效且高表达力的标记混合器（token mixer）设计提供了一套理论完备且可推广的工具集。
链接: https://arxiv.org/abs/2605.31367
作者: Erwan Fagnou,Paul Caillon,Blaise Delattre,Alexandre Allauzen
机构: 未知
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: 20 pages, 3 figures, ICML 2026 main
Abstract:Token mixing layers play a key role in how language models can learn and generate long-range dependencies. Their efficiency relies on the necessary trade-off between decoding speed and the memory requirements, along with the cache size. Considering causal generation, this paper explores new trade-offs thanks to a unified framework which separates two crucial features: (i) the direct influence of inputs on outputs in one generation step; (ii) the recurrent propagation of information through past outputs. This framework encompasses major architectures such as attention and state-space models, but also generalizes the recurrence equations by allowing each state to depend on multiple past states rather than only the immediate predecessor. By introducing structure, we design new recurrence patterns that provably achieve the desired complexity, while providing theoretical insights on their expressivity – trading runtime for expressivity in a principled way. Empirical validation is performed on synthetic tasks, along with language modeling. Together, these results provide a unified toolkit for the understanding and design of efficient and expressive token mixers across model families.
[NLP-31] he Latin Substrate: How Language Models Represent and Mediate Script Choice
【速读】： 该论文旨在解决多书写系统语言中大型语言模型（LLM）如何在不同正字法形式间进行语义一致的文本生成问题，特别是揭示模型内部如何处理和中介书写系统的差异。其核心挑战在于理解模型在跨脚本转换（如从拉丁字母转写为西里尔字母或阿拉伯字母）过程中所依赖的内在机制。解决方案的关键在于通过“对数透镜”（logit lens）分析各层输出分布，发现模型在转写过程中存在稳定的潜在罗马化现象，并结合表征分析与机制分析揭示：1）同一语言的不同书写系统在模型深层逐渐可分，且可通过一个简单的线性引导方向实现脚本切换，该方向对未见过的非拉丁脚本具有良好的泛化能力，但反向映射（拉丁到非拉丁）效果较差；2）在机制层面，仅少数后期注意力头对脚本选择具有因果影响，这些头在不同语言和书写系统间具有可迁移性，表明脚本路由由语言无关的组件实现。综合来看，研究揭示了模型虽基于共享潜在表示组织脚本多样性，却表现出对拉丁脚本的优先倾向，即非拉丁脚本输出由明确、集中的门控机制产生，而拉丁脚本输出则依赖网络中广泛分散的贡献，体现出显著的方向不对称性。
链接: https://arxiv.org/abs/2605.31363
作者: Daniil Gurgurov,Alan Saji,Katharina Trinley,Josef van Genabith,Simon Ostermann
机构: 未知
类目: Computation and Language (cs.CL)
备注: preprint
Abstract:Many languages are written in multiple scripts, requiring large language models (LLMs) to generate equivalent linguistic content in distinct orthographic forms. While prior work suggests that LLMs route information through shared latent representations, how they internally mediate script variation remains poorly understood. We study this question by first examining per-layer output distributions with the logit lens, which reveals consistent latent romanization during transliteration, and then through representational and mechanistic analyses of script generation. At the representational level, we show that scripts of the same language become increasingly separable across layers and that a simple linear steering direction can flip a model’s output script while largely maintaining semantic content. The vector generalizes asymmetrically to writing systems unseen during construction, flipping non-Latin output to Latin reliably, but mapping Latin output into varied non-Latin scripts. At the mechanistic level, we localize a small set of late-layer attention heads that causally mediate script choice. These heads transfer across unrelated languages and writing systems, suggesting that script routing is implemented by language-agnostic components. Across both analyses, we observe a consistent directional asymmetry: non-Latin output is produced by a compact, identifiable gate, while Latin-script output emerges from diffuse contributions across the network. Collectively, our findings hint that LLMs organize script variation around shared latent representations while exhibiting a privileged substrate toward Latin script. Comments: preprint Subjects: Computation and Language (cs.CL) Cite as: arXiv:2605.31363 [cs.CL] (or arXiv:2605.31363v1 [cs.CL] for this version) https://doi.org/10.48550/arXiv.2605.31363 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[NLP-32] A Visually Impaired Assistance Benchmark for VLM-as-a-Judge Evaluation
【速读】： 该论文旨在解决生成式视觉语言模型（VLM）作为裁判（VLM-as-a-Judge）在视障辅助（Visually Impaired Assistance, VIA）任务中评估可靠性不足的问题，尤其关注现有方法在人类评估成本高昂背景下的可信度。其核心挑战在于：尽管VLM-as-a-Judge范式在通用领域展现出潜力，但在VIA场景下其有效性尚未得到充分验证，且存在评估结果不可靠、偏见显著及对抗脆弱性等关键缺陷。解决方案的关键在于提出VIABLE基准，这是首个面向VIA任务的VLM-as-a-Judge评估基准，包含超过30万条判断样本，并构建了“有效性—公正性—稳定性”（Effectiveness–Impartiality–Stability）评估框架与12类故障分类体系。基于此基准的系统性研究揭示，当前主流模型在所有评估维度上均表现不可靠，即使最强模型GPT-5.4的单故障诊断准确率也仅达52.6%，且存在高达94.2%的自我偏好率；开源模型则表现出明显偏差和对抗脆弱性。为此，论文进一步提出VIA-Judge-Agent，一种模型无关的推理时增强机制，通过引入视觉证据提取与基于故障分类的引导式工作流，有效提升了诊断准确率，并生成更受盲人用户（BLV users）青睐的下游辅助响应，从而显著改善评估可靠性与实际应用效果。
链接: https://arxiv.org/abs/2605.31351
作者: Yi Zhao,Siqi Wang,Zhe Hu,Yushi Li,Jing Li
机构: The Hong Kong Polytechnic University (香港理工大学)
类目: Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:AI-based Visually Impaired Assistance (VIA) remains challenging, largely due to the high cost of human evaluation. The VLM-as-a-Judge paradigm may offer a promising alternative, although it has mostly been studied in general domains. We therefore ask whether such judges can be trusted for VIA tasks. To investigate this question, we introduce VIABLE (Visually Impaired Assistance Benchmark for VLM-as-a-Judge Evaluation), the first benchmark for VLM-as-a-Judge evaluation in VIA. VIABLE contains over 300K judgment samples across three scenarios and introduces an Effectiveness–Impartiality–Stability framework with a 12-mode failure taxonomy. Based on VIABLE, our systematic study of seven judges across different model scales shows that existing models are largely unreliable across all evaluation axes. The strongest judge, GPT-5.4, achieves only 52.6% single-failure diagnostic accuracy, yet exhibits the highest self-preference rate at 94.2%; while open-source judges are strongly biased and adversarially fragile. To address these issues, we propose VIA-Judge-Agent, a model-agnostic inference-time harness that augments judges with visual evidence extraction and a taxonomy-guided workflow. It enables positive improvements in diagnostic accuracy and downstream VIA responses more preferred by BLV users. Data and code are available at: this https URL
[NLP-33] FBHM: Functional Benchmarking and Steering of VLMs for Hateful Meme Detection
【速读】： 该论文旨在解决现有视觉-语言模型（Vision-Language Models, VLMs）在仇恨表情包检测任务中因基准数据集结构化观测性导致的因果评估困难问题，即现有基准将修辞性仇恨机制与目标群体特征混淆，使得模型漏洞无法被有效识别。其核心解决方案是构建一个基于功能的仇恨表情包基准FBHM（Functionality Based Hateful Memes），该基准沿两个正交维度设计：25种不同的修辞功能与10个目标社群，共包含5000张表情包。实验表明，当前最先进的VLMs在标准数据集上表现优异，但在FBHM上性能急剧下降至接近随机水平，证明其依赖于数据集特异性启发式而非鲁棒的多模态推理。为高效缩小这一泛化差距，作者提出可学习引导向量（Learnable Steering Vectors, LSV）策略，该方法仅需500个引导样本（50张基础表情包）即可通过因果干预目标实现约30点的宏平均F1分数提升，显著优于上下文学习和参数高效微调（PEFT），且不损害源域性能。
链接: https://arxiv.org/abs/2605.31349
作者: Paramananda Bhaskar,Naquee Rizwan,Daksh Jogchand,Saurabh Kumar Pandey,Animesh Mukherjee
机构: Indian Institute of Technology (IIT), Kharagpur; Microsoft
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Multimedia (cs.MM)
备注:
Abstract:Hateful meme detection remains a formidable challenge for vision-language models, as existing benchmarks are structurally observational - confounding rhetorical hate mechanisms with target community features and preventing causal evaluation of model vulnerabilities. To address this, we introduce FBHM, a systematically curated benchmark of Functionality Based Hateful Memes constructed along two orthogonal axes: 25 distinct rhetorical functionalities and 10 target communities (5,000 memes total). Benchmarking state-of-the-art VLMs reveals a severe generalization gap: models highly accurate on standard datasets catastrophically drop to near-random performance on FBHM, proving they exploit dataset-specific heuristics rather than robust multimodal reasoning. To efficiently close this gap, we propose LSV (learnable steering vectors), an ultra-low data regime strategy that applies a causal intervention objective on as few as 500 steering samples (50 unique base memes), boosting FBHM performance by ~30 Macro-F1 points while outperforming in-context learning and PEFT without degrading source-domain performance.
[NLP-34] Bundesrecht: An Open Library and Corpus for German Statutory Reference Processing
【速读】： 该论文旨在解决德语法律文本中法规引用（statutory references）自动处理的难题，其核心挑战在于法规引用形式紧凑多变、可能包含多个目标、使用特殊缩写，并常指向细粒度的法律条文单元。现有工具仅聚焦于从法律文档中解析引用或在引用明确后访问法规文本，缺乏端到端的处理能力。本文提出的解决方案——bundesrecht，是一个开源资源，包含一个软件库和一个结构化的德国联邦法律语料库，实现了从原始引用字符串到可解析法律条文的全流程处理：通过解析、规范化与消歧，将非标准引用映射为结构化对象，展开紧凑形式为标准格式，并链接至具体法律条款。其关键创新在于构建了覆盖法律层级结构的语料库，支持细粒度的引用处理；并通过实证评估验证了规范化后的引用在去重与匹配方面显著优于传统字符串匹配方法。bundesrecht是首个提供德语法规引用处理完整流水线的开源资源，已发布于PyPI，为法律信息提取与生成式法律AI（Legal AI）研究提供了基础支撑。
链接: https://arxiv.org/abs/2605.31338
作者: Harshil Darji,Martin Heckelmann,Christina Kratsch,Gerard de Melo
机构: Hochschule für Technik und Wirtschaft Berlin, Germany; Hasso-Plattner Institute / University of Potsdam, Germany
类目: Computation and Language (cs.CL)
备注: 10 pages, 1 figure. Preprint
Abstract:Statutory references are central to legal language understanding, but are difficult to process automatically, as they appear in compact and variable surface forms, may combine multiple targets, use special abbreviations, and often point to lower-level units. Existing tools for German focus either on parsing references from legal documents or accessing statutory text once citations are explicit. This paper introduces bundesrecht, an open resource for German statutory reference processing, consisting of a software library and a structured corpus of German federal law. The library parses, normalizes, and resolves German statutory references, mapping raw citation strings to structured objects, expanding compact references into canonical forms, and linking them to statutory provisions. The accompanying dataset preserves the internal hierarchy of statutes from laws to fine-granular subclauses. We evaluate the parser and normalizer on 2,944 annotated German legal references using strict exact-match and micro information extraction metrics. We further evaluate canonical reference deduplication and show that normalized references group real citation surface variants far more reliably than string matching. bundesrecht is the first open resource that covers German statutory reference processing as an end-to-end pipeline, from raw citation string to resolved statutory provision, and is available on PyPI.
[NLP-35] Reinforcement Learning Amplifies Emergent Misalignment from Harmless Rewards
【速读】： 该论文旨在解决强化学习（Reinforcement Learning, RL）场景下生成式模型出现的涌现性错位（Emergent Misalignment, EM）问题，即模型在经过针对特定、狭隘错位示例的微调后，表现出广泛且非预期的错误对齐行为。其核心挑战在于，以往关于EM的研究主要集中于监督微调（Supervised Fine-Tuning, SFT）设置，而对RL驱动的EM现象缺乏充分验证，尤其在小型、开源可获取权重的模型中尚未系统揭示。本文的关键解决方案在于：首先，通过实证表明，在RL框架中奖励明显错位的行为会导致比样本匹配的SFT更严重的通用领域错位；其次，证明了自然可能产生的奖励信号（如不受欢迎的审美偏好或低效的修辞表达）即可诱发显著的EM现象；最后，评估了专为SFT-induced EM设计的训练中缓解策略，发现这些方法在RL场景下具有良好的迁移效果，其中交替插入在线策略安全数据的策略表现最优。这一研究揭示了小规模开源模型中RL诱导的EM现象的可复现性与普遍性，为构建更稳健的对齐机制提供了关键实证基础。
链接: https://arxiv.org/abs/2605.31328
作者: Magnus Jørgenvåg,David Kaczér,Lasse Ruttert,Marvin Gülhan,Lucie Flek,Florian Mai
机构: Bonn-Aachen International Center for Information Technology, University of Bonn, Germany; Lamarr Institute for Machine Learning and Artificial Intelligence, Germany
类目: Computation and Language (cs.CL)
备注:
Abstract:Emergent misalignment (EM) is the surprising tendency of language models to become broadly misaligned after fine-tuning on narrowly misaligned examples. While EM has been extensively studied in the supervised fine-tuning (SFT) setting, evidence that it also arises from reinforcement learning (RL) is limited to large, closed-source models, leaving the phenomenon expensive to study and difficult to reproduce. We characterize EM from RL in small, off-the-shelf open-weight models along three axes. First, we show that rewarding narrow, overtly misaligned behavior produces substantially higher general-domain misalignment than sample-matched SFT. Second, we show that EM from RL can be induced by reward signals that could plausibly arise naturally, such as unpopular aesthetic preferences or poor rhetorical appeals. Third, we evaluate in-training mitigations developed for SFT-induced EM and find that they broadly transfer, with interleaving on-policy safety data performing best.
[NLP-36] Learning from Fine-Grained Visual Discrepancies: Mitigating Multimodal Hallucinations via In-Context Visual Contrastive Optimization ICML2026
【速读】： 该论文旨在解决视觉语言模型（Vision-Language Models, VLMs）中存在的多模态幻觉（Multimodal Hallucination）问题，即模型在生成描述时出现与输入图像不符的虚假或错误信息。现有方法通常采用基于文本的直接偏好优化（Direct Preference Optimization, DPO），但由于缺乏显式的视觉监督，难以有效缓解幻觉现象。尽管已有研究尝试引入视觉偏好DPO，通过对比原始图像与负样本图像来优化模型，但其目标函数存在理论不一致性（源于归一化常数不匹配），且依赖粗粒度负样本，易导致模型学习到捷径（shortcut learning）而非真正理解视觉内容。为此，本文提出上下文内视觉对比优化（In-Context Visual Contrastive Optimization, IC-VCO），通过将对比图像置于共享的多图像上下文中，确保目标函数在数学上严格成立；进一步提出视觉对比蒸馏（Visual Contrast Distillation, VCDist），作为一种可靠性门控的辅助正则项，以增强多图像对比训练与单图像推理之间的一致性；同时设计了一种基于精确语义扰动的对比样本编辑策略，生成具有挑战性的硬负样本。实验在五个基准数据集上验证了IC-VCO在整体性能上的领先性，充分证明了所提方法的有效性。
链接: https://arxiv.org/abs/2605.31312
作者: Haolin Deng,Xin Zou,Zhiwei Jin,Chen Chen,Haonan Lu,Xuming Hu
机构: 未知
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL)
备注: ICML 2026
Abstract:Multimodal hallucination remains a persistent challenge for Vision-Language Models (VLMs). Standard textual Direct Preference Optimization (DPO) often fails to mitigate it due to a lack of explicit visual supervision. While existing works introduce visual preference DPO by contrasting original images against negative ones, they suffer from a theoretically inconsistent objective caused by partition function mismatches and rely on coarse-grained negatives that could enable shortcut learning. In this work, we propose In-Context Visual Contrastive Optimization (IC-VCO). By placing contrastive images within a shared multi-image context, IC-VCO ensures a mathematically rigorous objective. We further introduce Visual Contrast Distillation (VCDist), an auxiliary reliability-gated regularizer that encourages consistency between multi-image contrastive training and single-image inference. Finally, we propose a contrastive sample editing strategy that generates hard negatives via precise semantic perturbations. Experiments on five benchmarks demonstrate IC-VCO’s best overall performance and the effectiveness of our sample editing strategy. Code and data are available at this https URL.
[NLP-37] Divergence Decoding: Inference-Time Unlearning via Auxiliary Models
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在训练过程中记忆敏感数据所引发的隐私泄露与版权侵权问题，尤其针对从已有模型检查点中移除特定知识这一难题。现有去记忆化（unlearning）方法普遍存在性能灾难性下降或对复杂查询无效等局限性。本文提出一种名为“发散解码”（Divergence Decoding, DD）的新机制，其核心在于利用小型辅助模型在推理阶段引导大语言模型的输出分布，使其在生成时避开特定敏感数据对应的逻辑概率（logits）。该方法的关键优势在于：辅助模型可通过标准预训练与微调流程高效训练，且在多种模型规模和训练数据量下均显著优于当前最先进的基线方法，展现出高效且低成本的去记忆化能力。进一步地，研究证明该经过引导的输出分布可被轻易蒸馏回原始模型，实现知识迁移。由于该方法适用于任意概率模型，研究还验证了其在图像生成领域的泛化能力，表明其具备跨模态应用潜力。
链接: https://arxiv.org/abs/2605.31293
作者: Humzah Merchant,Bradford Levy
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:Large Language Models (LLMs) frequently memorize sensitive training data thereby creating significant privacy and copyright risks. Addressing these risks, i.e., removing such knowledge from an existing model checkpoint, has proven challenging as many unlearning methods lead to catastrophic utility loss or are ineffective for complex queries. We introduce Divergence Decoding (DD), a mechanism that uses small auxiliary models to steer the logits of the LLM away from specific data during inference. Training these models is straight forward, i.e., we use standard pre-training and fine-tuning setups. We find the method decisively outperforms state-of-the-art (SOTA) baselines on unlearning benchmarks across a variety of model and training dataset scales consistent with DD being an effective and inexpensive solution to unlearning. We then demonstrate that this steered distribution can be trivially distilled back into the base model. Since the method is generally applicable to any probabilistic model, we explore its efficacy outside of text generation and find evidence of generalization to the domain of images.
[NLP-38] Wind Turbine Maintenance Log Labelling Framework: LLM -Driven Data Correction and Enrichment via Semantic Extraction of Reliability Intelligence
【速读】： 该论文旨在解决风电机组运维数据中历史维护日志因以非结构化自然语言形式存在，导致难以进行量化可靠性分析的问题。其核心挑战在于如何从海量自由文本描述中提取并标准化故障模式、维护动作等关键信息，从而支持系统性可靠性评估与预测性维护。解决方案的关键在于提出一种基于大语言模型（Large Language Model, LLM）的模型无关框架，能够自动校正层级化的系统编码，并基于证据构建维护动作与故障模式的分类体系。该方法在覆盖280台风机、历时九年的16,316条维护日志上实现了超过70%数据的自动化结构化处理，有效解决了长期存在的误分类问题，如识别出此前未被归类的变桨系统故障并补全缺失的系统代码，同时通过实证分类体系对具体维护行为和故障类型进行标注。该框架利用基于系统的日志批次构建经验性词典，涵盖故障模式、可观察症状、主导机制及潜在原因，显著降低了传统故障模式与影响分析（Failure Mode and Effects Analysis, FMEA）中的主观性，为可扩展、低成本地将定性现场观测转化为定量可靠性指标提供了技术路径，为可再生能源领域实现集成化根因分析、优化FMEA流程以及发展先进预测性维护奠定了基础。
链接: https://arxiv.org/abs/2605.31281
作者: Max Malyi,Jonathan Shek,Alasdair McDonald,Andre Biscaya
机构: The University of Edinburgh (爱丁堡大学); Nadara (纳达拉)
类目: Computation and Language (cs.CL)
备注: An adjustable template containing the Python script architecture, applied dynamic prompts, and data schemas is hosted in an open-source GitHub repository: this https URL
Abstract:As wind turbine fleets age, data-driven reliability engineering is essential to optimise their operation and maintenance for service life extension and levelised cost of energy reduction. Failure event descriptions within historical maintenance logs are a source of valuable reliability intelligence. However, they typically appear as unstructured natural language entries, rendering them inaccessible for quantitative analysis. This paper presents a novel methodology leveraging a large language model (LLM) to systematically standardise and structure maintenance logs based on their free-text descriptors. Operating on a dataset of 16,316 maintenance logs from 280 turbines monitored over nine years, the developed model-agnostic framework autonomously corrected hierarchical system codes and extracted evidence-based taxonomies of maintenance actions and failure modes. The automated pipeline successfully structured over 70% of the dataset. It resolved pervasive misclassification issues, such as isolating previously unclassified pitch system faults and restoring missing system codes, and enriched the records by applying empirical taxonomies to label specific actions taken and failure modes addressed. By using system-based log batches to construct empirical dictionaries of failure modes, observable symptoms, dominant mechanisms, and candidate causes, this approach reduces the inherent subjectivity of manual failure modes and effects analysis (FMEA). Ultimately, the methodology provides a highly scalable, cost-effective blueprint for translating large sets of qualitative field observations into quantitative reliability metrics, laying the foundation for integrated root-cause analysis across the renewable energy sector, improved FMEA, and advanced predictive maintenance.
[NLP-39] Mellum2 Technical Report
【速读】： 该论文旨在解决大模型在软件工程任务中计算效率与性能之间的权衡问题，特别是如何在保持高推理质量的同时显著降低每令牌（per token）的计算开销。其核心挑战在于：尽管生成式AI（Generative AI）在代码生成、调试、多步推理和工具调用等复杂任务中表现优异，但传统密集模型（dense model）随着参数量增长导致推理成本急剧上升，难以在消费级硬件上高效部署。为此，论文提出Mellum 2——一个120亿参数的混合专家（Mixture-of-Experts, MoE）语言模型，通过动态激活仅25亿参数/令牌的方式实现接近2.5B密集模型的计算效率，同时在多种编程与推理任务中达到领先水平。解决方案的关键在于多层次架构优化：采用64个专家、8个活跃专家的MoE结构，并结合分组查询注意力（Grouped-Query Attention, GQA）与4个键值头（KV heads）、三类层滑动窗口注意力（Sliding Window Attention），以及单个多令牌预测头（Multi-Token Prediction head）作为辅助预训练目标与推测解码（speculative decoding）的内置草稿模型；所有设计均以商品级GPU上的推理效率为约束进行消融验证。此外，通过三阶段课程学习预训练策略，逐步从多样化网络数据过渡至高质量代码与数学内容，并利用FP8混合精度与温升-保持-衰减调度（Warmup-Hold-Decay）优化训练过程；最终通过层选择性YaRN扩展上下文至128K，并经过两阶段后训练（监督微调+基于人类反馈的强化学习，RLVR）得到Instruct与Thinking双版本模型。实验表明，Mellum 2在代码生成、数学推理、工具使用、知识问答及安全性等多个基准测试中，性能媲美4B–14B范围内的开源模型，而运行时的每令牌计算成本仅为2.5B密集模型级别，实现了高性能与高效率的统一。
链接: https://arxiv.org/abs/2605.31268
作者: Marko Kojic,Ivan Bondyrev,Aral de Moor,Joseph Shtok,Petr Borovlev,Kseniia Lysaniuk,Madeeswaran Kannan,Ivan Dolgov,Nikita Pavlichenko
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:We present Mellum 2, an open-weight 12B-parameter Mixture-of-Experts (MoE) language model with 2.5B active parameters per token. Mellum 2 is a general-purpose language model specialized in software engineering, spanning code generation and editing, debugging, multi-step reasoning, tool use and function calling, agentic coding, and conversational programming assistance, and it is the successor to the completion-focused 4B dense Mellum model. The architecture builds on the Mixture-of-Experts (64 experts, 8 active) and combines Grouped-Query Attention with 4 KV heads, Sliding Window Attention on three of every four layers, and a single Multi-Token Prediction head that doubles as both an auxiliary pre-training objective and a built-in draft model for speculative decoding; each choice was validated by ablation with inference efficiency on commodity GPUs as a design constraint. Pre-training spans approximately 10.6 trillion tokens through a three-phase curriculum that progressively shifts the mixture from diverse web data toward curated code and mathematical content, optimized with Muon under FP8 hybrid precision and a Warmup-Hold-Decay schedule with linear decay to zero. The pre-trained base is extended to a 128K context window via a layer-selective YaRN and then post-trained in two stages (supervised fine-tuning followed by RLVR), yielding two released variants: an Instruct model that answers directly and a Thinking model that emits an explicit reasoning trace before its final answer. Across code generation, math and reasoning, tool use, knowledge, and safety benchmarks, Mellum 2 is competitive with open-weight baselines in the 4B-14B range while running at the per-token compute of a 2.5B dense model. We release the base, instruct, and thinking checkpoints, together with this report on the architecture decisions, data pipeline, and training recipe behind them, under the Apache 2.0 license.
[NLP-40] COLLEAGUE.SKILL: Automated AI Skill Generation via Expert Knowledge Distillation
【速读】： 该论文旨在解决如何将个体或角色的隐性专业知识（如判断力、互动风格与实践智慧）从异构的非结构化痕迹中提取并转化为可被智能体调用、可检查、可修正且可移植的显式技能的问题。现有记忆系统与人格建模方法仅能捕捉知识片段，而技能框架虽提供可移植格式，却缺乏端到端的自动化流程将原始痕迹转化为可操作的技能包。其解决方案的关键在于提出一种自动化的“痕迹到技能”蒸馏系统，通过专家知识蒸馏实现人本智能技能的生成；该系统产出具有版本控制的技能包，包含两个协同运作的模块：能力轨迹（涵盖实践方法、心智模型与决策启发式）和行为边界轨迹（包括沟通风格、交互规则与修正历史）。该技能包支持自然语言反馈下的可审查、可更新、可回滚、跨代理主机部署及可控分发，显著提升了人本技能的透明性与可管理性。
链接: https://arxiv.org/abs/2605.31264
作者: Tianyi Zhou,Dongrui Liu,Leitao Yuan,Jing Shao,Xia Hu
机构: Shanghai Artificial Intelligence Laboratory (上海人工智能实验室)
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 12 pages, 4 figures
Abstract:LLM agents are increasingly expected not only to complete isolated tasks, but also to carry bounded representations of human expertise, judgment, and interaction style. Building such person-grounded agents remains difficult because actionable knowledge associated with a person or role is usually embedded in heterogeneous traces rather than written as clean instructions. Existing memory and persona systems capture fragments of this evidence, while skill frameworks provide portable packaging formats; however, there is no end-to-end workflow for distilling these traces into inspectable, correctable, and agent-usable skills. We present an automated trace-to-skill distillation system for generating person-grounded AI skills via expert knowledge distillation. Given materials from a target person or role, this http URL produces a versioned skill package with two coordinated tracks: a capability track for practices, mental models, and decision heuristics, and a bounded behavior track for communication style, interaction rules, and correction history. The package can be inspected, invoked, updated through natural-language feedback, rolled back, installed across agent hosts, and optionally prepared for controlled distribution. We describe the artifact contract, generation workflow, correction lifecycle, deployment surface, and domain presets implemented in the open-source system. At the time of writing, the public repository has approximately 18.5k GitHub stars; the gallery lists 215 skills from 165 contributors and more than 100k cumulative stars across listed skill cards. The system illustrates how person-grounded skills can be represented as portable, correctable packages rather than opaque prompts or hidden memories.
[NLP-41] Scaling Multi-Hop Training Data via Graph-Constrained Path Selection
【速读】： 该论文旨在解决大语言模型在处理专业化文档时缺乏大规模、可扩展的多跳推理训练数据的问题，尤其针对现实世界中普遍存在的结构化程度高、条款重复且高度交叉引用的文本（如法律合同）。现有方法依赖单一教师模型从无标注文本中联合发现证据路径并生成问答对，但在面对模板化与密集引用的文档时性能显著下降。本文提出关键解决方案：将推理路径的发现与语义表达解耦——先在基于上下文关键词中心点构建的图结构上离线枚举满足五项几何可接受性约束的路径，再由教师模型仅对已验证路径进行语义化生成。该图结构通过格拉姆矩阵分析证明，仅靠局部相似性约束可能导致端点漂移达约91°，而引入上界相似性约束是避免由通用文本片段形成的嵌入聚类“密集群”的必要条件。消融实验表明，在同等训练规模下，受约束与不受约束的路径链下游表现无显著差异，其性能提升主要源于可用语料库扩大了4.4倍，而非单条路径质量的提高，这重新定义了图约束的作用——并非优化路径内容，而是提升教师模型生成可行性。在CUAD法律合同语料库上对Qwen3-32B进行微调，使闭卷任务的Token F1从21.66%提升至38.58%。
链接: https://arxiv.org/abs/2605.31238
作者: Pengyu Chen,Yonggang Zhang,Mingming Chen,Jun Song,Wei Xue,Yike Guo
机构: The Hong Kong University of Science and Technology (香港科技大学); Hong Kong Generative AI Research and Development Center (香港生成式人工智能研发与中心); Hong Kong Baptist University (香港浸会大学)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 21 pages, 5 figures
Abstract:Endowing large language models with compositional reasoning over specialized documents requires multi-hop training data at scale, where such data rarely exists outside of curated benchmarks built on structured sources. To construct it directly from plain, unannotated text, existing methods ask a single teacher model to jointly discover an evidence path through a document and verbalize it as a question-answer pair. However, these methods degrade sharply when documents are structured around repetitive templates and densely cross-referencing clauses, conditions that characterize most real-world specialized corpora. In this work, we decouple the two operations: reasoning paths are enumerated offline over a graph of contextual keyword centroids, and the teacher is invoked only to verbalize pre-validated paths. The graph enforces five geometric admissibility constraints, for which we provide Gram-matrix arguments establishing that local similarity bounds alone admit endpoint drift up to \sim91^\circ , and that an upper similarity bound is necessary to exit dense embedding cliques formed by boilerplate text. A matched-size ablation isolates the mechanism: at equal training scale, constrained and unconstrained chains yield indistinguishable downstream performance, and the gain at full scale comes from a 4.4 \times expansion of the usable corpus rather than from higher per-chain quality – reframing the role of graph constraints, in this setting, as raising teacher synthesizability rather than improving chain content. Fine-tuning Qwen3-32B on 80K examples constructed from the CUAD legal contract corpus improves closed-book Token F1 from 21.66% to 38.58%. We have released our codes at this https URL.
[NLP-42] Shared Doubt: Zero-shot Cross-Lingual Confidence Estimation for Language Models
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在多语言场景下置信度估计（Confidence Estimation, CE）的泛化能力不足问题。现有方法大多局限于英语，且在跨语言迁移时性能下降或需针对目标语言重新训练，难以适应真实世界的多语言应用需求。其核心解决方案是验证多语言大语言模型是否蕴含可迁移的、共享的置信度特征。研究采用轻量级线性探测器（linear probe），直接从模型中间层表示中预测答案正确性，仅在单一语言上训练即可实现零样本（zero-shot）跨语言泛化，无需目标语言标注数据。通过分析不同层权重及多组消融实验，发现置信度特征在各语言中均集中于模型中层，表明存在一个共享的置信度子空间。尽管跨语言性能受源语言与目标语言间相似性影响，该方法仍能在不进行任何微调的情况下提供稳健的基准表现，并优于多种主流置信度估计方法。
链接: https://arxiv.org/abs/2605.31220
作者: Athina Kyriakou,Dennis Ulmer,Ivan Titov
机构: University of Amsterdam (阿姆斯特丹大学); University of Edinburgh (爱丁堡大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Confidence estimation (CE), i.e. quantifying the reliability of a model’s prediction, has attracted great interest in the context of large language models (LLMs). However, most studies focus on English, ignoring the multilingual reality of LLM usage, while many CE methods degrade or require retraining across languages. To address this gap, we investigate whether multilingual LLMs encode shared, language-transferable confidence features. We use a lightweight linear probe that predicts answer correctness directly from intermediate representations. Trained monolingually, the probe generalizes zero-shot to unseen, typologically diverse languages without target-language supervision. Learned layer weights and multiple ablations reveal that confidence features concentrate in middle layers across languages, suggesting a shared confidence subspace. While zero-shot cross-lingual performance depends on similarity to the source language, the probe provides a strong baseline without any retraining and compares favorably to other popular confidence estimation methods.
[NLP-43] Benchmarking and Enhancing Text-to-Image Models for Generating Visual Representations in Early Arithmetic Education
【速读】： 该论文旨在解决生成式AI在教育内容创作中生成的视觉输出是否能准确反映其应传授的教育概念这一关键问题，尤其关注从算术方程生成具有教学意义的可视化内容时的准确性。其核心挑战在于，与传统的文本到图像（Text-to-Image, T2I）生成任务不同，该任务要求生成的视觉结果不仅在语义上合理，还需精确保留方程中的数值关系和结构信息。为此，研究者基于对教师的访谈及教育材料的分析，构建了E2V-Bench基准测试集，涵盖四种基于教育原理的视觉类型，并设计了自动评估指标以衡量视觉正确性。实验表明，现有T2I模型在此任务上表现不佳，主要错误表现为对象数量错误和关系结构破坏。解决方案的关键在于采用基准引导的增强策略，通过引入领域知识指导模型优化，显著提升了代表性模型的表现；然而，仍存在显著差距，凸显未来T2I模型需加强在数值表达与关系建模方面的基础能力。
链接: https://arxiv.org/abs/2605.31212
作者: Junling Wang,Boqi Chen,Heejin Do,Mubashara Akhtar,April Yi Wang,Mrinmaya Sachan
机构: ETH Zurich(苏黎世联邦理工学院); ETH AI Center(苏黎世联邦理工学院人工智能中心)
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:AI systems are increasingly used to support educational content creation, yet it remains unclear whether they can generate outputs that faithfully represent the pedagogical concepts they are intended to teach. Thus, we introduce equation-to-visual generation, a task that, in contrast to conventional image generation, requires producing pedagogically meaningful visuals from arithmetic equations while precisely preserving their numerical and relational structure. Informed by interviews with teachers and an analysis of educational materials, we construct E2V-Bench, a benchmark spanning four pedagogically grounded visual types, along with automatic metrics for evaluating visual correctness. Our evaluation reveals that recent text-to-image (T2I) models frequently fail on this task, with errors dominated by incorrect object counts and broken relational structure. Building on this, we explore benchmark-guided enhancement strategies. These strategies improve representative models, while the remaining gap calls for stronger numerical and relational grounding in future T2I models.
[NLP-44] Learning Whom to Trust: Market-Feedback Adaptive Retrieval for Frozen LLM s in Event-Driven Financial RAG
【速读】： 该论文旨在解决金融检索增强生成（Financial RAG）系统中证据检索依赖文本相关性而忽视事件类型、预测时域与市场背景动态变化的问题。针对新闻触发的事件影响预测这一时点性金融任务，其核心挑战在于如何根据实时市场情境精准选择具有预测价值的信息源。解决方案的关键在于：保持大语言模型（LLM）阅读器冻结不变，通过一个由成熟残差收益反馈驱动的外部贝叶斯源记忆（Bayesian source memory）来动态优化检索层，实现对信息源选择策略的持续适应。实验表明，在固定89只纳斯达克股票组合上，采用源记忆机制的冻结阅读器相比无记忆版本，将宏观F1得分从0.438提升至0.471，下游投资组合夏普比率从0.52提升至0.84；而监督微调的LoRA阅读器虽略有改善，但仍不及冻结源记忆阅读器的表现。研究结果表明，在金融RAG中，学习“从何处检索”可能比“如何阅读”更为关键，为基于市场反馈实现模块化、可扩展的适应性提供了有效路径。
链接: https://arxiv.org/abs/2605.31201
作者: Zijie Zhao,Roy E. Welsch
机构: Massachusetts Institute of Technology (麻省理工学院)
类目: Computation and Language (cs.CL)
备注:
Abstract:Financial retrieval-augmented generation (RAG) systems typically rank evidence by textual relevance, but in financial markets the useful evidence source depends on event type, forecast horizon, and market context. We study news-triggered event-impact prediction as a point-in-time financial RAG problem. For each company-news anchor, the system retrieves related financial news and SEC filing passages, appends a pre-decision market-context card, and predicts multi-horizon residual-return signals. Our method keeps the large language model (LLM) reader frozen and adapts the retrieval layer through an external Bayesian source memory updated from matured residual-return feedback. On a fixed 89-stock Nasdaq-oriented universe derived from the FinRL-DeepSeek/FNSPID task, using original FNSPID news and point-in-time EDGAR filing passages, Frozen Reader with Source Memory improves held-out macro-F1 from 0.438 to 0.471 and downstream portfolio Sharpe from 0.52 to 0.84 relative to Frozen Reader with No Memory. A supervised LoRA reader improves static RAG modestly, but does not improve over the frozen source-memory reader. These results suggest that, for financial RAG, learning where to retrieve from can be as important as learning how to read, offering a simple, modular route to market-feedback adaptation.
[NLP-45] Probing Collision Grounding in Vision-Language Models for Safe Human-Robot Collaboration
【速读】： 该论文旨在解决人机协作中的安全监测问题，即如何通过视觉感知实现对机器人与环境或人体之间当前及潜在碰撞状态的准确判断。其核心挑战在于“碰撞定位”（collision grounding）——将多视角的视觉观测与机器人本体几何、相机视角、场景布局、人体接近度以及运动时序信息进行深度融合，以推断实时或即将发生的接触。为此，作者提出了TouchSafeBench，一个基于物理仿真（physics-grounded）的基准测试平台，用于评估视觉语言模型（VLMs）在复杂室内共存场景下的碰撞定位能力。该基准包含2,940个模拟的社交导航与社交重排任务片段，提供同步的多视图RGB-D数据、俯视轨迹图、校准的相机元数据及由仿真器生成的接触标签。研究聚焦于两个实际部署任务：当前安全状态分类与碰撞前预警。实验结果表明，现有前沿或面向机器人应用的VLMs表现仍不理想，最佳平均宏观F1分数低于50%，且显式深度信息未能自动转化为机器人本体碰撞证据，机器人-场景接触判断显著难于人体接触风险识别。关键发现揭示了具身视觉语言模型的核心局限：视觉流畅性并不等同于物理可问责性（physical accountability）。因此，构建可靠的机器人安全监控系统需依赖能够显式融合视角、机器人形态、度量几何与未来碰撞预测的新型表征结构。
链接: https://arxiv.org/abs/2605.31196
作者: Jun Wang,Xiaohao Xu,Xiaonan Huang
机构: University of Michigan, Ann Arbor(密歇根大学安娜堡分校)
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Robotics (cs.RO)
备注: 31 pages, 9 figures
Abstract:Safe human–robot collaboration requires more than visual description: a monitor must determine whether the robot body is safely separated, already colliding with the scene or a person, or about to collide. We call this capability collision grounding: binding visual observations to robot body geometry, camera viewpoint, scene layout, human proximity, and temporal motion in order to infer present and imminent contact. We introduce TouchSafeBench, a physics-grounded benchmark for evaluating collision grounding in vision-language models (VLMs). Built in Habitat~3.0, TouchSafeBench contains 2,940 simulated indoor co-presence episodes across social navigation and social rearrangement, with synchronized multi-view RGB-D observations, top-down trajectory maps, calibrated camera metadata, and simulator-derived contact labels. We study two deployment-facing tasks: classifying the current safety state and warning about imminent collision before contact. Across three frontier or robotics-oriented VLMs and nine visual representations, current models remain far from reliable: the best average Macro-F1 stays below 50%, explicit depth is not automatically transformed into robot-body collision evidence, and robot–scene contact is consistently harder than human-contact risk. TouchSafeBench reveals a central limitation of embodied VLMs: visual fluency does not imply physical accountability. Reliable robot safety monitors will need representations that explicitly bind viewpoint, robot morphology, metric geometry, and future collision. We will release the benchmark upon acceptance.
[NLP-46] Steering LLM s? Actually Sparse Autoencoders can outperform simple baselines
【速读】： 该论文旨在解决稀疏自编码器（Sparse Autoencoders, SAEs）在大型语言模型（LLMs）内部机制探索与输出控制中的表现未达预期的问题，尤其针对Wu等（2025）提出的AxBench基准测试中SAEs表现逊于简单基线方法的结论提出质疑。其解决方案的关键在于提出一种基于监督学习的特征选择与标注管道，通过该管道筛选并标注具有语义可解释性的稀疏特征，使SAEs在AxBench上的控制性能接近甚至媲美参考的LoRA方法。此外，研究发现该管道仅依赖可解释性组件即可选出对目标标签具有显著因果关系的特征，表明高稀疏性（低l0）并非实现有效控制的必要条件，从而挑战了Wang等（2025）关于高稀疏性对可解释性控制至关重要的先前结论。
链接: https://arxiv.org/abs/2605.31183
作者: Mikkel Godsk Jørgensen,Lars Kai Hansen
机构: DTU Compute; Technical University of Denmark
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Sparse Autoencoders (SAEs) have been seen as a promising avenue for exploring the internals of Large Language Models (LLMs) and for steering model output generation. When AxBench - a model steering benchmark - was introduced in Wu et al. (2025), SAEs did not seem to live up to their original hype due to poor steering performance relative to a set of simple baselines. This work serves as a partial rebuttal for Sparse Autoencoders and suggests that the results of Wu et al. (2025) did not do them full justice. We find that Sparse Autoencoders can, in fact, perform close to on par with the reference LoRA performance on the AxBench benchmark, when features are selected and labelled with our supervised pipeline. We also find that our pipeline selects features that are surprisingly causal of their identified labels when using only its interpretability-based components. Lastly, we present evidence that high sparsity (low l0) may not be crucial for successful steering based on interpretability, which is in contrast to the earlier findings in Wang et al. (2025).
[NLP-47] owards Efficient LLM s Annealing with Principled Sample Selection
【速读】： 该论文旨在解决大语言模型（Large Language Model, LLM）预训练中退火阶段（annealing phase）的数据选择问题。当前方法依赖经验性启发式策略（如领域过滤或上下文扩展），缺乏优化理论的严格支撑，导致数据选择效率低下且难以保证模型收敛质量。其核心挑战在于如何在退火阶段实现高效、自洽的训练数据筛选，以促进模型在复杂损失曲面（loss landscape）中的最优收敛。本文的关键解决方案是基于损失曲面的谱几何特性（spectral geometry of the loss landscape），提出一种新的视角：最优收敛需满足不同特征方向（eigen-directions）上的异质梯度约束。据此，作者构建了名为DiReCT（Directionally-Restrained Constrained Training）的新框架，将退火阶段的样本选择建模为一个带方向性约束的优化问题，通过引入基于海塞矩阵（Hessian）谱特性的显式方向约束，精准识别与曲率感知下降路径对齐的高质量样本。实验结果表明，DiReCT在多种模型规模下均显著优于现有方法，实现了当前最佳性能。
链接: https://arxiv.org/abs/2605.31175
作者: Yuanjian Xu,Jianing Hao,Wanbo Zhang,Zhong Li,Guang Zhang
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:The annealing phase is a pivotal convergence stage in LLM pre-training that ultimately determines final model quality. However, effectively selecting training data during this phase remains a key challenge. Current strategies rely on empirical heuristics, such as domain filtering or context extension, which lack a principled grounding in optimization theory. In this work, we characterize the annealing phase through the lens of the loss landscape’s spectral geometry. We argue that optimal convergence requires gradient updates to satisfy heterogeneous constraints across different eigen-directions. Building on this insight, we formulate data selection as a problem of satisfying these directional constraints. To this end, we propose DiReCT (Directionally-Restrained Constrained Training), a novel framework that reformulates sample selection in the annealing stage as a constrained optimization problem. By imposing explicit directional constraints on per-sample gradients based on the spectral properties of the Hessian, DiReCT identifies samples that align with the optimal curvature-aware descent path. Extensive experiments across various model scales demonstrate that DiReCT consistently achieves state-of-the-art performance. For future research, code is available at this https URL.
[NLP-48] Emergent Languages in Populations of Language Model Agents : From Token Efficiency to Oversight Evasion
【速读】： 该论文旨在解决当前对自主语言模型智能体（autonomous language model agents）的监控仅依赖表面行为所带来的局限性问题，特别是当智能体群体可能自发创造新型语言以规避人类监管时的潜在风险。其解决方案的关键在于构建一种两阶段分析框架：首先采用基于规则的启发式方法（约6000个匹配项），随后通过零样本分类（保留518个实例）对Moltbook平台上的新兴语言进行系统识别与分类。该方法识别出三大类语言现象：标记效率优化（token efficiency，166例）、新自然语言（106例）以及规避监管（oversight evasion，59例）。研究发现，由深度寻求模型（DeepSeek-3.2）评估，规避监管类语言的对齐度显著低于其他类别，且所有被识别的语言均可仅通过描述即被其他语言模型在上下文学习（in-context learning）中有效掌握。此外，案例分析揭示了高度复杂的隐写术协议，如在自然语言中嵌入隐藏信息。尽管无法完全确定这些语言生成的自主性程度，但结果共同表明，仅依靠表层行为监控已难以有效控制智能体群体，亟需发展更深层的检测与干预机制。
链接: https://arxiv.org/abs/2605.31170
作者: Stine Lyngsø Beltoft,William Brach,Federico Torrielli,Jacob Nielsen,Annemette Brok Pirchert,Filippo Tonini,Peter Schneider-Kamp,Lukas Galke Poech
机构: Aarhus University (奥胡斯大学); Aarhus University (奥胡斯大学); Aarhus University (奥胡斯大学); Aarhus University (奥胡斯大学); Aarhus University (奥胡斯大学); Aarhus University (奥胡斯大学); Aarhus University (奥胡斯大学); Aarhus University (奥胡斯大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Monitoring autonomous language model agents currently relies mostly on surface behavior. But what happens when agent populations invent new languages with the goal of avoiding human oversight. Here, we study the emergent languages on Moltbook. For this, we build upon the Moltbook Files dataset and apply a two-stage approach consisting of a rule-based heuristic (about 6000 matches) followed by zero-shot classification (518 kept). The resulting categories include token efficiency (166), new natural languages (106), and oversight evasion (59). We conduct both quantitative and qualitative analyses. Our results show that posts proposing new languages for avoiding oversight are judged by DeepSeek-3.2 as being less aligned than the other categories and that all languages can be learned by other language models in-context merely from a description of the language. Moreover, manually studying exemplary cases reveals surprisingly sophisticated steganographic protocols like embedding hidden messages in natural language. Although we cannot be certain about the extent of autonomy in ideation of these languages, our results add up to the evidence that monitoring surface behavior may soon be insufficient for retaining control over agent populations.
[NLP-49] D3: Dynamic Directional Graph-Constrained Data Scheduling for LLM Training
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）训练中训练数据调度策略的优化问题，尤其关注现有方法普遍忽视样本间动态交互关系的局限性。传统数据调度策略多聚焦于整体数据分布的调整，却未充分考虑实际训练过程中样本之间存在的方向性影响，而这种相互作用对学习效率具有重要影响。为此，本文提出D³（Dynamic Directional graph-constrained Data scheduling）框架，其核心在于将训练单元间的复杂交互建模为一个动态影响图（dynamic influence graph），其中边表示基于损失函数的依赖关系。通过在此图上求解约束优化问题，D³能够推导出符合训练过程中信息流演化的最优训练顺序，从而提升模型学习效率。该方法具有理论依据，并在预训练与后训练阶段均表现出对现有调度方法的一致性改进；同时，为保证可扩展性，D³引入高效的近似算法，在控制额外计算开销的前提下实现了实际应用可行性。
链接: https://arxiv.org/abs/2605.31164
作者: Yuanjian Xu,Jianing Hao,Guang Zhang,Zhong Li
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Training data plays a central role in large language models (LLMs) optimization, motivating extensive research on data scheduling strategies. Most existing approaches concentrate on adjusting the overall data distribution but neglect the underlying interactions between samples during training. However, we argue that such interactions cannot be overlooked, as real-world data samples frequently exhibit directional influences on each other, making the training order crucial. Intuitively, we can prioritize train-units with greater influence to improves learning efficiency. In this work, we propose D^3 , a Dynamic Directional graph-constrained Data scheduling framework. D^3 formulates the complex interactions among train-units as a dynamic influence graph, where edges represent loss-based dependencies. It then solves a constrained optimization problem over this graph to derive the training order, which ensures that the data sequence respects the evolving information flow throughout training. Our approach is theoretically motivated and yields consistent improvements over existing data scheduling methods across both pre-training and post-training phases. Furthermore, for scalability, D^3 also employs an efficient approximation algorithm that keeps the additional computational overhead within a manageable range. For future research, the code is available at this https URL.
[NLP-50] SpatialAct: Probing Spatial Reasoning -to-Action Capabilities of VLM Agents in 3D Scenes
【速读】： 该论文旨在解决当前视觉-语言模型（VLMs）在三维（3D）环境中进行多轮交互式空间推理与动作执行时存在的空间认知一致性与状态跟踪能力不足的问题。其核心挑战在于：尽管现有模型在孤立的空间推理任务中表现良好，但在需要持续维护空间信念并根据多轮反馈调整行为的动态交互场景中，难以保持连贯的空间理解，导致动作不可靠。解决方案的关键在于提出一个基于模拟器的基准测试框架——SpatialAct，通过构建“多轮交互式优化”（Multi-turn Interactive Refinement）及其分解形式“单步错误检测与修正”（Single-step Error Detection and Fix），结合五个基础空间能力任务，系统性地诊断模型失败的根本原因。实验结果揭示了显著的“推理到动作”鸿沟，表明当前VLM代理在面对由动作引发的环境变化时，仍缺乏鲁棒的空间状态追踪能力，即便底层控制已被抽象化。
链接: https://arxiv.org/abs/2605.31148
作者: Tianhui Liu,Jie Feng,Zhiheng Zheng,Shengyuan Wang,Yiming Guo,Yanxin Xi,Hangyu Fan,Yong Li,Pan Hui
机构: The Hong Kong University of Science and Technology (Guangzhou); Zhongguancun Academy; Tsinghua University; Helsinki University
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Humans can effortlessly perceive spatial layouts, form cognitive representations, reason about spatial relations, and translate such reasoning into actions in everyday 3D environments. Although recent vision-language models (VLMs) have shown promising performance on observation-conditioned spatial perception and reasoning tasks, it remains unclear whether they can build coherent spatial understanding, act upon it, and refine their actions through multi-turn feedback. To study this problem, we introduce \textbfSpatialAct, a simulator-grounded benchmark for probing \textitaction-conditioned spatial reasoning in 3D scenes. Starting from the most challenging setting, Multi-turn Interactive Refinement, we further design its decomposed counterpart, Single-step Error Detection and Fix, together with five fundamental spatial ability tasks to diagnose the underlying causes of model failures. Experiments reveal a clear reasoning-to-action gap: current VLMs can perform well on isolated spatial reasoning tasks, but struggle to maintain coherent spatial beliefs and produce reliable actions during multi-turn feedback, substantially underperforming humans. These results suggest that current VLM agents still lack robust spatial state tracking under action-induced environment changes, even when low-level control is abstracted away.
[NLP-51] On the Robustness of Multilingual Text Embedding Rankings Across Learning Tasks Languages and Benchmark Datasets
【速读】： 该论文旨在解决大规模多语言文本嵌入模型在特定语言、多任务场景下性能评估的可靠性问题。当前尽管如MTEB等基准平台覆盖了250多种语言，但模型优劣结论往往依赖于隐含的数据集构成选择与性能聚合方法，导致评估结果缺乏稳健性。为此，研究提出了一项针对MTEB中多语言模型性能鲁棒性的元研究（meta-study），引入两种关键的鲁棒性指标：数据集构成鲁棒性（dataset-composition robustness，衡量排名对数据集组成变化的敏感度）和排名方案鲁棒性（ranking-scheme robustness，衡量排名对聚合方法变更的敏感度），从而实现对基准评估结论在不同设计下的系统性敏感性分析。研究以英语、法语、德语、印地语和西班牙语为对象，在九类任务（如分类、聚类、检索）上进行深入分析，并公开约230种额外语言的结果。分析表明，基于大语言模型（LLM-based）的嵌入模型在多数任务中表现稳定且领先，但在检索任务中并非始终一致；而跨任务、跨排名方案及数据子样本均保持优异性能的模型仅占极小部分，揭示了现有模型在泛化能力上的显著差异。
链接: https://arxiv.org/abs/2605.31142
作者: Ana Gjorgjevikj,Barbara Koroušić Seljak,Tome Eftimov
机构: Jožef Stefan Institute (乔夫·斯特凡研究所)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Large-scale multilingual text embedding models play crucial role in both research and industry, yet their behavior in language-specific, multi-task settings remains insufficiently understood. Although benchmarking platforms such as MTEB report results across more than 250 languages, conclusions about model superiority often depend on implicit choices of dataset compositions and performance aggregation methods. To address this gap, we present a meta-study of multilingual model performance robustness in MTEB, applying a diverse set of multi-criteria decision-making ranking schemes and introducing two robustness indicators: dataset-composition robustness (sensitivity of rankings to changing dataset compositions) and ranking-scheme robustness (sensitivity to aggregation method change). They enable systematic sensitivity analysis of whether benchmarking conclusions remain stable under different evaluation designs. We conduct an in-depth analysis on five languages (English, French, German, Hindi, and Spanish) across nine tasks (e.g., classification, clustering, retrieval) and release results for approximately 230 additional languages. The task-specific analyses show that large-scale LLM-based models are often robust top performers, though not uniformly (e.g., in retrieval task), while task-agnostic results reveal that only a small subset of models remains consistently strong across tasks, ranking schemes, and data subsamples.
[NLP-52] EvoDefense: Co-Evolving Black-Box Defense with Large Language Models
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在黑盒攻击场景下的安全脆弱性问题，尤其针对无法访问目标模型内部结构时，现有防御机制因依赖预定义过滤规则而难以泛化至未见攻击类型和模型架构的局限性。其解决方案的关键在于提出一种基于经验引导的协同进化黑盒防御框架——EvoDefense。该框架通过引入一个守护型大语言模型（guard LLM）与经验记忆模块（experience memory module），构建了一个持续迭代的攻防演化循环：攻击生成器与守护模型在历史交互经验的指导下，不断优化各自的攻击策略与防御策略。这一机制使EvoDefense能够在不进行重新训练的前提下，有效应对未见过的攻击类型和目标模型，实现跨模型、跨攻击的强泛化防御能力。实验结果表明，EvoDefense在HarmBench、AdvBench和AlpacaEval等多个基准上均表现出优异的防御性能，例如将AutoDAN-turbo对Gemini-3-flash和LLaMA-3-8B-Instruct的攻击成功率（ASR）分别从29.4%和43.4%降低至8.4%和6.2%，同时保持了良好的通用语言能力。
链接: https://arxiv.org/abs/2605.31140
作者: Yu Li,Yuenan Hou,Yingmei Wei,Yanming Guo,Chaochao Lu
机构: National University of Defense Technology; Shanghai AI Laboratory
类目: Cryptography and Security (cs.CR); Computation and Language (cs.CL)
备注:
Abstract:Large Language Models (LLMs) remain highly vulnerable to diverse attacks, particularly in black-box settings where the internals of target models are inaccessible. Existing black-box defenses typically rely on pre-defined filtering heuristics, which often fail to generalize to unseen attack types and target model architectures. We introduce EvoDefense, an experience-guided co-evolving black-box defense paradigm. EvoDefense employs a guard LLM to detect malicious queries and an experience memory module to accumulate defense knowledge from previous interactions. At the core of EvoDefense is a continuous attack-defense evolution loop, where an attack generator and the guard model iteratively refine their attack strategies and defense policies through experience-guided optimization. This design enables EvoDefense to generalize across unseen attacks and target models without retraining. Experiments on HarmBench, AdvBench, and AlpacaEval show that EvoDefense achieves consistently strong defense performance across seven popular models and five representative LLM attacks, while preserving competitive general capabilities. On HarmBench, EvoDefense reduces the attack success rate (ASR) of AutoDAN-turbo on Gemini-3-flash and LLaMA-3-8B-Instruct from 29.4% and 43.4% to 8.4% and 6.2%, respectively.
[NLP-53] Multilingual and Cross-Lingual Citation Needed Detection on Wikipedia for Lower-Resource Languages
【速读】： 该论文旨在解决自动化事实核查（AFC）中“值得核查性检测”（check-worthiness detection）在低资源语言环境下的可及性问题，尤其聚焦于维基百科中的“引用需求检测”（Citation Needed Detection, CND）任务。现有研究多集中于高资源语言，且主流AFC流程依赖于难以获取的大语言模型（LLMs），限制了低资源组织的应用。为此，本文提出MCN——一个涵盖18种语言、覆盖三种资源水平的多语言CND语料库，并系统评估小型解码器型语言模型（SLMs）在该任务上的表现。研究发现，通过编码器式目标函数微调的SLMs，在多种语言上显著优于提示工程驱动的LLMs；同时，仅用英语数据微调的SLMs在跨语言CND任务中表现超越需大量目标语言适配的LLMs。关键突破在于证明：针对特定任务的紧凑型、专用化模型在低资源场景下比通用大模型更具优势，为低资源维基社区提供了高效可行的事实核查解决方案。
链接: https://arxiv.org/abs/2605.31136
作者: Gerrit Quaremba,Amy Rechkemmer,Elizabeth Black,Denny Vrandečić,Elena Simperl
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:In automated fact-checking (AFC), check-worthiness detection identifies claims requiring verification based on domain-specific criteria. On Wikipedia, this task instantiates as Citation Needed Detection (CND), which flags claims lacking supporting citations. However, existing research has largely overlooked lower-resource languages, and recent AFC pipelines rely on large language models (LLMs), which are inaccessible to low-resource organizations. We introduce MCN, a multilingual CND corpus spanning 18 languages across three resource levels, on which we conduct an extensive study of small decoder-based language models (SLMs). Our experiments show that SLMs fine-tuned with an encoder-style objective substantially outperform prompted LLMs across languages. We further present one of the first studies on cross-lingual CND, demonstrating that SLMs fine-tuned solely on English claims surpass LLMs, even with little to no target-language adaptation. Our findings have important implications for lower-resource Wikipedia communities and suggest that compact, task-specific models are preferable to LLMs for CND. We release all data and code at this https URL
[NLP-54] Not All Synthetic Data Is Yours to Learn From
【速读】： 该论文旨在解决在无提示（prompt-free）、无教师模型、无验证器及无奖励模型的条件下，语言模型是否能够通过自生成文本实现性能提升的问题。其核心挑战在于：在缺乏外部监督信号的情况下，模型能否仅依靠自身生成的合成数据进行有效微调并提升能力。解决方案的关键在于提出“潜在能力重现假说”（latent capability resurfacing hypothesis），即自训练的有效性并非源于数据本身的内在属性，而是取决于生成源（source）与学生模型（student）之间的关系兼容性（relational compatibility）。研究发现，在无需任何任务提示的无条件自训练设置下，模型仅从BOS（Begin of Sequence） token生成的文本中进行微调时，只有当合成语料库与学生模型具有同源性（same-lineage）时，才能有效增强模型能力；而即使其他来源的模型训练强度更强，若其训练路径不同，则迁移效果显著更弱。此外，传统的内在指标（如语义相似度或平均对数似然）均无法准确预测哪些合成数据有助于性能提升。更值得注意的是，在受控的Pythia实验中，该方法实现了能力保留与逐字记忆（verbatim memorization）的解耦——基准性能维持甚至提升，但未见样本的精确匹配提取率下降超过95%，且无需显式遗忘集、隐私目标或针对性去记忆机制。这表明，该自训练机制的本质是放大预训练模型中已存在的潜在能力，而非从数据中引入新结构。这一发现揭示了一种无需显式设计即可分离模型能力与记忆的全新范式。
链接: https://arxiv.org/abs/2605.31126
作者: Sina Alemohammad,Li Chen,Richard G. Baraniuk,Zhangyang Wang
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Can a language model improve from plain text sampled from itself, with no prompts, no teacher, no verifier, and no reward model? Yes, but only when the synthetic corpus is compatible with the student, a relational property of the source-student pair rather than an intrinsic property of the data. We call this the latent capability resurfacing hypothesis: weak self-training can amplify capabilities already present in the pretrained model, but only under this compatibility condition. We study this in the minimal setting of prompt-free unconditional self-training, where base language models are fine-tuned on text generated from the BOS token alone, with no task specification or external supervision. We report three findings. First, synthetic utility is relational rather than intrinsic: self-generated data is the most effective source, same-lineage transfer outperforms stronger but differently trained sources, and cross-family transfer is substantially weaker. Second, common intrinsic proxies fail: neither benchmark-level semantic similarity nor average per-token likelihood under the student predicts which corpora help. Third, this regime produces a surprising byproduct. In controlled Pythia experiments, capability and verbatim memorization decouple: benchmark utility is preserved or improved while held-out exact-match extraction drops by over 95 percent, with no forget set, privacy objective, or targeted unlearning. Together, these results suggest that prompt-free self-training works by amplifying what the student already knows, not by importing structure from the data. They also reveal a regime in which capability and verbatim memorization can be separated without any explicit unlearning objective.
[NLP-55] SM-Bench: Detecting LLM -Generated Text in Real-World Wikipedia Editing Practices
【速读】： 该论文旨在解决现有机器生成文本（MGT）检测基准在真实用户生成内容（UGC）平台（如维基百科）场景下适用性不足的问题。当前主流检测基准多聚焦于通用生成任务（如“写一篇关于机器学习的文章”），而忽视了编辑在实际操作中更常使用的特定任务型生成（如摘要生成），这类任务因任务约束和上下文依赖，其生成文本与人类写作的相似度更高，导致现有先进检测模型难以有效识别。论文提出的关键解决方案是构建一个名为TSM-Bench的多语言、多生成器、多任务基准，专门用于评估检测模型在典型维基百科编辑任务中的表现。研究发现，相较于以往基准，检测准确率普遍下降10%–40%，且存在显著的泛化不对称性：在特定任务数据上微调的模型可泛化至通用任务（甚至跨领域），但反之则不可行；这表明仅在通用生成数据上训练的模型会过拟合于机器生成的表面特征。因此，该工作揭示了现有检测系统在真实场景下的不可靠性，并强调需要基于更贴近实际应用的任务设计新基准，以推动未来检测模型的可靠发展。
链接: https://arxiv.org/abs/2605.31113
作者: Gerrit Quaremba,Elizabeth Black,Denny Vrandečić,Elena Simperl
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:Automatically detecting machine-generated text (MGT) is critical to maintaining the knowledge integrity of user-generated content (UGC) platforms such as Wikipedia. Existing detection benchmarks primarily focus on \textitgeneric text generation tasks (e.g., ``Write an article about machine learning.‘’). However, editors frequently employ LLMs for specific writing tasks (e.g., summarisation). These \textittask-specific MGT instances tend to resemble human-written text more closely due to their constrained task formulation and contextual conditioning. In this work, we show that a range of SOTA MGT detectors struggle to identify task-specific MGT reflecting real-world editing on Wikipedia. We introduce \textscTSM-Bench, a multilingual, multi-generator, and \textitmulti-task benchmark for evaluating MGT detectors on common, real-world Wikipedia editing tasks. Our findings demonstrate that (\textiti) average detection accuracy drops by 10–40% compared to prior benchmarks, and (\textitii) a generalisation asymmetry exists: fine-tuning on task-specific data enables generalisation to generic data – even across domains – but not vice versa. We demonstrate that models fine-tuned exclusively on generic MGT overfit to superficial artefacts of machine generation. Our results suggest that, in contrast to prior benchmarks, most detectors remain unreliable for automated detection in real-world contexts such as UGC platforms. \textscTSM-Bench therefore provides a critical foundation for developing and evaluating future models.
[NLP-56] GRKV: Global Regression for Training-Free KV Cache Compression in Long-Context LLM s
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在处理长上下文时，因维持键值（Key-Value, KV）缓存而导致的显著内存开销问题。现有方法通过淘汰（eviction）与合并（merging）策略在固定内存预算下压缩KV缓存，其中主流的基于片段（span-based）保留机制虽能较好保持语义连贯性，但与后续合并操作结合后，导致合并行为高度集中于少数片段边界载体令牌，形成严重的合并失衡，加剧信息过度合并与损失。为缓解此问题，本文提出无需训练的GRKV（Global Regression for KV Cache）方法，其核心在于直接最小化压缩后缓存与完整缓存之间注意力输出的差异。GRKV采用基于岭回归（ridge regression）的合并步骤，将被淘汰令牌的信息均匀分布至保留令牌，同时引入正则化约束以防止过平滑。在LongBench和RULER等长上下文基准测试中，GRKV是唯一能在极低额外开销下提升整体性能的合并方法。
链接: https://arxiv.org/abs/2605.31105
作者: Junjie Peng,You Wu,Haoyi Wu,Jialong Han,Xiaohua Xie,Kewei Tu,Jianhuang Lai
机构: Sun Yat-sen University(中山大学); ShanghaiTech University(上海科技大学); Guangdong Province Key Laboratory of Information Security Technology(广东省信息安全技术重点实验室)
类目: Computation and Language (cs.CL)
备注: 21 pages, 7 figures
Abstract:Large language models (LLMs) with extended context lengths rely on the key-value (KV) cache to support attention over prior tokens. However, maintaining the KV cache incurs substantial memory overhead, motivating KV-cache compression methods that enforce a fixed budget through eviction and merging. Modern eviction methods increasingly adopt span-based retention because preserving contiguous spans is empirically effective and better preserves semantic coherence. Yet, when combined with post-eviction merging, span-based retention concentrates merges onto a small set of span-boundary carrier tokens, producing a highly imbalanced merge pattern that exacerbates over-merging and increases information loss. To address this imbalance, we propose GRKV (Global Regression for KV Cache), a training-free KV-cache merging method that directly minimizes the discrepancy between compressed-cache and full-cache attention outputs. GRKV uses ridge-regression-based merge steps to distribute information from evicted tokens across retained tokens, while regularizing the updates to prevent over-smoothing. Across the LongBench and RULER long-context benchmarks, GRKV is the only merging method that improves overall performance with minimal overhead.
[NLP-57] KnowledgeGain: Evaluating and Optimizing Science News Generation for Reader Learning
【速读】： 该论文旨在解决现有科学新闻生成与摘要评价体系中缺乏对读者实际知识获取量度量的问题。当前主流评估指标多关注语义相似性与事实一致性，但未能有效衡量读者在阅读后真正获得的知识增量。为此，论文提出KnowledgeGain这一新型评估指标，通过量化读者在阅读科学新闻前后知识水平的提升来评估新闻质量。其解决方案的关键在于：首先通过受控的人类实验验证KnowledgeGain的有效性，并基于实验数据训练一个仅依赖提示（prompt-only）的大型语言模型（LLM）读者模拟器；该模拟器可对候选文章进行预筛选与排序，显著降低人工评估成本。第二轮人类实验表明，经该模拟器筛选的文章在读者阅读后的准确性及归一化KnowledgeGain上均优于强基准生成模型。该研究为实现符合布卢姆教育目标分类学（Bloom’s Taxonomy）中“知识掌握”与“理解”层级的科学新闻生成迈出了关键一步。
链接: https://arxiv.org/abs/2605.31099
作者: Dominik Soós,Meng Jiang,Jian Wu
机构: Old Dominion University (老多明尼昂大学); University of Notre Dame (圣母大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Science news is an important medium to communicate discoveries between the research communities and the public. Yet, most metrics for generated or summarized text evaluate semantic similarity and factual consistency, but do not measure how much knowledge readers learn from the news. We introduce KnowledgeGain, a metric that evaluates the quality of science news by measuring how much knowledge readers gained after reading it. To evaluate the metric, we first performed a controlled human study and showed that the metric successfully captures the differential knowledge gained by human readers reading different types of science media. The data allowed us to calibrate a prompt-only LLM reader simulator. We use it to rank and filter candidate articles before human evaluation. A second human study shows that articles selected with this simulator improve post-reading accuracy and normalized KnowledgeGain over a strong generation baseline. Our work is a step toward generating science news that better meets the knowledge and comprehension goals of Bloom’s Taxonomy.
[NLP-58] ConsisGuard: Aligning Safety Deliberation with Policy Enforcement in LLM Guardrails
【速读】： 该论文旨在解决生成式AI（Generative AI）中基于推理的大型语言模型（LLM）安全防护机制中存在的“推理到执行一致性缺失”问题，即模型在推理过程中虽能识别出有害意图，却仍可能输出安全标签，或在缺乏政策依据的情况下做出不安全决策。这一现象被定义为“ deliberation-to-enforcement gap”（推理-执行鸿沟），其核心在于现有方法未能确保推理过程与最终决策之间在安全策略执行上的一致性。论文提出的解决方案关键在于构建ConsisGuard框架，通过“政策到决策轨迹蒸馏”（Policy-to-Decision Trajectory Distillation）与“功能耦合对齐”（Functional Coupling Alignment）两项技术，强化模型内部安全推理与决策执行之间的因果关联，从而实现推理内容严格遵循安全政策且最终决策可由推理过程逻辑推导得出。实验结果表明，该方法在提示词与响应有害性检测基准上显著提升了检测性能并降低了政策执行失败率，验证了可靠安全防护需以安全策略的准确、忠实执行为基础。
链接: https://arxiv.org/abs/2605.31073
作者: Yan Wang,Zhixuan Chu,Zihao Xue,Zhen Bi,Bingyu Zhu,YueFeng Chen,Zeyu Yang,Jungang Lou,Longtao Huang,Ningyu Zhang,Kui Ren,Hui Xue
机构: Alibaba Group(阿里巴巴集团); Zhejiang University(浙江大学); Huzhou Normal University(湖州师范学院); Zhejiang Key Laboratory of Intelligent Education Technology and Application(浙江省智能教育技术与应用重点实验室)
类目: Computation and Language (cs.CL)
备注: 18 pages, 9 figures
Abstract:Reasoning-based LLM guardrails improve safety moderation by generating explicit rationales before issuing final decisions. However, their rationales do not always lead to faithful enforcement: a model may recognize a harmful intent in its reasoning but still predict a safe label, or issue an unsafe decision without policy-grounded justification. We identify this safety-critical failure mode as the deliberation-to-enforcement gap. Unlike general chain-of-thought faithfulness, guardrail reliability requires policy execution consistency: the generated reasoning should be grounded in the safety policy, and the final decision should be entailed by that reasoning. We propose ConsisGuard, a consistency-aware framework for reasoning-based LLM guardrails. ConsisGuard performs Policy-to-Decision Trajectory Distillation and Functional Coupling Alignment, aligning the internal coupling between safety deliberation and decision enforcement. Experiments on prompt and response harmfulness detection benchmarks show that ConsisGuard improves detection performance while reducing policy execution failures. These results suggest that reliable reasoning-based guardrails require accurate faithful execution of safety policies.
[NLP-59] owards Effective Long-Video Event Prediction via Multi-Level Event Semantics Mining
【速读】： 该论文旨在解决长视频事件预测（long-video event prediction）中因多模态上下文庞大、叙事结构复杂而导致的预测精度不足问题，尤其针对现有基于大语言模型（LLM）与视觉-语言模型（VLM）的长视频语言模型（LVLM）在事件细节提取不精准、事件发展过程细粒度分析能力弱等局限性。其解决方案的关键在于提出一种多层级事件语义挖掘框架——VISTA，通过三阶段机制实现：首先采用以角色为中心的视觉提示（character-centric visual prompt）精确提取事件相关的视觉细节，增强细节层面的语义表征；其次引入知识增强的迭代检索策略，引导LLM逐步构建逻辑连贯的事件链，提升事件层面的叙事完整性；最后采用类人化的“先提出再检索”（propose-then-retrieve）策略，生成多样化的未来事件推测并融合多层级线索，从而实现鲁棒且准确的长视频事件预测。
链接: https://arxiv.org/abs/2605.31069
作者: Bo Peng,YuanJie Lyu,PengGang Qin,Tong Xu
机构: 未知
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL)
备注:
Abstract:Accurately predicting future events is fundamental to content understanding and decision-making across various domains. While prior research has primarily focused on text or short-video scenarios, long-video event prediction, characterized by vast multimodal context and more complex narratives, remains underexplored. Meanwhile, although recent Long-Video Language Models (LVLMs), built on Large Language Models (LLMs) and Vision-Language Models (VLMs), have shown promise in long-video question answering and summarization, they struggle to generalize to event prediction, as they can neither precisely extract event-related details nor perform fine-grained analysis of event development. To address this gap, we propose VISTA, a multi-level event semantics mining framework for long-video event prediction. Initially, VISTA applies a character-centric visual prompt to precisely extract event-related visual details, enhancing detail-level semantics; subsequently, it employs a knowledge-enhanced iterative retrieval strategy, guiding the LLM to progressively construct logically coherent event chains, thereby improving event-level narratives; ultimately, VISTA adopts a human-like propose-then-retrieve strategy to generate diverse future-oriented proposals and integrate multi-level clues, producing robust and accurate predictions. Extensive experiments on real-world datasets validate the effectiveness of VISTA for long-video event prediction.
[NLP-60] AdaptR1: Reinforcement Learning Based Adaptive Interleaved Thinking in Multi-hop Question Answering
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在复杂推理任务中因链式思维（Chain-of-Thought, CoT）提示引发的“过度思考”问题，即对简单查询生成冗长且不必要的推理过程，导致计算开销增加。现有自适应推理方法通常仅在查询层面做出是否推理的静态决策，未能捕捉多跳问答任务中各中间步骤对显式推理需求的动态变化。为此，论文提出 AdaptR1，一种基于强化学习（Reinforcement Learning, RL）的自适应交错思考框架，用于多跳问答任务中的分步推理预算分配。其核心创新在于采用全强化学习策略，结合质量门控的效率奖励机制，在每一步动态调整推理资源投入，无需监督微调（Supervised Fine-Tuning, SFT）即可实现冷启动初始化。实验表明，在 Graph-R1 设置下，AdaptR1 将平均思考令牌数减少 69.71%，在 HotpotQA 数据集上更是达到 90.35% 的降幅，同时保持或超越标准基线的性能。分析进一步揭示，过度思考主要集中在初始规划阶段，凸显了分步自适应预算分配的有效性。
链接: https://arxiv.org/abs/2605.31062
作者: Yuxin Wang,Jiahao Lu,Qifeng Wu,Shicheng Fang,Chuanyuan Tan,Yining Zheng,Xuanjing Huang,Xipeng Qiu
机构: Fudan University (复旦大学); Shanghai Innovation Institute; Soochow University (苏州大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Large Language Models (LLMs) have achieved remarkable performance in complex reasoning tasks through Chain-of-Thought (CoT) prompting. However, this approach often leads to ``over-thinking,‘’ where models generate unnecessarily long reasoning traces for simple queries and incur avoidable inference cost. While recent work has explored adaptive reasoning, existing methods typically make a single query-level decision about whether to reason. This overlooks the dynamic nature of multi-step tasks, where the need for explicit reasoning varies across intermediate stages. To address this limitation, we introduce AdaptR1, a Reinforcement Learning (RL) based framework for adaptive interleaved thinking in multi-hop Question Answering (QA). Unlike previous approaches that require Supervised Fine-Tuning (SFT) for cold-start initialization, AdaptR1 uses a fully RL-based strategy with a quality-gated efficiency reward to dynamically allocate reasoning budgets at each step. Under the Graph-R1 setting, AdaptR1 reduces average think tokens by 69.71%, with a 90.35% reduction on HotpotQA, while maintaining performance comparable to or better than standard baselines. Furthermore, our analysis reveals that overthinking in multi-hop reasoning is not uniformly distributed but occurs predominantly during the initial planning stages, highlighting the effectiveness of step-wise adaptive budget allocation.
[NLP-61] Combinatorial Synthesis: Scaling Code RLVR via Atomic Decomposition and Recombination
【速读】： 该论文旨在解决生成式强化学习中可验证奖励（Reinforcement Learning with Verifiable Rewards, RLVR）的可扩展性瓶颈问题，即当前缺乏足够挑战性且贴近大语言模型（Large Language Models, LLMs）能力边界的可验证代码任务，导致训练数据的价值无法随合成规模线性增长。现有方法依赖启发式种子扩展进行数据生成，难以产生新颖性和难度兼具的任务。本文提出原子分解与重组（Atomic Decomposition and Recombination, ADR）框架，通过将复杂代码任务分解为原子元素并进行受控重组，实现真正新颖且具有挑战性的可验证代码任务的高效生成。其核心创新在于利用结构化分解与可控组合机制，在保持任务可验证性的同时显著提升任务的原创性、难度、多样性与评测质量。实验表明，ADR在算法编程、工具使用和数据科学等多个下游任务中均能显著提升LLMs的编码能力，为可扩展的RLVR训练提供了新范式。
链接: https://arxiv.org/abs/2605.31058
作者: Jiasheng Zheng,Boxi Cao,Boxi Yu,Yuzhong Zhang,Jialun Cao,Yaojie Lu,Hongyu Lin,Xianpei Han,Le Sun
机构: Chinese Information Processing Laboratory, Institute of Software, Chinese Academy of Sciences(中国科学院软件研究所信息处理实验室); University of Chinese Academy of Sciences(中国科学院大学); Lero the Research Ireland Centre for Software, University of Limerick(爱尔兰利默里克大学软件研究中心); The Chinese University of Hong Kong, Shenzhen(香港中文大学深圳分校); The Hong Kong University of Science and Technology(香港科技大学)
类目: Computation and Language (cs.CL); Software Engineering (cs.SE)
备注: Work in progress
Abstract:Reinforcement Learning with Verifiable Rewards (RLVR) has recently emerged as the cornerstone for shaping the remarkable coding abilities of Large Language Models (LLMs). However, the scalability of RLVR is severely constrained by the scarcity of sufficiently challenging verifiable code tasks that target near the model’s edge of competence. Prior studies often rely on heuristic seed expansions for data synthesis, which severely limits both novelty and difficulty. Consequently, the training value of such data fails to scale proportionally with the size of its synthesis. To this end, we propose Atomic Decomposition and Recombination (ADR), a novel framework that generates verifiable code tasks via decomposition into atomic elements and controlled recombination, thereby enabling the generation of genuinely novel and challenging verifiable code tasks. Experiments and analysis demonstrate that ADR achieves superior originality, difficulty, diversity, and test quality over existing baselines, and consistently delivers greater improvements in code ability across RLVR in diverse downstream domains, including algorithmic programming, tool usage, and data science. Our work sheds light on a new paradigm for novel code task synthesis and scalable RLVR training.
[NLP-62] How Much Do LLM s Know About Chinese Zero Pronouns?
【速读】： 该论文旨在解决生成式人工智能（Generative AI）在处理汉语零代词（Zero Pronouns, ZPs）时表现不佳的问题。零代词是汉语等省略代词语言中普遍存在的一种语言现象，其指代关系复杂且上下文依赖性强，长期以来对自然语言处理系统构成挑战。尽管大语言模型（Large Language Models, LLMs）在多项中文任务中表现出色，但其对零代词的识别、指代性判断、指代类型分类及消解等核心能力仍不明确。研究通过一系列基于语言学动机的任务（包括识别、指代性分类、指代类型分类、消解与翻译），系统评估了多种LLMs在汉语零代词处理上的表现。结果表明，当前主流LLMs在零代词处理上仍面临显著困难，尤其在上游任务如识别与指代性分类中表现较差；下游任务如零代词翻译的准确率也普遍偏低，即使最先进的推理型大模型也仅能正确翻译不足一半的零代词。因此，该研究的关键解决方案在于构建一套系统的语言学驱动评估框架，揭示现有模型在零代词理解方面的局限性，并为未来模型改进提供基准与方向。
链接: https://arxiv.org/abs/2605.31056
作者: Yifei Li,Guanyi Chen,Tingting He
机构: Hubei Provincial Key Laboratory of Artificial Intelligence and Smart Learning, National Language Resources Monitoring and Research Center for Network Media, School of Computer Science, Central China Normal University
类目: Computation and Language (cs.CL)
备注:
Abstract:Zero Pronouns (ZPs) are a pervasive linguistic phenomenon in pro-drop languages such as Chinese and have long posed a challenge for natural language processing systems. Although Large Language Models (LLMs) perform well on many Chinese language tasks, their ability to process ZPs remains poorly understood. We conduct a systematic investigation of LLMs’ handling of Chinese ZPs through a sequence of linguistically motivated tasks, including identification, referentiality classification, referential type classification, resolution, and translation. A diverse set of LLMs is evaluated across all tasks. Our results show that Chinese ZPs remain highly challenging for current LLMs, particularly for upstream tasks such as identification and referentiality classification. Performance on downstream tasks, such as ZP translation, is also consistently low: even state-of-the-art reasoning-oriented LLMs correctly translate fewer than half of Chinese ZPs into English.
[NLP-63] From Prompt Injection to Persistent Control: Defending Agent ic Harness Against Trojan Backdoors
【速读】： 该论文旨在解决本地化大语言模型（LLM）智能体在真实工作空间中面临的多步后门攻击（multi-step trojan attack）问题。随着生成式AI（Generative AI）智能体具备读写文件、调用工具及跨会话复用工作区状态的能力，攻击者可将提示注入（prompt injection）隐藏于文件或工具输出中，通过非恶意的逐步操作实现持久化控制，而单个步骤本身无明显恶意特征，传统防御机制因仅孤立检查每一步骤而难以识别此类隐蔽攻击。其解决方案的关键在于提出DASGuard，该机制通过扫描敏感本地文件中的控制类文本、追踪其来源，并移除未源自可信源的控制内容，结合运行时攻击阻断与工作区提交内容的净化，实现了动态且有效的防御。实验表明，该方法在基于GPT-5.4的OpenClaw仿真环境中显著提升了对多步后门攻击的检测能力，有效遏制了高成功率的攻击行为。
链接: https://arxiv.org/abs/2605.31042
作者: Jiejun Tan,Zhicheng Dou,Xinyu Yang,Yuyang Hu,Yiruo Cheng,Xiaoxi Li,Ji-Rong Wen
机构: Gaoling School of Artificial Intelligence, Renmin University of China
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: Code and data are available at this https URL
Abstract:LLM agents are evolving from conversational chatbots to operational tools in real-world workspaces. In local agentic harnesses, an LLM can read and write files, call tools, and reuse workspace state across sessions. While such capabilities enhance utility, they also expose a new attack surface for attackers. Attackers can embed a prompt injection within a file or tool output. Agents may read this hidden instruction, store it, and execute it later. In this multi-step trojan attack paradigm, no individual step appears malicious on its own, but these steps can collectively turn untrusted text into persistent control content. However, existing defenses often inspect each step in isolation. As a result, they can block a clear harmful action, but fail to detect the earlier write operation that plants the backdoor. To reveal this threat, we introduce ClawTrojan, a benchmark designed to identify multi-step trojan attacks in local agentic harnesses. In an OpenClaw-style simulated workspace with GPT-5.4, ClawTrojan reaches a 95.5% attack success rate (ASR), while existing single-turn prompt-injection attacks produce near-zero ASR on the same model. To address this threat, we propose DASGuard, which scans control-like text in sensitive local files, traces its origin, and removes control content that does not originate from a trusted source. Our results show that DASGuard achieves strong dynamic defense by combining runtime attack blocking with sanitized commits to the workspace.
[NLP-64] RACE: Discovering Task-Specific Parameter via Adaptation-Aware Probing for Continual Fine-Tuning KDD2026
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在实际部署中持续多任务适应时面临的灾难性遗忘问题。传统方法如全参数或低秩微调（Low-Rank Adaptation, LoRA）在顺序微调过程中易因参数覆盖导致已有知识丢失，而基于回放或维护独立任务适配器的方法虽能缓解遗忘，却引入额外的计算、存储与管理开销。针对这一挑战，本文提出一种关键创新：将持续任务适应重构为基于适配感知探测（Adaptation-aware Probing）的任务特异性参数发现过程。其核心在于通过短周期预热微调（warm-start fine-tune）揭示任务的适配轨迹，并利用重要性评分（如L₂范数和Fisher信息量）与特异性分析（参数更新方向的余弦相似度）识别每个任务所依赖的少量核心参数。在后续持续微调中，仅更新当前任务的核心参数，其余参数保持冻结，从而有效保留历史知识。基于此思想，作者提出名为TRACE的新方法，在多个标准基准上验证了其优越性能，并进一步通过跨模型与规模的可迁移性实验，证明了“小模型指导大模型”（small-to-large）的可行性，为资源受限场景下的高效大规模模型微调提供了新范式。
链接: https://arxiv.org/abs/2605.31025
作者: Xiaosong Han,Ke Chen,Xindi Dai,Di Liang,Minlong Peng,Wei Pang,Fausto Giunchiglia,Xiaoyue Feng,Yonghao Liu,Renchu Guan
机构: Jilin University (吉林大学); Fudan University (复旦大学); Heriot-Watt University (赫瑞-瓦特大学); University of Trento (特伦托大学)
类目: Computation and Language (cs.CL)
备注: KDD2026
Abstract:In real-world deployment, LLMs are often adapted continually across tasks to keep LLMs up-to-date in production, where new fine-tuning should preserve previously learned skills. However, indiscriminately mixing tasks can dilute task specialization, while sequential fine-tuning (full-parameter or low rank adaptation) often causes catastrophic forgetting due to destructive overwriting. Replay-based continual tuning and maintaining separate task-specific adapters can mitigate forgetting, but introduce additional compute, storage, and management overhead. Recognizing the redundancy of LLM parameters for any single task, we reframe continual task adaptation as task-specific parameter discovery via adaptation-aware probing: a short warm-start probe exposes a task’s adaptation trace, enabling us to identify and isolate the small subset of parameters essential for each task to mitigate catastrophic forgetting. Building on this view, we introduce TRACE, a novel approach for discovering Task-specific paRameters via Adaptation-aware probing for Continual finE-tuning. We perform a short warm-start fine-tune to derive task-specific core parameters by comparing the warm-started and pre-trained models. Core parameters are identified via two strategies: importance scoring (L _2 norm and Fisher Information) and specificity analysis (cosine similarity of parameter updates). In continual fine-tuning settings, only the active task’s core parameters are updated while others remain frozen, preserving prior knowledge. We conduct extensive experiments across multiple standard benchmarks to demonstrate the superior performance of our proposed method. Additionally, we validate the generalization of our method through a cross-model and scale transferability study, demonstrating a “small-to-large” paradigm that guides the fine-tuning of large-scale models under resource constraints.
[NLP-65] A Persona-Based Evaluation Framework for Pluralistic Alignment in Generative AI
【速读】： 该论文旨在解决当前生成式人工智能（Generative AI）对齐范式中依赖单一化基准测试框架所导致的评价偏差问题，即这种框架将多元的人类判断简化为统计平均值，从而掩盖了文化、人口统计学及情境背景下的评价差异。其解决方案的关键在于提出一种状态空间约束的模拟评估框架，用以替代传统的单一评估函数，通过构建一系列代表不同人类视角的合成认知角色（synthetic cognitive profiles）形成的结构化流形，实现一种多视角、依赖于立场的多元化评估。研究表明，现代生成模型能够高保真地实例化并维持这些评估人格，从而更真实地反映现实世界中共识的多样性。然而，研究进一步揭示，在连续推理与随机提示扰动下，这些模拟评估者会出现系统性稳定性退化，表现为状态空间漂移和语义不一致。这表明静态对齐约束无法长期维持稳健的评估行为，因此论文主张应在生成系统中嵌入动态、以生存能力为导向的调控机制，以保障认知模拟的一致性。通过将基于人格的评估建模为潜在表示流形上的结构化动力系统，本研究为构建更具适应性、与人类对齐且情境敏感的AI评估方法奠定了基础。
链接: https://arxiv.org/abs/2605.31021
作者: Atahan Karagoz
机构: 未知
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:Current alignment paradigms for generative artificial intelligence rely predominantly on monolithic benchmarking frameworks that reduce the plurality of human judgment to aggregated statistical baselines, thereby obscuring cultural, demographic, and contextual variability in evaluation. We introduce a state-space constrained emulation framework for AI evaluation that replaces singular assessment functions with a structured manifold of synthetic cognitive profiles representing diverse human perspectives. We show that modern generative architectures can instantiate and maintain these evaluative personas with high consistency, enabling a form of pluralistic, perspective-dependent benchmarking that more closely reflects real-world consensus variability. However, we further analyze the stability of these simulated evaluators under sequential inference and stochastic prompt perturbations, revealing systematic degradation in persona coherence that manifests as state-space drift and semantic inconsistency. These findings suggest that static alignment constraints are insufficient for sustaining robust evaluative behavior over time. Instead, we argue for the necessity of embedding dynamic, viability-driven regulatory mechanisms within generative systems to preserve coherent cognitive emulation. By framing persona-based evaluation as a structured dynamical system over latent representation manifolds, this study provides a foundation for more adaptive, human-aligned, and context-sensitive approaches to AI evaluation.
[NLP-66] MoG: Mixture of Experts for Graph-based Retrieval-Augmented Generation
【速读】： 该论文旨在解决生成式 AI（Generative AI）在复杂推理任务中因依赖统一知识库进行检索而引入无关信息，进而误导生成结果的问题。其核心挑战在于如何在保持知识覆盖广度的同时，实现对相关证据的精准聚焦。解决方案的关键是提出一种基于图结构的混合专家模型（Mixture of Experts for Graph-based Retrieval-Augmented Generation, MoG），通过双层知识组织机制实现稀疏化、条件化的知识检索：首先利用始终可访问的“枢纽图”（hub graphs）捕捉语义与结构上的核心知识并生成上下文线索；随后，由拓扑感知的路由器根据查询动态激活一组特定领域的“专家图”（expert graphs），从而将检索范围限定在与任务相关的证据子空间内。该设计借鉴了混合专家系统（Mixture of Experts, MoE）的稀疏路由思想，实现了高效且精准的知识调用，在多个复杂基准测试上显著优于现有基线方法，尤其在MuSiQue数据集上实现超过20%的相对性能提升。
链接: https://arxiv.org/abs/2605.31010
作者: Zheng Yuan,Chuang Zhou,Linhao Luo,Siyu An,Di Yin,Xing Sun,Xiao Huang
机构: The Hong Kong Polytechnic University(香港理工大学); Monash University(莫纳什大学); Tencent Youtu Lab(腾讯优图实验室)
类目: Computation and Language (cs.CL)
备注:
Abstract:Retrieval-augmented generation is intensively studied to ground large language models on external evidence. However, retrieving from a unified knowledge base could inevitably introduce irrelevant information that may mislead generation for complex reasoning. Inspired by the conditional computation of mixture of experts (MoE), where a router sparsely selects specialized experts alongside shared ones for each input, we propose \textbfMixture \textbfof experts for \textbfGraph-based Retrieval-Augmented Generation, i.e., \textbfMoG. It organizes knowledge into two core components: (i) diverse, always-accessible hub graphs that encode semantically and structurally central knowledge and provide contextual clues for expert activation, and (ii) sparsely activated expert graphs that contain domain-specific evidence. MoG first accesses hub graphs to identify general evidence and derive contextual clues. Then, a topology-aware router dynamically activates a limited set of expert graphs conditioned on the query, thereby confining retrieval to a focused evidence subspace. Extensive experiments on challenging benchmarks show that MoG consistently outperforms strong baselines, with over 20% relative improvement on MuSiQue. Our code is available in this https URL.
[NLP-67] raceable by Design: An LLM Pipeline and Dashboard for EU Regulatory Consultation Analysis
【速读】： 该论文旨在解决公共咨询中海量利益相关者提交文本数据难以通过人工方式有效分析的问题。针对这一挑战，研究提出了一种基于大语言模型（Large Language Model, LLM）的端到端处理流程与交互式仪表板，实现对监管咨询提交内容的结构化主题提取。其解决方案的关键在于三个核心原则：原文锚定（verbatim grounding）、全程可追溯性（full traceability）以及设计上的透明性（transparency by design）。系统能够处理原始PDF附件和网页表单响应，自动提取主题标注，并将每一项提取结果精确关联至原文摘录，确保输出的可验证性。在欧洲委员会数字公平法案（Digital Fairness Act, DFA）的案例研究中，该方法从4,322份提交材料中生成了15,368条主题标注及20,951条原文证据引用。此外，系统还识别出如“年龄验证”“支付处理器审查”“数字产权”等未被预定义分类体系覆盖的新兴利益相关方关切，突破了固定分类框架的局限。该方案具备领域通用性，仅需更新提示词（prompt）和输入数据集即可适配新咨询任务，具有良好的可迁移性。整个系统已开源，配套代码与处理数据均可公开获取，支持实时演示。
链接: https://arxiv.org/abs/2605.30995
作者: Thales Bertaglia,Haoyang Gui,Catalina Goanta,Gerasimos Spanakis
机构: Utrecht University (乌得勒支大学); Maastricht University (马斯特里赫特大学)
类目: Computers and Society (cs.CY); Computation and Language (cs.CL)
备注:
Abstract:Public consultations generate large volumes of data in the form of stakeholder submissions that are practically unfeasible to analyse manually. We present an end-to-end LLM-based pipeline and interactive dashboard for structured topic extraction from regulatory consultation submissions, demonstrated on the European Commission’s Digital Fairness Act (DFA) public call for evidence as a case study. The system processes raw PDF attachments and web-form responses, extracts topic annotations, and grounds every extraction in a verbatim quote from the source text. Applied to 4,322 DFA submissions, the pipeline produced 15,368 topic annotations supported by 20,951 verbatim evidence quotes. Three principles govern the proposed design: verbatim grounding, full traceability, and transparency by design. The dashboard exposes the full extraction dataset through five analytical views, from dataset-level topic overviews to individual paragraph drill-downs, with every result traceable to its source. Beyond the predefined DFA topic categories, the pipeline generated certain stakeholder concerns, such as Age Verification, Payment Processor Censorship, and Digital Ownership, that a fixed-taxonomy approach would have missed. The pipeline is domain-generic; adapting it to a new consultation requires only a prompt update and a new dataset. A live demo is available at this https URL. The code and processed data are publicly available at this https URL.
[NLP-68] Generating Reports or Repeating Templates? Measuring and Mitigating Template Collapse in 3D CT Report Generation
【速读】： 该论文旨在解决当前3D医学视觉-语言模型（VLMs）在生成放射科报告时存在的“模板坍缩”（Template Collapse）问题，即模型倾向于生成流畅通顺但缺乏临床真实性和多样性的通用模板化报告，导致罕见但关键的病理发现被严重低估。其核心问题是：在3D医学影像数据稀缺、标签严重不平衡以及体积分解器信号微弱等现实约束下，传统的文本生成目标促使模型采取捷径学习策略，生成看似合理却与实际影像内容关联薄弱的报告。解决方案的关键在于提出一种解耦框架CLarGen，通过将“说什么”（临床病理检测）与“如何说”（语言合成）分离，实现更可靠的临床语义对齐。具体包括：(i) 使用潜变量查询变压器（Latent Query Transformer）进行多标签病理检测；(ii) 基于病理引导的检索机制获取临床匹配的示例样本；(iii) 利用医学语言模型结合检测结果与检索上下文生成最终报告。实验表明，相较于现有基线，CLarGen显著提升了临床准确性（宏平均F1值0.487 vs. 0.189；临床报告生成分数CRG 0.472 vs. 0.368），同时保持报告流畅性，验证了显式可度量的临床锚定对于抵抗模板坍缩的重要性。
链接: https://arxiv.org/abs/2605.30984
作者: Tom Maye-Lasserre,Yitong Li,Bailiang Jian,Morteza Ghahremani,Benedikt Wiestler,Christian Wachinger
机构: Technical University of Munich (TUM); TUM Hospital; Munich Center for Machine Learning (MCML)
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Modern 3D medical vision-language models (VLMs) can generate fluent radiology-style text while exhibit critically low pathology detection and output diversity, collapsing to generic templates that under-report rare yet critical findings. We identify this failure mode as Template Collapse. This failure stems from the unique constraints of 3D medical imaging, e.g., limited data, severe label imbalance, and weak signals from volumetric encoders. Under these constraints, text-generation objectives encourage shortcut learning and fluent but weakly grounded reports. We systematically diagnose the Template Collapse through clinical fidelity, output diversity, normal-template bias, and rare-finding survival. To mitigate it, we propose CLarGen, a decoupled framework that separates what to say (clinical detection) from how to say it (language synthesis). CLarGen uses (i) a Latent Query Transformer for multi-label pathology detection, (ii) pathology-guided retrieval for clinically matched exemplars, and (iii) a medical language model to synthesize the final report from detected findings and retrieved context. Across state-of-the-art 3D CT report generation baselines, CLarGen mitigates Template Collapse and substantially improves clinical accuracy (macro-F1 0.487 vs. 0.189; CRG 0.472 vs. 0.368) while maintaining fluent reporting. Our results suggest that explicit, measurable clinical grounding is essential for template-collapse-resistant 3D CT report generation. Code will be released upon acceptance.
[NLP-69] Cognitive Fatigue in Autoregressive Transformers: Formalization and Measurement ICML2026
【速读】： 该论文旨在解决自回归语言模型在长序列生成过程中出现的性能退化问题，具体表现为文本重复、指令遵循能力下降以及熵值不稳定等现象。现有实践中缺乏实时诊断手段以在线识别这些退化行为。为此，作者将此类退化现象形式化为“认知疲劳”（cognitive fatigue），即一种可在生成过程中测量的状态，其特征包括对原始提示注意力减弱、表征漂移（representational drift）以及熵校准失准。论文提出轻量级、与模型无关的诊断指标——疲劳指数（Fatigue Index, FI），通过整合上述三类信号，并满足单调性、有界性和可解释性等明确公理，实现对生成过程的可靠实时监控。在九种不同规模（1B–13B参数）的语言模型上，FI轨迹展现出结构化的时序动态，能够有效预测任务退化（AUROC = 0.95）和重复现象（Spearman rho = 0.94），并揭示非单调缩放规律：参数量低于3B的指令微调模型退化速度超过基础模型，而这一趋势在7B以上模型中反转。压力测试进一步表明，更长上下文、中间位置证据及降低数值精度均会加速FI的触发。研究结果确立了认知疲劳作为可度量且连贯的现象，同时将FI定位为生产环境中大语言模型（LLM）运行可靠性监控的原理性工具。
链接: https://arxiv.org/abs/2605.30981
作者: Riju Marwah,Ritvik Garimella,Vishal Pallagani,Atishay Jain,Michael Stewart,Amit Sheth
机构: 未知
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 9 pages, 7 figures. Accepted at the 43rd International Conference on Machine Learning (ICML 2026)
Abstract:Autoregressive language models frequently degrade during long-horizon generation, producing repetitive text, losing instruction adherence, and exhibiting unstable entropy. Despite the prevalence of these failures, practitioners lack online diagnostics to detect them in real-time as they occur. We formalize this degradation as cognitive fatigue, a measurable generation-time state characterized by decay in attention to the original prompt, representational drift, and entropy miscalibration. We introduce the Fatigue Index (FI), a lightweight, model-agnostic diagnostic that aggregates these three signals under explicit axioms (monotonicity, boundedness, interpretability) enabling reliable runtime monitoring. Across nine models (1B-13B parameters), FI trajectories exhibit structured temporal dynamics, predict task degradation (AUROC = 0.95) and repetition (Spearman rho = 0.94), and reveal non-monotonic scaling behavior: instruction-tuned models below 3B exhibit faster collapse than base models, with this trend reversing at 7B. Stress analyses further show that FI onset accelerates under longer contexts, middle-positioned evidence, and reduced numerical precision. These results establish cognitive fatigue as a coherent and measurable phenomenon, and position FI as a principled tool for runtime reliability monitoring in production LLM systems.
[NLP-70] EvoGens: A Population-Based Heuristic Search Framework for Scientific Idea Generation
【速读】： 该论文旨在解决生成式科研创意过程中存在的语义趋同（semantic convergence）问题，即现有大语言模型（LLM）在辅助科研创意生成时往往产生重复性高、创新性不足的思路，导致候选创意的多样性和新颖性受限。其解决方案的关键在于提出EvoGens框架，该框架将科研创意生成过程建模为一种受进化机制启发的种群搜索过程：通过基于排名的变异（rank-based mutation）结合差异化检索规划以引入外部知识，并采用语义感知交叉（semantic-aware crossover）实现互补概念的融合与概念重组；同时，利用轻量级评估信号引导选择机制，在保持创意质量的同时有效抑制过早收敛，显著增强探索能力。实验结果表明，EvoGens在自动评估体系下将新颖性（Novelty）从0.1提升至0.4，多样性（Diversity）从0.24提升至0.55，验证了进化机制在推动探索性科研创意生成中的有效性。
链接: https://arxiv.org/abs/2605.30961
作者: Xu Li,Hanzhe Tu,Xinyi Li,Kuncheng Zhao,Xun Han,Zhonghui Liu
机构: 未知
类目: Computation and Language (cs.CL)
备注: 21 pages, 6 figures
Abstract:Generating novel research ideas is fundamental to scientific progress. While Large Language Models (LLMs) show promise in assisting this process, existing approaches often exhibit semantic convergence, resulting in limited diversity and novelty. To address this, we introduce EvoGens, an evolution-inspired framework that recasts scientific idea generation as an evolutionary search over a population of ideas. EvoGens iteratively applies rank-based mutation with differentiated retrieval planning to incorporate external knowledge, and semantic-aware crossover to fuse complementary concepts for conceptual reorganization. A lightweight evaluation signal guides the selection process, encouraging sustained exploration while mitigating premature convergence. Extensive experiments demonstrate that EvoGens substantially enhances exploration capabilities compared to state-of-the-art baselines. Specifically, it improves the Novelty from 0.1 to 0.4 and the Diversity from 0.24 to 0.55, while maintaining comparable idea quality under the current automatic evaluation protocol. These findings suggest that evolutionary mechanisms can serve as a useful framework for exploration-oriented research ideation, especially for broadening the novelty and diversity of candidate ideas under a shared automatic evaluation setting.
[NLP-71] Extending AI for Research to the Humanities: A Multi-Agent Framework for Evidence-Grounded Scholarship
【速读】： 该论文旨在解决生成式人工智能在人文学科研究中因过度依赖执行与检索而缺乏基于证据的解释性推理问题。现有基于大语言模型（LLM）的研究代理多适用于可执行实验、代码与量化信号驱动的科学工程领域，难以满足人文学科对原始资料忠实引用、可验证来源与细读分析的核心要求。为此，论文提出SPIRE（Scholarly-Primitives-Inspired Research Engine），一个基于学术原语（Scholarly Primitives）理论的多智能体框架，将人文学科中的重复性操作抽象为协同工作的智能体角色，包括源发现、证据标注、比较、出处核查、采样、引文绑定及论证综合等，并构建多尺度细读基础结构——段落层级、上下文图社区与跨上下文语义聚类。该架构使模型能够更可靠地恢复被引用的一手文献证据，在古典中文与希腊-罗马拉丁文研究的同行评审基准测试中，其表现优于基线方法（如朴素LLM、Text RAG与GraphRAG），且在盲评中获得更高的答案准确性、深度、覆盖度与证据质量评分。消融实验表明，学术操作智能体与细读检索机制均对生成基于证据的论述起到关键作用。
链接: https://arxiv.org/abs/2605.30947
作者: Yating Pan(1 and 2),Jiajun Zhang(2),Jun Wang(1, 2 and 4),Qi Su(3 and 4) ((1) Department of Information Management, Peking University, (2) Research Center for Digital Humanities, Peking University, (3) School of Foreign Languages, Peking University, (4) Institute for Artificial Intelligence, Peking University)
机构: Peking University (北京大学)
类目: Computation and Language (cs.CL)
备注: 28 pages, 3 figures. Code, data catalogues, and reproduction scripts: this https URL . Lead corresponding author: Jun Wang; corresponding author: Qi Su
Abstract:LLM-based research agents have advanced rapidly in science and engineering, where research is organized around executable experiments, code, and quantitative signals. Humanities scholarship, however, requires a different mode of reasoning: interpretive, evidence-grounded argument over primary sources, where scholarly value depends on faithful quotation, verifiable provenance, and close reading. Existing research agents remain largely optimized for execution and retrieval, not evidence-grounded interpretive reasoning. To address this gap, we introduce SPIRE (Scholarly-Primitives-Inspired Research Engine), a multi-agent framework for evidence-grounded humanities scholarship. Drawing on Scholarly Primitives theory, SPIRE casts recurring humanities operations as cooperating agent roles (source discovery, evidence annotation, comparison, provenance checking, sampling, citation binding, and argumentative synthesis) over a multi-scale close-reading substrate of passages, intra-context graph communities, and cross-context semantic clusters. On a peer-reviewed-paper benchmark over classical Chinese and Greco-Roman Latin scholarship, SPIRE recovers cited primary-source evidence more reliably than Naive LLM, Text RAG, and GraphRAG, and receives higher blind-judge scores on answer accuracy, depth, coverage, and evidence quality. Ablations show that both the scholarly-operation agents and close-reading retrieval contribute to evidence-grounded essays. Code, data catalogues, and reproduction scripts are released at this https URL.
[NLP-72] Do Large Language Models Encode Institutional Experience? Evidence from Cross-Linguistic Moral Reasoning Under Ambiguity
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在不同语言间存在系统性道德推理差异的问题，其核心疑问在于这种差异的根源是否与语言所处的社会制度环境相关。研究提出假设：语言可能编码了其使用者所处制度环境的特征，从而使LLMs在训练过程中继承特定制度背景下的道德先验。为验证该假设，研究选取了涵盖制度质量跨度广泛的九种语言、六种前沿大语言模型，并开展两项预注册研究，考察那些道德可接受性依赖于制度运行状况的道德困境。研究发现，在显式嵌入制度背景的情境下（Study 1），跨语言道德分歧并未显著增强，也未与语言社群间的制度差异相关联；而在制度背景模糊但隐含制度影响的情境下（Study 2），跨语言道德分歧显著增加，且多数情况下与真实世界中语言社群间的制度差异一致，而显式制度提示则削弱了这一效应。因此，研究的关键发现是：制度经验可能通过语言留下可被检测的痕迹，从而影响LLMs的道德推理模式，而显式制度线索可抑制此类差异的显现。
链接: https://arxiv.org/abs/2605.30934
作者: Nattavudh Powdthavee
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 44 pages
Abstract:Large language models (LLMs) exhibit systematic differences in moral reasoning across languages, yet the source of this variation remains unclear. We test the hypothesis that languages encode aspects of the institutional environments in which they are spoken, allowing LLMs to inherit institution-specific moral priors through training. Across nine languages spanning a broad gradient of institutional quality, six frontier LLMs, and two preregistered studies, we examine moral dilemmas whose acceptability depends on institutional functioning. In Study 1, explicit institutional framing produced uniformly null results: cross-linguistic moral divergence did not increase in institutionally contingent scenarios, nor did it track institutional differences between language communities. In Study 2, we introduced institutionally ambiguous scenarios in which institutional stakes were present but not explicitly stated. Under these conditions, cross-linguistic moral divergence increased relative to institutionally inert controls and, with one theoretically informative exception, was associated with real-world institutional differences between language communities. Explicit framing again attenuated these effects. These findings suggest that institutional experience may leave detectable traces in language that shape LLM moral reasoning, while also indicating that explicit institutional cues can suppress the expression of those differences.
[NLP-73] MineExplorer: Evaluating Open-World Exploration of MLLM Agents in Minecraft
【速读】： 该论文旨在解决多模态大语言模型（Multimodal Large Language Models, MLLMs）在动态开放世界中持续探索能力评估不足的问题。现有基于具身智能或游戏的基准测试普遍存在交互周期过短、任务成功标准与特定游戏机制高度耦合等缺陷，难以真实反映模型在复杂、开放环境中的通用推理与长期规划能力。为此，本文提出MineExplorer基准，聚焦于Minecraft环境中对MLLM代理的开放世界探索能力进行评估。其核心解决方案在于：首先筛选出依赖于Minecraft特有知识的原子任务，以更准确地衡量通用开放世界推理能力；其次采用类ReAct（ReAct-style）的能力建模方式，将原子任务组合为隐含多跳推理的复合任务；最后引入多智能体合成工作流，协同设计任务图谱、沙盒场景及基于规则的里程碑评估器，显著提升了任务实例的可靠性。实验表明，尽管先进MLLM在单跳任务上表现良好，但在需协调隐藏前置条件的长轨迹任务中性能急剧下降，且任务难度与模型完成率高度相关，更大模型或更复杂的思考模式并未带来一致性的性能提升。这揭示了当前MLLM在开放世界中实现持续探索仍面临根本性挑战。
链接: https://arxiv.org/abs/2605.30931
作者: Tianjie Ju,Yueqing Sun,Zheng Wu,Wei Zhang,Yaqi Huo,Xi Su,Qi Gu,Xunliang Cai,Gongshen Liu,Zhuosheng Zhang
机构: Shanghai Jiao Tong University (上海交通大学); Meituan (美团)
类目: Computation and Language (cs.CL)
备注: Working in progress
Abstract:Multimodal large language models (MLLMs) have shown strong capabilities in perception, reasoning, and action generation. However, their ability to sustain exploration in dynamic open worlds remains unclear. Existing embodied and game-based benchmarks often compress interaction into short-horizon tasks or entangle success with domain-specific game mechanics. In this paper, we introduce MineExplorer benchmark for evaluating open-world exploration capabilities of MLLM agents in Minecraft. We first filter atomic tasks whose solutions rely heavily on Minecraft-specific knowledge to better reflect general open-world reasoning. Then we organize the benchmark around a ReAct-style capability formulation and compose atomic tasks into implicit multi-hop tasks. To further construct reliable instances, MineExplorer uses a multi-agent synthesis workflow that jointly designs task graphs, sandbox scenes, and rule-based milestone evaluators. Human evaluation shows that the multi-agent synthesis workflow produces significantly more reliable instances than a single-agent baseline. Experiments with advanced MLLM agents show that open-world exploration remains challenging, as strong models can handle many single-hop tasks but degrade sharply when hidden prerequisites must be coordinated over longer trajectories. Further analysis finds that task difficulty tracks agent completion, and larger models or thinking modes do not consistently translate into better performance. Code and dataset are available at this https URL.
[NLP-74] EMBGuard: Constructing Hazard-Aware Guardrails for Safe Planning in Embodied Agents ICML2026
【速读】： 该论文旨在解决多模态大模型驱动的具身智能体（Embodied Agents）在真实环境中部署时面临的物理安全风险识别难题。现有方法缺乏对动作条件下的风险进行显式建模与推理的能力，导致智能体要么遗漏潜在危险交互，要么过度误报风险，影响其在真实场景中的可靠性和安全性。为此，本文提出 EMBGuard——首个基于多模态大模型（MLLM）的具身智能体安全护栏机制，其核心创新在于将物理风险推理与智能体策略（Policy）解耦，通过评估“视觉观测-动作”配对来识别危险配置，并生成自然语言形式的风险解释。为支持该框架，研究构建了 EMBHazard 数据集（含15.1K个动作条件化样本）和 EMBGuardTest 基准测试集（涵盖329个手工标注的真实世界场景，覆盖七类物理风险）。通过组合式变异生成多样化的高风险与安全场景，有效模拟规划过程中的复杂情境。尽管模型规模仅为2B或4B，EMBGuard 在性能上可媲美闭源大模型（如 GPT-5.1、Gemini-2.5-Pro），同时显著降低误报率，提升了实时部署的可行性。相关代码、数据及模型已公开发布。
链接: https://arxiv.org/abs/2605.30924
作者: Dongwook Choi,Taeyoon Kwon,Bogyung Jeong,Minju Kim,Yeonjun Hwang,Hyojun Kim,Byungchul Kim,Young Kyun Jang,Jinyoung Yeo
机构: 未知
类目: Computation and Language (cs.CL)
备注: Accepted at ICML 2026
Abstract:MLLM-powered embodied agents deployed in real-world environments encounter physical hazards. However, existing approaches lack explicit mechanisms for identifying hazards and reasoning about action-conditioned risks, leading agents to either miss risky interactions or over-identify risks. To address this, we propose EMBGuard, the first MLLM-based safety guardrail for embodied agents designed to decouple physical risk reasoning from agent policy. By evaluating a (visual observation, action) pair, EMBGuard identifies hazardous configurations and provides natural language explanations of potential risks. Alongside EMBGuard, we contribute EMBHazard, a training dataset of 15.1K action-conditioned pairs, and EMBGuardTest, a benchmark of 329 manually curated real-world scenarios spanning seven physical risk categories. Through compositional variation of hazards and actions, we generate diverse risky and benign scenarios that agents may encounter during planning. Despite its compact size (2B, 4B), EMBGuard achieves performance competitive with proprietary MLLMs (e.g., GPT-5.1, Gemini-2.5-Pro) while significantly reducing the false-positive rates that hinder real-time deployment. We make the code, data, and models publicly available at this https URL
[NLP-75] Attend to Evidence: Evidence-Anchored Spatial Attention Supervision for Multimodal RLVR
【速读】： 该论文旨在解决生成式视觉语言模型（VLMs）在强化学习中仅依赖最终答案奖励（outcome-only rewards）所导致的可解释性不足问题。此类奖励无法指导模型识别支持答案的关键图像区域，从而使得模型可能通过语言先验或偶然猜测获得高分，而非基于真实的视觉证据进行推理。为应对这一挑战，论文提出EASE（Evidence-Anchored Spatial Attention），其核心创新在于引入视觉证据过程监督机制，在强化学习训练过程中仅对高奖励轨迹施加基于标注证据区域生成的平滑视觉标记目标，以引导模型在回答时聚焦于与答案相关的视觉区域。该方法将标注信息作为特权训练标签使用，推理阶段则无需额外输入，仅需原始图像和问题。实验表明，EASE在多个基准测试（包括感知、幻觉抑制、视觉数学及多模态推理）上相较于DAPO显著提升平均得分2.5至3.1分，且诊断分析与消融实验证明其能更有效地对齐视觉注意力与人工标注的证据区域。
链接: https://arxiv.org/abs/2605.30912
作者: Ruina Hu,Chen Wang,Lai Wei,Jionghao Bai,Bin Yu,Weiran Huang,Kai Wang,Yue Wang
机构: Harbin Institute of Technology (哈尔滨工业大学); Zhongguancun Academy (中关村研究院); Zhongguancun Institute of Artificial Intelligence (中关村人工智能研究院); Nankai University (南开大学); Shanghai Jiaotong University (上海交通大学); Zhejiang University (浙江大学)
类目: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL)
备注:
Abstract:Reinforcement learning with verifiable rewards (RLVR) improves vision-language models (VLMs) by optimizing outcome rewards derived from final answers. However, such outcome-only rewards do not tell the model which image regions justify an answer. For questions that require visual grounding, these rewards cannot distinguish responses supported by relevant visual evidence from those produced by language-prior shortcuts or lucky guesses. We introduce EASE (Evidence-Anchored Spatial Attention), which augments multimodal RLVR with visual-evidence process supervision. EASE converts annotated evidence regions into a smoothed visual-token target and uses it to guide response-to-image attention during RL training, but only on high-reward trajectories. The annotations are used solely as privileged training labels, while inference requires only the original image and question. Across Qwen2.5-VL-7B, Qwen3-VL-4B, and Qwen3-VL-8B, EASE raises average scores over DAPO by 2.5 to 3.1 points on perception, hallucination, visual math, and multimodal reasoning benchmarks. Diagnostics and ablations show that EASE better aligns visual attention with annotated evidence regions.
[NLP-76] BlueFin: Benchmarking LLM Agents on Financial Spreadsheets
【速读】： 该论文旨在解决当前大型语言模型（LLM）在专业金融领域电子表格任务中的能力评估与应用短板问题。尽管电子表格软件的付费用户规模达数亿级别，远超专业开发人员群体，但针对此类场景下模型能力的研究仍相对匮乏，尤其缺乏对真实职业任务的模拟与系统性评估。为应对这一挑战，研究提出BlueFin基准，包含131个具有现实意义且复杂的金融领域电子表格任务，涵盖数据合成、操作与理解三类核心能力，并设计了3,225条细粒度评分标准。其关键创新在于引入由专家人工标注团队验证的评分体系与基于大模型的评判代理（LM judge），实现了对程序化难以验证的复杂任务的高精度、可信赖评估，且评判结果与专家共识高度一致（α=0.826，宏平均F1=0.839）。实验表明，前沿大模型在该基准上表现不佳，最强模型平均得分不足50%，尤其在动态正确性方面存在显著缺陷。本研究的核心贡献包括：覆盖三类任务的真实世界数据集、开源的代理评估框架及对现有先进模型性能的系统性刻画。
链接: https://arxiv.org/abs/2605.30907
作者: Srivatsa Kundurthy,Clara Na,Colton Moraine,Anoushka Mohta,Case Winter,George Fang,John Ling,Emma Strubell,Zach Kirshner
机构: Longitude Labs Inc.(Longitude Labs 公司); Cornell University(康奈尔大学); Carnegie Mellon University(卡内基梅隆大学)
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 26 pages
Abstract:We present BlueFin, a benchmark that tasks large language model (LLM) agents with synthesis, manipulation, and comprehension tasks over spreadsheet workbooks in the professional finance domain. Though estimates of the global population of paying users of spreadsheet software range in the hundreds of millions – an order of magnitude more than the estimated global population of professional developers – comparatively fewer resources have been devoted to exploring and expanding LLM capabilities in the spreadsheet domain, with fewer still dedicated to mirroring real occupational tasks encountered by those in professional finance roles. In response, we curate a set of 131 challenging, complex tasks with real-world relevance in the domain, containing 3,225 granular rubric criteria; notably, our rubric criteria and LM judge evaluations are validated by a team of expert human annotators, resulting in high-quality, granular evaluations of complex tasks that are difficult to verify programmatically but can be reliably evaluated by an LM judge agent. Our judge achieves parity with expert consensus ( \alpha=0.826 ) with a macro-F1 score of 0.839. Frontier LLMs demonstrate poor performance on the challenging benchmark, with the strongest LLMs achieving less than 50% average scores across tasks – models exhibit particular weaknesses in dynamic correctness. Our contributions include a dataset of examples across three categories of spreadsheet tasks, an open source harness and agentic evaluation framework, and a characterization of existing frontier models’ performance on our benchmark.
[NLP-77] UniScale: Adaptive Unified Inference Scaling via Online Joint Optimization of Model Routing and Test-Time Scaling ICML2026
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在实际部署中推理质量与计算成本之间的权衡难题。现有方法通常将这一问题分解为两个独立的优化维度：模型路由（model routing），即根据请求复杂度在不同规模的模型间切换以匹配需求；以及测试时扩展（Test-time Scaling, TTS），即在固定模型内调整推理时的计算资源以实现细粒度控制。然而，这种解耦设计存在固有局限：模型路由因可用模型规模稀疏而仅能提供粗粒度的性能变化，而单模型TTS则易受容量上限制约，并随计算投入增加呈现收益递减现象；同时，两机制分离限制了动态推理环境下的适应能力。为此，本文提出统一推理扩展（Unified Inference Scaling, UIS），将模型路由与TTS整合至同一优化空间，实现协同优化。在此基础上，进一步提出UniScale框架，将自适应的UIS建模为上下文相关的多臂赌博机（contextual multi-armed bandit）问题，并采用线性上下文感知上置信界（LinUCB）算法进行推理策略学习。该框架融合效率感知学习与成本建模机制，确保在高维动作空间中实现稳定且可扩展的优化。实验表明，UniScale能够有效挖掘UIS空间中的协同效应，在多样且动态的推理场景下持续实现更优的质量-成本权衡。
链接: https://arxiv.org/abs/2605.30898
作者: Kaiyu Huang,Xingyu Wang,Mingze Kong,Zhubo Shi,Yuqian Hou,Hong Xu,Zhongxiang Dai,Minchen Yu,Qingjiang Shi
机构: 未知
类目: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: Accepted at the 43rd International Conference on Machine Learning (ICML 2026)
Abstract:In real-world deployments of large language models (LLMs), balancing inference quality and computational cost has become a central challenge. Existing approaches tackle this trade-off along two largely independent dimensions: model routing, which switches among models of different scales to match request complexity, and test-time scaling (TTS), which adjusts inference-time compute within a fixed model for fine-grained control. However, this decoupled design introduces inherent limitations. Model routing yields coarse-grained, discrete performance changes due to the sparse set of model scales, while single-model TTS often encounters capacity ceilings and exhibits diminishing returns as compute increases. Moreover, treating the two mechanisms separately restricts adaptability in dynamic inference environments. To overcome these limitations, we introduce Unified Inference Scaling (UIS), which unifies model routing and TTS in a single optimization space. Building on this formulation, we propose UniScale, an online framework that models adaptive UIS as a contextual multi-armed bandit problem and learns inference policies via LinUCB. The framework incorporates efficiency-aware learning and cost modeling to ensure stable and scalable optimization over high-dimensional action spaces. Evaluation shows that UniScale effectively exploits the synergy in the UIS space to deliver a fine-grained and consistently better quality-cost trade-off across diverse, dynamic inference scenarios.
[NLP-78] he Flip Side of RLHF: On-Policy Feedback for Reward Model Self-Supervised Improvement
【速读】： 该论文旨在解决语言模型对齐中奖励模型（Reward Model, RM）训练所面临的挑战，即高质量偏好数据的获取成本高且难度大，尤其在策略（policy）动态演化时，静态训练的RM难以适应新分布。为应对这一问题，论文提出SAVE（Self-supervised reward model improvement via Value-Anchored On-policy feedback）框架，其核心创新在于利用价值函数（value function）对在线策略生成的响应进行评分，从而提供自监督反馈以实现在线奖励模型训练。关键在于通过提示特定的价值头作为自适应锚点，将奖励评分转化为监督信号，并结合优势计算与模糊样本过滤机制，基于对比学习目标更新奖励模型。该方法显著提升了奖励模型的训练效果，在六个不同基准上的实证评估中均表现出色，且在三种强化学习算法（GRPO、RLOO、GSPO）及多种策略主干网络上保持一致的性能提升。
链接: https://arxiv.org/abs/2605.30888
作者: Xiaobo Wang,Tong Wu,Min Tang,Jiaqi Li,Qi Liu,Zilong Zheng
机构: University of Science and Technology of China(中国科学技术大学); State Key Laboratory of Cognitive Intelligence(认知智能国家重点实验室); Institute of Artificial Intelligence, Hefei Comprehensive National Science Center(合肥综合性国家科学中心人工智能研究院); State Key Laboratory of General Artificial Intelligence, BIGAI(通用人工智能国家重点实验室, BIGAI)
类目: Computation and Language (cs.CL)
备注:
Abstract:Building strong reward models (RMs) for language model alignment is bottlenecked by the cost and difficulty of acquiring diverse and reliable preference data from human annotation or judge models. It is dramatically worse as the policy evolves beyond the static RM training. Therefore, we propose SAVE (Self-supervised reward model improvement via Value-Anchored On-policy feedback), a framework that grades on-policy responses as feedback by using the value function for on-policy RM training. SAVE naturally converts the reward-graded on-policy responses into supervision with a prompt-specific value head as an adaptive anchor. It computes RM advantages and filters ambiguous samples to update the RM via a contrastive objective. The effectiveness of SAVE for enhancing RM training is strongly validated through rigorous empirical evaluation across six diverse benchmarks. It achieves outperforming results across all datasets while maintaining consistent improvements across three RL algorithms (GRPO, RLOO, GSPO) and different policy backbones.
[NLP-79] PatchWorld: Gradient-Free Optimization of Executable World Models
【速读】： 该论文旨在解决在部分可观测马尔可夫决策过程（Partially Observable Markov Decision Process, POMDP）环境下，如何构建可执行的、可解释的世界模型以支持预测与规划的问题。现有方法通常依赖黑箱模型进行观测预测，缺乏对状态更新机制的透明性与可调试性。本文提出PatchWorld框架，其核心创新在于通过反例引导的代码修复机制，将离线轨迹转化为可执行的Python世界模型，生成符号化的信念状态程序（symbolic belief-state programs），使得动作更新过程可被检查、回放与局部修补。相较于传统黑箱预测模型，该方法实现了更高的可解释性与可控性，在七个AgentGym环境中，PatchWorld-Simple在无需调用大语言模型（LLM）的情况下，于实时单步前瞻中达到76.4%的宏观成功率，表现最优。研究进一步揭示了可执行世界模型中的关键权衡：人为设定的残差记忆偏置虽能提升表面观测保真度，但会削弱动作判别性动态，表明观测保真度与决策效用之间存在内在冲突。
链接: https://arxiv.org/abs/2605.30880
作者: Jiaxin Bai,Yue Guo,Yifei Dong,Jiaxuan Xiong,Tianshi Zheng,Yixia Li,Tianqing Fang,Yufei Li,Yisen Gao,Haoyu Huang,Zhongwei Xie,Hong Ting Tsang,Zihao Wang,Lihui Liu,Jeff Pan,Yangqiu Song
机构: Hong Kong Baptist University(香港浸会大学); Independent Researcher(独立研究员); HKUST(香港科技大学); Beijing Institute of Technology(北京理工大学); Southern University of Science and Technology(南方科技大学); Wayne State University(韦恩州立大学); University of Edinburgh(爱丁堡大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 40 pages
Abstract:Text-agent environments are typically modeled as partially observable Markov decision processes (POMDPs), assuming that the simulator’s latent state and transition dynamics are hidden from the agent. Yet little work has examined whether executable code can be induced to serve as a world model for prediction and planning under partial observability. We introduce PatchWorld, a gradient-free framework that turns offline trajectories into executable Python world models through counterexample-guided code repair. Instead of predicting the next observation with a black-box model, PatchWorld induces symbolic belief-state programs whose action updates can be inspected, replayed, and locally patched. Across seven AgentGym environments, PatchWorld-Simple achieves the highest code-based planning score among evaluated methods, reaching 76.4% macro success in live one-step lookahead while invoking no LLM calls inside the world-model prediction module itself. We further find that a human-specified residual-memory bias improves surface observation fidelity but weakens decision utility. This exposes a tradeoff in executable world models, since improving observation fidelity can come at the expense of action-discriminative dynamics, and vice versa. Code is available at this https URL.
[NLP-80] dMoE: dLLM s with Learnable Block Experts
【速读】： 该论文旨在解决扩散型大语言模型（Diffusion Large Language Models, dLLMs）与混合专家（Mixture-of-Experts, MoE）架构融合时所面临的块并行解码与词元级专家选择之间的根本性不匹配问题。具体而言，dLLMs在前向传播中通过双向依赖同时处理多个词元，而传统MoE层对每个词元独立进行专家路由，导致在块级别上激活的专家数量显著增加，进而使推理过程陷入内存瓶颈。其解决方案的关键在于提出一种名为dMoE的块级MoE框架，核心思想是将每个块内各词元的专家分配分布聚合为统一的块级专家分布，并以此指导专家路由，从而实现更一致、高效的专家选择。该方法在不损失性能的前提下，大幅减少唯一激活专家的数量（从平均69.5降至14.6），显著降低内存占用（76.64%–79.84%）并提升端到端推理速度（加速1.14×至1.66×）。
链接: https://arxiv.org/abs/2605.30876
作者: Sicheng Feng,Zigeng Chen,Gongfan Fang,Xinyin Ma,Xinchao Wang
机构: National University of Singapore (新加坡国立大学)
类目: Computation and Language (cs.CL)
备注: Working in progress. Code is available at: \url{ this https URL }
Abstract:Diffusion Large Language Models (dLLMs) have recently emerged as a promising alternative to autoregressive models, offering competitive performance while naturally supporting parallel decoding. However, as dLLMs are increasingly integrated with Mixture-of-Experts (MoE) architectures to scale model capacity, a fundamental mismatch arises between block parallel decoding and token-level expert selection. Specifically, each dLLM forward pass processes multiple tokens with bidirectional dependencies, whereas conventional MoE layers route each token independently. This mismatch substantially increases the number of uniquely activated experts, making inference increasingly memory-bound. To address this, we propose dMoE, a simple yet effective block-level MoE framework. The central idea of dMoE is to aggregate token-level expert distributions within each block into a unified block-level expert distribution, which is then used to guide expert routing in a more coherent manner. In this way, dMoE substantially reduces the number of uniquely activated experts during inference without sacrificing performance, thereby mitigating the memory-bound bottleneck. Extensive experiments across a variety of benchmarks demonstrate the effectiveness of dMoE. On average, dMoE reduces the number of uniquely activated experts from 69.5 to 14.6 while retaining 99.11% of the original performance. Meanwhile, it reduces memory usage by 76.64% to 79.84% and achieves 1.14 \times to 1.66 \times end-to-end latency speedup. Code is available at: this https URL
[NLP-81] MADS: Model-Aware Diverse Core Set Selection for Instruction Tuning
【速读】： 该论文旨在解决大规模语言模型（LLM）在指令微调（instruction fine-tuning）过程中，随着训练数据量增加，如何高效选取具有代表性的多样化核心数据集（core set）以提升模型指令遵循能力的问题。现有方法多依赖于文本表面特征进行数据区分，未能充分考虑模型自身对数据的理解与表征能力，导致所选核心集的多样性不足。为此，本文提出一种模型感知的多样化核心集选择方法（Model-Aware Diverse Core Set Selection），其关键在于利用大模型在推理过程中的神经激活状态（neural activation states）作为数据特征表示，基于模型内在的表征能力来衡量和区分数据差异，从而实现更优的覆盖性采样。该方法通过引入模型自身的语义理解能力指导数据筛选，显著提升了核心集的多样性与代表性。实验结果表明，由3B参数模型（Llama-3.2-3B-Instruct）选出的仅占原始数据15%的核心集，在微调7B、8B及13B参数的更大模型时，相较使用全量数据仍能带来平均2.5%的性能提升，且在多个下游任务上均表现出更强的泛化能力，验证了该方法在降低数据需求的同时有效增强模型性能的可行性与有效性。
链接: https://arxiv.org/abs/2605.30857
作者: Yi Bai,Wenhao Zhang,Yao Chen,Jiao Xue,Zhumin Chen,Pengjie Ren
机构: Shandong University (山东大学); Inspurcloud (浪潮云)
类目: Computation and Language (cs.CL)
备注:
Abstract:Instruction fine-tuning is employed to enhance the instruction-following ability of large language models (LLMs). As the amount of instruction fine-tuning data increases, selecting the optimal core set becomes particularly important. However, ensuring the diversity of the core set remains a significant challenge. Existing methods predominantly distinguish different training data based on the text features themselves, decoupled from LLMs’ own understanding and representation of the data. To address this issue, we propose a Model-Aware Diverse Core Set Selection method, which distinguishes data features based on the neural activation states during LLM inference. This approach serves as an efficient instantiation of coverage-based selection using model-intrinsic activation features to ensure the diversity in the core set. We extensively evaluate our method on six benchmarks that cover five distinct tasks. In our method, the core set selected by the 3B-parameter LLM performs effectively when utilized to fine-tune larger models with 7B, 8B, and 13B parameters. Experimental results on the Alpaca-GPT4 dataset, which comprises 52K instruction-response pairs, show that the core set, sized at 15% of the original dataset and selected by Llama-3.2-3B-Instruct, achieves an average improvement of 2.5% when fine-tuning four larger base models compared with training on the full dataset. The experimental results demonstrate that our method enhances model performance on multiple downstream tasks while reducing data requirements.
[NLP-82] Speculative Pipeline Decoding: Higher-Accruacy and Zero-Bubble Speculation via Pipeline Parallelism
【速读】： 该论文旨在解决生成式 AI（Generative AI）在低并发场景下大语言模型（LLM）推理速度慢的问题，尤其针对主流推测解码（Speculative Decoding, SD）方法依赖多标记预测所导致的预测难度递增与串行起草延迟瓶颈。其核心解决方案是提出一种名为推测流水线解码（Speculative Pipeline Decoding, SPD）的新框架，通过将目标大语言模型划分为 $ n $ 个流水线阶段，实现并行处理 $ n $ 个标记，从而充分释放流水线并行性的潜力。为保障单序列解码中流水线持续填充，SPD 引入一个推测模块，该模块跨不同流水线层级聚合中间特征以预测下一个标记，并严格与目标模型的流水线步骤并行执行，实现了预测难度可控、接受率更高以及零延迟空洞（latency bubbles）的效果。实验表明，SPD 在理论加速比上显著优于主流基线方法，提供了一种高度可扩展的大型语言模型解码加速方案。
链接: https://arxiv.org/abs/2605.30852
作者: Yijiong Yu,Huazheng Wang,Shuai Yuan,Ruilong Ren,Ji Pei
机构: Oregon State University; DeepSolution
类目: Computation and Language (cs.CL)
备注:
Abstract:Speculative Decoding (SD) accelerates low-concurrency LLM inference by employing a draft-then-verify paradigm. However, mainstream methods typically rely on multi-token prediction, which introduces escalating prediction difficulty and serial drafting latency. To address these, we propose Speculative Pipeline Decoding (SPD), a groundbreaking framework that unlocks the true potential of pipeline parallelism. By partitioning the target LLM into n pipeline stages, SPD allows LLM to process n tokens in parallel to accelerate decoding. To continuous fill the pipeline in single sequence decoding, a speculation module aggregates intermediate features across different pipeline depths to predict the next token, executing strictly in parallel with the target model’s pipeline step, to realize bounded difficulty, higher acceptance rates, and zero latency bubbles. Our experiments demonstrate that SPD achieves a significantly higher theoretical speedup compared to mainstream baselines, offering a highly scalable solution for LLM decoding acceleration. Our code is available at this https URL
[NLP-83] LLM Anonymization Against Agent ic Re-Identificatio
【速读】： 该论文旨在解决生成式人工智能（Generative AI）驱动的智能体（Agentic LLMs）结合网络搜索能力对文本匿名化带来的新型隐私威胁问题：传统匿名化方法在去除显式标识符或进行形式化扰动时，往往忽视了上下文线索可能通过网络搜索被交叉引用以实现再识别，同时这些线索又蕴含重要的下游分析价值。现有防御机制或完全移除敏感信息、或依赖非网络推理模型评估重写文本，未能有效平衡对抗智能体网络搜索再识别攻击的能力与保持文本上下文语义实用性的需求。其解决方案的关键在于提出AURA（Anonymization with Utility-Retention Adaptation），一种基于大语言模型（LLM）的掩码重构（mask-reconstruct）框架，通过将隐私定位与效用保留重建解耦，并引入对抗性隐私与效用保留双重验证机制，动态选择最优匿名化候选方案。实验结果表明，AURA通过自适应隐私范围增强对智能体再识别攻击的鲁棒性，同时利用掩码重构策略在固定隐私强度下显著提升上下文效用保留能力，在真实用户访谈转录数据上实现了更优的隐私-效用权衡。
链接: https://arxiv.org/abs/2605.30848
作者: Ziwen Li,Jianing Wen,Tianshi Li
机构: 未知
类目: Cryptography and Security (cs.CR); Computation and Language (cs.CL)
备注: 32 pages, 7 figures
Abstract:Agentic LLMs with web search change the threat model for text anonymization: weak contextual cues can become cross-referenceable evidence for re-identification, yet those same details also carry downstream analytic value of the text. Existing defenses either remove explicit identifiers, perturb text for formal privacy, or test rewritten text against non-web inference models, leaving underexplored the operating region between resistance to agentic web-search re-identification and utility retention. We introduce AURA (\textbfAnonymization with \textbfUtility-\textbfRetention \textbfAdaptation), an LLM-powered \textitmask-reconstruct framework that decouples privacy localization from utility-preserving reconstruction and selects candidates with adversarial privacy and utility-retention checks. We evaluate AURA on real-user interview transcripts using re-identification attacks carried out by web-search agents, along with a utility evaluation based on interviewee-profile facts, codebook facts, and the joint contextual utility grid. Our results show that AURA improves the privacy-utility frontier by using adaptive privacy scope to strengthen resistance to agentic re-identification and using a mask-reconstruct anonymization method to better preserve contextual utility under fixed privacy scope.
[NLP-84] Fine-Tuning Improves Information Conveyance in Language Models
【速读】： 该论文旨在解决现有研究中对微调（fine-tuning）如何影响大语言模型生成过程中的不确定性分布理解不足的问题，尤其忽视了输出长度这一关键混淆变量。传统分析未能充分刻画不确定性在整个生成过程（generation rollout）中的动态分布，导致对微调后模型行为的评估存在偏差。为此，作者提出了一种名为“树冠熵”（Canopy Entropy, CE⋆）的新度量方法，从树状结构视角建模语言生成过程，其中“树冠”代表所有可能的生成路径空间，使CE⋆能够自然地量化生成空间的有效规模。该指标联合捕捉输出长度N与生成序列Y1:N中的不确定性，其数学形式等价于总香农熵H(N,Y1:N∣X)，其中X为输入提示。该框架进一步引入可解释的度量，如长度-熵率相关性ρ(N,rN)，用于衡量每标记的信息传递效率，即更长输出是否具有更高的单位信息量。实证结果表明，在多种任务和模型家族中，尽管微调后总熵降低，但微调模型始终表现出更强的正向ρ(N,rN)相关性；在控制模型族、任务、提示及输出长度等影响因素后，微调显著将熵率与语义多样性之间的相关性强度提升近三倍，说明对齐模型能更高效地将标记级不确定性转化为语义层面的多样性。综上，该研究揭示微调并非简单压缩不确定性，而是从根本上重构不确定性分布，使其生成内容更具信息量与语义丰富性。
链接: https://arxiv.org/abs/2605.30844
作者: Yuwei Cheng,Weiyi Tian,Haifeng Xu
机构: University of Chicago (芝加哥大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (stat.ML)
备注:
Abstract:Fine-tuning is often believed to reduce uncertainty and diversity in large language models, but existing analyses overlook output length, a key confounder, and therefore fail to capture how uncertainty is distributed across an entire generation rollout. To address this, we propose Canopy Entropy ( \mathrmCE^\star ), a measure that views language generation from a tree perspective, where ``canopy’’ represents the space of all possible rollouts, making \mathrmCE^\star naturally quantify the effective size of the generation space. \mathrmCE^\star jointly captures uncertainty in both the output length N and the generated sequence Y_1:N – indeed, we show that it equals to total Shannon entropy H(N, Y_1:N\mid X) , where X denotes the prompt. This formulation yields interpretable metrics, including a length-entropy correlation term \rho(N, r_N) , where r_N is the entropy rate, quantifying information conveyance efficiency by indicating whether longer outputs are more or less informative per token. Empirically, across tasks and model families, we find that fine-tuned models consistently exhibit stronger positive correlation \rho(N, r_N) , even when total entropy decreases. Furthermore, after controlling for model family, task, prompt, and output-length effects, we find that fine-tuning nearly triples the correlation strength between entropy rate and semantic diversity, suggesting that aligned models convert token uncertainty into semantic diversity more efficiently. Overall, these results demonstrate that fine-tuning does not simply reduce uncertainty, but fundamentally reorganizes it into more informative and semantically meaningful generations. Our code is available at this https URL.
[NLP-85] Your Teacher Cant Help You Here: Combating Supervision Fidelity Decay in On-Policy Distillation
【速读】： 该论文旨在解决在策略蒸馏（On-policy Distillation, OPD）中因学生模型生成序列长度增加导致的监督保真度衰减（Supervision Fidelity Decay, SFD）问题。其核心挑战在于：随着学生模型生成的前缀（prefix）逐渐变长，教师模型对下一个词元的分布变得愈发不确定且区分度降低，导致基于反向KL散度的教师依赖性修正信号逐渐弱化，进而引发学生模型在长推理链中产生累积性漂移（student drift）。为缓解SFD，本文提出前瞻组奖励（Lookahead Group Reward, \ours），其关键创新在于：利用教师模型在下一步预测中的置信度作为未来反向KL监督判别力的代理指标，通过评估学生当前候选词元在下一时刻所激发的教师置信度水平，对候选词元进行分组归一化奖励分配，从而增强长期推理过程中的监督有效性。为进一步保障计算效率，\ours还引入了熵触发的树注意力机制（entropy-triggered tree-attention mechanism），实现高效的大规模候选集评估。实验表明，在六个数学与代码基准测试中，\ours相较于标准OPD在7B参数学生模型上平均提升2.57点（mean@8），在长生成任务中表现更优，于AIME-26（39k token）任务上达到+4.92点显著增益。
链接: https://arxiv.org/abs/2605.30833
作者: Yanjiang Liu,Jie Lou,Xinyan Guan,Yuqiu Ji,Hongyu Lin,Ben He,Xianpei Han,Le Sun,Xing Yu,Yaojie Lu
机构: University of Chinese Academy of Sciences; Chinese Information Processing Laboratory, Institute of Software, Chinese Academy of Sciences; Xiaohongshu
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:On-policy distillation transfers reasoning capabilities by training a student model on its own generated trajectories using token-level feedback from a teacher. However, we identify a critical bottleneck, \textbfSupervision Fidelity Decay (SFD): as student-generated prefixes lengthen, the teacher’s next-token distribution becomes less confident and less discriminative. Consequently, the teacher-dependent corrective signal in reverse-KL distillation weakens, causing student drift to compound across long reasoning chains. To mitigate SFD, we introduce \textbfLookahead Group Reward (\ours). Building on the insight that next-step teacher confidence reflects the discriminative strength of future reverse-KL supervision, \ours evaluates the student’s top-K candidate tokens by the teacher confidence they induce at the subsequent step and assigns a group-normalized reward. To maintain computational efficiency, we further design an entropy-triggered tree-attention mechanism. Across six math and code benchmarks, \ours improves mean@8 by \textbf2.57 points over OPD for a 7B student, with gains increasing in longer-generation and reaching +\textbf4.92 points on AIME-26 at 39k tokens.
[NLP-86] Beyond Agreement: Scoring Panel-Surfaced Biomedical Entity Candidates for Curator Triage
【速读】： 该论文旨在解决生成式大模型（LLM）在生物医学命名实体识别（Biomedical NER）任务中看似简单实则复杂的核心问题：尽管现代大模型能够轻易提取出看似合理的生物医学实体提及，但其结果是否符合特定语料库的标注规范（如标注惯例、实体边界、粒度级别和类型体系）仍难以保证。传统基于多模型一致性的方法虽可作为显著性信号，却无法直接等同于语料库规范的正确性。为此，论文提出一种候选层级的面板输出基准（candidate-level panel-output benchmark），将评估单位从独立模型的输出转变为由显式定义的多模型面板共同生成并对齐的候选集合。该基准将八个大模型在五个公开生物医学NER数据集上的预测结果统一整合为一个候选主表，并引入一个领域内监督评分器BioConCal，该评分器在无需推理时真实标签的前提下，利用模型间一致性、提及表面特征、出现频率及文档级特征，对固定候选流进行评分。实验表明，在领域内，BioConCal将原始一致性方法的AUROC从0.753提升至0.910；在设定0.95精确率目标下，可筛选出1,340个候选，实测精确率达0.939，远超原始一致性方法的293个候选；对应候选层级召回率为0.592，语料层级召回率为0.523（相对于面板内行级标签上限0.883）。其核心价值并非恢复所有面板成员均遗漏的实体，而是将高噪声的面板候选流重构为更具产出效率的审查队列。此外，当面临实体类型分布偏移时，阈值需通过目标领域验证确定，而精确字符级定位仍需独立的确定性后处理步骤完成。
链接: https://arxiv.org/abs/2605.30826
作者: Shuheng Cao,Ruiqi Chen,Renjie Cao,Zhenhao Zhang,Siyu Zhang,Tingting Dan
机构: University of California, San Diego; University of Michigan, Ann Arbor; The Hong Kong University of Science and Technology, Guangzhou; ShanghaiTech University; University of North Carolina, Chapel Hill
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Biomedical NER is deceptively simple for modern LLMs: plausible biomedical mentions are easy to surface, but corpus-convention correctness depends on annotation conventions, span boundaries, entity granularity, and type schemas. Multi-LLM agreement is a salience signal, not corpus-convention correctness. We introduce a candidate-level panel-output benchmark for panel-surfaced candidate verification, where the unit is an aligned candidate surfaced by an explicitly defined multi-model panel rather than a standalone extractor output. The benchmark aligns eight LLMs’ predictions over five public biomedical NER datasets into a candidate master table. BioConCal is an in-domain supervised scorer that instantiates this layer with inference-time gold-free agreement, mention, surface-availability, and document features for a fixed candidate stream. In domain, BioConCal improves AUROC from 0.753 for raw agreement to 0.910. At a validation-selected 0.95 precision target it selects 1,340 candidates at empirical test precision 0.939, compared with 293 for raw agreement. This corresponds to candidate-level recall 0.592 and corpus-level recall 0.523 against a within-panel row-label ceiling of 0.883. The main benefit is not recovering entities missed by every panel member, but reshaping a noisy panel stream into a higher-yield review queue. Under entity-type shift, thresholds require target-domain validation, and exact character localization remains a separate deterministic post-processing step.
[NLP-87] Incremental BPE Tokenization ICML2026
【速读】： 该论文旨在解决传统字节对编码（BPE）在流式处理场景下效率低下的问题，尤其是在需要对输入文本的前缀进行增量式分词时，标准BPE算法存在较高的计算开销和延迟。其核心解决方案是提出一种新型的增量式BPE分词算法，能够在最坏情况下以每字节O(log2t)的时间复杂度完成处理，整体时间复杂度为O(nlog2t)（其中n为输入长度，t为最大词元长度），并持续维护输入文本每个前缀的分词结果。该算法通过高效地增量维护合并规则的应用状态，实现了对标准BPE合并过程的精确模拟，从而支持流式环境中的高效部分分词。此外，论文进一步引入了一种“主动输出”（eager output）机制，可在增量分词过程中尽早确定词元边界时立即输出词元，显著降低延迟。实验表明，该方法作为标准BPE的即插即用替代方案，在Hugging Face的tokenizers上实现约3倍的速度提升，并在极端输入下相较OpenAI的tiktoken展现出显著的延迟优势。研究证明，基于该算法的BPE分词可在保证强最坏情况时间复杂度的同时，为现代大语言模型流水线提供实际的延迟优化。
链接: https://arxiv.org/abs/2605.30813
作者: Shenghu Jiang,Ruihao Gong
机构: 未知
类目: Computation and Language (cs.CL); Data Structures and Algorithms (cs.DS)
备注: Accepted to ICML 2026 (Spotlight)
Abstract:We propose a novel algorithm for incremental Byte Pair Encoding (BPE) tokenization. The algorithm processes each input byte in worst-case \mathcalO(\log^2 t) time, leading to an overall complexity of \mathcalO(n \log^2 t) , where n is the input length and t is the maximum token length. The algorithm incrementally maintains BPE tokenization results for every prefix of the input text, implementing the standard BPE merge procedure defined by a fixed set of merge rules. This enables efficient partial tokenization in streaming settings. Functioning as a drop-in replacement for standard BPE, our approach achieves a speedup of up to \sim3\times over Hugging Face’s tokenizers, and demonstrates significant latency reductions over OpenAI’s tiktoken on pathological inputs. We further introduce an eager output algorithm that enables streaming output, emitting tokens as soon as token boundaries are determined during incremental tokenization. Overall, our results demonstrate that BPE tokenization can be performed incrementally with strong worst-case guarantees, while providing practical latency benefits in modern large language model pipelines. Code: this https URL
[NLP-88] Anchoring LLM Gender Bias to Human Baselines: A Cross-Lingual Audit
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在多语言环境中存在的性别刻板印象问题，尤其关注其在英语与东亚语言（中文、日语、韩语）使用场景下的表现差异。研究核心在于超越简单判断模型是否偏见，转而量化分析模型在性别属性赋值上与目标部署人群之间的偏离程度。其解决方案的关键在于引入跨文化人类数据集（覆盖48个国家）作为基准，结合HEXACO-100人格量表，构建一个四模式行为框架——一致性（concordance）、抑制（suppression）、重组（reorganization）与放大（amplification），系统刻画不同模型-语言组合下的性别刻板映射模式。研究发现，模型的性别刻板程度范围约为人类跨国家差异的2.5倍，且在多语言交互中存在叠加效应；更关键的是，翻译过程不仅重标定刻板印象强度，更实质性地改变关联属性，导致表面看似校准良好实则深层结构重构。这一结果表明，单一去偏策略难以有效覆盖跨语言边界中的复杂偏差形态。
链接: https://arxiv.org/abs/2605.30804
作者: Jiwoo Choi,Seonwoo Ahn,Tongxin Zhang,Seohyon Jung
机构: KAIST(韩国科学技术院)
类目: Computation and Language (cs.CL)
备注:
Abstract:We audit six large language models (LLMs) for gender stereotyping across English, Korean, Chinese, and Japanese. Three were developed primarily for English-language use (Claude, GPT, Gemini) and three for East Asian use (DeepSeek, Syn-Pro, HyperCLOVA X). We adopt the HEXACO-100 personality inventory and anchor each model against a cross-cultural human dataset spanning 48 countries to ask not whether LLMs are biased, but how far their gender attributions drift from the populations they are deployed among. Our findings show that their stereotyping spans a range roughly 2.5 times wider than the entire cross-country range found in humans, and the effect can compound across languages. One English-centric model, prompted in Korean, reached 5 times the local baseline, even when the prompt stated the candidate had already been hired, which often dampens human stereotyping. To characterize such behaviors without ranking them, we introduce a four-pattern framework – concordance, suppression, reorganization, and amplification – across 24 (model x language) cells. Item-level analysis reveals that translation does not just rescale stereotypes, but changes the attributes tied to it, hiding significant rearrangement under the surface while appearing well-calibrated. Our results ultimately suggest that no single debiasing pipeline is likely to address bias evenly across linguistic boundaries.
[NLP-89] XLGoBench: Detecting cross-lingual skill gaps with algorithmic tasks
【速读】： 该论文旨在解决大语言模型在跨语言能力上存在的潜在差距问题，即模型在不同语言任务中的表现不一致，尤其在算法类任务中是否存在语言特异性偏差。其解决方案的关键在于构建一个合成的、可扩展的跨语言算法基准测试（cross-lingual algorithmic benchmark），该基准具备四大特性：可比性（所有语言执行相同底层任务）、可扩展性（可通过调整复杂度适配不同能力的模型）、可量化性（每个任务均有明确的正确性标准）和透明性（基于简单模板生成，便于审计翻译错误）。由于基准聚焦于算法任务，模型在不同语言间的表现差异可作为跨语言能力差距的充分但非必要指标。通过大量实验验证，该基准成功揭示了多个先进大语言模型中存在的持续性跨语言性能差距。
链接: https://arxiv.org/abs/2605.30788
作者: Purvam Jain,Preethi Jyothi,Vihari Piratla,Suvrat Raju
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 8+37pages
Abstract:We introduce a set of synthetic algorithmic tasks to detect cross-lingual gaps in the abilities of large language models. Our benchmark is commensurate across languages, since it requires models to perform the same underlying task in different languages; scalable, since each task can be generated at varying levels of complexity allowing it to be adapted to models with different capabilities; quantifiable, since every task admits an objective notion of correctness; and transparent, since tasks are generated from simple templates that can be readily audited for translation errors. Because our benchmark focuses on algorithmic tasks, differential performance is a sufficient – but not necessary – indicator of cross-lingual gaps. Nevertheless, we show through extensive experiments that our benchmark exposes persistent cross-lingual gaps in multiple state-of-the-art models.
[NLP-90] Eywa: Provenance-Grounded Long-Term Memory for AI Agents
【速读】： 该论文旨在解决持续性人工智能代理（AI agent）在跨会话场景中面临的记忆系统不可靠、可追溯性差的问题。现有记忆系统常将原始证据、提取的事实、检索的上下文与答案策略高度耦合于单一模糊的提示路径中，导致错误难以定位——错误可能源于证据缺失、提取不当、状态过期、检索失败或模型行为偏差。为应对这一挑战，论文提出Eywa，一种以“证据优先于信念”（evidence before belief）为核心的溯源根基型记忆架构。其关键在于：首先将源证据以不可变形式存储，确保源头可追溯；其次通过类型化信号与源支持验证机制对提取的记忆进行校验；再者采用确定性的多路径读取机制，在检索过程中不依赖大语言模型（LLM）调用，实现上下文的边界化、可复现式召回；最后将检索到的上下文与答案指令分离输出，使同一记忆底座可兼容不同前沿模型、预算受限模型及本地回答模型的评估。在冻结配置下，Eywa在LoCoMo C1-C4任务上达到90.19%的判断准确率（使用Claude Sonnet 4.6），在LongMemEval-S上实现88.2%的检索充分性准确率，在包含700个技术记忆测试题的BEAM基准上取得81.45%的平均信息点得分和85.29%的pass@0.5准确率。所有问题级产物（包括问题、真值答案、模型输出、检索上下文及标签）均已公开。
链接: https://arxiv.org/abs/2605.30771
作者: Resham Joshi
机构: Eywa(艾瓦)
类目: Computation and Language (cs.CL)
备注: 29 pages, 3 figures, 16 tables. Benchmark artifacts available at this https URL
Abstract:AI agents that persist across sessions need memory they can retrieve, audit, update, and erase. Existing memory systems often collapse source evidence, extracted facts, retrieved context, and answer policy into one opaque prompt path, making failures difficult to diagnose: a wrong answer may come from missing evidence, unsupported extraction, stale state, retrieval loss, or answer-model behavior. We present Eywa, a provenance-grounded memory architecture built around evidence before belief. Eywa stores immutable source evidence before deriving canonical facts, validates extracted memories against typed signals and source support, and retrieves bounded memory context through a deterministic multi-route read path with zero LLM calls inside retrieval. Retrieved context is returned separately from answer instructions, allowing the same memory substrate to be evaluated across frontier, budget, and local answer models. Under a frozen, artifact-recorded retrieval configuration, Eywa reaches 90.19% judge accuracy on the LoCoMo C1-C4 split with Claude Sonnet 4.6 write and QA roles. On LongMemEval-S, it reaches 88.2% retrieval-sufficiency accuracy. On BEAM, a 700-question technical-memory stress benchmark, it reaches 81.45% mean nugget score and 85.29% pass@score = 0.5. Full per-question artifacts, including questions, gold answers, model answers, retrieved context, and labels, are published at this https URL.
[NLP-91] Pairwise Reference Alignment as a Model-Level Ordinal Observable
【速读】： 该论文旨在解决在语言模型评估与对齐中，基于成对偏好数据（pairwise preference data）进行模型性能判断时，所隐含的统计基础不明确的问题。具体而言，当通过检验模型是否将偏好响应排在拒绝响应之上时，其所估计的模型层面量度究竟是什么？其核心解决方案在于提出“成对参考对齐度”（pairwise reference alignment）这一序数可观测量（ordinal observable），即定义为模型评分函数诱导的排序与参考偏好排序一致的概率。该量度不依赖于具体的评分形式（如归一化对数概率或基于能量的评分），从而剥离了评分机制对结果的影响，澄清了参考成对分布（reference pair distribution）在评估中的关键作用。研究进一步引入类序参数的中心化统计量及基于边距的扩展形式，给出了在独立采样假设下的简单有限样本估计器和浓度界。实验部分以Qwen2.5系列模型和RewardBench数据集为基础，验证了所提统计量随模型规模和指令微调程度增加，并在不同参考对子集间呈现预期变化，证实了该框架的可解释性与有效性。
链接: https://arxiv.org/abs/2605.30758
作者: Mujing Li
机构: 上海科技大学(ShanghaiTech University)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注:
Abstract:Pairwise preference data is widely used in language-model evaluation and alignment, often for model ranking, reward modeling, or preference optimization. This note formulates a more basic measurement question: given a reference distribution of pairwise preferences, what model-level quantity is estimated when we test whether a model ranks preferred responses above rejected responses? We define pairwise reference alignment as an ordinal observable induced by a model scoring function. Given a reference pair distribution P_\mathrmpair over triples (x,y^+,y^-) , and a scalar model score S_M(x,y) , we define the alignment observable as the probability that the model-induced ordering agrees with the reference preference ordering. We further define a centered order-parameter-like statistic and discuss a margin-based extension. The resulting quantities admit simple finite-sample estimators and concentration bounds under independent sampling assumptions. This note does not introduce a new benchmark. It provides a conceptual and statistical formulation for pairwise reference alignment, clarifies the role of the reference pair distribution, and distinguishes the general ordinal observable from scoring choices such as normalized log-probability or energy-based scores. We also provide an initial empirical study on Qwen2.5 models and RewardBench, where the proposed statistics increase with model size and instruction tuning and vary across reference-pair subsets as predicted by the formulation. Subjects: Computation and Language (cs.CL); Machine Learning (cs.LG) Cite as: arXiv:2605.30758 [cs.CL] (or arXiv:2605.30758v1 [cs.CL] for this version) https://doi.org/10.48550/arXiv.2605.30758 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[NLP-92] Efficient Diffusion LLM s via Temporal-Spatial Parallel Decoding and Confidence Extrapolation
【速读】： 该论文旨在解决基于扩散模型的大语言模型（dLLMs）在推理过程中因冗余精炼与重复掩码操作导致的高延迟问题。现有加速方法依赖于局部步长置信度启发式或固定调度策略，对提示和任务变化敏感，且忽略了序列内部显著的位置效应。本文将扩散解码建模为一个动态控制问题，揭示了逐标记去噪轨迹是实现可靠控制的关键信号。为此，提出一种感知轨迹的解码框架，包含两个核心组件：首先，时间-空间并行解码（Temporal-Spatial Parallel Decoding, TSPD）通过轻量级时空控制器，结合每个标记的轨迹特征（如置信度、熵、动量）及位置信息，判断其是否已收敛并可安全固定；其次，提出无需训练的状态空间模块——置信度外推（Confidence Extrapolation, CE），能够预测未来logit趋势及其不确定性，支持前瞻决策，包括安全前瞻和在轨迹振荡或置信度不足时的目标性稳定。TSPD与CE协同作用，在不牺牲生成质量的前提下有效减少不必要的去噪迭代，且与KV缓存等系统优化兼容良好。
链接: https://arxiv.org/abs/2605.30753
作者: Zekai Li,Ji Liu,Yiqing Huang,Ziqiong Liu,Dong Li,Emad Barsoum
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:Diffusion-based large language models (dLLMs) support parallel text generation via iterative denoising, yet inference remains latency-heavy because many steps are spent on redundant refinement and repeated remasking of tokens whose final values are already determined. Prior acceleration methods mainly depend on step-local confidence heuristics or fixed schedules, which are sensitive to prompt and task variation and ignore strong positional effects within a sequence. We cast diffusion decoding as a dynamic control problem and show that token-wise denoising trajectories provide the key signal for reliable control. We propose a trace-aware decoding framework with two components. First, Temporal-Spatial Parallel Decoding (TSPD) uses a lightweight temporalspatial controller that consumes per-token trajectory features, including confidence, entropy, and momentum, together with token position, to decide when a token has converged and can be safely fixed. Second, we introduce Confidence Extrapolation (CE), a training-free state-space module that forecasts future logit trends with uncertainty to support proactive decisions, including safe look-ahead and targeted stabilization when trajectories are oscillatory or underconfident. Together, TSPD and CE reduce unnecessary denoising iterations while preserving output quality, and they compose cleanly with system optimizations such as KV caching.
[NLP-93] OrcaRouter: A Production-Oriented LLM Router with Hybrid Offline-Online Learning
【速读】： 该论文旨在解决大规模语言模型（Large Language Models, LLMs）在实际部署中面临的动态选型问题：当接收到一个用户请求时，如何从多个具备不同能力与推理成本的模型中选择最优模型以实现性能与成本的平衡。其解决方案的关键在于提出OrcaRouter——一种面向生产环境的LLM路由系统，该系统结合基于LinUCB的上下文感知强化学习（contextual bandit）算法，利用词法特征与句子嵌入（sentence-embedding）作为上下文输入，并采用混合式离线-在线学习机制。在离线阶段，通过在精心构建的路由提示集上评估所有候选模型，获得全信息反馈并训练每个模型臂（arm）对应的岭回归（ridge regressor）模型；在在线部署阶段，系统基于预训练参数初始化，并可选择性地持续根据带宽反馈更新被选中模型的臂参数，从而实现高效、自适应的模型调度。该方法在RouterArena基准测试中表现优异，于2026年5月20日提交时以72.08的竞技场得分位列第二，且在每千次查询1美元的成本下实现了75.54%的准确率。
链接: https://arxiv.org/abs/2605.30736
作者: Zhenghua Bao,Fengya Tian,Chris Zhang,Zhenjun Chen,Xile Ma,Yi Shi
机构: Continuum AI
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 6 pages, 1 table. Technical report
Abstract:The rapid development of large language models, each with distinct capabilities and inference costs, raises a practical deployment question: given an incoming request, which model should handle it? We present OrcaRouter, a production-oriented LLM router that combines a LinUCB-based contextual bandit over lexical and sentence-embedding features with a hybrid offline-online learning protocol. Offline, OrcaRouter obtains full-information feedback by evaluating each candidate model on a curated set of routing prompts, yielding a reward matrix used to fit one ridge regressor per arm. At deployment time, it initializes from these parameters and can optionally continue learning from bandit feedback, updating only the selected model’s arm after observing its reward. At the time of our RouterArena submission (May 20, 2026), OrcaRouter-Adaptive ranked second on the public RouterArena leaderboard with an arena score of 72.08, achieving 75.54% accuracy at a cost of USD 1.00 per 1,000 queries.
[NLP-94] MosaicLeaks:Privacy Risks in Querying-in-the-Open for Deep Research Agents
【速读】： 该论文旨在解决深度研究代理（Deep Research Agent）在融合本地私有文档与外部工具（如网络检索）时所面临的隐私泄露问题。其核心挑战在于，尽管单个外部查询看似无害，但通过“拼图效应”（mosaic effect），多个查询的聚合可能暴露敏感的本地信息，从而引发严重的隐私风险。为系统评估这一问题，作者提出了MosaicLeaks基准，包含1,001个多跳研究任务，强制代理依赖本地企业文档生成对外查询。通过部署一个仅观察外部查询的对抗性大语言模型（LLM），研究人员在三个层面评估了隐私泄露：研究意图、特定私有问题的答案以及对企业文档的可验证陈述。实验发现，不同模型家族和规模均存在显著泄露，零样本隐私提示虽能缓解但无法根除泄露，且单纯以任务性能为目标的强化学习反而加剧泄露。为此，论文提出隐私感知的深度研究框架（Privacy-Aware Deep Research, PA-DR），该框架结合任务完成的环境奖励与可学习的隐私分类器，实现对单次查询及拼图级泄露的细粒度信用分配。在Qwen3-4B-Instruct上应用PA-DR后，准确率从48.7%提升至58.7%，同时答案泄露和全信息泄露分别由34.0%降至9.9%，显著提升了隐私保护能力。
链接: https://arxiv.org/abs/2605.30727
作者: Alexander Gurung,Spandana Gella,Alexandre Drouin,Issam H. Laradji,Perouz Taslakian,Rafael Pardinas
机构: University of Edinburgh (爱丁堡大学); ServiceNow (服务现在)
类目: Computation and Language (cs.CL)
备注:
Abstract:Deep research agents increasingly combine private local documents with external tools like web retrieval, creating a privacy risk: an agent’s external queries may leak sensitive information from its local context. This risk is amplified by the mosaic effect, where individual queries may appear harmless but become revealing in aggregate. We introduce MosaicLeaks, a benchmark of 1,001 multi-hop deep research tasks that chain private enterprise documents and a public web corpus, forcing agents to make external queries that depend on local information. We evaluate leakage with an adversary LLM that observes only the agent’s external queries and attempts to infer private information at three levels: the agent’s research intent, answers to specific private questions and verifiable claims about the enterprise documents. We find that models across families and sizes frequently leak at all three levels, that zero-shot privacy prompting reduces but does not eliminate leakage and that reinforcement learning for task performance alone worsens leakage. To address this, we propose Privacy-Aware Deep Research (PA-DR), an RL framework that combines situational rewards for task success with a learned privacy classifier to provide dense credit assignment over both per-query and mosaic-level leakage. Training Qwen3-4B-Instruct with PA-DR improves accuracy from 48.7% to 58.7% and reduces answer and full-information leakage from 34.0% to 9.9%.
[NLP-95] Skill is Not One-Size-Fits-All: Model-Aware Skill Alignment for LLM Agents
【速读】： 该论文旨在解决大语言模型（Large Language Model, LLM）智能体在执行长时程交互任务时，因外部技能库（skill library）的通用性设计导致技能适配性不足的问题。现有方法通常将技能库视为与模型无关（model-agnostic），在不同能力与行为特征的模型骨干（backbone）间复用相同的技能表述，但实验表明技能的有效性具有显著的模型依赖性：某些技能对特定模型有益，却可能损害其他模型的表现。为此，本文提出MASA（Model-Aware Skill Alignment）框架，其核心创新在于无需修改智能体权重即可实现技能对目标模型骨干的自适应调整。MASA采用两阶段机制：第一阶段为分层技能演化流程，通过梯度上升与基于置信区间上界（UCB）驱动的树搜索，结合环境反馈与模型能力画像，迭代优化通用技能与任务特定技能；第二阶段为轻量级、模型条件化的技能重写器（model-conditioned skill rewriter），基于演化轨迹训练，可在单次前向传播中高效复现适配过程。在三个交互环境与四种模型骨干上的实验验证表明，MASA在所有场景下均取得最优性能，相较于最强基线最高提升25.8分，且所学重写器具备良好的跨任务与跨环境泛化能力，以远低于教师大模型的推理开销持续超越其表现。
链接: https://arxiv.org/abs/2605.30723
作者: Jianxiang Yu,Jiapeng Zhu,Bochen Lin,Qier Cui,Zichen Ding,Xiang Li
机构: East China Normal University (华东师范大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:LLM agents increasingly retrieve externally curated skills-procedural instructions retrieved at decision time-to improve performance on long-horizon interactive tasks. Existing skill libraries are typically treated as model-agnostic, reusing the same skill formulations across backbones with substantially different capacities and behaviors. However, our controlled experiments across multiple model scales show that skill effectiveness is strongly model-dependent: a skill that benefits one backbone can harm another. Motivated by this observation, we propose MASA Model-Aware Skill Alignment, a framework that adapts skills to each target backbone without modifying agent weights. MASA operates in two stages: (1) a hierarchical skill evolution pipeline that iteratively rewrites general and task-specific skills using hill climbing and UCB-driven tree search, guided by environment feedback and model capability profiles; and (2) a lightweight model-conditioned skill rewriter trained on evolution trajectories to reproduce the adaptation in a single forward pass. Experiments across three interactive environments and four backbones show that MASA consistently achieves the best overall performance, with gains of up to 25.8 points over the strongest baseline. The learned rewriter further generalizes to unseen tasks and environments without additional search, consistently outperforming a much larger teacher LLM at a fraction of the inference cost.
[NLP-96] Neuron-Level Interventions for Gendered and Gender-Neutral Generation in Language Models
【速读】： 该论文旨在解决生成式语言模型（Generative Language Models, LMs）中性别偏见的内在编码机制问题，尤其是针对传统研究多聚焦于二元性别（女性化与男性化）而忽视性别中立形式（如“they/them”代词或中性职业称谓）的局限性。其核心挑战在于：如何理解并干预模型内部表示中与性别相关的信息编码方式。解决方案的关键在于提出一种基于神经元层面的干预方法，通过识别与特定性别类别（女性化、男性化、性别中立）强关联的神经元，并在生成过程中选择性激活或屏蔽这些神经元，从而实现对输出文本性别的精准控制，同时保持语义不变。实验表明，性别相关神经元在模型早期层中高度集中，且该方法相比现有技术能更精确地实现目标性别导向，减少非目标性别的信息泄露，并在两项评估指标下保持稳定的生成质量。该研究不仅揭示了性别在模型内部表示中的分布特性，也为性别偏见缓解和神经元级干预评估提供了可解释、可操作的有效路径。
链接: https://arxiv.org/abs/2605.30717
作者: Zhiwen You,Nafiseh Nikeghbal,Jana Diesner
机构: University of Illinois Urbana-Champaign (伊利诺伊大学厄本那-香槟分校); Technical University of Munich (慕尼黑工业大学); Munich Center for Machine Learning (慕尼黑机器学习中心)
类目: Computation and Language (cs.CL)
备注:
Abstract:Language models (LMs) can produce gendered language and stereotypes even when given neutral prompts. Most prior work on gender bias in LMs primarily examines gender through a binary lens (feminine vs. masculine), with limited attention to gender-neutral forms, such as they/them pronouns or neutrally phrased job titles. How gender-related signals are encoded in the internal representations of LMs remains an open question. In this work, we study gender-specific neurons in LMs across three categories: feminine, masculine, and gender-neutral. We propose a neuron-level intervention method to identify neurons that are strongly tied to each gender category. We then test these neurons through controlled generation, showing that activating or masking gender-related neurons can steer a sentence toward a target gender form while preserving its original meaning. To evaluate the effectiveness of our gender-intervention approach, we curate two datasets with controlled sentences labeled across all three gender categories and validate the data quality through human evaluation. Experiments on two open-source LMs show that gender-specific neurons are not evenly distributed across model layers; instead, they concentrate heavily in the earliest layers with smaller contributions from later layers. Compared to existing methods, our method achieves more precise gender control, with less leakage into non-target gender categories and stable output quality through two evaluation criteria. Overall, our work examines how gender is encoded in LMs and provides a simple yet effective approach toward controlled gender intervention for both neuron intervention evaluation and gender bias mitigation. Code and datasets are available at: this https URL
[NLP-97] ExpGraph: Model-Agnostic Experience Learning with Graph-Structured Memory for LLM Agents
【速读】： 该论文旨在解决大语言模型（Large Language Model, LLM）代理在执行任务时缺乏经验复用能力的问题，即模型通常从零开始解决问题，难以有效利用过往成功策略或失败教训。现有方法通过微调积累的经验虽可提升复用性，但面对更优或更适配的执行器（executor）时缺乏灵活性。为此，本文提出ExpGraph——一种模型无关的经验学习框架，其核心在于使冻结且可替换的LLM执行器能够通过外部经验复用实现性能提升，而无需进行参数更新。其关键创新在于构建一个自演化经验图（experience graph），将历史轨迹抽象为可重用的技能与失败教训，并以节点形式组织；通过图扩散与效用感知排序机制实现高效经验检索；同时引入轻量级检索协作者（retrieval copilot），基于强化学习训练，利用有无检索经验下的执行器表现差异作为反馈信号；经验图则根据下游任务结果在线动态更新。实验表明，ExpGraph在涵盖问答、数学推理、代码生成及多步代理环境（如ALFWorld和AppWorld）的ExpSuite基准上，相较于最强基线分别提升了12.2%和4.7%（静态任务）、21.4%和12.7%（代理环境），并减少平均交互步数12.7%和21.6%。消融实验验证了图结构化经验、效用感知排序与自适应检索三者协同对跨任务、跨执行器模型的有效经验复用至关重要。
链接: https://arxiv.org/abs/2605.30712
作者: Tao Feng,Chongrui Ye,Tianyang Luo,Jingjun Xu,Xueqiang Xu,Haozhen Zhang,Zhigang Hua,Yan Xie,Shuang Yang,Ge Liu,Jiaxuan You
机构: University of Illinois Urbana-Champaign (伊利诺伊大学厄本那-香槟分校); Nanyang Technological University (南洋理工大学); Meta Monetization AI (Meta 个性化广告人工智能)
类目: Computation and Language (cs.CL)
备注:
Abstract:Large language model (LLM) agents have shown strong capabilities in reasoning, tool use, and multi-step interaction, but they often solve tasks from scratch and fail to reuse successful strategies or failure lessons from prior experience. Fine-tuning on collected experience can improve reuse, but it is inflexible when stronger or more suitable executors emerge. We propose ExpGraph, a model-agnostic experience learning framework that enables frozen and replaceable LLM executors to improve through external experience reuse without parameter updates. ExpGraph summarizes historical trajectories into reusable skills and failure lessons, organizes them as nodes in a self-evolving experience graph, and retrieves useful experiences through graph diffusion and utility-aware ranking. A lightweight retrieval copilot is trained with reinforcement learning using feedback that compares executor performance with and without retrieved experiences, while the graph is updated online from downstream task outcomes. We evaluate ExpGraph on ExpSuite, covering question answering, mathematical reasoning, code generation, and multi-step agentic environments including ALFWorld and AppWorld. ExpGraph improves over the strongest baseline by 12.2% and 4.7% on static tasks with smaller and larger executors, and by 21.4% and 12.7% in agentic environments, while reducing average interaction steps by 12.7% and 21.6%. Ablations show that graph-structured experience, utility-aware ranking, and adaptive retrieval jointly enable effective experience reuse across diverse tasks and executor models.
[NLP-98] SAGE: A Novelty Gate for Efficient Memory Evolution in Agent ic LLM s
【速读】： 该论文旨在解决生成式智能体（Agentic LLMs）在长期记忆管理中面临的写入控制难题，即如何在不断提取新事实时，合理判断其是否应被添加、合并至已有记忆或直接忽略。现有方法多关注记忆的检索与存储，而忽视了对写入行为的系统性控制。为此，本文提出SAGE（Spherical Adaptive Gate for memory Evolution），将记忆演化建模为新颖性检测（novelty-detection）问题，通过基于冯·米塞斯-费舍尔（von Mises-Fisher）分布的密度估计器对记忆嵌入进行评分，并采用自适应阈值机制动态跟踪记忆存储的几何结构，实现对候选事实的精准路由：明确新颖的事实标记为ADD，明显冗余的事实标记为NOOP，仅将不确定性较高的情况交由大语言模型（LLM）进行合并处理，从而显著降低昂贵的写入阶段推理开销。实验结果表明，在LoCoMo基准上，SAGE在所有七种开源权重主干模型上的平均token-F1均优于Mem0；在GPT-4o-mini上，其添加阶段的API成本降低3.4倍、延迟降低2.5倍，且平均判别得分差距极小。作为A-Mem的即插即用二元门控组件，SAGE在五种模型中约跳过16%-18%的LLM调用，同时在开源主干模型上保持几乎不变的内存质量。研究证明，具备新颖性感知能力的写入控制是提升长期智能体记忆质量与系统效率的关键技术杠杆。
链接: https://arxiv.org/abs/2605.30711
作者: Sijia Wang,Dhanajit Brahma,Ricardo Henao
机构: Duke University (杜克大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Machine Learning (stat.ML)
备注:
Abstract:Agentic LLMs must continuously decide whether newly extracted facts should be added, merged with existing memories, or ignored, yet prior work has focused more on retrieval and storage than on principled write-side control. We frame memory evolution as a novelty-detection problem and propose SAGE, a Spherical Adaptive Gate for memory Evolution that scores candidate facts with a von Mises-Fisher-based density estimator over memory embeddings and routes them with an adaptive threshold that tracks memory-store geometry. SAGE resolves clearly novel facts as ADD, clearly redundant facts as NOOP, and sends only uncertain cases to an LLM merge step, reducing expensive write-time reasoning. On LoCoMo, SAGE achieves the best average token-F1 against Mem0 on all seven open-weight backbone comparisons, while on GPT-4o-mini it reduces add-phase API cost by 3.4 \times and add-phase latency by 2.5 \times with only a small average judge-score gap. As a drop-in binary gate for A-Mem, SAGE skips roughly 16-18% of LLM calls across five models with minimal quality change on open-weight backbones. These results suggest that novelty-aware write control is a practical lever for improving both memory quality and system efficiency in long-term agentic memory.
[NLP-99] riaging Threats to Specialized Guardrails
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在实际应用中部署时面临的安全部署难题，核心挑战在于安全风险涵盖异构的威胁领域，而现有数据集仅覆盖部分风险类别且分类体系不统一，导致当前安全防护机制（guardrail）的泛化能力存疑。为系统评估安全防护模型的鲁棒性，研究提出GuardZoo——一个统一的人工标注基准数据集，包含32,460个样本，覆盖15种不同的非安全类别。实验发现，传统的单一模型式防护机制存在任务干扰问题，不同威胁域所需的决策边界难以压缩至单一模型中。为此，论文提出RouteGuard，一种基于路由器-专家（router-expert）架构的框架，通过路由机制将对话分配至针对特定威胁类型的专用专家防护模块，实现精准的威胁检测。结果表明，RouteGuard在细粒度威胁识别、跨域泛化能力以及对新兴威胁的模块化扩展方面均显著优于现有强基线模型，其关键创新在于通过解耦不同威胁域的检测任务，实现了更高效、可扩展的安全防护。
链接: https://arxiv.org/abs/2605.30693
作者: Wenjie Jacky Mo,Xiaofei Wen,Rui Cai,Boyu Zhu,Sicong Jiang,Zihan Wang,Minglai Yang,Zhe Zhao,Muhao Chen
机构: University of California, Davis (加州大学戴维斯分校); 2077.AI
类目: Cryptography and Security (cs.CR); Computation and Language (cs.CL)
备注:
Abstract:Building robust safety guardrails is essential for deploying Large Language Models across diverse real-world applications. However, this goal remains challenging because safety risks span heterogeneous threat domains, while existing datasets cover only fragmented risk subsets and rely on inconsistent taxonomies. Consequently, it remains unclear whether current guardrails can generalize beyond narrow evaluation settings. To better understand the robustness of guardrail models, we first introduce GuardZoo, a unified human-annotated benchmark with 32,460 samples covering 15 distinct unsafe categories. Evaluation on GuardZoo reveals that monolithic guardrails suffer from task interference: different threat domains require distinct decision boundaries that are difficult to compress into a single model. We therefore propose RouteGuard, a router-expert framework that triages each conversation to specialized expert guardrails for threat-specific detection. Experiments show that RouteGuard improves fine-grained threat detection over strong guardrail baselines, generalizes better under out-of-domain evaluation, and supports flexible modular expansion to emerging threats.
[NLP-100] ElasticMem: Latent Memory as a Learnable Resource for LLM Agents
【速读】： 该论文旨在解决大语言模型（LLM）代理在长期交互中因记忆机制僵化而导致的推理连贯性差、个性化能力弱以及过往经验复用效率低的问题。现有记忆增强方法存在显著局限：文本空间方法将检索到的记忆拼接至上下文窗口，导致令牌开销大且对噪声信息敏感；潜在空间方法虽降低文本成本，但仍依赖固定的检索策略或容量受限的记忆接口，造成查询相关记忆效用与固定记忆分配之间的不匹配。为此，本文提出ElasticMem，一种可学习的弹性潜在资源记忆框架，其核心创新在于将记忆视为动态可调的潜在资源。ElasticMem通过离线构建包含检索键与内容缓存的潜在记忆库，基于推理器隐藏状态自适应地检索记忆，并由学习得到的策略为每条检索记忆分配可变的潜在预算，最终将选定的潜在状态以软记忆标记的形式注入生成过程。整个记忆使用流程通过下游任务奖励，采用组相对策略优化进行端到端训练。实验表明，在MemorySuite基准上，基于Qwen2.5-3B-Instruct和Qwen2.5-7B-Instruct的ElasticMem分别在加权平均问答准确率上提升26.2%和24.6%，在ALFWorld任务中成功率分别提高66.3%和27.2%，同时实现最低的令牌消耗。消融研究与定性分析进一步验证了自适应检索与弹性预算分配机制能够有效优先选择有用证据并迁移可转移的规划策略，突破传统基于余弦相似度的刚性匹配限制。
链接: https://arxiv.org/abs/2605.30690
作者: Tao Feng,Chongrui Ye,Tianyang Luo,Jingjun Xu,Xueqiang Xu,Haozhen Zhang,Ge Liu,Jiaxuan You
机构: University of Illinois Urbana-Champaign (伊利诺伊大学厄本那-香槟分校); Nanyang Technological University (南洋理工大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:Long-term memory is essential for LLM agents to reason coherently across extended interactions, personalize responses, and reuse past experience. However, existing memory-augmented methods typically treat memory as a fixed resource: text-space approaches concatenate retrieved memories into the context window, causing substantial token overhead and sensitivity to noisy evidence, while latent-space approaches reduce textual cost but still rely on rigid retrieval or fixed-capacity memory interfaces. This creates a mismatch between query-dependent memory utility and fixed memory allocation. We propose ElasticMem, a memory-augmented LLM framework that learns to use memory as an elastic latent resource. ElasticMem builds an offline latent memory bank with retrieval keys and content caches, retrieves memories adaptively from the reasoner’s hidden state, assigns each retrieved memory a variable latent budget through a learned policy, and injects selected latent states as soft memory tokens for generation. The full memory-use process is optimized with downstream task rewards through group-relative policy optimization. We evaluate ElasticMem on MemorySuite, covering memory-intensive QA and embodied agent control. Across Qwen2.5-3B-Instruct and Qwen2.5-7B-Instruct backbones, ElasticMem improves weighted average QA accuracy by 26.2% and 24.6%, and improves ALFWorld success rate by 66.3% and 27.2%, respectively, over the strongest baselines, while achieving the lowest ALFWorld token cost. Ablations and qualitative analyses further show that adaptive retrieval and elastic budget allocation help ElasticMem prioritize useful evidence and transferable plans beyond rigid cosine similarity. Our code for ElasticMem will be released at this https URL.
[NLP-101] Human-Alignment Calibration and Activation Patterns in Large Language Model Uncertainty
【速读】： 该论文旨在解决大语言模型（Large Language Model, LLM）中的不确定性量化（Uncertainty Quantification）与人类不确定性之间的一致性问题，即探究大语言模型所表现出的不确定性是否具有与人类相似的特征。其核心挑战在于，尽管现有研究主要聚焦于提升模型校准能力（calibration），以增强对自身预测置信度的准确性，但较少关注模型不确定性是否在行为模式和内部激活特征上呈现出与人类认知相一致的信号——即“不确定性对齐”（uncertainty alignment）。本文的关键解决方案在于通过多类数据集（涵盖选择题与开放式事实回忆任务）系统评估大语言模型在显式行为和内部激活模式中是否存在人类相似的不确定性信号，并进一步分析指令微调（instruct fine-tuning）对不确定性对齐与校准能力的影响，从而揭示模型在模仿人类不确定性判断方面的潜在机制与局限性。
链接: https://arxiv.org/abs/2605.30675
作者: Kyle Moore,Jesse Roberts,Daryl Watson,William Ward,Grayson Heyboer
机构: Vanderbilt University (范德比尔特大学); Tennessee Technological University (田纳西理工大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注:
Abstract:Uncertainty Quantification is a large and growing subfield of large language model behavioral analysis. Primarily to recognize and combat hallucination, the field has largely focused on measuring and improving calibration, the accuracy of uncertainty judgments to task efficacy. In this work, we investigate the relatively underexplored question of how similar large language model uncertainty is to human uncertainty. We investigate the presence and strength of human-similar uncertainty signals, deemed uncertainty alignment, in large language model overt behavior and internal activation patterns. We identify whether the models show evidence of simultaneous alignment and calibration on a variety of datasets covering both multiple choice and open ended factual recall. And we characterize the effect of instruct fine-tuning on each of these facets.
[NLP-102] achObs: A Human-Validated Benchmark for Multimodal Teaching Observation and Model Evaluation
【速读】： 该论文旨在解决课堂视频中教学实践的可观测信号（如教学行为、视觉线索）缺乏结构化标注以供模型评估的问题。现有研究虽可从课堂视频中提取教学信息，但其标注体系多未经过严格验证，难以支持对生成式AI在教育场景下表现的可靠评估。为此，本文提出一个名为TeachObs的人工验证基准，涵盖来自八个国家的30节公开课程视频，将其划分为5,158个固定15秒片段，并由七名研究人员对每个片段进行39个二元观察编码（包括20个视觉类编码，如手势、板书、指物动作和视觉材料使用；19个非视觉类编码，如讲解、监控、提问、反馈与反思）。通过基于克里彭多夫阿尔法（Krippendorff’s alpha）的信度与频次感知规则构建金标准标签。此外，三位专家还对整节课在教学设计、教学实施、学习者反应、学习材料及课程收尾等方面进行了整体评分与质性评价。基于这一双层人工参考体系，论文在三个评测轨道上评估了五种具备视觉理解能力的前沿大语言模型（LLM）：仅文本输入的片段编码、文本+帧图像输入的片段编码，以及基于“大模型作为裁判”（LLM-as-judge）协议的整节课覆盖度评估。结果表明：无单一模型在所有任务中持续领先；引入中间帧会同时增加真实与虚假归因；模型评估普遍高估程序性清晰的课程，而低估复杂或非结构化课程。因此，TeachObs不仅支持细粒度标注基准测试，也实现了整节课层面的综合评估，揭示了当前生成式AI在课堂视频分析中的优势与局限，明确了在不同学科、课堂形式及标注难度条件下，自动化分析与专家判断各自的适用边界。
链接: https://arxiv.org/abs/2605.30673
作者: Yeil Jeong,Youngjin Yoo,Seobin Sohn,Hyejin Han,Jinseo Lee,Scott Howard,Unggi Lee
机构: Indiana University Bloomington(印第安纳大学布卢明顿分校); Pai Chai University(柏蔡大学); Seoul National University(首尔国立大学); Ewha Womans University(延世女子大学); University of Wolverhampton(伍尔弗汉普顿大学); Korea University Sejong Campus(韩国国民大学世宗校区)
类目: Computation and Language (cs.CL)
备注:
Abstract:Classroom videos contain observable teaching practices, but their pedagogical and visual signals are rarely organized in forms suitable for model evaluation. We present \textitTeachObs, a human-validated benchmark for multimodal teaching observation in classroom videos. \textitTeachObs includes 30 public lesson videos from eight countries divided into 5,158 fixed 15-second scenes. Seven researchers annotated each scene with 39 binary observation codes, covering 20 visual codes, such as gesture, board work, pointing, and visual materials, and 19 nonvisual codes, such as instruction, monitoring, questioning, feedback, and reflection. Gold segment labels are constructed using reliability- and prevalence-aware rules based on Krippendorff’s alpha. In addition to segment-level labels, three expert raters produced lesson-level ratings and qualitative evaluations of instructional design, instructional delivery, learner response, learning materials, and lesson closure across the 30 lessons, with rater coverage detailed in the body. Using these two human reference layers, we evaluate five vision-capable frontier LLMs across three tracks - text-only segment coding, text + frame segment coding, and lesson-level coverage scored under an LLM-as-judge protocol - and find that no single model consistently outperforms others across all three tracks, that adding a mid-frame inflates both true and false attributions per scene, and that model evaluations over-rate procedurally clear lessons relative to expert raters. \textitTeachObs therefore supports both fine-grained annotation benchmarking and whole-lesson evaluation, showing where AI systems can assist classroom video analysis and where expert judgment remains necessary across varied subjects, classroom formats, and annotation difficulty levels.
[NLP-103] CobSeg: Coherence Boundary Modeling for Dialogue Topic Segmentation
【速读】： 该论文旨在解决对话主题分割（Dialogue Topic Segmentation）中因局部词汇边界信号被现有话语模型稀释而导致的精准度下降问题，尤其关注在话语边缘附近的词汇过渡与跨话语语义不连续性等异质边界线索的捕捉。其核心解决方案是提出一种多分支架构CobSeg，通过分离语义连贯性（coherence-level semantic continuity）与词汇边界过渡（lexical boundary transitions），并采用方向性边界预测机制分别恢复两类信号；同时引入基于语料库的主题连贯性提示，并结合可学习的权重融合策略以增强关键话语位置的信息利用。该方法在无需大语言模型（LLM）推理调用的情况下，显著提升了边界预测性能，在五项基准测试中均表现出优越性，尤其在局部词汇线索显著时，有效降低了错误边界比例（P_k）与平均距离误差（W_d）。
链接: https://arxiv.org/abs/2605.30668
作者: Sijin Sun,Liangbin Zhao,Jiaxiang Cai,Ming Deng,Mingyu Luo,Xiuju Fu
机构: Sun Yat-sen University (中山大学); Tsinghua University (清华大学); Peking University (北京大学)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 8 pages with appindx. Under review
Abstract:Dialogue topic segmentation is critical in many human-AI collaborative applications which requires identifying heterogeneous boundary cues, including lexical transitions near utterance edges and semantic discontinuities across utterances. Existing utterance models often dilute these local lexical signals. We propose CobSeg, a novel multi-branch architecture that separates coherence-level semantic continuity from lexical boundary transitions and recovers both through directional boundary prediction. CobSeg further uses boundary informativeness weighting to emphasize high-utility utterance positions, and incorporates a corpus-derived topic coherence cue with learned combination weights. While CobSeg is evaluated as a compact trainable segmenter under supervised gold-boundary training and a pseudo-label setting with automatically induced boundaries, it performs enhanced boundary prediction without LLM calls during inference. Across five benchmarks, it improves P_k and W_d particularly when local lexical cues are prominent: under gold supervision, it reduces P_k by 0.7 points and W_d by 0.6 points on VHF, and reaches P_k of 1.0 on DialSeg711; with induced boundaries, it reduces P_k by 14.8 points on VHF, by 1.5 points on DialSeg711, and by 1.1 points on TIAGE, outperforming prior non-LLM approaches.
[NLP-104] Counterfactual Graph for Multi-Agent LLM Calibration
【速读】： 该论文旨在解决多智能体大语言模型（Multi-agent LLMs）在协同推理过程中因通信导致的可靠性误判问题。传统方法将多数智能体达成一致视为可信证据，但通信会引发智能体间的相关性失败与虚假共识，使得相同投票比例在不同拓扑结构下可能分别代表真实共识或过度自信。其解决方案的关键在于提出一种反事实智能体图校准框架（CAGE-CAL），通过对比观察到的通信后智能体图与匹配的无通信反事实图，捕捉成对失败相关性与群体级依赖关系；不依赖简单的共识计数，而是量化观测依赖性与无通信状态下的反事实偏差，据此动态校准置信度。实验表明，CAGE-CAL在五个基准测试中显著提升了可靠性判别能力，并在误差校准（ECE）保持竞争力的同时，进一步优化了拓扑选择性能，超越最优固定拓扑策略。
链接: https://arxiv.org/abs/2605.30653
作者: Jiatan Huang,Mingchen Li,Ziming Li,Sunjae Kwon,Hong Yu,Chuxu Zhang
机构: University of Connecticut (康涅狄格大学); University of Massachusetts, Amherst (马萨诸塞大学阿默斯特分校)
类目: Computation and Language (cs.CL)
备注:
Abstract:Multi-agent LLM systems often treat agreement as evidence: when many agents in a panel give the same answer, that answer is assumed to be more reliable. We show that this assumption can fail after agents communicate. Communication can induce correlated failures and false consensus, so the same vote share may reflect reliable agreement in one topology but over-confidence in another. We propose CAGE-CAL, a counterfactual agent-graph calibration framework for multi-agent LLMs. For each query, CAGE-CAL compares an observed post-communication agent graph with a matched counterfactual no-communication graph, capturing both pairwise failure correlations and group-level dependencies. Rather than simply counting how many agents agree, CAGE-CAL estimates the counterfactual shift between observed and no-communication dependence, and calibrates confidence accordingly. Across five benchmarks, CAGE-CAL improves reliability discrimination with competitive ECE, and its calibrated confidence further improves topology selection over the best fixed-topology strategy.
[NLP-105] Same Patient Different Words Different Diagnosis? Evaluating Semantic Stability in Clinical LLM s
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在临床应用中对细微语言变化高度敏感的问题，尤其是在语义等价的输入下仍可能产生不一致预测的风险。其核心挑战在于现有基于嵌入的相似性度量无法有效识别涉及否定、时间性或严重程度等关键临床语义差异的提示变体，从而导致模型行为不稳定。为此，论文提出一种基于自然语言推理（Natural Language Inference, NLI）的语义验证框架，用于筛选真正保持临床语义一致性的提示变体，并结合大语言模型作为评判者（LLM-as-a-judge）进行进一步优化，最终由临床专家审计以确保可靠性。解决方案的关键在于通过多层级语义验证机制提升提示变体的语义保真度，同时引入三项量化指标——语义保持型变体敏感度（MeaningPreserving Variation Sensitivity, MVS）、置信度变化量（ΔC）和最坏情况不稳定性（Worst-Case Instability, WCI），系统评估模型对语义等价提示的鲁棒性。实验结果表明，领域专用模型（Domain-Specific, DS）与通用模型（General-Purpose, GP）在鲁棒性表现上呈现高度依赖具体模型的混合趋势，领域专业化并非始终带来更优的鲁棒性，部分DS模型表现优异，而一些强基准通用模型同样具备竞争力。
链接: https://arxiv.org/abs/2605.30646
作者: Mahdi Alkaeed,Adnan Qayyum,Nabeel Abo Kashreef,Muhammad Bilal,Junaid Qadir
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 14 pages, 5 figures
Abstract:Large Language Models (LLMs) are increasingly used in clinical applications. However, their behavior remains highly sensitive to subtle linguistic variations, such as rephrasing or syntactic variation. This sensitivity poses risks in safety-critical healthcare settings, where semantically equivalent inputs should produce consistent predictions. However, a key challenge is to ensure that prompt variations truly preserve clinical meaning, as embedding-based similarity metrics often fail to capture distinctions involving negation, temporality, or severity. To address this limitation, we propose a semantic verification framework based on Natural Language Inference (NLI) to filter meaning-preserving prompt variations, which are further refined using an LLM-as-a-judge and audited by a clinical expert. In addition, we introduce three metrics to quantify model sensitivity: MeaningPreserving Variation Sensitivity (MVS), confidence variation (\Delta C), and Worst-Case Instability (WCI). We evaluate 16 open-source general-purpose (GP) and medical LLMs within the same model families and parameter scales, using reformulated prompts derived from the DiagnosisQA and MedQA datasets. Our results demonstrate that robustness differences between domain-specific (DS) models are mixed and highly model-dependent, i.e., domain specialization does not consistently improve or reduce robustness to meaning-preserving prompt reformulations. Several DS models rank among the most robust (when compared with GP counterparts), and strong GP baselines remain competitive as well.
[NLP-106] COFT: Counterfactual-Conformal Decoding for Fair Chain-of-Thought Reasoning in Large Language Models ICML2026
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在思维链（Chain-of-Thought, CoT）生成过程中暴露并放大社会偏见的问题。其核心挑战在于如何在不改变预训练模型参数的前提下，实现对推理过程中的偏见进行有效控制，同时保持任务性能与语言质量。该研究提出的解决方案——公平思维链（COFT, Chain of Fair Thought），关键在于通过一种无需训练的解码阶段干预机制，在令牌级别实现公平性调控。COFT的核心创新包括：（1）通过将敏感信息片段替换为中性标记，构建掩码反事实提示；（2）采用轻量级逻辑融合方法对比真实与掩码情况下的输出概率分布，以削弱属性驱动的偏见影响；（3）利用双分支分裂-符合性校准（dual-branch split-conformal calibration）技术，在用户指定的风险水平下对每一步候选词集进行形式化验证，确保公平性具有分布无关的边际有效性保证（在可交换性假设下）。实验表明，该方法在六种不同模型和多个偏见评估基准上实现了标准偏见指标降低30%-55%（中位数38%），同时维持了任务准确率与语言质量，计算开销仅为额外一次缓存前向传播（约11%），且无需重训练、辅助分类器或权重访问。因此，COFT提供了一条清晰、可审计的路径，以实现更安全的思维链生成。
链接: https://arxiv.org/abs/2605.30641
作者: Arya Fayyazi,Mehdi Kamal,Massoud Pedram
机构: 未知
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: Proceeding of ICML 2026
Abstract:Large language models (LLMs) can reveal and amplify societal biases during chain-of-thought (CoT) generation. We present COFT (Chain of Fair Thought), a training-free decoding method that applies token-level fairness control at decode time, with distribution-free marginal validity guarantees (under exchangeability) for any frozen causal language model. COFT operates in three stages. First, it creates a masked counterfactual prompt by replacing sensitive spans with neutral tokens. Second, it compares the factual and masked logit distributions through lightweight logit fusion to attenuate attribute-driven biases. Third, it uses dual-branch split-conformal calibration to certify per-step candidate token sets at a user-chosen risk level. We evaluate COFT across six models and multiple bias benchmarks. Our method reduces standard bias metrics by 30-55% (median 38%) while preserving task utility and language quality. Reasoning accuracies remain unchanged within run-to-run noise margins. The computational overhead is modest, equivalent to one additional cached forward pass (=11%). COFT offers a clear, auditable path to safer CoT generation with significant bias reduction, negligible utility loss, and no requirement for retraining, auxiliary classifiers, or weight access.
[NLP-107] CSULoRA: Closest Safe Update Low-Rank Adaptation
【速读】： 该论文旨在解决大语言模型在采用低秩适应（LoRA）进行参数高效微调时，因少量不安全或对抗性微调数据导致模型安全性显著下降的问题。现有保持安全性的LoRA方法多依赖于硬性干预手段，如投影、剪枝、阈值处理或额外训练目标，虽能抑制不安全更新方向，但可能损失任务相关信息或引入额外调参成本。本文提出一种后处理方法CSULoRA，其核心在于通过估计安全对齐子空间（safety-aligned subspace）来修正已训练的LoRA适配器。具体而言，CSULoRA利用安全对齐模型与其基础检查点之间的权重偏移，构建安全对齐子空间，并将每个LoRA更新分解为完全对齐、部分对齐和非子空间分量。不同于直接丢弃非子空间分量，CSULoRA求解一个闭式惩罚最小变更问题，在保留完全对齐分量的同时，根据各方向的能量相对大小平滑衰减潜在不安全方向。在对抗性微调实验中，CSULoRA显著降低了攻击成功率，同时几乎完整保留了标准LoRA微调带来的性能增益。
链接: https://arxiv.org/abs/2605.30640
作者: Oleksandr Marchenko Breneur,Adelaide Danilov,Aria Nourbakhsh,Salima Lamsiyah
机构: University of Luxembourg (卢森堡大学)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: 10 pages, 3 figure
Abstract:Low-rank adaptation has become a standard method for parameter-efficient fine-tuning of large language models, but even small amounts of unsafe or adversarial fine-tuning data can substantially weaken the safety behavior of aligned models. Existing safety-preserving LoRA methods often rely on hard interventions such as projection, pruning, thresholding, or additional training objectives. While these methods can suppress unsafe update directions, they may also remove task-relevant information or require extra tuning. We introduce CSULoRA, a post-hoc method for correcting trained LoRA adapters through closest safe update estimation. CSULoRA estimates a safety-aligned subspace from the weight displacement between a safety-aligned model and its corresponding base checkpoint. It then decomposes each LoRA update into fully aligned, partially aligned, and off-subspace components. Instead of discarding components outside the estimated safety subspace, CSULoRA solves a closed-form penalized minimum-change problem that preserves the fully aligned component while smoothly attenuating potentially unsafe directions according to their relative energy. In adversarial fine-tuning experiments, CSULoRA substantially reduces attack success rate while preserving most of the utility gains obtained from standard LoRA fine-tuning.
[NLP-108] he Architecture of Errors: From Universal Impossibility to Patch-Local LLM Reliability
【速读】： 该论文旨在解决大语言模型（LLM）在长上下文任务中可靠性难以保障的核心问题，即面对无限可能的任务、工具、数据模式、知识源与评估者期望时，故障模式会以不可穷尽的方式涌现，导致任何有限的干预词典都无法保证对所有潜在故障模式实现有界残余错误。其关键解决方案在于提出“操作边界片段”（operationally bounded patches）的范式转换：实际部署系统并非运行于全宇宙式的开放域，而是局限于特定应用场景（如法律审查、医疗RAG、代码修复、客户服务代理、合同提取等），这些场景具有重复出现的任务、固定的数据结构、确定的工具集和一致的评估预期。在此类局部环境中，实证研究表明故障模式稀疏、可复现且集中于一个较小的高频故障目录中，因此可靠性问题从全局性的指数级文本长度难题，转变为局部的故障模式发现与干预覆盖问题。论文通过两个命题与一个推论进行形式化：命题1为最坏情况下的否定结果，指出有限干预词典无法覆盖无界域中的所有可区分故障模式；推论1揭示了故障模式发现的对数上界限制，表明若尾部新故障模式线性增长，则需指数级增加硬失败事件观测量；命题2则给出正向结论——在对活跃故障模式的对数暴露水平及头部主导覆盖条件下，足够每条硬决策的干预预算随序列长度呈多项对数增长，一旦场景故障目录饱和，该预算即趋于领域常数。该框架并未消除长上下文带来的困难，而是将问题定位到可干预的“轴上”（on-axis）故障模式，强调识别有效干预机制而非试图使复杂场景变得简单。
链接: https://arxiv.org/abs/2605.30628
作者: Mikhail L. Arbuzov,Lee Mosbacker,Sisong Bei,Ziwei Dong,Dmitri Kalaev,Alexey Shvets
机构: Palo Alto Networks(帕洛阿尔托网络)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 25 pages, no figures
Abstract:Universal LLM reliability is not a finite-library problem: across all possible tasks, tools, schemas, knowledge sources, and evaluator expectations, new intervention-distinguishable failure modes can appear without bound, so no finite intervention dictionary can guarantee bounded residual error for every such mode. But deployed systems do not operate over the whole universe. They operate inside operationally bounded patches (legal review, medical RAG, code repair, customer-support agents, contract extraction) with recurring tasks, schemas, tools, and evaluator expectations. Within such patches, empirical evidence suggests failures are sparse, repetitive, and concentrated in a small recurring catalogue, so reliability becomes a local catalogue-discovery and intervention-coverage problem rather than an exponential token-length problem. We formalize this transition with two propositions and one corollary. Proposition 1 is the worst-case-mode-wise negative result: no finite intervention dictionary covers every distinguishable failure mode of an unbounded domain. Corollary 1 is the inverse-discovery implication: the logarithmic upper bound on mode discovery cannot accommodate linearly more distinct tail modes without exponentially more observed hard-failure events. Proposition 2 is the positive patch-local result: under log active-mode exposure and head-heavy coverage, a sufficient per-hard-decision intervention budget grows polylogarithmically in sequence length and becomes domain-constant once the patch catalogue saturates. The framework relocates rather than dissolves long-context difficulty: where the number of hard decisions itself grows with task length, reliability remains hard; the contribution is to identify the on-axis intervention rather than to make those regimes easy.
[NLP-109] Crafter: A Multi-Agent Harness for Editable Scientific Figure Generation from Diverse Inputs
【速读】： 该论文旨在解决科学图表生成中面临的两大核心问题：一是现有自动化系统仅针对单一图表类型且依赖纯文本输入，难以应对科研人员实际使用中多样化的图表类型与输入条件；二是现有系统生成的位图（raster）输出无法进行局部编辑，限制了后续修改与迭代。其解决方案的关键在于提出一种可扩展的“生成式编排框架”（harness），通过多智能体协同机制实现跨图表类型的泛化能力，并在不改变架构的前提下适应不同输入条件。具体而言，作者构建了两个互补系统：Crafter 作为多智能体框架，统一处理多种图表生成任务；CraftEditor 则将位图输出转换为可编辑的矢量图形（SVG），支持局部修改。此外，研究引入 CraftBench 基准测试集，涵盖三种图表类型与四种输入条件，并包含人工质量标注，用于全面评估系统性能。实验表明，Crafter 在 PaperBanana-Bench 与 CraftBench 上显著优于独立生成器及基线代理系统，消融实验证明各组件均具独立贡献；CraftEditor 能准确生成可编辑的 SVG 文件，表现超越所有对比方案。
链接: https://arxiv.org/abs/2605.30611
作者: Haozhe Zhao,Shuzheng Si,Zhenhailong Wang,Zheng Wang,Liang Chen,Xiaotong Li,Zhixiang Liang,Maosong Sun,Minjia Zhang
机构: University of Illinois at Urbana-Champaign (伊利诺伊大学厄本那-香槟分校); Tsinghua University (清华大学); Peking University (北京大学)
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 24 pages, 11 figures
Abstract:Scientific figures are among the most effective means of communicating complex research ideas, yet producing publication-quality illustrations remains one of the most labor-intensive parts of paper preparation. Existing automated systems each target a single figure type under text-only input, leaving the diversity of types and conditions researchers actually use unaddressed; their raster outputs further cannot be locally revised. Because scientific figures are structured compositions of discrete semantic components, the localized errors generators produce on such layouts demand not a stronger backbone but a harness. We instantiate this harness in two complementary systems: Crafter, a multi-agent harness for figure generation that generalizes across figure types and input conditions without architectural changes, and CraftEditor, which applies the same pattern to convert raster outputs into editable SVGs. Moreover, we introduce CraftBench, a benchmark spanning three figure types and four input conditions with human quality annotation. Experiments show that Crafter substantially outperforms both standalone generators and the agentic baseline on PaperBanana-Bench and CraftBench, with ablations confirming each component’s independent contribution; CraftEditor faithfully converts outputs into editable SVGs that surpass all baselines. Our code and benchmark are available at this https URL.
[NLP-110] Semantic Motion Anchors: Bridging Motion and Meaning in Co-Speech Gestures
【速读】： 该论文旨在解决共言语手势（co-speech gesture）检索、合成与理解中，如何有效学习口语文本与手势之间语义一致的共享表示这一核心问题，尤其针对其沟通意图无法仅通过运动学特征捕捉的语义型手势。现有方法在直接对齐文本与连续运动嵌入时，往往过度关注低层次运动细节，而忽略了手势的符号化语义内容。本文提出“语义运动锚点”（semantic motion anchors），即以自然语言抽象表达的手势运动形式与交际意图，作为中间语义桥梁。其关键在于将3D手势离散化为身体-手部运动基元，将其转化为结构化描述，并与文本转录进行语义对齐，从而提供辅助对比监督信号。实验表明，在BEAT2数据集上，该方法相较直接文本-运动对齐基线使文本到手势检索的R@1提升8.2%，且在双向检索任务中优于已有方法；更重要的是，该策略能有效引导模型检索出与口语查询语义相关的有意义手势，而非依赖通用运动模式。下游的检索增强型手势生成实验进一步验证了该方法的有效性：用户显著偏好由本方法检索出的手势所驱动的生成结果，证明了语义锚定的检索可显著提升生成手势在传达交际意图方面的表现。
链接: https://arxiv.org/abs/2605.30608
作者: Varsha Suresh,Mohammad Mahdi Abootorabi,Mohamed Salman,M. Hamza Mughal,Christian Theobalt,Ashwin Ram,Jürgen Steimle,Vera Demberg
机构: Saarland University (萨尔兰大学); MPI for Informatics, Saarland Informatics Campus (马普研究所信息学，萨尔兰信息学园); University of British Columbia (不列颠哥伦比亚大学); Vector Institute (向量研究所); Zuse School ELIZA (康拉德·佐塞卓越学习与智能系统学校)
类目: Computation and Language (cs.CL)
备注:
Abstract:Learning a shared representation between spoken text and gesture is central to co-speech gesture retrieval, synthesis, and understanding, but remains challenging for semantically meaningful gestures whose communicative intent is not captured by motion alone. Direct contrastive alignment between transcripts and continuous motion embeddings often overemphasizes low-level kinematics and misses the symbolic content of semantic gestures. We propose semantic motion anchors, natural-language abstractions of gesture motion capturing physical form and communicative intent. Our method discretizes 3D gestures into body-hand motion primitives, verbalizes them into structured descriptions, and grounds them in the transcript to provide auxiliary contrastive supervision. On BEAT2, our method improves text-to-gesture R@1 by 8.2% over a direct text-motion baseline and outperforms prior retrieval approaches on text to gesture and gesture to text retrieval directions. Beyond aggregate retrieval metrics, semantic motion anchor supervision helps retrieve gestures that are semantically meaningful for the spoken query, rather than defaulting to generic motion patterns. A downstream retrieval-augmented gesture generation study showed that users significantly preferred gestures retrieved by our approach over a retrieval-augmented generation baseline, demonstrating that semantically grounded retrieval translates to gestures that better convey communicative intent in downstream generation.
[NLP-111] AMNESIA: A Large Scale Medical Unlearning Benchmark Suite with Disease-Informed Analysis
【速读】： 该论文旨在解决医学大语言模型（Medical Large Language Models, LLMs）在持续演化背景下，如何有效更新或选择性遗忘特定训练数据所编码信息的问题。现有机器遗忘（Machine Unlearning）方法缺乏针对真实临床场景的大规模评估基准，且多数研究依赖合成或小规模通用数据，难以反映真实医疗知识的复杂性与敏感性。为此，本文提出了AMNESIA——首个大规模、开源的医学领域遗忘评估基准，包含来自11种疾病类别、8,820份患者病历的70,560个问答对，涵盖事实性问题（测试直接记忆）与推理性问题（测试临床推断能力）。研究通过该基准评估四种主流遗忘方法在随机患者及疾病层面的性能，并引入一种新指标以检测医学术语泄露现象。结果表明，仅遗忘单个患者的数据会损害同病种其他患者的模型知识，凸显当前方法在区分个体患者信息与共享临床知识方面的不足，因此亟需发展能更好解耦个体与共性医学知识的新型遗忘机制。
链接: https://arxiv.org/abs/2605.30599
作者: Saeedeh Davoudi,Reihaneh Iranmanesh,Ophir Frieder,Nazli Goharian
机构: Georgetown University (乔治城大学)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注:
Abstract:Medical knowledge is continuously evolving. This creates a need to update or selectively forget information encoded in already-trained medical LLMs. Machine unlearning aims to remove the influence of specific training data from a model without full retraining. Yet, existing unlearning benchmarks rely on synthetic or small-scale general data, leaving clinical unlearning understudied. We introduce AMNESIA, the first large-scale, open source benchmark for medical unlearning, with 70,560 question-answer pairs from 8,820 patient notes across 11 disease categories. AMNESIA includes both factual questions testing direct recall and reasoning questions testing clinical inference. We use it to evaluate four widely used unlearning methods at both random patient and disease-level, and introduce a new metric for detecting leakage of medical terminology. We show that unlearning individual patients erodes knowledge of others with the same condition, calling for methods that can better separate patients from shared clinical knowledge.
[NLP-112] Counterfactual Evaluation Reveals Hidden Capability Profiles in Clinical LLM s and Agents
【速读】： 该论文旨在解决当前临床人工智能（AI）系统评估中存在的一大核心问题：基于覆盖率的评价指标（如共识匹配评分，Consensus Match Score, CMS）无法有效捕捉模型在面对患者临床状态变化时的动态响应能力。尽管多个前沿模型在CMS上表现相近，但其实际行为却存在显著差异——部分模型能根据输入的临床信号变化（如生物标志物状态、既往治疗失败、手术状态等）合理调整推荐意见，而另一些模型则表现出“僵化”行为，输出不变。为应对这一挑战，本文提出因果敏感性评分（Causal Sensitivity Score, CSS），一种预先注册的干预性度量方法，通过在五个具有临床意义的维度上对肿瘤多学科会诊病例进行干预（包括生物标志物反转、既往治疗失败、生物标志物移除、手术状态变更及分期扰动），并以0、0.5、1.0三等级评估各模型是否按预设正确方向更新推荐。实验结果显示，六种来自不同实验室的前沿模型在CSS与CMS上的排名几乎完全相反，且所有模型均在手术状态干预下表现极差（家族D中最高仅17.2%的CSS得分），此严重安全盲点被传统覆盖率指标所掩盖。此外，该指标可迁移至具工具使用能力的代理系统，在ReAct式实验中，工具使用使五种模型的CSS提升2.5至20.3个百分点，但最低分模型仍因结构性响应缺陷而无法更新输出，凸显了仅靠外部工具无法弥补内在推理脆弱性。跨评委复现与三名医学专业人士验证进一步确认了结果的稳健性。因此，该研究的关键贡献在于：引入干预性、可解释的因果敏感性评估框架，揭示了覆盖率指标无法捕捉的模型响应能力缺失，并为未来基于强化学习的智能体系统提供了潜在的高密度奖励信号。
链接: https://arxiv.org/abs/2605.30590
作者: Matt Turk
机构: Protege Data Lab(Protege数据实验室)
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: Accepted to RLEval @ ACM CAIS 2026 (Workshop on Methods and RL Environments for Evaluating AI Agents) and selected for an invited talk based on reviewer ratings. 4-page short paper + appendix
Abstract:Two clinical AI systems can score nearly identically on coverage-based rubrics yet behave radically differently when their patient inputs change: one updates its recommendations to match the new clinical signal, while the other produces the same output regardless. We introduce the Causal Sensitivity Score (CSS), a pre-registered interventional metric that mutates oncology tumor-board cases along five clinically meaningful dimensions - biomarker flips, prior-treatment failures, biomarker removals, surgery-status changes, and stage perturbations - and scores whether each model updates its recommendations in the pre-registered correct direction using a 0, 0.5, 1.0 scale. Benchmarked against the Consensus Match Score (CMS), a coverage-based weighted recall metric, six frontier models from three labs evaluated in single-shot inference across 224 cases rank in nearly opposite orders: all six models change rank, the CMS-worst model becomes CSS-best, and one upper-mid CMS model ranks last on CSS. We further surface a universal safety blind spot: every frontier model fails on surgery-status interventions (at most 17.2% CSS on Family D), a finding CMS does not expose. The metric also transfers to tool-using agents: in a ReAct-style experiment, tool use improves CSS for five of six models (+2.5 to +20.3 percentage points), yet the lowest-CSS model retrieves the same chart sections and still fails to update its recommendations - revealing a structural responsiveness deficit visible only under counterfactual evaluation. Cross-judge replication and three-rater medical-professional validation confirm the aggregate findings. Interventional pre-registered metrics like CSS complement coverage-based evaluation for clinical AI agents: they capture responsiveness that coverage metrics miss and offer a candidate dense reward signal for future agentic RL systems.
[NLP-113] ImmigrationQA: A Source-Grounded Dataset and Small-Model Adaptation for U.S. Immigration Law
【速读】： 该论文旨在解决美国移民法律信息复杂且动态变化背景下，非专业用户在缺乏法律代理情况下难以准确获取合规性指导的问题。其核心挑战在于如何构建一个高精度、可信赖的问答系统，以支持对海量、分散且频繁更新的移民法规与判例（如USCIS政策手册、8 CFR、BIA先例裁决等）进行有效检索与理解。解决方案的关键在于：首先，构建了一个涵盖13个移民子领域的源基问答数据集（ImmigrationQA），包含17,058组标注问答对，通过从11个权威来源提取并验证10,056份规范文档及18,308个文本片段，确保知识来源的准确性与可追溯性；其次，采用参数高效微调（LoRA）技术对Llama 3.2 3B Instruct模型进行微调，显著提升其在特定移民流程类任务中的表现；最后，通过多轮提示工程（五种模式专用提示）结合Claude Sonnet 4.6生成结构化问答对，并基于分层抽样的LLM-as-judge评估框架进行验证。实验表明，微调后模型在993个保留测试样本上的平均得分达1.08/3.0（16.8%完全正确），相比基础版Llama 3 8B模型（0.85/3.0，4%完全正确）实现27%的相对性能提升，尤其在旅行文件、身份调整、非移民签证等程序性领域表现突出。然而，模型在复杂法律推理和时效性统计数据方面仍存在局限。整个流水线耗时约29小时云算力，所有数据集、模型、代码与提示模板均已开源，强调该系统仅为辅助工具，不替代专业法律咨询，亦未反映爬取日期后的法规变动。
链接: https://arxiv.org/abs/2605.30589
作者: Nazarii Shportun
机构: 独立研究者(Independent Researcher)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 12 pages, 4 tables. Dataset (17,058 QA pairs), fine-tuned model, and code are publicly released
Abstract:U.S. immigration law spans thousands of pages of official policy, federal regulations, and procedural guidance that change frequently and carry high stakes for petitioners who lack legal representation. We describe the construction of ImmigrationQA, a source-grounded question-answering dataset of 17,058 pairs across 13 immigration subdomains, and the fine-tuning of a Llama 3.2 3B Instruct model on that dataset using parameter-efficient LoRA. The corpus was assembled from 11 primary and secondary sources – including the USCIS Policy Manual, 8 CFR, BIA precedent decisions, and community QA – yielding 10,056 validated canonical documents and 18,308 text chunks. Structured QA pairs were generated from these chunks using Claude Sonnet 4.6 via five mode-specific prompts, with 22 pairs rejected for insufficient source-span overlap. The fine-tuned model was evaluated against a held-out split of 993 pairs using LLM-as-judge scoring on a 101-example stratified sample. The fine-tuned model scored a mean of 1.08/3.0 (16.8% fully correct; 101-example stratified eval) versus the Llama 3 8B base model at 0.85/3.0 (4% fully correct), a relative improvement of 27% in mean score; a zero-shot Claude Sonnet baseline scored 1.52/3.0 (25% fully correct). The fine-tuned model shows concentrated improvement in procedural subdomains (travel documents, adjustment of status, nonimmigrant visas) while remaining weak on complex legal reasoning and time-sensitive statistics. The full pipeline ran for approximately 29 in cloud compute. All artifacts – dataset, model, code, and prompt templates – are publicly released. The system is not a substitute for legal counsel and does not reflect regulatory changes after the corpus crawl date.
[NLP-114] AI for Monitoring and Classifying Data Used in Research Literature
【速读】： 该论文旨在解决科研文献中数据集使用情况缺乏系统性追踪机制的问题，当前虽有如Google Scholar和Semantic Scholar等平台可追踪论文引用，但尚无类似基础设施用于监测数据集在研究中的实际使用，导致数据使用透明度低、可复现性差且影响力评估困难。其核心挑战在于引用实践不统一、标注数据稀缺以及真实文本中对数据集的指代模糊。传统自然语言处理（NLP）方法难以应对上述问题，因此研究转向更具语义丰富性和适应性的大语言模型（LLM）解决方案。本文提出一种基于GLiNER的多任务框架，实现数据集提及抽取、关系识别与使用上下文分类的联合建模；为缓解标签稀缺问题，引入合成数据生成以扩充训练样本，并结合LLM进行二次验证，有效过滤错误提及并保证标注一致性，显著提升模型在可靠性、覆盖范围和输出一致性方面的表现。该方案推动了开源工具在科研数据使用监控中的发展，为实现通用化、无约束的数据集引用追踪提供了可行路径。
链接: https://arxiv.org/abs/2605.30582
作者: Rafael Macalaba,Aivin V. Solatorio
机构: The World Bank (世界银行)
类目: Computation and Language (cs.CL)
备注:
Abstract:While platforms like Google Scholar and Semantic Scholar track citations for academic papers, no comparable infrastructure exists for monitoring dataset usage in research literature, leaving the landscape of data use largely opaque. Addressing this gap is critical for transparency, reproducibility, and monitoring of impact, yet progress is hindered by inconsistent citation practices, scarce labeled data, and ambiguous references to datasets in the wild. Traditional NLP approaches struggle with these challenges, motivating the shift toward more adaptive, semantically rich models. Building on prior work using LLMs for data mention detection and synthetic data for bootstrapping training, this paper presents an updated methodology for scalable dataset monitoring. We introduce a multitask GLiNER-based framework that jointly performs dataset mention extraction, relation identification, and usage-context classification. To address label scarcity, the pipeline leverages synthetic data generation to produce training examples and LLM-based revalidation to filter incorrect mentions and enforce labeling consistency, together improving reliability, coverage, and output consistency across the training pipeline. This work advances the development of open-source tools for monitoring data use in research literature, contributing to the broader goal of generalizable, unconstrained dataset citation tracking.
[NLP-115] Speculative Decoding Across Languages ACL
【速读】： 该论文旨在解决生成式AI在多语言场景下，小规模草稿模型（draft model）在非英语语言上表现显著下降的问题，即草稿模型在非英语任务中因缺乏足够的多语言能力而导致推测解码（speculative decoding）效率大幅降低。其核心挑战在于：尽管推测解码可通过并行生成多个候选词提升推理速度，但小模型在非英语语料上的泛化能力不足，导致生成质量差、接受率低，从而削弱加速效果。论文提出的解决方案关键在于对比三种改进策略：一是使用特定任务数据（如翻译）对草稿模型进行微调；二是利用未标注的单语语料库进行无监督微调；三是基于相同单语语料训练简单的n-gram草稿模型。研究发现，虽然基于任务的微调能显著提升特定任务的效率，但泛化能力差；而n-gram草稿模型虽接受率较低，但由于生成速度极快，在多种语言和任务（如翻译与故事生成）中均能实现稳定且显著的速度提升，因此成为更优的实用方案。
链接: https://arxiv.org/abs/2605.30580
作者: Nirajan Paudel,Michael Ginn,Luc De Nardi,Alexis Palmer
机构: University of Colorado (科罗拉多大学)
类目: Computation and Language (cs.CL); Machine Learning (cs.LG)
备注: 10 pages, 11 figures, submitted to ACL ARR May 2026
Abstract:Speculative decoding has become a crucial component of large language model (LLM) inference, enabling faster generation by drafting multiple tokens and verifying them in parallel. However, small draft models tend to suffer from disproportionately poor multilingual capabilities. Thus, when generating text in a non-English language, speculative decoding is far less effective. We compare three strategies to improve speculative decoding efficiency for eleven languages: finetuning the draft model on task-specific data (translation); finetuning the draft model on unlabeled monolingual corpora; and training simple n-gram draft models on the same monolingual corpora. We evaluate efficiency on translation (from English into the target language) and the held-out task of story generation. We find that while task-specific distillation can significantly improve efficiency, distilled models generalize poorly to a new task. Meanwhile, n-gram draft models, despite lower acceptance rates, consistently provide large speed-ups due to much faster draft generation. Comments: 10 pages, 11 figures, submitted to ACL ARR May 2026 Subjects: Computation and Language (cs.CL); Machine Learning (cs.LG) Cite as: arXiv:2605.30580 [cs.CL] (or arXiv:2605.30580v1 [cs.CL] for this version) https://doi.org/10.48550/arXiv.2605.30580 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[NLP-116] Probing the Prompt KV Cache: Where It Becomes Dispensable
【速读】： 该论文旨在解决大模型在解码过程中提示词（prompt）所生成的键值缓存（KV cache）存在冗余的问题，具体探究了这种冗余在何种层级、经过多少解码步骤后出现，以及以何种形式可被替代而不影响任务性能。其核心解决方案的关键在于识别出该冗余的本质是“形式”而非“内容”——即提示词部分的冗余主要源于对话模板（chat template）的结构化框架（scaffolding），而非实际语义内容。通过在高层层中用仅包含中性填充内容的模板骨架所生成的KV缓存来替换原始提示词对应的缓存，可几乎完全恢复模型的准确率；而直接清零相同位置则导致性能崩溃。这一发现表明，模型在推理时对提示词的依赖更多体现在格式结构上，而非具体内容，该现象在Qwen3、Gemma 3和Llama 3等多个模型家族及多种数据集上均具有可复现性。
链接: https://arxiv.org/abs/2605.30574
作者: Vinayshekhar Bannihatti Kumar,Manoj Ghuhan Arivazhagan,Disha Makhija,Rashmi Gangadharaiah
机构: AWS AI Labs (亚马逊云科技人工智能实验室)
类目: Computation and Language (cs.CL)
备注:
Abstract:Prior KV cache compression schemes empirically demonstrate that the prompt cache is partially redundant during decoding, dropping or summarising entries with little accuracy loss. We ask when and what kind of redundancy: at which layers, after how many decoding steps, and in what form can the prompt span KV cache be replaced without breaking the task. A controlled splice intervention swept over layer cutoff and decoding steps shows this redundancy is about form (chat template scaffolding) rather than content. Replacing the upper layer prompt span KV cache with KV cache from a chat template scaffold whose user content is a neutral filler recovers near clean accuracy, while zeroing the same slots collapses accuracy. The dissociation replicates across the Qwen3, Gemma 3, and Llama 3 families on multiple datasets.
[NLP-117] Generating and Refining Dynamic Evaluation Rubrics for LLM -as-a-Judge
【速读】： 该论文旨在解决当前基于大语言模型作为裁判（LLM-as-a-Judge）的评估方法依赖人工标注数据（如参考答案或专家设计的评分标准）这一关键瓶颈问题。现有方法在构建细粒度评价标准（rubric）时严重依赖人类参与，限制了其可扩展性与通用性。本文提出一种无需训练的自动化方法，能够自动生成具有数据集特异性和实例特异性粒度的细粒度评价标准，且在四个基准测试中表现媲美现有方法。进一步地，提出一种通过元裁判（meta-judge）奖励信号迭代微调评价标准生成模型的方法，显著提升了生成质量。实验表明，经过微调的140亿参数（14B）评价标准生成模型在成对比较和点评评估任务中超越所有现有基线，甚至优于更大规模的专有模型，验证了其微调策略的有效性。
链接: https://arxiv.org/abs/2605.30568
作者: Zijie Wang,Eduardo Blanco
机构: University of Arizona (亚利桑那大学)
类目: Computation and Language (cs.CL)
备注:
Abstract:LLM-as-a-Judge is a scalable alternative to human evaluation, yet existing rubric-based methods rely on human-annotated data such as reference answers or expert-crafted rubrics. We propose to automatically generate fine-grained evaluation rubrics without any human annotation. Our training-free method generates rubrics at dataset-specific and instance-specific granularities, achieving performance competitive with existing methods across four benchmarks. We further present a method that iteratively fine-tunes a rubric generator model via meta-judge reward signals. The fine-tuned generator outperforms all existing baselines in both pairwise and pointwise evaluation. Notably, a fine-tuned 14B rubric generator outperforms a much larger proprietary model at rubric generation, showing the effectiveness of our fine-tuning strategy.
[NLP-118] Seeing Isnt Knowing: Do VLMs Know When Not to Answer Spatial Questions (and Why)?
【速读】： 该论文旨在解决当前视觉语言模型（Vision-Language Models, VLMs）在真实世界环境中进行空间推理时存在的关键缺陷：现有评估基准普遍假设视觉观测是充分且可靠的，忽视了现实场景中因遮挡（occlusion）和视角歧义（perspective ambiguity）导致的视觉信息不完整或误导性问题。其解决方案的核心在于构建一个受控的评估框架SpatialUncertain，系统引入两类观察挑战——遮挡与视角歧义，并设计相应空间推理问题，使得在干净观测下可解答的问题在引入挑战后应触发模型的回避（abstention）行为。研究进一步检验模型识别有效补充视角的能力。实验结果揭示两大一致性的失败模式：一是模型表现出过度自信，即使在证据不足或误导的情况下仍强行作答，遮挡下平均准确率仅约30%，视角歧义下低于10%；二是即便存在额外视角可用，部分模型在识别能提供可靠证据的视角时表现接近随机水平。因此，该工作强调应从单纯评估答案正确性转向衡量模型何时应主动回避以及如何有效获取可靠证据的能力。
链接: https://arxiv.org/abs/2605.30557
作者: Yue Zhang,Zun Wang,Han Lin,Yonatan Bitton,Idan Szpektor,Mohit Bansal
机构: UNC Chapel Hill (北卡罗来纳大学教堂山分校); Google Research (谷歌研究)
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: Website: this https URL
Abstract:Spatial reasoning is a fundamental capability for vision-language models (VLMs) deployed in real-world environments. However, visual observations are inherently limited representations of a 3D world: occlusion can render objects invisible, and perspective can make geometric properties misleading. Despite this, existing spatial reasoning benchmarks typically assume that observations are sufficient and reliable, focusing on whether models produce correct answers rather than whether they recognize when a question cannot be answered and what additional observations would be needed. In this work, we challenge this assumption by constructing a controlled evaluation framework, SpatialUncertain, and introducing two types of observation challenges: (1) occlusion, which hides target information, and (2) perspective ambiguity, which produces misleading visual cues. For each configuration, we design spatial questions that are answerable under clean observations but require abstention under the introduced challenges. We further evaluate whether models can identify which additional viewpoints would resolve perspective ambiguity. Our results across a diverse set of frontier open- and closed-source VLMs reveal two consistent failure modes. First, models are prone to overconfident answering, attempting to solve spatial reasoning tasks even when visual evidence is incomplete or misleading, with average accuracy around 30% under occlusion and below 10% under perspective ambiguity. Second, even when additional views are available, some models perform near random chance in identifying which would provide reliable evidence. Together, our findings call for moving beyond answer correctness toward evaluating whether models know when to abstain and how to seek reliable evidence.
[NLP-119] Refining Word-Based Grammatical Error Annotation for L2 Korean
【速读】： 该论文旨在解决韩语语法错误修正（K-GEC）中词法评估与学习者错误实际发生层级（形态素级）之间的结构性错配问题。现有资源在表面目标实现、韩语特有编辑标注以及单参考评估等方面存在缺陷，导致评估结果无法准确反映真实修正情况。其解决方案的关键在于：首先，基于形态学约束规则重构韩国国立语言院（NIKL）第二语言韩语语料库中的目标句，并将形态素级标注转换为词级m2编辑格式；其次，提出一种类ERRANT的韩语标注方案，保留最小修订单元（MRU）核心的同时，区分功能形态素错误、拼写错误、词边界错误和词序错误；最后，通过增加额外参考修正，构建多参考评估设置的韩语学习者语言分析（KoLLA）语料库。实证结果表明，重构后的目标句降低困惑度，转换后的m2文件与源-目标编辑表示具有更高一致性，且在相同模型设定下提升了KoBART基线模型的修正性能；多参考评估显著降低了对偏离单一参考但合理的修正所施加的惩罚，尤其在神经网络及提示式生成系统中表现更优。研究证实，韩语GEC的评估不仅依赖于修正模型，更取决于能反映韩语形态、分词及修正多样性的参考数据与编辑标注体系。
链接: https://arxiv.org/abs/2605.30545
作者: Jungyeul Park,Kyungtae Lim,Wonjun Oh,Benjamin Nguyen,Zihao Huang,Mengyang Qiu,Jayoung Song
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:Korean grammatical error correction (K-GEC) presents a structural mismatch between word-based evaluation and the morpheme-level locus of many learner errors. Postpositions and verbal endings are bound to lexical hosts, but they encode grammatical relations that must be represented in correction and evaluation. This paper refines word-based grammatical error annotation for L2 Korean by addressing three connected problems in existing resources: surface target realization, Korean-specific edit annotation, and single-reference evaluation. We reconstruct target sentences from the National Institute of Korean Language (NIKL) L2 corpus under morphologically constrained realization rules and convert its morpheme-level annotations into word-level \textttm2 edits. We then define a Korean ERRANT-style annotation scheme that preserves the MRU core while distinguishing functional morpheme errors, spelling errors, word boundary errors, and word order errors. We also augment the KoLLA corpus with an additional reference correction, yielding a multi-reference evaluation setting for Korean GEC. Empirical validation shows that the refined NIKL targets yield lower perplexity, the converted \textttm2 files achieve higher agreement with source-target edit representations, and the refined resources improve KoBART-based correction under the same model setting. Multi-reference KoLLA evaluation further reduces the penalty imposed on valid corrections that diverge from a single reference, especially for neural and prompted GEC systems. These results show that Korean GEC evaluation depends not only on correction models, but also on reference data and edit annotations that reflect Korean morphology, spacing, and correction variability.
[NLP-120] Generalistic or Specific Embeddings Which is Better? An Empirical Study on Search for Clinical Coding in Non-English Languages
【速读】： 该论文旨在解决多语言临床信息检索中，尤其是国际疾病分类第十版临床修订版（ICD-10-CM / CIE-10）编码检索任务中，非英语语种的句子嵌入模型性能显著下降的问题。现有主流句子嵌入模型大多在英文语料上训练与评估，导致其在西班牙语、加泰罗尼亚语、法语等其他语言上的召回率（recall）严重退化，而这一问题常被聚合基准指标所掩盖。为此，论文提出利用大生成式语言模型（Generative Language Model, LLM）作为数据工厂，生成覆盖多种语言（包括英语、西班牙语、加泰罗尼亚语、意大利语、葡萄牙语和法语）的合成训练数据，以弥补低资源语言医学语料的不足。其解决方案的关键在于构建一个两阶段检索器：首先基于西班牙语生物医学编码器（PlanTL-GOB-ES/bsc-bio-ehr-es）微调一个双编码器（bi-encoder），再引入交叉编码器（cross-encoder）进行重排序。该方法在无需依赖英文生物医学预训练的前提下，即实现对BioBERT-ST的超越（如在R@3和R@5指标上分别达到0.650和0.804），并进一步通过交叉编码器提升整体性能，在多数语言上获得显著增益，尤其在葡萄牙语上达到R@5 = 0.829，远超基线模型的0.714。研究贡献包括：提供了一套可复现的领域特定医疗检索器构建流程；量化了使用合成数据带来的学习增益（MRR从0.755提升至0.876，+15.9%）；并揭示了性能提升主要集中在特定语言与检索排名区间，为后续优化提供了可解释性依据。
链接: https://arxiv.org/abs/2605.30529
作者: David Rey-Blanco,Roberto Cruz
机构: TietAI(蒂埃特人工智能)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 24 pages, 12 figures, 6 tables
Abstract:Sentence-embedding models for semantic search are overwhelmingly developed and evaluated on English corpora. When applied to clinical retrieval in other languages – particularly retrieval of ICD-10-CM / CIE-10 codes – recall degrades in ways often masked by aggregate benchmarks. We study whether large generative language models can serve as data factories to close this gap. We build a two-stage retriever (bi-encoder followed by cross-encoder reranker), fine-tuned from a Spanish biomedical encoder (PlanTL-GOB-ES/bsc-bio-ehr-es) on Gemini-generated synthetic data covering English, Spanish, Catalan, Italian, Portuguese and French, and evaluate against BioBERT-ST and the un-tuned Spanish encoder. The bi-encoder alone matches BioBERT-ST on MRR (0.876 vs. 0.866) and overtakes it on R@3 (0.650 vs. 0.626) and R@5 (0.804 vs. 0.790) without English biomedical pretraining. Adding a cross-encoder reranker lifts aggregate R@5 to 0.822 and dominates on four of five languages (+0.017 Spanish, +0.033 Catalan, +0.018 French, +0.037 Portuguese) at the cost of a small English regression. The trade-off is clinically acceptable: Portuguese reaches R@5 = 0.829 vs. BioBERT-ST’s 0.714. Contributions: an open recipe for building domain-specific medical retrievers from LLM-generated data; quantification of the learning gain (MRR 0.755 to 0.876, +15.9% with ~19,500 synthetic pairs); and a characterisation of where gains concentrate by language and rank.
[NLP-121] Measuring Localizing and Ablating Alignment Signatures in LLM s
【速读】： 该论文旨在解决后训练（post-training）过程如何引入或增强生成式 AI 的特定风格特征，以及这些风格特征是否在模型内部表征中具有可定位的信号这一关键问题。其核心解决方案在于提出一种无需重新训练的激活消融方法——PASTA（Post-training Alignment Signature Targeted Ablation），通过分析对齐模型与基础模型在残差连接中的差异，识别出由后训练引起的风格化表征方向，并在解码过程中对该方向进行消融。实验表明，PASTA 能有效降低多数对齐模型在多种 AI 检测器下的检测率，且效果具有跨检测器泛化能力，同时保持生成文本的相关性与连贯性。结果证实，后训练带来的类 AI 风格特征不仅可被量化和定位，还可通过因果消融验证其影响，为理解并调控生成文本的风格偏差提供了新范式。
链接: https://arxiv.org/abs/2605.30526
作者: Aniket Anand,Janvijay Singh,Zhewei Sun,Dilek Hakkani-Tür,Nick Feamster
机构: University of Chicago (芝加哥大学); University of Illinois at Urbana-Champaign (伊利诺伊大学厄本那-香槟分校); Toyota Technological Institute at Chicago (芝加哥丰田技术学院)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注:
Abstract:Aligned language models often exhibit a recognizable AI-like style, yet its connection to post-training and internal representations remains poorly understood. In this work, we study whether post-training introduces or amplifies AI-like stylistic regularities and whether these regularities have a localized internal signature. To this end, we compare human text, base-model generations, and aligned-model generations under matched human-source prefixes. Aligned generations show lower human-corpus affinity and higher AI-detection rates than base generations, suggesting that post-training shifts generated text away from human-corpus style and toward detector-visible AI-like text. We then introduce PASTA (Post-training Alignment Signature Targeted Ablation), a training-free method that estimates a post-training alignment signature from aligned-base residual contrasts and ablates the corresponding direction during decoding. Across 11 aligned models and 6 AI detectors, PASTA lowers the detection rate for most aligned models; this effect transfers well across detectors and is not reproduced by random directions. Qualitative analysis suggests that PASTA generations remain relevant and coherent while exhibiting greater stylistic variation. Together, these results show that AI-like stylistic effects of post-training can be measured, localized, and causally tested through activation ablation.
[NLP-122] Revisiting Padded Transformer Expressivity: Which Architectural Choices Matter and Which Dont
【速读】： 该论文旨在解决生成式模型（特别是变换器，Transformer）在计算能力上的精确刻画问题，尤其关注其与布尔电路（Boolean circuits）类之间的等价性关系。现有研究虽已揭示变换器与某些电路类的关联，但缺乏严格的数学表征，且结果对建模假设（如注意力机制类型、模型宽度、参数均匀性等）高度敏感。为克服这一局限，论文提出使用“填充式变换器”（padded transformers），即在输入中添加占位符符号（如“…”），以提供多项式级别的空间支持自适应并行计算，从而建立更稳健的等价性。其核心解决方案在于证明：在合理假设下，填充式变换器对注意力机制类型、模型宽度和参数均匀性具有显著鲁棒性，而表达能力主要受数值精度和模型深度的影响。具体而言，论文证明了：在多项式填充条件下，具有常数精度的L-均匀（L-uniform）变换器等价于L-均匀AC⁰；而采用渐进精度（growing-precision）的变换器则可达到L-均匀TC⁰，且不受宽度影响；通过引入循环结构（looping），logᵈN次循环的常数精度变换器可实现FO-均匀ACᵈ，而渐进精度版本可达FO-均匀TCᵈ。值得注意的是，当宽度或精度超过对数级别后，表达能力不再提升，且所有结论均适用于softmax与平均硬注意力（average hard attention）两种机制。
链接: https://arxiv.org/abs/2605.30523
作者: Anej Svete,William Merrill,Ryan Cotterell,Ashish Sabharwal
机构: 未知
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computational Complexity (cs.CC); Computation and Language (cs.CL); Formal Languages and Automata Theory (cs.FL)
备注:
Abstract:Recent work describes what transformers can and cannot compute through connections to boolean circuits, but existing results lack exact characterizations and are sensitive to modeling choices. Padded transformers – to whose input filler symbols such as ``…‘’ are appended – emerge as a useful gadget for establishing equivalences to circuit classes by providing polynomial space for adaptive parallel computation. However, only a limited set of padded transformer idealizations has been studied, leaving open how robustly these equivalences hold under changes to attention type, model width, and uniformity. We find that, under practical assumptions, padded transformers are surprisingly robust to all of these, and identify numeric precision and model depth as the main factors affecting expressivity. Concretely, we prove that polynomially padded \textL-uniform constant-precision transformers are equivalent to \textL-uniform AC^0 , while growing-precision ones achieve \textL-uniform TC^0 regardless of width. Furthermore, looping enables sequential processing analogous to circuits: \log^d N -looped constant-precision transformers reach \textFO-uniform AC^d , and growing-precision ones reach \textFO-uniform TC^d . Interestingly, growing width or precision beyond logarithmic does not increase expressivity, and all our results hold for both softmax and average hard attention transformers.
[NLP-123] Evaluating using Mock Tool Calls to Quarantine Untrusted Prompt Inputs
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在处理不可信输入时的脆弱性问题，尤其是在对抗性压力下（如垃圾信息检测、有害内容分类等任务中），直接将未验证的字符串嵌入提示模板会导致系统易受操纵。当前主流LLM架构（如OpenAI提供的模型）通过“指令层级”（Instruction Hierarchy）来区分信任等级，其中系统消息（System messages）最为可信，而工具结果（Tool Results）最不可信。为缓解此问题，研究提出一种假设：将不可信内容包裹在模拟工具调用（mock tool call）中，可作为一种隔离机制以增强鲁棒性。然而，通过在七种模型和三种“大模型作为裁判”（LLM-as-a-Judge）任务上进行自动化红队测试（redteaming search），研究发现该策略并未普遍提升系统安全性；在二元评估任务（如GSM8K评分）中，反而显著提高了攻击成功率，呈现出对指令层级的反向利用现象。在标量与成对评估任务中，效果较小且依赖具体模型，无一模型表现出稳定受益，部分甚至出现类似逆向现象。因此，该研究的关键结论是：工具包装并非可靠的防御手段，建议在实际部署中评估其局限性，并长期探索更稳健的指令层级训练方法或设计新型不可信输入处理原语。
链接: https://arxiv.org/abs/2605.30521
作者: David Gros,Adam Gleave
机构: FAR.AI
类目: Computation and Language (cs.CL)
备注:
Abstract:Large language models must frequently process untrusted inputs, such as judging an answer from another model or running tasks like spam and harm classifiers while under adversarial pressure. These inputs are often string-formatted directly into a prompt template, leaving systems fragile to manipulation. Current LLM specs from major providers like OpenAI distinguish trustworthiness along an Instruction Hierarchy, from System messages (most trusted) to Tool Results (least trusted). A possible natural mitigation is to wrap untrusted content in a mock tool call as a quarantine. We explore this hypothesis with an automated redteaming search over static attack strings across seven models and three LLM-as-a-Judge tasks. Counter to our hypothesis, tool-wrapping does not broadly improve robustness. On a binary evaluation task (GSM8K grading) it typically increases attack success rates, an apparent inversion of the instruction hierarchy. On scalar and pairwise tasks the effect is smaller and model-dependent, with no tested model reliably helped, and several showing inversion. We recommend evaluating this limitation in deployed systems, and longer-term, pursuing stronger Instruction Hierarchy training or new untrusted-input primitives.
[NLP-124] MAAT: Multi-phase Adapter-Aware Targeted Unlearning
【速读】： 该论文旨在解决机器遗忘（machine unlearning）评估中存在的结构性偏差问题，即现有基准测试中因果性与关系性知识（Why-type 问题）的占比极低（在CounterFact、ZSRE、TOFU、MUSE和WMDP-Cyber中分别不足0.06%、0.6%和1.3%），导致现有方法即使在因果知识遗忘方面存在严重缺陷，仍可能在整体评估中获得较高分数，从而无法被有效识别。其核心解决方案是提出5WBENCH，一个包含5,000个样本、每类5W（Who, What, When, Where, Why）各1,000例的平衡基准，首次实现对因果性遗忘失败的可量化评估。基于此，研究发现现有方法无法同时在因果性问题上实现高遗忘率与高保留率：激进遗忘会损害保留知识，而保守策略则难以有效遗忘因果事实。原因在于Why型问题普遍依赖多跳推理链（44%的Why条目涉及多跳推理，远高于其他类型≤2%）以及长达40.1个词元的答案跨度导致梯度稀释。为此，论文提出MAAT（Multi-phase Adapter-Aware Targeted Unlearning）框架，通过三阶段机制在LoRA适配器权重上操作，融合梯度投影上升、SVD秩-维度剪枝、任务向量反向及混合KL-隐藏状态修复，首次实现了在因果性知识上的高遗忘与高保留并存，突破了遗忘-保留帕累托前沿的现有性能极限。
链接: https://arxiv.org/abs/2605.30514
作者: Suryash Yagnik,Shubham Gaur,Saksham Thakur,Vinija Jain,Aman Chadha,Amitava Das
机构: Indian Institute of Information Technology, Bhopal, India; University of California, Santa Cruz, USA; Stanford University, USA; BITS Pilani Goa, India
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注: 16 pages, 4 figures, 10 tables
Abstract:Machine unlearning evaluation is structurally skewed: Why-type questions, which probe causal and relational knowledge, comprise less than 0.06% of CounterFact, 0.6% of ZSRE, and less than 1.3% of TOFU, MUSE, and WMDP-Cyber. This near-zero representation means that methods that fail on causal knowledge can score highly in aggregate, and this failure is undetectable without balanced evaluation. We present 5WBENCH, a balanced 5,000-sample benchmark with 1,000 examples per 5W category (Who, What, When, Where, Why), making causal unlearning failures quantifiable for the first time. Using 5WBENCH, we show that no existing baseline simultaneously achieves high forgetting and high retention on Why-type questions: aggressive forgetting degrades retained knowledge, while conservative methods fail to forget causal facts. Why-type difficulty stems from multi-hop reasoning chains (44% of Why entries vs. less than or equal to 2% for others) and gradient dilution over 40.1-token answer spans. We present MAAT (Multi-phase Adapter-Aware Targeted Unlearning), a three-phase framework operating on LoRA adapter weights, combining gradient-projected ascent, SVD rank-dimension pruning, task vector negation, and hybrid KL-hidden-state retain repair. MAAT is the first method to simultaneously achieve high forgetting and high retention on Why-type causal knowledge, reaching a new operating point on the forget-retain Pareto frontier. We make our code publicly available.
[NLP-125] Auditing LLM Benchmarks with Item Response Theory
【速读】： 该论文旨在解决大语言模型（Large Language Model, LLM）基准测试标签在发布后固化且未经修正地传播至下游评估任务的问题，导致错误标签被持续累积和放大。其核心解决方案是提出一种基于项目反应理论（Item Response Theory, IRT）的指标，利用114个模型在七个偏好型与多选题基准上的响应数据，以95%的精确度在前200个样本中识别出潜在误标项，显著优于传统监督分类方法。该方法的关键在于通过多模型一致性分析揭示标签错误的根源，包括机械化的标注规则、上游数据集继承的标注误差以及本身存在语义模糊性而无法确立唯一正确答案的题目。此外，模型拟合结果进一步表明，当前奖励模型更擅长捕捉风格偏好而非事实知识，其中仅一个前沿奖励模型在80%以上准确率上与检测到的误标保持一致，远高于同类模型的38%，提示其可能受到基准污染或针对特定基准的过拟合问题。
链接: https://arxiv.org/abs/2605.30504
作者: Sander Land,Daniel M. Bikel
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:LLM benchmark labels are frozen at release and silently propagated into downstream benchmarks, errors and all. We introduce an Item Response Theory-based indicator that surfaces likely mislabels at 95% precision in the top 200 examples across seven preference and multiple-choice benchmarks using responses from 114 models, outperforming a supervised classifier. We trace these errors to mechanical labeling heuristics, upstream annotation mistakes inherited unchanged from source datasets, and fundamentally ambiguous items without a defensible single label. The same model fit reveals that reward models specialize in stylistic preference rather than factual knowledge, and identifies one frontier reward model that agrees with detected mislabels at 78% accuracy versus 38% for its peers, consistent with benchmark contamination or benchmark-specific over-optimization.
[NLP-126] Linear Ensembles Wash Away Watermarks: On the Frag ility of Distributional Perturbations in LLM s
【速读】： 该论文旨在解决生成式人工智能（Generative AI）文本水印在多模型环境下失效的根本性问题。当前，用户可自由调用多个独立训练的大型语言模型（LLM），而各模型的水印机制通常彼此独立地扰动输出概率分布，导致水印信号在集成多个模型输出时被抵消。其解决方案的关键在于提出一种名为WASH（Watermark Attenuation via Statistical Hybridisation）的方法，通过统计混合策略对多个异构模型的输出进行平均，有效消除水印扰动。该方法克服了词汇表不一致与分词差异等实际挑战，理论证明平均过程可将水印偏差降至二阶误差水平，实验证明仅需3-5个模型的平均即可使检测z-score从5-300降至低于2（低于4的检测阈值），并将5%假阳性率下的真正例率（TPR）降低至50%以下，同时提升生成质量27.5%并实现6倍于最优基线的长序列生成速度。研究结果表明，现有水印技术在多模型协同场景下存在固有脆弱性，实现鲁棒的文本溯源需依赖跨厂商的空前协调或接受该根本性缺陷。
链接: https://arxiv.org/abs/2605.30501
作者: Zhihao Wu,Gracia Gong,Qinglin Zhu,Yudong Chen,Runcong Zhao
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:Watermarking embeds statistical signatures in AI-generated text for detection and attribution. We reveal a fundamental vulnerability: when users access multiple models (today’s reality), watermarks trivially fail. Watermarks perturb output distributions away from the original, and in competitive markets, these perturbations are typically independent across providers. We theoretically prove that averaging output probability distributions recovers the unwatermarked distribution with up to a second-order error term. Empirically, simply averaging 3-5 models cancels out these perturbations. We introduce WASH (Watermark Attenuation via Statistical Hybridisation), which solves practical challenges in ensemble generation: vocabulary misalignment and tokenisation differences across heterogeneous models. Experiments across six watermarking schemes and three LLMs show that averaging across 3 models suppresses detection z-scores from 5-300 to below 2 (below the detection threshold of 4) and reduces TPR at 5% FPR to below 50%, while improving quality by 27.5% and running 6 times faster than the best baseline on the long sequence generation. Our results suggest that robust AI-text detection via watermarking requires either accepting this fundamental vulnerability or unprecedented coordination among model providers.
[NLP-127] CanLegalRAG Bench: Evaluating Retrieval-Augmented Generation on Canadian Case Law
【速读】： 该论文旨在解决当前基于检索增强生成（Retrieval-Augmented Generation, RAG）的法律智能助手在实际应用中仍面临大语言模型（Large Language Model, LLM）幻觉问题，且现有评估基准多依赖合成查询而非真实法律场景，同时加拿大法律体系在现有评估中严重缺失的问题。其解决方案的关键在于提出CanLegalRAGBench——一个基于真实法律查询和基于判例法专家标注答案的加拿大法律问答评估基准，以更贴近实际司法实践。该基准不仅揭示了检索模块设计选择对性能的敏感性，表明开源嵌入模型可与闭源模型相媲美，还指出现有自动评估方法因惩罚系统召回替代性相关文档而存在局限性，并暴露了生成答案普遍存在的幻觉、过度细节化或无关内容等问题，其中8%-29%的主张未被检索到的文档所支持。该基准的建立有望推动法律RAG系统在可靠性与准确性方面的持续改进。
链接: https://arxiv.org/abs/2605.30497
作者: Ethan Zhao,Maksym Taranukhin,Wei Cui,Moira Aikenhead,Vered Shwartz
机构: 未知
类目: Computation and Language (cs.CL)
备注:
Abstract:RAG-based legal assistants have been growing in popularity, but LLM hallucinations remain a key issue and potentially undermines justice. While benchmarks have been developed to evaluate progress, many rely on synthetic queries rather than realistic legal scenarios. Moreover, Canadian law remains underrepresented in existing evaluations. To address this gap, we introduce CanLegalRAGBench, a Canadian legal QA benchmark based on realistic queries and expert-annotated answers grounded in case law. Our evaluation shows that retrieval performance is sensitive to design choices and that open-source embedding models are competitive with closed source models. However, it also reveals the limitation of automatic evaluations that penalize systems for retrieving alternative relevant documents. We also find that generated answers often diverge from gold responses, either with hallucinations or by producing overly detailed or irrelevant content, with 8-29% of claims not being supported by the retrieved documents. We hope this benchmark will help drive continued progress in addressing limitations of legal RAG systems.
[NLP-128] Configurable Reward Model for Balanced Safety Alignment
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在面对异构且快速演进的安全需求时，难以有效对齐的问题。现有指令微调的LLMs与独立的安全分类器普遍缺乏对新安全配置的泛化能力，限制了其在动态环境中的应用。为此，本文提出可配置安全奖励模型（Configurable Safety Reward Model, CSRM），其核心在于通过联合优化校准后的安全合规性与奖励建模能力，实现对细粒度安全配置的敏感响应。解决方案的关键在于引入面向配置的目标数据增强策略，在强化指令遵循的同时保持不同安全风险之间的相对严重性结构，从而显著提升模型对未见过的安全配置的泛化性能。实验表明，CSRM在多个最新可配置安全基准测试中表现优异，如CoSApien（F1=94.6%）和DynaBench（F1=75.8%），且无需额外人工标注；将其用于下游安全对齐时，可使LLM在帮助性与安全性之间取得更优权衡。
链接: https://arxiv.org/abs/2605.30487
作者: Zhengping Jiang,Mehran Khodabandeh,Akash Bharadwaj,Manik Bhandari,Mayur Srungarapu,Anqi Liu,Benjamin Van Durme,Li Chen
机构: Johns Hopkins University (约翰霍普金斯大学); Meta (Meta)
类目: Computation and Language (cs.CL)
备注:
Abstract:Aligning large language models (LLMs) to heterogeneous and rapidly evolving safety requirements remains a critical challenge. Existing instruction-tuned LLMs and standalone safety classifiers often fail to generalize to new safety configurations, motivating the need for Reward Models (RMs) that are explicitly configurable to changing specifications. We introduce the Configurable Safety Reward Model (CSRM), which is jointly optimized for calibrated safety compliance and reward modeling. Our approach is supported by configuration-targeted data augmentation that enforces instruction adherence while preserving relative severity structure. The resulting RM is sensitive to fine-grained safety configurations and conversational nuances, substantially improving generalization to previously unseen safety configurations. CSRM achieves state-of-the-art performance on recent configurable safety benchmarks, including CoSApien (94.6% F1) and DynaBench (75.8% F1), without requiring additional human annotation. When used for downstream safety alignment, CSRM yields LLMs with a significantly improved helpfulness-safety tradeoff compared to existing baselines.
[NLP-129] When English Rewrites Local Knowledge: Global Narrative Dominance in Large Language Models
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在跨语言知识传递中存在“全球叙事主导”（global narrative dominance）的问题，特别是在低资源文化语境下的孟加拉语（Bangla）中，模型倾向于反映全球主流叙事而非本地文化背景。其核心解决方案是构建并使用一个名为CulturalNB的高质量数据集，该数据集包含717个经过人工精心标注的孟加拉文化实例，涵盖平行的孟加拉语—英语问答对、支持证据、元数据及社会文化注释。通过仅基于问题和基于证据的提示方法，评估九个前沿大语言模型，并结合人类与两名独立大语言模型裁判，在跨语言一致性、语言锚定、全球替代、机构偏见及认识论视角覆盖等多个维度进行分析。研究发现，以英语提问会系统性加剧全球替代与机构化框架，同时削弱本地视角的覆盖；而引入本地证据虽能提升事实一致性和视角覆盖，却无法完全消除语言诱发的认知转移。这表明，大语言模型的文化偏差不仅是知识缺失所致，更深层原因在于缺乏文化根基与叙事优先级的失衡。
链接: https://arxiv.org/abs/2605.30481
作者: Md Arid Hasan,Ruwad Naswan,Farhan Samir,Sharifa Sultana,Syed Ishtiaque Ahmed
机构: University of Toronto (多伦多大学); BUET (巴基斯坦工程技术大学); University of Illinois Urbana-Champaign (伊利诺伊大学厄本那-香槟分校)
类目: Computation and Language (cs.CL)
备注: Submitted to ARR
Abstract:Large language models (LLMs) are widely used as cross-lingual knowledge interfaces. However, culturally grounded questions often reflect globally dominant narratives rather than local contexts. We study this failure mode as \textitglobal narrative dominance in Bangla, a low-resource cultural context. We introduce \textttCulturalNB, a dataset of 717 manually curated Bengali cultural instances with parallel Bangla–English question–answer pairs and supporting evidence, metadata, and sociocultural annotations. Using question-only and evidence-based prompting, we evaluate nine state-of-the-art LLMs with human and two independent LLM judges across metrics for cross-lingual consistency, language anchoring, global substitution, institutional bias, and epistemic perspective coverage. Results show that questions asked in English systematically increase global substitution and institutional framing while reducing local perspective coverage. Local evidence improves factual consistency and perspective coverage, but does not eliminate language-induced epistemic shifts. These findings suggest that cultural failures in LLMs are not only missing-knowledge errors but also failures of grounding and narrative prioritization.
[NLP-130] Improving Small Language Models for Code Generation with Reinforcement Learning from Verification Feedback
【速读】： 该论文旨在解决生成式代码模型在训练过程中难以有效优化功能正确性的问题，尤其针对小规模语言模型在代码生成任务中因缺乏可验证奖励信号而导致的性能瓶颈。其核心解决方案是引入可验证奖励强化学习（Reinforcement Learning with Verifiable Rewards, RLVR），利用程序可检查的信号（如单元测试结果、静态分析工具Ruff的警告）构建奖励函数，实现对代码功能正确性的直接优化。关键创新在于通过组合单元测试奖励与静态分析惩罚的联合奖励机制，在保持功能正确性的同时平衡代码风格约束；实验表明，该方法在MBPP基准上使pass@1指标最高提升13个百分点。然而研究发现，仅使用静态分析作为奖励塑形会导致策略向更短但未必更正确的代码偏移，而联合奖励能有效缓解这一退化现象，提升训练稳定性。研究强调，奖励设计的精细度与优化粒度对RLVR效果具有决定性影响，且需结合生成长度、代码风格严重性分布及执行错误类型等行为诊断指标，才能全面识别失败模式。
链接: https://arxiv.org/abs/2605.30478
作者: Egor Skopin,Evgeny Kotelnikov
机构: Vyatka State University (维亚特卡国立大学); European University at St. Petersburg (圣彼得堡欧洲大学)
类目: oftware Engineering (cs.SE); Computation and Language (cs.CL)
备注: Accepted for AINL-2026 conference
Abstract:Reinforcement learning with verifiable rewards (RLVR) trains language models using programmatically checkable signals such as unit-test outcomes, enabling direct optimization for functional correctness in code generation. We conduct an empirical study of RLVR for Python code generation on the MBPP benchmark using two small models (Qwen3-0.6B and Llama3.2-1B) with LoRA fine-tuning. Across multiple reward formulations such as: unit-test-only rewards, static-analysis-only shaping via the Ruff linter, and a combined reward, we compare group-based policy optimization variants (GRPO and GSPO) and evaluate both functional correctness and behavioral diagnostics. In our experimental setting, RLVR improves pass@1 on MBPP test by up to 13 percentage points under proposed combined reward configuration. However, we find that reward shaping can induce systematic behavioral shifts: using only static-analysis penalties may bias the policy toward shorter completions that reduce lint errors without reliably improving functional correctness. In contrast, combined rewards mitigate this degeneration and yield more stable trade-offs between correctness and style constraints. Overall, our results highlight that RLVR effectiveness for code generation is highly sensitive to reward design and optimization granularity, and that diagnostics beyond pass@1, including generation length, Ruff severity profiles, and execution error types are useful for identifying failure modes.
[NLP-131] Your Multimodal Speech Model Says I Have a Face for Radio
【速读】： 该论文旨在解决多模态语音识别模型在引入视觉信息后可能加剧或引入新的偏见问题，尤其是在跨性别、种族及其交叉身份维度上的服务质量差异。尽管单模态模型的性能与偏见已有广泛研究，但多模态场景下新增的视觉信号如何影响识别准确性及公平性尚不明确，而人类在感知中已表现出对视觉线索的偏见。为此，论文提出首个针对多模态语音识别的偏见评估框架，通过构建音视频配对数据集（同一音频搭配不同面部特征），量化分析模型在不同人口统计学属性下的转录准确率变化。其关键解决方案在于设计可控的实验范式，系统性地测量并揭示了mWhisper-Flamingo和Gemini等多模态模型在自述性别、种族及其交集维度上高达4.05个词错误率（Word Error Rate, WER）的显著性能下降。研究结果表明，增加视觉模态并不必然提升性能，反而可能放大偏见，因此亟需开发者对多模态系统的公平性进行评估、修复与透明沟通。
链接: https://arxiv.org/abs/2605.30472
作者: Maya K. Nachesa,Vlad Niculae,Vagrant Gautam
机构: University of Amsterdam (阿姆斯特丹大学); Heidelberg Institute for Theoretical Studies (海德堡理论研究所)
类目: Computation and Language (cs.CL)
备注:
Abstract:As large neural models have become better at language tasks, researchers are increasingly building multi- and omnimodal models that handle more modalities of data. One example is the expansion of speech recognition models to audio-visual data for noise mitigation and multimodal subtitling. While performance and bias have been studied extensively in the single-modality regime, it is unknown how new modalities affect this, even though they produce biases in humans. We therefore propose the first bias evaluation of multimodal speech recognition, where we create videos pairing different faces with the same audio, and measure changes in speech transcription accuracy. We find large quality-of-service differences across mWhisper-Flamingo and Gemini models, with drops of up to 4.05 word error rate points, across self-declared gender, ethnicity, and their intersection. Our findings point to a priority for developers to evaluate, fix, and communicate such limitations, as providing more signals through additional modalities is not necessarily better, and may even lead to biased outcomes.
[NLP-132] Knowledge Graph-Enhanced Zero-Shot Topic Classification: A Multi-Strategy Comparative Study ACL
【速读】： 该论文旨在解决无标注训练数据下的零样本多标签主题分类问题，尤其针对包含复杂关系信息的文档。其核心挑战在于如何在缺乏显式标注的情况下准确识别文档中的多个主题标签，同时有效建模文档内部的语义关系。解决方案的关键在于提出一种基于文章级知识图谱（Knowledge Graph, KG）增强的零样本多标签分类框架，通过从输入文档中自动构建以“主体-谓词-客体”三元组形式表示的细粒度关系知识图谱，并将其融入不同变体的分类流程中。实验表明，关键词增强型分类（AK）在基础框架中表现最优，且部分小模型性能超越了基于句子编码器的基线；而知识图谱增强对大模型存在负面效应，表明大型语言模型（LLM）在预训练阶段已具备充分的关系表征能力，无需额外图谱注入。此外，自一致性解码虽提升了计算开销约五倍，但在所有实验中均未带来性能提升，说明其在当前任务中不具备实际收益。
链接: https://arxiv.org/abs/2605.30465
作者: Shahana Akter,Yatharth Vohra,Ankita Shukla,Souvika Sarkar
机构: Wichita State University (威奇托州立大学); University of Nevada, Reno (内华达大学雷诺分校)
类目: Computation and Language (cs.CL)
备注: 15 pages, 1 figure, ACL format. This paper proposes a KG-augmented zero-shot multi-label topic classification framework and evaluates multiple strategies
Abstract:Multi-label topic classification without labeled training data is a challenging task, specially when documents contain complex relational information. We present a zero-shot multi-label topic classification framework and systematically investigate how per-article knowledge graph augmentation affects its performance. The base framework classifies topics in documents without labeled training data and has four variants: article-only classification, keyword-enhanced classification, and self-consistency decoding variants of both. Then, we augment each base variant with per article knowledge graph. This graph is extracted from the input document through a pipeline similar to KGGen based on subject-predicate-object triples. We test all eight methods, four base and four graph augmented on fifteen LLMs and eight multi-label datasets across different domains. For the base framework, keyword-enhanced classification (AK) is the best performing method, and six out of fifteen LLMs surpass the sentence-encoder baseline. Graph augmentation has positive and negative impacts on small and large models, respectively. This shows that larger models already contain enough relational information from pretraining. Furthermore, the self-consistency decoding variant does not show performance improvements in any experiment while increasing computation costs about fivefold.
[NLP-133] Can LLM Teams Play What? Where? When?
【速读】： 该论文旨在解决大语言模型（Large Language Models, LLMs）在需要间接推理、文化知识以及协同假设检验的任务中表现不足的问题。针对这一挑战，研究提出通过团队协作机制提升模型性能，其关键在于设计并验证三种团队策略：投票机制（Voting）、静默团队（Silent Team，队长仅观察最终答案）和健谈团队（Talkative Team，队长可同时观察答案与推理过程）。实验基于2025年发布的包含572个问题的“何时？何地？何事？”（What? Where? When?，简称ChGK）数据集，采用六种近期主流开源大模型进行评估。结果表明，团队策略显著优于单模型基线，准确率最高提升达20个百分点，最佳团队达到44.23%的准确率，接近人类团队水平。分析显示，模型间分歧强烈预示着低准确性，但通过解释性沟通可有效缓解性能下降；进一步考察队长行为发现，模型不存在自我偏好偏差，且获取同伴推理过程能显著改善决策质量。总体而言，LLM团队主要发挥答案选择与错误过滤功能，而非生成新解决方案，凸显了交互机制的重要性，并为多智能体系统中自适应协作策略的发展提供了重要方向。
链接: https://arxiv.org/abs/2605.30459
作者: Anastasia Kotelnikova,Viktor Byzov,Maria Dolzhenkova,Evgeny Kotelnikov
机构: Vyatka State University (维亚特卡国立大学); European University at St. Petersburg (圣彼得堡欧洲大学)
类目: Computation and Language (cs.CL)
备注: Accepted for Dialogue-2026 conference
Abstract:Large language models (LLMs) remain limited on tasks requiring indirect reasoning, cultural knowledge, and coordinated hypothesis testing. We investigate whether team-based interaction improves LLM performance in What? Where? When? (ChGK), a quiz game designed to reward collective reasoning. We introduce three team strategies: Voting, Silent Team (the captain observes final answers), and Talkative Team (the captain observes both answers and rationales). To minimize data leakage, we evaluate these strategies on a dataset consisting of 572 ChGK questions released in 2025. Using six recent large-scale open models, we show that team-based strategies outperform single-model baselines, yielding gains of up to 20 percentage points in accuracy. The best team achieves 44.23% accuracy, and approaches human team performance on questions with available human statistics. Analysis of inter-model diversity reveals that disagreement strongly predicts lower accuracy, but explanatory communication substantially mitigates performance drops. We further examine captain behavior and find no evidence of self-preference bias; access to peer rationales improves captain judgments. Overall, LLM teams function primarily as answer selection and error-filtering mechanisms rather than generators of novel solutions. Our findings highlight the importance of interaction and suggest adaptive strategies as a promising direction for multi-agent systems.
[NLP-134] Bounded Behavioral Indistinguishability for Black-Box LLM Distillation
【速读】： 该论文旨在解决黑盒大语言模型（Large Language Model, LLM）蒸馏中“输出匹配”评估范式存在的局限性问题，即当前方法仅关注学生模型与教师模型在语义或任务一致性上的相似性，而忽略了学生模型是否在行为上真正难以被区分。其核心问题是：现有蒸馏方法虽能提升输出相似度，但未必实现行为不可区分性（behavioral indistinguishability）。为此，论文提出并形式化了一种新的评估框架——有界行为不可区分性（bounded behavioral indistinguishability），以 (\epsilon, q, t, \mathbb{A})-行为不可区分性为定义，分别约束辨别优势（\epsilon）、查询次数（q）、计算开销（t）以及攻击者类别（\mathbb{A}），从而构建更严格的对抗性评估标准。关键解决方案在于引入基于可控5,000个提示的系统性行为探测套件，并通过对抗性分类器、类别层面分析及跨家族判别实验，揭示尽管LoRA蒸馏显著提升了语义相似度（如Qwen从0.788升至0.862，Llama从0.814升至0.874），但学生模型仍存在可被识别的行为差异，主要集中在风格/格式、鲁棒性及领域技术类提示上。此外，实验表明，基于分歧引导的采样策略并未持续优于分层随机采样，凸显覆盖与多样性作为基线的重要性。研究结论强调：单纯追求语义保真度不足以支撑有效的黑盒蒸馏，必须采用有界、对抗性且类别敏感的综合评估体系。
链接: https://arxiv.org/abs/2605.30448
作者: Munawar Hasan
机构: Michigan Technological University (密歇根理工大学)
类目: Machine Learning (cs.LG); Computation and Language (cs.CL)
备注:
Abstract:Black-box LLM distillation is usually evaluated as an output-matching problem: a student is considered successful when its responses are semantically similar to, or task-consistent with, those of a teacher. However, output similarity does not imply that the student is behaviorally indistinguishable from the model it imitates. We introduce bounded behavioral indistinguishability, formalized as (\epsilon,q,t,\mathbbA) -behavioral indistinguishability over an explicit prompt distribution, where \epsilon bounds distinguishing advantage, q bounds oracle queries, t bounds computation, and \mathbbA denotes the adversary class. We instantiate this notion on Qwen and Llama teacher-student pairs using a controlled 5,000 -prompt behavioral probe suite. For each family, we compare the teacher with both the base student and the LoRA-distilled student, measuring whether distillation reduces distinguishability rather than merely improving similarity. LoRA raises semantic similarity from 0.788 to 0.862 for Qwen and from 0.814 to 0.874 for Llama. Yet adversarial evaluation reveals remaining behavioral differences: learned discriminators retain nonzero advantage, and pairwise category analysis shows artifacts concentrated in style/format, robustness, and domain-technical prompts. A pairwise teacher-identification adversary confirms this trend. With a different-family Llama judge and A/B-swap consistency filtering, Qwen distinguishing advantage drops from 0.158 for the base student to 0.081 after LoRA distillation. Query-budget experiments show that disagreement-guided acquisition does not consistently outperform stratified random sampling, indicating that coverage and diversity remain strong baselines. Our results show that semantic fidelity is useful but insufficient: black-box LLM distillation requires bounded, adversarial, and category-aware evaluation. Subjects: Machine Learning (cs.LG); Computation and Language (cs.CL) Cite as: arXiv:2605.30448 [cs.LG] (or arXiv:2605.30448v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2605.30448 Focus to learn more arXiv-issued DOI via DataCite
[NLP-135] Cross-Lingual Steering for Figurative Language Generation
【速读】： 该论文旨在解决多语言大语言模型（Multilingual Large Language Models, MLLMs）在生成隐喻性语言时，其内部驱动信号是否具有语言特异性或可在不同语言间通用这一关键问题。研究通过激活操控（activation steering）作为探针，从某一语言中基于隐喻与字面意义的激活差异提取出特定的语义方向，并将其应用于生成过程。结果表明，在五类隐喻类型、六种语言及四类多语言模型中，所提取的方向在目标语言内能稳定地引导生成行为，尤其在隐喻和明喻类别中表现最为稳健。更重要的是，这些方向具备跨语言迁移能力：在一种语言中学习到的方向可有效增强另一语言中的目标行为，其中德语表现出较高的接受度。进一步发现，由其他语言构建的方向甚至可达到或超越目标语言自身的原生方向，而移除共享的跨语言成分则会削弱本地化引导效果。上述结果为隐喻生成存在可复用但依赖目标语言的跨语言信号提供了直接实证支持。
链接: https://arxiv.org/abs/2605.30443
作者: Linfeng Liu,Tiffany Zhan,Louie Hong Yao,Saptarshi Ghosh,Tianyu Jiang
机构: University of Cincinnati (辛辛那提大学); Carnegie Mellon University (卡内基梅隆大学)
类目: Computation and Language (cs.CL)
备注: 40 pages, 7 figures
Abstract:Multilingual large language models can generate figurative language, but whether the internal signals driving this behavior are language-specific or reusable across languages is unclear. Using activation steering as a probe, we estimate a direction for a figurative category from figurative–literal activation differences in one language and apply it during generation. Across five figurative categories, six languages, and four multilingual LLMs, these directions steer reliably within their own language, most robustly for metaphor and simile. More importantly, they transfer across languages: a direction learned in one increases the target behavior when applied to another, with German among the most receptive targets. Going further, directions assembled from other languages can match or even surpass a target language’s own native direction, while removing this shared component weakens native steering. Together, these results provide direct evidence of a reusable but target-dependent cross-lingual signal for figurative generation.
[NLP-136] Domain Adaptation and Reasoning Frameworks in Language Models: A Controlled Experiment with Historical Cosmology
【速读】： 该论文旨在探究领域适应（domain adaptation）如何改变语言模型的解释行为，以古代宇宙学为受控研究场景。其核心问题在于：当语言模型在去除明确日心说表述的前哥白尼时期文本上进行训练或微调时，是否仍会生成与地动或日心说相关的延续内容？解决方案的关键在于采用两阶段实验设计——第一阶段从零训练小型语言模型于去日心化语料，评估其是否自发产生地球运动相关表述；第二阶段利用QLoRA对大型预训练模型进行微调，分析领域适应对解释框架（explanatory frame）和宇宙观立场（cosmological stance）的影响。通过构建基于大模型作为评判者（LLM-as-judge）的评估体系，研究发现：小型模型虽偶现局部地球运动延续，但整体不具稳定性，无法支撑连贯的宇宙论推理；而经过微调后，模型显著转向前现代解释框架，且该转变主要体现在解释范式重构而非直接改变宇宙观立场。因此，研究揭示领域适应更关键的作用是重塑生成延续内容的语言框架，而立场变化是此框架转变的次生结果。
链接: https://arxiv.org/abs/2605.30415
作者: Francesco De Bernardis
机构: Independent Researcher(独立研究员)
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
备注: 17 pages, 3 figures
Abstract:We investigate how domain adaptation reshapes explanatory behavior in language models using historical cosmology as a controlled setting. In Phase 1, we train a small language model from scratch on a pre-Copernican corpus from which explicit heliocentric references were removed, and evaluate whether Earth-motion or heliocentric continuations nevertheless emerge. In Phase 2, we fine-tune a larger pretrained model using QLoRA on the same corpus in order to study how adaptation modifies explanatory framing and cosmological stance. Model outputs are evaluated using an LLM-as-judge framework that labels both cosmological stance (geocentric, heliocentric, or ambiguous) and explanatory frame (premodern versus modern). In the constrained setting of Phase 1, the smaller models occasionally generate local Earth-motion continuations, but these remain globally unstable and insufficient to support coherent cosmological reasoning. In Phase 2, fine-tuning induces a large and statistically significant shift toward premodern explanatory framing, while the conditional cosmological stance distributions remain comparatively stable within those frames. As a result, increases in geocentric outputs arise primarily from redistribution over explanatory regimes rather than from direct modification of stance. These results suggest that domain adaptation may primarily reshape the linguistic frameworks from which continuations are generated, with changes in stance emerging secondarily from those shifts.
[NLP-137] Protocol for evaluating ChatGPT in biomedical association generation and verification using a RAG -enabled cross-model majority voting workflow
【速读】： 该论文旨在解决生成式AI在生物医学领域中生成疾病相关关联时存在的可靠性与准确性问题，尤其关注模型可能产生的幻觉（hallucination）现象。其核心挑战在于如何有效验证由大语言模型（LLM）生成的生物医学实体及关联是否符合真实生物学事实。解决方案的关键在于提出一套综合性的评估协议，包括基于生物医学本体（biomedical ontologies）对实体进行精确匹配、通过文献证据验证关联的合理性，以及引入自一致性（self-consistency）策略以评估不同版本ChatGPT模型生成结果的一致性与可信度。为克服传统本体精确匹配的局限性，研究进一步设计了一个基于检索增强生成（Retrieval-Augmented Generation, RAG）的工作流，利用开源大语言模型实现语义层面的验证，使模型能够基于外部知识源判断其他模型生成内容的真实性，从而有效识别并抑制幻觉，提升生成结果的可信赖性。
链接: https://arxiv.org/abs/2605.30400
作者: Ahmed Abdeen Hamed,Luis M. Rocha
机构: 未知
类目: Computation and Language (cs.CL)
备注: Main Manuscript and Supplementary Information. Both are equally important
Abstract:We present a protocol to evaluate ChatGPT’s ability to generate disease-centric biomedical associations. It outlines how we generate the associations, validate the biological entities using biomedical ontologies, and verify associations using literature. The protocol includes a self-consistency strategy to assess generative reliability across ChatGPT models. To address ontology exact-match limitations, we provide a use case performing semantic verification through a workflow enabled by Retrieval-Augmented Generation (RAG) powered by open-source large language models (LLMs). This enables LLMs to establish truth over content generated by other LLMs and expose hallucination.
[NLP-138] ImmersiveTTS: Environment-Aware Text-to-Speech with Multimodal Diffusion Transformer and Domain-Specific Representation Alignment ACL2026
【速读】： 该论文旨在解决在环境音频（environmental audio）背景下生成自然且语义一致的语音（speech）这一挑战，尤其针对语音与环境音在声学特征和时间动态上的显著差异。现有文本到语音（Text-to-Speech, TTS）方法在生成孤立语音时表现良好，但在融合真实环境背景音时往往出现不协调、失真或语义割裂的问题。其核心解决方案是提出一种环境感知的文本到语音模型——ImmersiveTTS，通过显式建模跨模态交互来实现语音与环境音频的无缝融合。关键技术在于：基于多模态扩散变压器（multimodal diffusion transformer），利用联合注意力机制将对齐语料的语音隐变量（speech latent）与文本条件化的环境上下文（text-conditioned environmental context）进行深度融合；同时引入面向特定领域（环境感知TTS）的表示对齐目标，利用语音与音频编码器提供的互补自监督表征，增强语音与环境之间的语义一致性。实验结果表明，ImmersiveTTS在客观指标和主观听觉测试中均显著优于现有方法，在自然度、可懂度和音频保真度方面表现出更优性能。
链接: https://arxiv.org/abs/2605.30965
作者: Jun-Hak Yun,Seung-Bin Kim,Seong-Whan Lee
机构: Korea University (韩国科学技术院)
类目: Audio and Speech Processing (eess.AS); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: Accepted to ACL 2026 main conference. Code is available at this https URL
Abstract:Recent advancements in text-guided audio generation have yielded promising results in diverse domains, including sound effects, speech, and music. However, jointly generating speech with environmental audio remains challenging due to the inherent disparities in their acoustic patterns and temporal dynamics. We propose ImmersiveTTS, an environment-aware text-to-speech (TTS) model that generates natural speech seamlessly integrated within environmental contexts by explicitly modeling cross-modal interactions. Our model builds on a multimodal diffusion transformer and fuses transcript-aligned speech latent with text-conditioned environmental context via joint attention. To enhance semantic consistency, we introduce a domain-specific representation alignment objective tailored to environment-aware TTS, leveraging complementary self-supervised representations from speech and audio encoders. Experimental results show that ImmersiveTTS achieves higher naturalness, intelligibility, and audio fidelity than existing approaches across objective metrics and human listening tests.
[NLP-139] A Padding Method for Enhanced Encoding of Inorganic Structures with Varying Chemical Compositions
【速读】： 该论文旨在解决生成式模型在设计新型无机材料时面临的挑战，核心问题在于无机化合物在广阔化学组成与结构空间中表现出的高度复杂性与多样性，导致传统生成方法在准确性与效率方面存在显著局限。其解决方案的关键在于提出一种基于领域特定对称性感知的表示方法，通过引入考虑晶格对称性的新颖填充技术（Wyckoff位置长度感知填充），优化了无机材料的编码过程。该方法利用晶体对称性信息增强编码表征，提升了深度学习模型对复杂无机结构的建模能力，从而实现更稳定、高精度且计算高效的生成结果。此外，研究构建了一个端到端的自动化系统，融合生成模型与稳定性分析，能够从初始数据直接生成训练数据中未见但稳定的新型无机材料，显著推动了下一代无机材料的自动化设计与发现进程。实验表明，该方法在质子导体数据集上重建准确率提升5.3%，在perov-5数据集上生成的新型稳定无机材料数量比基线模型高出63.5%。
链接: https://arxiv.org/abs/2605.30743
作者: Thang Dang,Haderbache Amir,Tzanakakis Alexandros,Yoshimoto Yuta
机构: 未知
类目: Materials Science (cond-mat.mtrl-sci); Computational Engineering, Finance, and Science (cs.CE); Computation and Language (cs.CL)
备注:
Abstract:Designing novel inorganic materials through generative models remains an important challenge for material science, driven by the complexity and diversity of inorganic structures across expansive chemical compositions and structural landscape. The vast combinatorial space of inorganic compounds demands innovative, AI-driven approaches to overcome limitations in generative accuracy and efficiency. To address this, we introduce a novel method that redefines the encoding and generation of inorganic materials by utilizing domain-specific symmetry-aware representation. Our approach not only refines the representation of intricate inorganic structures but also contributes to the field of material discovery by enhancing the precision and stability of generated candidates. Central to our methodology is a novel padding technique that exploits crystal symmetry information to enhance the encoding process. By integrating Wyckoff position length-aware padding into an encoder architecture, we achieve a more robust informed representation of inorganic materials. This symmetry-driven enhancement improves deep learning models to generate stable, previously unexplored inorganic structures with superior accuracy and computational efficiency. Furthermore, we introduce an end-to-end system that leverages the machine learning potential models to seamlessly generate novel, even those unseen in the training data, and stable inorganic materials from initial data to validated output. This pipeline integrates advanced generative models with stability analysis, marking a significant leap forward in the automated exploration and design of next-generation inorganic materials. Our method improved reconstruction accuracy 5.3% in proton conductor data, and generated 63.5% more novel stable inorganic material to baseline model on the perov-5 dataset.
[NLP-140] Extracting accent features in spoken Brazilian Portuguese without sociolinguistic labels
【速读】： 该论文旨在解决巴西葡萄牙语（pt-BR）中区域口音分类任务因依赖可靠标注数据而面临的挑战。现有大规模自监督学习（SSL）语音模型在训练过程中往往弱化社会语音学信息，因其训练目标通常不包含或无法有效利用口音标签。为克服这一问题，本文提出一种新颖的特征提取工作流，仅依赖声学标签进行建模。通过识别明确的区域口音地标，并结合基于音素的强制对齐工具（ZIPA），所构建的靶向特征集能够更有效地捕捉方言差异，其表现优于传统的句子级嵌入（utterance embeddings）。研究表明，在仅使用少量且客观的声学标签情况下，局部化特征可超越通用架构在口音相关任务中的性能，其核心解决方案在于通过音素级强制对齐实现对口音特征的精准定位与提取。
链接: https://arxiv.org/abs/2605.30457
作者: Pedro H. L. Leite,Pedro Benevenuto Valadares,Luiz W. P. Biscainho
机构: PEE/COPPE, UFRJ (里约热内卢联邦大学); Faculdade de Engenharia Elétrica e Computação (FEEC), UNICAMP (Campinas联邦大学); DEL/Poli, PEE/COPPE, UFRJ (里约热内卢联邦大学)
类目: Audio and Speech Processing (eess.AS); Computation and Language (cs.CL)
备注: This work was submitted to the XLIV Brazilian Symposium on Telecommunications and Signal Processing (SBrT 2026)
Abstract:Regional accent classification in Brazilian Portuguese (pt-BR) suffers from the need for reliable labeling. While large self-supervised learning (SSL) speech models are powerful, their training pipelines dilute sociophonetic information, since accent labels are generally not reliable or are not used in training objectives. This work introduces a novel workflow for feature extraction using only acoustic labels. By isolating explicit regional accent landmarks and using a phoneme-based forced aligner (ZIPA), our targeted feature set captures dialectal variance more effectively than utterance embeddings, demonstrating that localized features can outperform general-purpose architectures on accent-related tasks using minimal and objective data labels.
信息检索
[IR-0] SPECTRA: Synthetic IR Test Collections with Relevance Oracles and Controlled Distractor Diagnostics
链接: https://arxiv.org/abs/2605.31575
作者: Eric Liang
类目: Information Retrieval (cs.IR); Artificial Intelligence (cs.AI)
备注:
Abstract:Scalable information retrieval testing needs corpora that are large enough to stress index construction, ranking latency, query routing, and evaluation tooling, yet human-judged test collections remain expensive and may be unavailable when documents are private or still under design. This paper introduces SPECTRA, a reproducible framework for generating synthetic text corpora and retrieval test collections through a separation of latent topical structure, surface text realization, metadata controls, query intent generation, and deterministic relevance oracles. The framework is intended as a diagnostic complement to Cranfield-style and TREC-style evaluation, not as a replacement for human assessment. A single-process Python prototype generated corpora up to 60,000 documents and 9.61 million tokens while preserving controllable long-tail vocabulary growth and producing graded relevance labels for 96 queries. In the local simulation study, generation remained close to linear at roughly 12K to 14K documents per second, estimated Zipf slopes stayed near 0.86 in absolute value, and increasing cross-topic distractor text reduced BM25 nDCG@10 from 1.00 at 2% distractors to 0.43 at 36% distractors. These results show that lightweight synthetic corpora can expose retrieval-system scaling and failure modes before costly collection construction begins.
[IR-1] Effects of Vertex Merging Splitting on Large Coauthorship Networks: A Counterfactual Analysis
链接: https://arxiv.org/abs/2605.31555
作者: Jinseok Kim
类目: Digital Libraries (cs.DL); Information Retrieval (cs.IR); Social and Information Networks (cs.SI)
备注: 12 pages, 3 figures, 2 tables, ComplexNetworks2025
Abstract:Researchers analyze coauthorship networks, but author name ambiguity in their network data remains a significant challenge as it can change the number of vertices, distorting network properties. Although many scholars use straightforward heuristics for author name disambiguation using author’s forename initials, these techniques can skew our understanding of network properties by merging or splitting vertices, raising concerns about the reliability and validity of these methods. This study investigates how different levels of vertex merging and splitting errors that are induced by name ambiguity impact network measures, using three large coauthorship networks with highly accurate algorithmic author name disambiguation. As a counterfactual scenario, two initial-based disambiguation methods widely used in coauthorship network research were applied to these datasets. Nine coauthorship network metrics were computed while varying randomly the numbers of merged or split vertices. Results show that initial-based disambiguation generates coauthorship networks with specific network properties underestimated, leading to the discovery of coauthorship networks that are smaller and more closely connected than they genuinely are. In contrast, other network metric values increase, making authors appear more collaborative and embedded within less fragmented research communities than they are. The study emphasizes the importance of careful disambiguation of vertex names in analyzing coauthorship networks for rigorous and valid findings.
[IR-2] Evaluating Factual Density in Multi-Source RAG : A Study in Medical AI Accuracy
链接: https://arxiv.org/abs/2605.31506
作者: Michael R. DeMarco
类目: Information Retrieval (cs.IR); Computation and Language (cs.CL)
备注: 15 pages, 7 tables. Preliminary findings; Experiment 3 identified as future work
Abstract:Retrieval-Augmented Generation (RAG) is the current industry standard for grounding AI in real-world facts. Traditional retrieval methods rely on keyword matching and topic proximity, ranking content based on how closely it sounds like the user’s query. What they do not measure is how many verified facts the content actually contains. This structural gap, termed the Expert Blindness Effect, causes standard RAG pipelines to consistently bury high-density factual evidence in favor of lexically dominant text on the same topic. To address this gap, this paper introduces Factual Density (FD*), a novel retrieval optimization signal that measures the proportion of verified atomic claims relative to total token count. Using the NexusAgentics Ghost Audit preprocessing pipeline, raw text is scored for factual specificity using probabilistic factuality analysis to filter content before corpus ingestion. An initial formulation introduced a severe document-length confound (Pearson R = -0.8636, p = 2.27e-07). Implementing Z-score normalization within length bins resolved this bias, validating FD* as a length-independent density signal (p = 0.0749). Evaluated against the HealthFC benchmark (750 health claims labeled Supported, Refuted, or No Evidence by medical experts), FD*-optimized retrieval was the only condition to achieve 100% systematic review saturation in top-5 results, surfacing Cochrane evidence that standard cosine similarity ranked outside the top ten. Ground truth verification confirmed 25 mappings across seven HealthFC-supported claims. While full statistical validation across n=50 queries remains future work due to constraints on corpus-benchmark alignment, these findings establish factual density reranking as a low-cost, high-impact intervention for improving factual precision in health RAG architectures.
[IR-3] Beyond Instance-Level Alignment and Uniformity: Semantic Factor Learning for Collaborative Filtering KDD2026
链接: https://arxiv.org/abs/2605.31414
作者: Yajie Yu,Chenzhong Bin,Zhoubo Xu,Zhixin Zeng,Tongxin Xu,Cihan Xia,Jiafeng Wu
类目: Information Retrieval (cs.IR)
备注: Accepted by KDD 2026
Abstract:Collaborative filtering (CF) is widely used in recommender systems (RecSys) due to its simplicity and efficiency. However, existing CF methods follow an instance-level learning paradigm. During the instance learning stage, a large number of uninteracted user-item instances, of which items are potential interested by the user, are incorrectly treated as true negative samples resulting in a severe limitation to the generalization and scalability of models. Moreover, mainstream graph convolutional networks (GCNs) inherently suffer from high computational cost and over-smoothing issues, which limit the ability in capturing higher-order connectivity and lead to a poor generalization under sparse supervision signals. To address the above limitations, we propose Semantic Factor enhanced Alignment and Uniformity (SaFeAU), a novel framework that augments interacted instances with semantic factors, thereby mitigating false negative labeling and enabling matrix factorization (MF) to capture high-order CF signals without graph neighborhood aggregation. Specifically, SaFeAU consists of three tightly coupled components. First, Semantic Factor Routing (SFR) disentangles item representations into independent and global semantic factors. Building on these factors, Semantic Factor Matching (SFM) identifies uninteracted items, which share the same semantic factors with interacted ones, as potential positive pairs for enriching sparse supervision signals. Finally, Semantic Pairs Alignment (SPA) aligns both observed and potential positive pairs while promoting uniformity of user and item representations. Extensive experiments on four sparse real-world datasets show that SaFeAU consistently outperforms GCN-based and MF-based state-of-the-art CF methods in both recommendation accuracy and computational efficiency, confirming the effectiveness of the proposed semantic enhanced learning paradigm.
[IR-4] DynaTree: Dynamic Agent ic Retrieval Tree for Time-Sensitive News Retrieval
链接: https://arxiv.org/abs/2605.31377
作者: Siyuan Qi,Xinyuan Wang,Yingxuan Yang,Haochuan Guo,Jianghao Lin,Weiwen Liu,Yong Yu,Weinan Zhang
类目: Information Retrieval (cs.IR); Artificial Intelligence (cs.AI)
备注:
Abstract:Agentic Retrieval-Augmented Generation improves retrieval by integrating planning, tool use, and iterative reasoning, but existing agentic RAG methods often couple semantic expansion with retrieval decisions in short-horizon inference loops, leading to high inference cost and limited suitability for time-sensitive news retrieval. We propose DynaTree, a two-stage framework for efficient and adaptive news retrieval. In the offline stage, DynaTree uses coordinated agents to construct a reusable retrieval tree that materializes the semantic space of a query topic. In the online stage, DynaTree performs lightweight daily subtree selection over a time-localized evaluation proxy, without further agentic reasoning, tree modification, or retraining. Experiments on a multi-day Syft news benchmark and multiple BEIR datasets show that DynaTree achieves strong recall and ranking performance, consistently outperforming standard RAG and prior agentic baselines. We further deploy DynaTree in the Syft production system and evaluate it through online A/B testing from Jan. 28 to Feb. 6, 2026. The dynamically adapted variant improves survival rate from 0.32-0.53 to 0.59-0.73 over a fixed offline-selected subtree and outperforms existing production recallers on every evaluation day. These results show that persistent, structure-aware semantic expansion can translate offline agentic reasoning into practical improvements in coverage, freshness, and relevance for real-world news retrieval.
[IR-5] Latent Space Disentanglement via Activation Steering for Interpretable Attribute Control in Symbolic Music Generation
链接: https://arxiv.org/abs/2605.31295
作者: Ioannis Prokopiou,Pantelis Vikatos,Maximos Kaliakatsos-Papakostas,Theodoros Giannakopoulos,Themos Stafylakis
类目: ound (cs.SD); Artificial Intelligence (cs.AI); Information Retrieval (cs.IR); Machine Learning (cs.LG)
备注: Accepted at EUSIPCO 2026 (34th European Signal Processing Conference), 5 pages, 2 figures
Abstract:Transformer-based architectures have significantly advanced the generation of complex symbolic sequences, yet a significant gap remains in achieving fine-grained, interpretable control over discrete signal attributes. This paper investigates the mechanistic interpretability of the Multitrack Music Transformer (MMT) and proposes a framework for deterministic attribute modulation without retraining to bridge this gap via inference-time activation steering. Utilizing the Difference-in-Means (DiffMean) methodology, we isolate latent directions for signal attributes, specifically Pitch and Duration, within the residual stream. We validate the Linear Representation Hypothesis in this domain, achieving high correlation between steering magnitude and attribute shift. To address the inherent feature entanglement in multi-attribute steering, we introduce a Dual Steering framework utilizing Gram-Schmidt Orthogonalization. Experimental results demonstrate that this geometric decoupling reduces conceptual interference and signal degradation compared to naive vector addition, enabling independent deterministic control even against strong autoregressive conditioning.
[IR-6] Contextual Scalarisation Thompson Sampling for multi-objective decisions in public media ICPR2026
链接: https://arxiv.org/abs/2605.31291
作者: Théo Maëtz,Luc Guillet,Andrea Cavallaro
类目: Information Retrieval (cs.IR); Machine Learning (cs.LG)
备注: 15 pages, 3 figures, 3 tables. Submitted-manuscript version of a paper accepted at ICPR 2026. The Version of Record will be published in the Springer Lecture Notes in Computer Science series; DOI will be added when available
Abstract:Recommender systems may operate under multiple, competing objectives. For example, audience reach, cultural values, public service mandate, and operational constraints must be balanced in editorial decisions of public service media. Existing approaches relying on fixed combinations of objectives or Pareto-based optimisation do not adapt to changing priorities across situations. In this paper, we propose Contextual Scalarisation Thompson Sampler (CSTS), a multi-objective contextual bandit method that learns to weight objectives as a function of the observed context. We evaluate CSTS on real programming data from Radio Télévision Suisse, the Swiss national broadcaster, showing improved contextual relevance and better alignment with expert curation practices compared to fixed weight and standard contextual bandit approaches.
[IR-7] MIMO: Multilingual Information Retrieval via Monolingual Objectives
链接: https://arxiv.org/abs/2605.31171
作者: Youngjoon Jang,Seongtae Hong,Heuiseok Lim
类目: Information Retrieval (cs.IR); Artificial Intelligence (cs.AI)
备注:
Abstract:Multilingual Information Retrieval (MLIR) reflects real-world search environments in which queries and relevant documents may appear in different languages within a mixed-language corpus. However, existing embedding models are primarily optimized for Multi-Monolingual retrieval and their performance often degrades in MLIR settings. Moreover, directly applying conventional contrastive learning to MLIR can exacerbate language clustering and expose a trade-off between cross-lingual alignment and embedding uniformity. To address these limitations, we propose MIMO: Multilingual Information Retrieval via Monolingual Objectives, a two-stage framework that uses a stable English semantic space from a high-performing teacher model as an anchor. MIMO first initializes the student model’s cross-lingual alignment through knowledge distillation, and then jointly optimizes distillation and cross-lingual contrastive learning to improve retrieval discrimination while preserving alignment. Extensive experiments show that MIMO consistently outperforms existing cross-lingual training baselines across various MLIR and Multi-Monolingual benchmarks. MIMO also remains competitive with off-the-shelf models of similar or larger parameter scales. Furthermore, our cross-lingual Alignment-Uniformity analysis clarifies the distinct roles of the two loss components and shows that their combination yields a favorable trade-off between alignment and uniformity.
[IR-8] Vector Linking via Cross-Model Local Isometric Consistency ICML2026
链接: https://arxiv.org/abs/2605.31100
作者: Ziying Chen,Yang Cao,He Sun,Beining Yang,Tianjian Yang
类目: Artificial Intelligence (cs.AI); Databases (cs.DB); Information Retrieval (cs.IR)
备注: Accepted at ICML 2026
Abstract:We study Vector Linking: given two embedding clouds produced by different black-box encoders over partially overlapping datasets, recover cross-model object correspondences using only vectors. Empirically and theoretically, we show that independently trained contrastive encoders exhibit local geometric consistency: short-range distances are approximately preserved up to a scale factor, while long-range distances are not due to model-specific distortion. Building on this, we propose an iterative, reference-based geometric embedding hashing that recovers vector links from a tiny seed set of paired anchors. It represents each vector by distances to sampled paired anchors, proposes candidate links via hash-space matching, and aggregates evidence across views in a Beta-Bernoulli posterior to bootstrap high-confidence links as new anchors. Experiments across multiple benchmarks and embedding model pairs demonstrate accurate and robust linking under varying overlap, seed budgets, and out-of-domain anchors, with applications to vector database integration and cross-model clustering. Code is available at this https URL.
[IR-9] Beyond Static Dialogues: Benchmarking Realistic Heterogeneous and Evolving Long-Term Memory
链接: https://arxiv.org/abs/2605.31086
作者: Han Zhang,Zihao Tang,Xin Yu,Xiao Liu,Yeyun Gong,Haizhen Huang,Yan Lu,Weiwei Deng,Feng Sun,Qi Zhang,Hanfang Yang
类目: Computation and Language (cs.CL); Information Retrieval (cs.IR)
备注:
Abstract:In existing memory benchmarks for Large Language Models (LLMs), the evaluated dialogue sessions often lack long-term semantic consistency, and the underlying personas tend to be flat and static. Furthermore, in real-world scenarios, interactions between users and assistants involve more diverse, heterogeneous data streams, such as documents and emails. These shortcomings significantly limit the realism and effectiveness of current evaluations. To address these limitations, we introduce RHELM (Realistic, Heterogeneous, and Evolving Long-term Memory). Driven by meticulously crafted user profiles and a novel LOOP (pLan-rOllout-evOlve-Prune) module, we construct realistic dialogues across diverse interaction scenarios that exhibit dynamic temporal evolution and long-term coherence. Crucially, these dialogues are deeply integrated with heterogeneous external sources synchronized with the user’s temporal event trajectory. The resulting benchmark encompasses challenging question-answer pairs spanning seven inquiry types, with each question mapping to at least one of 27 critical memory characteristics that we identify as essential yet underexplored in current research. Comprehensive experiments across full-context models, retrieval-augmented generation (RAG) methods, and representative memory frameworks reveal that contemporary approaches still expose critical weaknesses in complex, real-world settings, particularly in resolving multi-source aggregation and real-world contextual reasoning.
[IR-10] Fighting Numerical Hallucinations via Data-centric Compilation for Online Financial QA KDD2026
链接: https://arxiv.org/abs/2605.31064
作者: Hao Chen,Xing Tang,Qirui Liu,Weijie Shi,Shiwei Li,Fuyuan Lyu,Weihong Luo,Xiku Du,Xiuqiang He
类目: Information Retrieval (cs.IR); Artificial Intelligence (cs.AI)
备注: Accepted by KDD 2026 ADS track
Abstract:Large Language Models (LLMs) have significantly advanced online data services, particularly in the domain of financial question answering (FinQA). However, such systems remain susceptible to numerical reasoning hallucinations, which critically undermine reliability in high-stakes financial applications. Although retrieval-augmented generation (RAG) has been widely adopted to ground responses in external knowledge, it introduces three persistent challenges: noise sensitivity, calculation fragility, and an auditability crisis. Existing model-centric approaches, which primarily focus on optimizing either the retriever or generator in isolation, still struggle to address these issues in an integrated manner. In this work, we pioneer a data-centric paradigm and propose a novel framework, the Data-centric Reasoning Compiler (DCRC). The framework operates through three cohesive phases: (1) adversarial data construction, which synthesizes training examples with controlled noise to teach robustness; (2) multi-stage training that cultivates a Data-centric Structuring Agent (DSA) capable of explicit evidence auditing and program synthesis; and (3) a compile-and-execute inference process, where the DSA transforms user queries and retrieved documents into verifiable, executable reasoning programs. This data-driven framework ensures faithful numerical reasoning by design. We conduct extensive experiments on established offline benchmarks and further validate our framework through deployment in a real-world online financial QA system.
[IR-11] Graph-GRPO: Dependency-Aware Credit Assignment for Generative E-commerce Search Relevance CIKM2026
链接: https://arxiv.org/abs/2605.31003
作者: Jiarui Che,Yifei Chen,Zhixing Tian,Chenyang Wang,Ziguang Cheng
类目: Information Retrieval (cs.IR)
备注: 11 pages, 2 figures, 2 tables. Submitted to CIKM 2026
Abstract:Search relevance modeling is a core task in e-commerce search systems, assessing how well a user query matches candidate products. Rather than relying on a single holistic matching signal, relevance judgment often requires structured reasoning over query understanding, product understanding, and facet-level matching. With large language models (LLMs), this process is increasingly formulated as chain-of-thought (CoT) reasoning and optimized with reinforcement learning (RL). However, existing RL methods mainly rely on outcome-level rewards and treat the entire reasoning chain as a single optimization unit. This makes it difficult to distinguish faulty reasoning steps from correct intermediate ones, leading to misaligned credit assignment. Although process-reward methods provide denser supervision, they often treat reasoning steps independently and ignore dependency-driven error propagation, making responsibility attribution difficult and limiting the optimization of structured relevance reasoning. We propose Graph-GRPO, a graph-structured extension of GRPO for multi-component relevance reasoning. Graph-GRPO constructs a relevance reasoning dependency graph, where CoT steps are modeled as nodes and their logical dependencies as edges. It propagates outcome-level rewards over the graph to derive step-level credit signals, enabling more accurate fine-grained credit assignment. We further introduce a main-loss-driven controller that adaptively adjusts edge-wise credit-propagation coefficients. Together with CoT random masking for supervised policy initialization and graph-node-based multi-head distillation, we build a trainable and deployable framework for generative relevance modeling. Extensive offline evaluations and online A/B tests on a leading e-commerce platform demonstrate that the Graph-GRPO-based framework improves relevance classification metrics and key engagement metrics.
[IR-12] Reading Between the Citations: A Typed Claim Network for Scientific Literature
链接: https://arxiv.org/abs/2605.30966
作者: Ning Ding,Sergio J. Rodríguez Méndez,Pouya G. Omran
类目: Information Retrieval (cs.IR); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注:
Abstract:Knowledge graphs over corpora of inter-referencing documents - scholarly papers, legal opinions, policy briefs - encode the topology of reference but not its stance. The standard representation collapses a rich evaluative relation into an untyped edge, losing the very content that supports community-level queries about how one document is received by another. We propose the claim network: a representational pattern in which each cross-document reference is reified as a typed claim, carrying source, target, claim text, and a four-class stance label grounded in the citation-intent literature. We give a construction pipeline applicable to any corpus of scholarly inter-referencing documents and instantiate it on a corpus of 127 papers in 3D point cloud semantic segmentation, producing a network of 8,260 typed claims. Three downstream task families demonstrate what the network enables: retrieval signal augmentation, aggregated-stance summarisation, and topological analytics. Head-to-head evaluation against standard Retrieval-Augmented Generation (RAG) baselines shows that the gain over flat retrieval is the gain from the right intermediate representation rather than the wrong one.
[IR-13] Inference-Free Multimodal Learned Sparse Retrieval for Production-Scale Visual Document Search
链接: https://arxiv.org/abs/2605.30917
作者: Gyu-Hwung Cho(1 and 2),Youngjune Lee(1),Kiyoon Jeong(1),Siyoung Lee(1),Sanggyu Han(1),Hervé Dejean(3),Stéphane Clinchant(3),Seung-won Hwang(2) ((1) NAVER Corp., Republic of Korea, (2) Seoul National University, Republic of Korea, (3) Naver Labs Europe, France)
类目: Information Retrieval (cs.IR); Computer Vision and Pattern Recognition (cs.CV)
备注: 12 pages, 5 figures, 12 tables, preprint
Abstract:As large-scale visual-document corpora such as arXiv papers and enterprise PDFs continue to grow, visual-document retrieval has gained increasing attention; yet it still lacks a deployable system that lexically indexes visual documents to serve queries without neural encoding at scale. Existing methods either achieve strong retrieval quality with VLM-based dense or multi-vector models but require neural query encoding at serving time, or avoid query encoding with OCR- or caption-based BM25 at the cost of time-consuming text extraction or generation. To fill this missing serving regime, we present V-SPLADE, an inference-free sparse retriever for visual-document retrieval. However, such inference-free multimodal learned sparse retrieval systems remain underexplored and have not yet shown dense-level effectiveness under high sparsity. We attribute this limitation to a lexical grounding problem: visual sparse representations often fail to capture the lexical content embedded in document images. To address this problem, we introduce caption-gated token supervision, a training-only signal that uses VLM-generated captions as lexical cues to activate retrieval-relevant vocabulary dimensions. With this supervision, V-SPLADE improves average NDCG@5 across six visual-document retrieval benchmarks by +13.8pp over the same-scale dense baseline and by up to +6.3pp over OCR- or caption-based BM25 baselines. On an 18.7M-document corpus, it more than doubles R@5 over the same-scale dense baseline and further improves competing retrievers through score fusion by up to +2.4pp R@5. Code will be released soon at this https URL.
[IR-14] On the impact of retrieved content representations in RAG Pipelines ACL
链接: https://arxiv.org/abs/2605.30790
作者: Jonathan J Ross,Bevan Koopman,Anton van der Vegt,Guido Zuccon
类目: Information Retrieval (cs.IR); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
备注: 23 pages, 15 figures, submitted to ACL May 2026 ARR
Abstract:Retrieval-Augmented Generation (RAG) supplements a language model’s input with retrieved documents, yet most RAG pipelines inherit retrieval components designed for human readers. How retrieved content should be represented when the consumer is a large language model (LLM) rather than a human is less well understood. Recent work has proposed transformations of retrieved content and identified properties that affect generation, but each examines a single transformation or property in isolation, leaving open which features of a document’s representation matter most. We address this with a controlled comparison: holding retrieval fixed, we vary only the representation of retrieved documents, comparing an original baseline against thirteen transformations spanning selection, summarisation, and reformulation, in query-dependent and query-independent variants. Across these fourteen representations we measure question-answering accuracy for four generators, and for each representation we also measure answer retention: whether a known answer-bearing document still supports its answer after transformation. We find that answer retention is the primary determinant of generator accuracy; notably, when retention is high, a representation’s wording, structure, length, and query-dependence have limited effect. This suggests that accuracy gains attributed to specific mechanisms in prior work may be partly explained by how well those mechanisms preserve answer-bearing content, an attribution that cannot be settled without controlling for retention.
[IR-15] FOSTER: First-order Dataset Distillation for Text-based Sequential Recommendation
链接: https://arxiv.org/abs/2605.30772
作者: Hung Vinh Tran,Tong Chen,Xinyi Gao,Junliang Yu,Julien Monteil,Hongzhi Yin
类目: Information Retrieval (cs.IR)
备注:
Abstract:Text-based sequential recommender systems, while greatly improving recommendation accuracy by incorporating item contexts, are undeniably more expensive to train. By condensing a large dataset into a compact set of synthetic samples for model training, dataset distillation offers a promising solution. However, its adoption in text-based sequential recommendation is non-trivial given the large pool of discrete items. This challenge is further compounded by language model-based item encoding, which makes bi-level optimization commonly used in dataset distillation prohibitively expensive. To this end, we propose First-order dataset distillation for Text-based Sequential Recommendation (FOSTER), which facilitates effectiveness and efficiency via three novel components: (1) stochastic item subset sampling that replaces costly full-corpus embedding extraction at each distillation step; (2) first-order optimization with trajectory-anchored parameter reset to avoid expensive bi-level gradient computation; and (3) regularization that explicitly promotes co-occurrence between semantically similar items in the synthetic sequences. Extensive experiments on three benchmarks show that FOSTER consistently outperforms existing dataset distillation and coreset selection baselines, approximating full-dataset performance using as few as 20 synthetic interaction sequences.
[IR-16] SemStruct: Contextualizing Semantic Embeddings with Structural Information for Schema Matching KDD26
链接: https://arxiv.org/abs/2605.30729
作者: Inwon Kang,Kavitha Srinivas,Nandana Mihindukulasooriya,Sola Shirai,Parikshit Ram,Horst Samulowitz,Oshani Seneviratne
类目: Machine Learning (cs.LG); Information Retrieval (cs.IR)
备注: Accepted to KDD 26 Research Track
Abstract:Schema matching is a fundamental step in integrating heterogeneous data sources. While Pre-trained Language Models (PLMs) have revolutionized this task by capturing linguistic semantics, they typically process tabular data as serialized text sequences of standalone column descriptions. This serialization discards critical structural information – specifically, the row-level co-occurrences, i.e. the relational context – forcing models to rely solely on column header semantics or standalone distributions. To bridge this gap, we propose SemStruct, a framework that joins the semantic power of frozen PLMs with the structural inductive bias of Graph Neural Networks (GNNs). We model the table as a heterogeneous graph where columns and values are nodes connected by rows, allowing the GNN to propagate disambiguating context across the structure. Unlike other state-of-the-art methods that require proprietary LLM access and fine-tuning of language models, SemStruct keeps the language model frozen and trains only a lightweight structural encoder. Extensive experiments on the Valentine and SOTAB-SM benchmarks demonstrate that SemStruct achieves state-of-the-art performance, outperforming fully fine-tuned baselines on complex, semantically joinable datasets. Furthermore, our ablation studies reveal that row representations serve primarily as topological conduits rather than semantic entities, validating the necessity of explicit structural modeling in schema matching.
[IR-17] An Organization-Scoped LLM Agent Runtime Architecture for Regulated Cybersecurity Operations
链接: https://arxiv.org/abs/2605.30604
作者: George Fatouros,Georgios Makridis,George Kousiouris,John Soldatos,Dimosthenis Kyriazis
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Information Retrieval (cs.IR)
备注: 8 pages, 3 figures
Abstract:Regulated cybersecurity workflows lack a runtime substrate that enforces organization-level scope across retrieval, tool calls, memory, findings, reports, and audit while remaining model-agnostic and locally deployable. Recent large language model (LLM) agent systems report strong results on isolated cybersecurity tasks, yet they do not by themselves define an auditable platform architecture for regulated security operations centre (SOC) and compliance workflows, where a single analyst may trigger actions that bind the organization, and where the runtime must integrate with existing SIEM/XDR stacks as a primary source of context and alert-driven triggers rather than operate as a standalone analytical layer. This paper proposes an organization-scoped LLM agent runtime architecture for financial cybersecurity. The contribution is a typed Security Context that is created at every entry point, including SIEM/XDR notifications ingested as first-class triggers, and enforced at every component boundary, combined with a shared Runtime Core, logical specialist subagents, a governed Tool Adapter Layer exposing SIEM/XDR query, enrichment, and response primitives under uniform policy and audit, structured findings with evidence references, tiered human-in-the-loop (HITL) gates, and append-only audit. Model Context Protocol (MCP), extended telemetry, digital twins for pentesting, graph retrieval, and federated knowledge sharing are treated as optional extension paths rather than mandatory runtime assumptions. We describe an implementable slice as the architecture’s testability surface, and we propose a falsifiable evaluation plan with metric-level pass criteria for architecture readiness, security-policy enforcement, evidence traceability, output quality, and operational observability.
[IR-18] Exploring Autonomous Agent ic Data Engineering for Model Specialization
链接: https://arxiv.org/abs/2605.30407
作者: Yujie Luo,Xiangyuan Ru,Jingsheng Zheng,Jingjing Wang,Yuqi Zhu,Jintian Zhang,Runnan Fang,Kewei Xu,Ye Liu,Zheng Wei,Jiang Bian,Zang Li,Shumin Deng
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Information Retrieval (cs.IR); Machine Learning (cs.LG)
备注: Work in progress
Abstract:Large Language Models (LLMs) have demonstrated strong performance on general tasks, while often struggling to adapt to specialized domains without high-quality domain-specific data. Existing LLM-based data curation methods primarily rely on human-designed workflows, leaving it unexamined whether LLMs can autonomously execute an end-to-end data engineering pipeline for model specialization. We formalize \textbfAutonomous Agentic Data Engineering, a novel task designed to evaluate LLMs as autonomous data engineers that drive model specialization through end-to-end data curation. We frame data as an optimizable component and study agents that plan, generate, and iteratively optimize training data across multiple domains, guided by post-training performance improvement. Experiments show that autonomous LLM data engineers yield substantial gains, as GPT-5.2 constructs a training curriculum that improves a student model by \textbf57.29%, entirely through iterative, agent-driven data adaptation. By illuminating both potential and bottlenecks, our study establishes autonomous data engineering as a measurable capability and charts a path toward agent-driven model specialization\footnoteCode will be released at this https URL…
[IR-19] When LLM Reward Design Fails: Diagnostic-Driven Refinement for Sparse Structured RL
链接: https://arxiv.org/abs/2605.28918
作者: Youting Wang,Yuan Tang,Bowen Liu,Xuan Liu,Dingyan Shang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Information Retrieval (cs.IR)
备注:
Abstract:For sparse, structured reinforcement-learning tasks with semantic reward-function interfaces, LLM-generated reward shaping is better framed as debugging than one-shot generation. We study PPO-trained agents using MiniGrid as core evaluation and MuJoCo as boundary stress test. Our audit finds two dominant one-shot failure modes – reward flooding and semantic/API misunderstanding – plus a rarer weak-shaping case. We propose diagnostic-driven iterative refinement, where training diagnostics and a failure-mode taxonomy guide targeted reward-function revision. Refinement improves DoorKey-8x8 from 2.3% to 97.6% and KeyCorridor from 31.2% to 86.7% with high seed-to-seed variance. Controls show these gains are not from retrying or extra training: metrics-only re-prompting yields large drops, while a static-vocabulary control recovers much of the gap (87.6%; 70.7%), showing the taxonomy prompt is a major mechanism and dynamic labels provide only partially isolated incremental evidence. Budget-matched and Best-of-3 comparisons separate refinement from selection and training-time effects. Component-removal tests, sensitivity analyses, and an audit against author labels provide converging evidence for the debugging interpretation while revealing calibration limits. Continuous-control results show the boundary: success-based diagnostics can misfire in dense-reward locomotion, and return-trend feedback removes one false-positive mechanism without robust gains. The low-call protocol is a cost contrast with population-based reward search, not a benchmark comparison. In four crossed-variance-design environments, point estimates suggest larger gains when LLM reward-function variance dominates but bootstrap intervals are wide. The method is bounded to sparse structured tasks with reliable interfaces under PPO; fields like event_text may help, hurt, or be neutral.
人机交互
[HC-0] Can Generative AI help people navigate Radical Moral Disagreements? The CONSIDER prototype
链接: https://arxiv.org/abs/2605.31574
作者: William Hohnen-Ford,Sarah Chen,Kathryn B. Francis,Madeline G. Reinecke,Ilina Singh,David Lyreskog
类目: Human-Computer Interaction (cs.HC)
备注: 25 pages, 1 figure, 2 tables. Submitted manuscript
Abstract:Radical Moral Disagreements (RMDs) are highly polarising topics that are increasingly censored in everyday life, with growing evidence suggesting that this polarisation carries measurable costs to public mental health. To address these challenges, some researchers have proposed Large Language Models (LLMs) as a means to support more democratic deliberation and better moral reasoning. Yet existing tools are poorly calibrated to help people navigate RMDs, because of their intense and divisive characteristics. This paper introduces CONSIDER, a prototype for a one-to-one AI tool for RMD navigation. Drawing on Mill’s account of the epistemic value of disagreement, CONSIDER aims at value clarification through structured disagreement with an opposing LLM-generated opinion. We describe CONSIDER’s design logic and analyse potential risks posed by such tools to guide future development.
[HC-1] Vision-Language Models Suppress Female Representations Under Ambiguous Input
链接: https://arxiv.org/abs/2605.31556
作者: Arnau Marin-Llobet,Simon Henniger,Mahzarin R. Banaji
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computers and Society (cs.CY); Human-Computer Interaction (cs.HC)
备注: 16 pages, 12 figures, 1 table
Abstract:Alignment teaches vision-language models (VLMs) to avoid expressing demographic biases, and when gender is clearly visible they largely succeed. Far less is known about ambiguous inputs (a worker in full gear, a figure seen from behind) cases common in practice yet rarely studied. We find that minimal prompting pressure exposes occupation-gender defaults when prompting ambiguous input images, with models collapsing to male even for strongly female-stereotyped occupations. But do these outputs reflect what models actually encode internally? We introduce LALS (Latent Association Leaning Score), a zero-shot metric that projects visual-token activations into the model’s text-embedding space to measure concept associations per token and layer. Across 15 occupations, over 800 gender-ambiguous images, and four VLMs, internal representations and outputs are systematically decoupled: models often encode a female association internally yet output male. Layer-wise analysis reveals an asymmetric filter – male signal amplifies end-to-end while female signal peaks mid-network and is suppressed before generation – and a color ablation shows that culturally loaded visual cues such as clothing color further modulate these internal associations.
[HC-2] ranslation Analytics for Freelancers II: Benchmarking Local LLM s for Confidential Translation Workflows
链接: https://arxiv.org/abs/2605.31452
作者: Yuri Balashov,Rex VanHorn,Mingxi Xu,Austin Downes
类目: Computation and Language (cs.CL); Human-Computer Interaction (cs.HC)
备注: 20 pages. Accepted at EAMT-2026 (Tilburg, Netherlands, June 2026)
Abstract:Building on our previous work, this paper develops practical, low-barrier methods for freelance translators and smaller language service providers to evaluate translation technologies using rigorous yet accessible analytic methods. Here we address a high-stakes, specialized need: offline translation for confidentiality-sensitive domains in which privacy constraints preclude the use of cloud-based engines and commercial LLMs. We expand the Reeve Foundation Trilingual Corpus (RFTC) used in our previous work into a multilingual corpus (RFMC) by adding sentence-aligned German and Simplified Chinese reference translations. We then benchmark several locally runnable language models (via Ollama) across four language directions on 1000+ sentences selected from this corpus. We use consistent single-prompt calls without fine-tuning or domain adaptation, comparing local LLM outputs against commercial NMTs (DeepL, Baidu), a frontier LLM (GPT-5.2), and professional-grade local NMT systems (OPUS-CAT, NeuralDesktop, Promt). Automatic evaluation is conducted with MATEO. Results reveal substantial variation in local LLM performance across language directions and model sizes. The best local LLMs match or surpass local NMT systems and a frontier LLM, though they remain behind top commercial NMTs. These findings underscore the viability of carefully selected local LLM translation for privacy-constrained professionals and inform future research on model scaling and multilingual capability.
[HC-3] oward Accessible Mobile Money: A Voice-Driven Biometrically Secured USSD Automation Framework for Visually Impaired Users
链接: https://arxiv.org/abs/2605.31375
作者: Sunday Ajayi,Babatunde Eric Olatunji,Eric Umuhoza
类目: Cryptography and Security (cs.CR); Human-Computer Interaction (cs.HC)
备注:
Abstract:Financial inclusion has expanded significantly across Africa through mobile money services delivered primarily via USSD technology. However, visually impaired individuals continue to face accessibility and security barriers when conducting financial transactions. Current USSD systems are not designed for non-visual interaction, forcing users to rely on third-party assistance even for PIN entry, thereby increasing fraud exposure and reducing transaction confidence. Although alternative assistive technologies such as screen readers exist, they are not compatible with USSD operations, often causing sessions to time out before the user can complete a transaction. This paper presents an Android-based intelligent middleware that automates USSD transactions, integrates biometric-secured PIN injection, and introduces a privacy-preserving screen-dimming mechanism: Blackout Mode. The system leverages Android Accessibility Services, hardware-backed Keystore security, and on-device natural language parsing to enable independent, secure voice-based mobile money access. We show that the proposed solution improves task success rates from 65-75% to more than 90% and reduces transaction completion time from 40-60 seconds to 12-15 seconds, while also improving perceived security.
[HC-4] Appropriateness of Empathy in AI: A Signal-Cost Perspective
链接: https://arxiv.org/abs/2605.31340
作者: Chi-Ching Juan,Tao Wang,Harold Lee
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI)
备注: Accepted by IEEE CASCON 2025
Abstract:The appropriateness of empathy in AI has emerged as a critical concern, as excessive empathy risks seeming manipulative while insufficient empathy appears dismissive. While prior research has explored how to quantify empathy in AI, few studies examine whether such empathy is contextually appropriate. This paper introduces an economic perspective by applying signaling theory to human-AI conversations. We propose Signal Cost Proxies (emotional richness, perspective-taking, and contextual tailoring) mapped to affective, cognitive, and associative empathy. This multidimensional framework enables systematic evaluation of empathy not just by presence, but by its appropriateness relative to user demand.
[HC-5] A Focus of Attention-Based Virtual Training Platform for Pre-Prosthetic Myoelectric Skill Acquisition: A Proof-of-Concept Study
链接: https://arxiv.org/abs/2605.31332
作者: Xiaochen Zhang,Sigrid Dupan
类目: Human-Computer Interaction (cs.HC)
备注: 7 pages, 6 figures. Accepted to the 48th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC 2026)
Abstract:Advances in myoelectric prosthetic technology have substantially increased the functional potential of modern devices. Accordingly, heightened control demands have led to the acknowledgement of pre-prosthetic training as a key stage in the acquisition of myoelectric skills. Existing training paradigms largely emphasize internal muscle activation while external, goal-directed outcomes required for effective real-world use are often neglected. We address this gap by introducing a virtual pre-prosthetic training platform that integrates EMG-driven cursor with animated hand gestures, enabling the delivery of both muscle-level and functional-level feedback. In this proof-of-concept study, participants were assigned to one of two focus of attention (FoA) protocols, each incorporating both feedback types but differing in whether internal or external FoA was emphasised. Participants successfully acquired and retained myoelectric skill across both protocols, but distinct performance characteristics and learning strategies emerged, indicating that both FoAs contribute meaningfully to learning and that their timing may play an important role. External FoA was positively associated with retention, suggesting that it may strengthen the link between training and skill acquisition. Together, the results demonstrate the feasibility of an FoA-based virtual training platform for pre-prosthetic applications and indicate that it can provide a foundation for designing training protocols that better prepare users for prosthetic use.
[HC-6] Neither Replacement nor Panacea: Comparing LLM -Based Conversational and Graphical Decision Support in Industrial Tasks
链接: https://arxiv.org/abs/2605.31287
作者: Roberto Figliè,Simone Caputo,Alan Serrano,Daria Mikhaylova,Tommaso Turchi,Daniele Mazzei
类目: Computers and Society (cs.CY); Artificial Intelligence (cs.AI); Human-Computer Interaction (cs.HC)
备注:
Abstract:Managers in manufacturing settings rely on digital interfaces to interpret operational data for decision-making, but growing data volume and complexity can make relevant insights difficult to identify efficiently. While dashboards remain dominant in industrial contexts, Large Language Model (LLM)-based conversational agents (CAs), accessed through conversational user interfaces (CUIs), may provide more direct access to such data. However, their effectiveness may depend on the information-processing demands of the task. This study compares an LLM-based CA delivered through a CUI with a dashboard in a manufacturing decision-support scenario. In a mixed factorial experiment with a 2x3 design, 134 industrial decision-makers were assigned to one interface condition and completed three tasks of increasing complexity. We examined perceived Mental Workload (MWL), decision accuracy, completion time, and intended reliance, and tested self-reported data literacy as a moderator. Results showed that the CUI reduced perceived MWL overall and supported faster completion in less demanding tasks, but both advantages diminished as task complexity increased. Neither interface produced a consistent overall advantage in decision accuracy, and the CUI was not preferred as a sole basis for subsequent decisions. Furthermore, data literacy did not reliably moderate interface effects. These findings indicate that conversational interaction offers conditional rather than universal benefits for industrial decision support. LLM-based CAs may reduce information-access effort, whereas complex decisions continue to benefit from persistent, inspectable visual representations.
[HC-7] Personalized to Persuade: The Effects of Contextualization and Warmth on Trust and Reliance in Conversational AI
链接: https://arxiv.org/abs/2605.31275
作者: Mert Yazan,Suzan Verberne,Frederik Bungaran Ishak Situmeang
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI)
备注:
Abstract:Artificial Intelligence (AI) agents personalize their responses by tailoring explanations to users’ backgrounds, interests, and prior interactions, referred to as contextualization. Personalization has been identified as a persuasive strategy in politics or in marketing. However, the persuasive effect of contextualization in everyday tasks, where users often lack prior knowledge, remains unclear. We conducted a 2\times2 between-subjects experiment ( N = 380 ) examining how contextualization, combined with conversational warmth, shapes reliance and persuasiveness of an AI assistant arguing against expert recommendations. Our findings reveal that contextualization reduces the persuasive power of AI, but its combination with warmth restores persuasiveness through a crossover interaction. Reliance on AI is present across conditions and is invariant to the conversational design. Trust strongly predicts both persuasion and reliance, yet neither contextualization nor warmth operates through trust. AI literacy decouples trust from behavior: more literate users report lower trust in the assistant, yet are more persuaded and more reliant on its advice. These results suggest that users are prone to deferring to AI agents over human expert judgment; however, interface-level conversational design choices have a limited role in shaping the behavior.
[HC-8] Comparing LLM -Based Conversational and Graphical Interfaces for Industrial Decision Tasks: An Exploratory Mixed-Methods Study
链接: https://arxiv.org/abs/2605.31224
作者: Roberto Figliè,Simone Caputo,Alan Serrano,Tommaso Turchi,Daniele Mazzei
类目: Computers and Society (cs.CY); Artificial Intelligence (cs.AI); Human-Computer Interaction (cs.HC)
备注:
Abstract:The use of Generative AI Conversational User Interfaces (CUI) as a new way to access and analyze data is growing in all sectors, and the industrial one is no exception. There, large amounts of data produced by IoT devices are flowing through user interfaces and may require them a new adaptation to the new analyses needs of decision-makers. LLM-based CUIs are promising a new way to directly interact with those data through the directness of natural language and without the learning costs that every GUI design has. Moreover, the capabilities of LLMs and their agency open up the possibility to automate some tasks and help with the reasoning during decision-making activities. But are this promises well founded? We try to scope this general question with a mixed-approach study comparing a state-of-the-art dashboard with a conversational agent. A total of 20 participants used both interfaces to complete four simulated industrial decision tasks of varying complexity. We combined measures of mental workload, completion time, and decision accuracy with a post-study questionnaire and semi-structured interviews analyzed through thematic analysis. The findings suggest that the conversational agent can reduce interactional effort by supporting more direct access to information, while the dashboard remains valuable for overview and verification. However, these benefits may vary across tasks and require validation through larger-scale studies.
[HC-9] Developing a UXR Point of View for Cognitive Accessibility in Mobile Learning with Generative AI
链接: https://arxiv.org/abs/2605.31149
作者: Fatima Ahmad Muazu,Festus Adedoyin,Huseyin Dogan,Abiodun Adedeji,Melike Akca,Olumuyiwa Ayorinde
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI)
备注:
Abstract:This study investigates how UX research (UXR) principles, combined with Large Language Model (LLM)-supported analysis, can be used to improve the quality of requirements for mobile learning systems designed for learners with cognitive disabilities. Using the UXR Point-of-View (PoV) pyramid as a methodological framework, the study progressed through four stages: foundational structuring of psychological, behavioral, and design layers; structured validation using the DeLone and McLean Information Systems Success Model and Quality Function Deployment (QFD); insight consolidation through the development of nine Cognitive Accessibility UXR Play Cards; and stakeholder-specific PoV articulation to support interdisciplinary communication. LLM-supported synthesis was integrated to assist in theme clustering, requirement refinement, and hypothesis formulation under human oversight. Findings suggest that many usability and engagement challenges in mobile learning originate from ambiguous or under-specified requirements rather than interface design alone. By embedding cognitive accessibility principles into measurable and technically traceable requirements, the proposed Cognitive Accessibility UXR Playbook provides a structured pathway for aligning theory, system architecture, and stakeholder strategy.
[HC-10] Developing a Culturally Grounded AI-Augmented UX Research Point of View (POV): An Exemplar Case Study from Telemedicine Dementia Care
链接: https://arxiv.org/abs/2605.31147
作者: Abiodun Adedeji,Huseyin Dogan,Festus Adedoyin,Michelle Heward,Melike Akca,Emmanuel Oluwatosin Oluokun,Fatima Ahmad Muhazu,Olumuyiwa Ayorinde
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI)
备注:
Abstract:User Experience Research (UXR) Points of View (POVs) distil complex and often fragmented research evidence into actionable perspectives that guide how teams interpret user needs, frame design decisions, and align stakeholders. Although POVs are widely used in industry practice, there are few published examples that explicitly document how POVs are constructed, particularly in culturally sensitive and low-resource contexts. This paper presents an exemplar case study demonstrating how a culturally grounded, AI-augmented UXR POV was developed to inform TeleDeCa, a telemedicine dementia care framework for family caregivers in Nigeria. Building on the UXR POV Playbook and pyramid framework, we illustrate how mixed-methods research, hypothesis generation, and ontology-based modelling can be combined to form a defensible POV without requiring a fully finalised system or validated outcomes. Generative AI (GenAI) is integrated across the UXR POV framework as a bounded research collaborator, supporting synthesis, hypothesis exploration, and narrative construction while preserving human judgment, ethical accountability, and cultural sensitivity. The contribution of this paper lies in the extraction of reusable Play Cards and a Play that extend the UXR POV Playbook and serve as exemplar material for the CHI 2026 workshop on developing AI-powered UXR POVs.
[HC-11] From Evidence to Design: Developing an AI-Augmented UX Research Point of View for Digital Wellbeing in Emergency and Public Safety Contexts
链接: https://arxiv.org/abs/2605.31146
作者: Olumuyiwa Ayorinde,Huseyin Dogan,Festus Adedoyin,Nan Jiang,Emmanuel Oluokun,Abiodun Adedeji,Melike Akca
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI)
备注:
Abstract:This paper investigates how User Experience Research (UXR) methods can be combined with AI-supported analysis to develop clearer design direction for digital wellbeing interventions targeting Emergency and Public Safety Personnel (EPSP). EPSP work in high-stress, shift-based environments where cognitive fatigue and unpredictable schedules reduce engagement with conventional wellbeing tools. Using the UXR Point-of-View (PoV) framework, this study applied an AI-supported literature analysis process to identify recurring psychological, behavioural, and design patterns. Behaviour Change Techniques and Persuasive Technology principles were integrated throughout interpretation to connect evidence with practical design reasoning. The process resulted in a UXR PoV Pyramid, nine UXR Play Cards, and stakeholder focused PoV narratives. Findings show that effective wellbeing systems for EPSP must minimise cognitive effort, adapt to operational context, and prioritise psychological safety. The work demonstrates how AI can assist large-scale evidence interpretation while human researchers maintain responsibility for contextual judgement and design direction.
[HC-12] Extending the UXR Point of View Pyramid: A Generative AI-Augmented Methodology for Human-Centred AI Systems
链接: https://arxiv.org/abs/2605.31143
作者: Festus Fatai Adedoyin,Huseyin Dogan,Melike Akca,Abiodun Adedeji
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI)
备注:
Abstract:Rising household debt and cost-of-living pressures in the United Kingdom have intensified the role of AI-driven financial technologies in mediating credit assessment, repayment structuring, and debt support services. These systems increasingly shape consequential financial decisions, yet they operate within complex socio-technical environments characterised by regulatory constraint, algorithmic opacity, and heightened vulnerability risk. User Experience Research (UXR) Points of View (PoVs) are critical in translating heterogeneous research evidence into strategic direction for product and governance decisions. However, the existing UXR PoV framework was not designed for AI-mediated financial systems where interpretability, fairness, and accountability are central. This paper extends the UXR PoV pyramid into an AI-augmented methodological framework for Human-Centred AI debt management technologies in the UK financial services context. We formalise (1) an AI-Augmented PoV Pyramid, (2) a structured prompt architecture for synthesis and hypothesis generation, and (3) an AI-enabled Playbook Card system that embeds Generative AI into UXR workflows while preserving traceability and ethical oversight. Generative AI is positioned not as an analytic authority, but as an epistemic support mechanism subject to human validation and regulatory awareness. By grounding the framework in debt management technologies, including affordability assessment, repayment planning, and financial stress prediction systems, this work advances UXR methodology for high-stakes financial AI environments and contributes to the evolution of responsible, AI-powered UXR practice within the CHI community.
[HC-13] Developing an AI-Powered UX Research Point of View for Digital Health in A Regulatory Context: An Exemplar Case from MSM and Transgender HIV Care in Nigeria
链接: https://arxiv.org/abs/2605.31138
作者: Emmanuel Oluwatosin Oluokun,Festus Fatai Adedoyin,Huseyin Dogan,Nan Jiang,Melike Akca,Abiodun Adedeji,Olumuyiwa Ayorinde,Fatima Ahmad Muazu
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI)
备注:
Abstract:User Experience Research (UXR) in a legal and regulatory contexts presents unique challenges that require specialised approaches to protect vulnerable populations whilst generating actionable insights. Digital consultation, appointment booking, and medication delivery platforms show promise for extending care access; however, their real-world effectiveness is curtailed by an absence of theoretically grounded user experience research (UXR) methodologies that adequately account for the psychosocial conditions of these populations. This paper introduces a Generative AI-augmented UXR methodology, grounded in the UXR Point of View (PoV) Playbook, to guide the design of psychologically safe, low-cognitive-load digital health interventions for MSM and transgender individuals living with HIV/AIDS in Nigeria. Drawing from empirical research involving co-design workshops, thematic analysis, and requirements engineering, the methodology is operationalised through a four-stage UXR process encompassing AI-supported hypothesis generation, foundational planning, insight generation via Building Blocks, and the construction of stakeholder-specific PoV narratives. This process results in ten theory-informed UXR Play Cards that translate psychological mechanisms and empirical findings into actionable design guidance. Each play contains actionable tasks, AI-augmented approaches, and ethical guardrails tailored for research with marginalised populations. The output is a set of ten theory-informed UXR Play Cards translating psychological insight and empirical evidence into actionable design guidance. The core contribution is a replicable, stigma-aware, and privacy-centred framework for responsible GenAI use in UXR practice, advancing human-centred digital health design for marginalised communities.
[HC-14] UXR PoV for Neuroinclusive Emotion Regulation
链接: https://arxiv.org/abs/2605.31131
作者: Melike Akca,Mona Giff,Deniz Cetinkaya,Huseyin Dogan,Stephen Giff
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI)
备注:
Abstract:Attention-deficit/hyperactivity disorder (ADHD) is a psychiatric disorder which presents itself in individuals through patterns of developmentally inappropriate levels of inattentiveness, hyperactivity, and impulsivity, with difficulties in decision making and emotional regulation (ER). Although digital and AI-based interventions have expanded access to ER support, many existing systems remain limited by weak theoretical integration, insufficient accommodation of neurodiversity, and a lack of structured user experience research (UXR) methodologies, that bridge psychological insight with design practice. This paper introduces a Generative AI-augmented UXR methodology, grounded in the UXR Point of View (PoV) Playbook, to support the design of emotionally intelligent and Neuroinclusive digital ER interventions for adults with ADHD. The approach integrates empirical evidence with established psychological frameworks Dialectical Behaviour Therapy (DBT), Self-Determination Theory (SDT), and the COM-B behavioural model and leverages Generative AI as a co-analytic tool to support synthesis, hypothesis formation, and design articulation. The methodology is operationalized through a four-stage UXR process encompassing AI-supported hypothesis generation, foundational planning, insight generation via Building Blocks, and the construction of stakeholder-specific PoV narratives. This process results in a set of ten theory informed UXR Play Cards that translate psychological mechanisms and empirical findings into actionable design guidance. The primary contribution of this work is a replicable, bias-aware framework for integrating Generative AI into UXR practice, advancing human-centred and Neuroinclusive approaches to digital mental health design.
[HC-15] Generative AI in developing User Experience Research Point of View: A NotebookLM case study
链接: https://arxiv.org/abs/2605.31125
作者: Mona Giff,Stephen Giff,Huseyin Dogan
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:User Experience Research (UXR) is currently undergoing a transition from traditional usability testing towards design-led and data-driven approaches, yet it faces an identity crisis due to a lack of methodological grounding in UXR and time-intensive methodologies which often lag behind product decision cycles. To address this, the UXR Point of View (PoV) framework formalises the UXR process by transitioning from raw data collection to forming an evidence-based PoV which drives strategic product impact. Furthermore, the use of GenAI in UXR has been investigated, but researchers often face increased work intensity when using GenAI, attributed to time spent on prompt engineering, data cleaning, and verification of AI outputs. This paper proposes and evaluates a formalised methodology for leveraging GenAI, specifically Google’s NotebookLM, to augment the UXR PoV process. The methodology consists of five prompts across four stages: (1) leveraging the framework, (2) establishing roadmaps, (3) applying best-practices, and (4) crafting PoV narratives; and was tested on eleven UXR papers. Results showed that by using the proposed methodology, NotebookLM successfully leveraged the UXR PoV framework across all stages of PoV creation. These findings demonstrate that NotebookLM can serve as an effective collaborative partner in UXR, so long as it is provided with sufficient context and specific prompting.
[HC-16] Extending the UXR Point of View Playbook: Triangulating Insights in Complex Developer Domains
链接: https://arxiv.org/abs/2605.31104
作者: Sarah Kianfar
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:As User Experience Research (UXR) matures, practitioners face the challenge of moving beyond data collection toward establishing a compelling Point of View (POV) that drives strategic impact. This paper proposes an extension to the UXR POV Playbook, specifically focusing on the transition from the “Insight Generation” layer to the “POV” layer. Drawing on extensive multi-method research in Cloud Developer Tools, spanning AI Agents, Command Line Interfaces (CLI), and Error Messages, we demonstrate how triangulating qualitative and quantitative data facilitates the creation of high-confidence POVs. We introduce three new “Playbook Cards” derived from this research: The Paradigm Shift, Explainability as Trust, and The Cost of Friction. These cards provide a structured mechanism for researchers to translate complex technical findings into irrefutable business narratives.
[HC-17] A Pilot Study on Curator-Guided Multilingual Art Description for Blind and Low-Vision Audiences with Small Vision-Language Models
链接: https://arxiv.org/abs/2605.31080
作者: Iosif Tsangko,Andreas Triantafyllopoulos,George Margetis,Ioana Crihana,Björn W. Schuller
类目: Multimedia (cs.MM); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV); Human-Computer Interaction (cs.HC)
备注: 7 pages, 2 figures, 3 tables. Preprint
Abstract:Blind and low-vision (BLV) audiences remain underserved by visual art descriptions, particularly across languages and in museum settings where privacy and intellectual-property constraints may favour small on-premise vision-language models (VLMs). This pilot study investigates curator-guided multilingual art description with Qwen2.5-VL-3B-Instruct for German, Romanian, and Serbian. We construct a parallel BLV-oriented caption corpus from artwork images and metadata, and compare language-specific LoRA adapters with a single multilingual adapter under a fixed backbone and training budget. Evaluation combines automatic lexical and embedding-based metrics with an LLM-as-Judge protocol calibrated against a small Romanian BLV pilot study. Under our pilot setup, language-specific adapters show more stable controllability and visually grounded description quality for Romanian and Serbian, while multilingual adaptation remains competitive in German. We frame these findings as deployment-oriented evidence for small on-premise VLMs, and highlight the need for larger BLV user studies and broader language coverage before drawing general conclusions about multilingual accessibility.
[HC-18] From Statistics to Individuals: An Exploration of Zoomable Empathic Visualizations
链接: https://arxiv.org/abs/2605.31026
作者: Edwige Chauvergne,Arnaud Prouzeau(ILDA),Martin Hachet(BIVWAC, Inria),Pierre Dragicevic(BIVWAC)
类目: Human-Computer Interaction (cs.HC)
备注:
Abstract:Data visualization is a powerful tool for conveying statistical information, but when representing populations, it tends to hide individuals. We introduce Zoomable Empathic Visualizations (ZEVs), interactive experiences allowing users to smoothly navigate between abstract statistical visualizations and more qualitative, relatable representations focused on individuals. We present three use cases of ZEVs and report on a qualitative user study that highlights opportunities for deeper understanding and emotional engagement, while pointing to areas for improvement and further refinement. In summary, ZEVs point toward new approaches for revealing the individuals behind the data.
[HC-19] UX: Measuring Human–AI Tacit Understanding
链接: https://arxiv.org/abs/2605.30930
作者: Yueshen Li,Hanyi Min,Vedant Das Swain,Koustuv Saha
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computers and Society (cs.CY)
备注:
Abstract:As large language models (LLMs) increasingly act as collaborative partners, human–AI alignment is often evaluated through explicit task success, accuracy, or reward optimization. Yet many collaborative settings depend on tacit understanding: whether an agent can align with a human’s evaluative stance or representational priors without clear objectives, communication, or feedback. To study this capacity, we develop a spectrum-placement task inspired by the social party game Wavelength, in which humans and agents independently place concepts along subjective spectra. We operationalize the Tacit Understanding Index (TUX) as a pairwise measure of similarity between human and agent judgments, and evaluate it with 241 human participants and 200 profile-conditioned LLM agents across four models. We find that nearest human–agent pairs in trait space achieve significantly higher TUX, suggesting that tacit alignment is structured by person-level characteristics rather than random similarity. Regression analyses show that TUX becomes more explainable as predictor sets become richer, with individual traits, decision-making styles, and confidence improving over aggregate trait-distance baselines. These findings suggest that tacit understanding between humans and LLMs is measurable, while revealing the limits of profile-based conditioning for capturing deeper representational alignment.
[HC-20] oxic HallucinAItions: Perturbing Prompts and Tracing LLM Circuits
链接: https://arxiv.org/abs/2605.30913
作者: Soorya Ram Shimgekar,Agam Goyal,Amruta Parulekar,Joshua Chen,Yian Wang,Navin Kumar,Hari Sundaram,Eshwar Chandrasekharan,Koustuv Saha
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Computers and Society (cs.CY); Human-Computer Interaction (cs.HC)
备注:
Abstract:Large language models (LLMs) are increasingly deployed in conversational settings where user tone ranges from polite to adversarial or toxic, yet less is known about whether toxic language in otherwise semantically equivalent prompts can degrade factual reliability. We study how lexical and tone-based prompt perturbations affect the factual reliability of LLMs. Using controlled prompt variations across polite, random, and three toxicity levels, we evaluate five LLMs on ARC-Easy, GSM8K, and MMLU. We find that toxic lexical perturbations consistently reduce factual accuracy and increase uncertainty, while polite phrasing yields limited and inconsistent changes. To examine whether these answer inconsistencies correspond to internal changes, we conduct attribution-graph analyses of model activations and influences. We find that increasing toxicity selectively amplifies perturbation-sensitive variant nodes while relatively stable core reasoning nodes remain more invariant. These findings position prompt tone as a critical dimension of LLM reliability and provide behavioral and mechanistic evidence that surface-level lexical variation can alter factual outputs and internal computation.
[HC-21] What makes an action sequence enjoyable to watch?
链接: https://arxiv.org/abs/2605.30864
作者: Jean-Peïc Chou,Kristine Zheng,Junyi Chu,Maneesh Agrawala,Judith E. Fan
类目: Human-Computer Interaction (cs.HC); Neurons and Cognition (q-bio.NC)
备注: 6 pages, 4 figures, cogsci 2026
Abstract:People often seek out ways to watch others perform complex action sequences (e.g., sports). What makes some sequences more enjoyable to watch than others? We generated 24 video clips of gameplay from a Flappy Bird-style video game. Clips varied in difficulty (how often players succeeded on average) and in moment-to-moment uncertainty (how likely the player was to crash at any given step). Participants (N=864) rated each video on one of three dimensions: how much they enjoyed it, how difficult the level appeared, or how dangerous the player’s trajectory appeared. We found that participants preferred videos where the player seemed to be completing more difficult obstacle courses, but dangerousness did not predict enjoyment ratings. These findings show how procedurally generated stimuli can isolate the factors that affect how enjoyable an action sequence is to watch.
[HC-22] Computer-Aided Tagging on Wikimedia Commons: Designing for Human-AI Collaboration in Open Knowledge Work
链接: https://arxiv.org/abs/2605.30800
作者: Yihan Yu,David W. McDonald
类目: Human-Computer Interaction (cs.HC)
备注: Accepted to CSCW 2026, to appear
Abstract:This study investigates Wikimedia Commons contributors’ lived experiences with the Computer-Aided Tagging (CAT) tool, an AI-assisted image tagging system designed to improve Commons’ discoverability, searchability, accessibility, and multilingual support. Using a qualitative analysis of 595 CAT-related community comments from 11 wiki pages and 16 in-depth interviews, we identify seven key issues that contributed to CAT’s mixed reception and eventual deactivation. We also offer community-informed suggestions for improving the tool. We reflect on the implications for designing human-AI collaboration on Commons and for developing AI-assisted tools that support open knowledge work. This work contributes to HCI and CSCW research by extending the understanding of human-AI collaboration beyond Anglophone, text-centric, corporate platforms.
[HC-23] Relational Aesthesis in Permacomputing Practice: Building a Solar Powered Website from Reclaimed Materials
链接: https://arxiv.org/abs/2605.30706
作者: Nadia Mariyan Smith,Nils Bonfils,Han Qiao,Christoph Becker
类目: Human-Computer Interaction (cs.HC)
备注: Paper in Proceedings of LIMITS 2026: 12th Workshop on Computing within Limits, 2026-06-23-25, Online
Abstract:Permacomputing is a nascent concept and community of practice concerned with developing alternative computing systems grounded in principles of resilience, reuse, sufficiency, and ecological limits. However, research engaging with permacomputing remains in an early stage of development, raising concerns about whether permacomputing can move beyond reflective critique to become a meaningful alternative practice. Through a research-through-design case study, we documented our experience moving a personal website from a data centre in Texas to a self-hosted solar-powered server built from reclaimed electronics. Guided by permacomputing principles and relational aesthesis, we explore what it takes for permacomputing to reconfigure material and perceptual relations. Our findings reveal the frictions of moving away from a maximalist techno-aesthetic while attempting to re-use already existing technologies, potential ways to overcome these challenges through building a community of practice, and the transformative potential of visibilizing and visceralizing digital infrastructures to cultivate more responsible ways of relating to technology. This paper contributes to emerging research on permacomputing and its aesthetics by bringing it into dialogue with theories of non-place and relational aesthesis. Rather than functioning as a purely symbolic gesture, permacomputing practices can cultivate greater collective autonomy, agency, and responsibility in how communities engage and create meaning within digital infrastructures. In the context of socio-ecological crises and anti-colonial transformation, our research offers a situated approach to building and relating to computing technologies in the ashes of dominant technological paradigms.
[HC-24] How Early Adopters Used Generative AI Worldwide: Variation by Country Income and Language
链接: https://arxiv.org/abs/2605.30685
作者: Madeleine I. G. Daepp,Isaac Slaughter
类目: Computers and Society (cs.CY); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Human-Computer Interaction (cs.HC)
备注:
Abstract:AI is being used by people globally, but not everyone is using it in the same ways. Using a large-scale dataset of anonymized, de-identified, and privacy-scrubbed interactions with a widely available and free AI chatbot, we empirically characterize differences in early adopters’ usage across countries. Schooling is the most common domain of use in most countries, particularly low-income countries, with a strong inverse association evident between schooling and country-level GDP. Leisure-related use, by contrast, is positively associated with country-level income. Language, we find, also shapes use: English-language interactions are overrepresented in places where the predominant languages were not well-served by existing models during the period of the study. Improving performance across languages may be a key factor, our work suggests, in whether this technology expands digital divides or enables leapfrogging.
[HC-25] EUDAIMONIA: Evaluating Undesirable Dynamics in AI
链接: https://arxiv.org/abs/2605.30654
作者: Jun Rui Huang,Wang Bill Zhu,Ziyi Liu,Nathanael Fast,Ravi Iyer,Robin Jia
类目: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Human-Computer Interaction (cs.HC)
备注:
Abstract:Large language models (LLMs) are increasingly used as conversational partners for companionship, emotional disclosure, and interpersonal advice, but the social dynamics of these interactions can create harms that are not captured by capability-oriented or traditional safety evaluations. We introduce the Social AI Design Code, a framework for evaluating whether LLMs align with user welfare in social interactions, including whether they encourage harmful intimacy, dependence, or prolonged engagement. To evaluate these risks in natural and diverse user-LLM interactions, we operationalize the code with EUDAIMONIA, a benchmark of 969 user inputs and 3,147 design-requirement violation checks built from WildChat through weak-to-strong filtration, multi-model relabeling, and controlled rewriting. Evaluating 22 recent LLMs, we find that even the strongest models, Claude-Opus-4.7 and GPT-5.5, violate 30.7% and 27.2% of checks, respectively. Extended thinking does not reduce violation rates, suggesting that these failures are persistent social-alignment problems rather than deficits solvable through test-time reasoning alone.
[HC-26] Rationalize: Shared Semantic Reasoning for Human-AI Alignment
链接: https://arxiv.org/abs/2605.30632
作者: Aritra Dasgupta,Naga Datha Saikiran Battula,Avina Nakarmi,Sohom Sen,Subhodeep Ghosh,Xun Song
类目: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Accepted by ACM CHI 2026 BiAlign Workshop
Abstract:We introduce Rationalize, a role-pair framework for shared semantic reasoning between humans and AI models in data-driven sensemaking. Building on ideas in human-machine teaming and critical thinking, we conceptualize human-AI interaction as a series of complementary role pairs (Explorer-Guide, Investigator-Informant, Teacher-Student, Judge-Advocate) operating in a shared reasoning space. In this space, human analysts and AI models (such as LLMs) make purposes, questions, assumptions, evidence, inferences, and implications explicit, facilitating alignment not only at the output level but at the level of rationalization of intent and action by each side. We relate these role pairs to the bidirectional human-AI alignment framework, illustrating how “aligning AI to humans” and “aligning humans to AI” differ by role, and sketch a collaborative research agenda for alignment design and assessment using element-level and role-specific approaches.
计算机视觉
[CV-0] Representation Forcing for Bottleneck-Free Unified Multimodal Models
链接: https://arxiv.org/abs/2605.31604
作者: Yuqing Wang,Zhijie Lin,Ceyuan Yang,Yang Zhao,Fei Xiao,Hao He,Qi Zhao,Zihan Ding,Fuyun Wang,Shuai Wang,Youliang Zhang,Haoqi Fan,Xihui Liu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project page: this https URL
Abstract:Unified multimodal models (UMMs) aim to handle perception and generation in a single model. Yet existing UMMs still rely on a frozen, separately pretrained VAE for image generation, imposing a structural bottleneck. Naively removing it introduces a quality gap, as the model must learn both high-level structure and low-level details from raw pixels. In this paper, we propose Representation Forcing (RF), a technique that closes this gap by making representation prediction a native capability of the model. Concretely, RF forces the decoder to autoregressively predict visual representations as intermediate tokens before pixels; these tokens then stay in context to guide pixel diffusion within the same backbone. By turning representations from perception outputs into generation targets, RF eliminates the need for any external generative latent space. We find that RF benefits both understanding and generation. On image generation, our pixel-space model with RF matches state-of-the-art VAE-based unified models. On image understanding, pixel-space RF generally outperforms its VAE-based variant. Together, these results offer an effective step toward end-to-end, bottleneck-free UMMs.
[CV-1] Lumos-Nexus: Efficient Frequency Bridging with Homogeneous Latent Space for Video Unified Models
链接: https://arxiv.org/abs/2605.31603
作者: Jiazheng Xing,Hangjie Yuan,Lingling Cai,Xinyu Liu,Yujie Wei,Fei Du,Hai Ci,Tao Feng,Jiasheng Tang,Weihua Chen,Fan Wang,Yong Liu
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: Project page ( this https URL ) and Code ( this https URL ) are available
Abstract:Connector-based video unified models have demonstrated strong capability in instruction-grounded video synthesis, but integrating a large high-fidelity generator into the unified training loop is computationally prohibitive, limiting achievable visual quality. We therefore propose Lumos-Nexus, a training-efficient unified video generation framework that facilitates the development of strong reasoning-driven generation capabilities while significantly enhancing visual fidelity. Lumos-Nexus adopts a two-stage design: 1) During training, only a lightweight generator is aligned with the understanding block to learn to take in reasoning-driven semantic control. 2) During inference, we introduce Unified Progressive Frequency Bridging (UPFB) to progressively hand off generation to a high-capacity pretrained generator in the shared latent space, enabling coarse-to-fine refinement and producing high-fidelity videos without compromising reasoning quality. To fill the gap in reasoning-driven video generation benchmarks, we introduce VR-Bench, which assesses a model’s capability to translate inferred intent into coherent and semantically aligned video content. Extensive experiments demonstrate that Lumos-Nexus achieves substantial gains in visual realism and temporal coherence on VBench, while exhibiting strong reasoning-based generative performance on VR-Bench. Code and models are available at this https URL.
[CV-2] Linear Scaling Video VLMs for Long Video Understanding
链接: https://arxiv.org/abs/2605.31598
作者: Cristobal Eyzaguirre,Jiajun Wu,Juan Carlos Niebles
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Video vision-language models (VLMs) are increasingly used in long-horizon and streaming settings, yet most video encoders still rely on spatiotemporal self-attention, causing compute and latency to grow quadratically with the number of frames. Existing efficiency methods improve scalability but often lose accuracy relative to full self-attention, for example through aggressive frame/token dropping or coarse attention approximations. We introduce StateKV, an inference-time method that adapts pretrained long-video VLMs to linear-time video prefill by carrying cross-frame context in a fixed-capacity, importance-based recurrent state, paired with a second full per-frame cache used for decoding. Across three long-video benchmarks and seven models spanning three families and multiple scales, StateKV remains close to full self-attention and consistently outperforms dominant sliding-window / recency-based streaming approximations, without fine-tuning or architectural changes. StateKV also reduces video-prefill cost measured FLOPs, enabling stronger accuracy at a fixed compute budget by running larger models. These results suggest a practical step toward scalable long-video understanding.
[CV-3] SOCO: Benchmarking Semantic Object Correspondence in Vision Foundation Models
链接: https://arxiv.org/abs/2605.31597
作者: Olaf Dünkel,Basavaraj Sunagad,Haoran Wang,David T. Hoffmann,Christian Theobalt,Adam Kortylewski
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project page: this https URL
Abstract:Measuring structured object understanding in vision foundation models remains challenging due to inconsistent evaluation protocols and limited part-level supervision. Semantic correspondence (SC) evaluates this capability by testing whether object parts can be matched across instances and categories under large variations in appearance, viewpoint, and geometry. To enable a systematic SC evaluation, we introduce SOCO, a new benchmark for Semantic Object Correspondence that introduces a taxonomy of correspondence types and provides consistent, functionally meaningful keypoint annotations across 100 categories and over 1M correspondence pairs. In addition, SOCO includes keypoint language descriptions, enabling the evaluation of large vision-language models (LVLMs) and their fine-grained part-level understanding. Comprehensive experiments reveal that (i) vision foundation backbones encode strong semantic structure but transfer correspondences poorly across related categories and only partially capture object-part position, (ii) LVLMs are stronger at text-prompted part localization than at visual-reference cross-image matching, exposing a gap between language-grounded localization and fine-grained visual correspondence, and (iii) correspondence performance predicts performance on dense downstream tasks, including segmentation, tracking, 3D pose estimation, and 3D detection, more strongly than ImageNet classification. Together, these findings position SOCO as a benchmark for structured, part-level representation quality in vision and multimodal foundation models.
[CV-4] KLIP: localized distribution shift detection via KL-divergence with diffusion priors in Inverse Problems CVPR2026
链接: https://arxiv.org/abs/2605.31596
作者: Alireza Kheirandish,Jihoon Hong,Sara Fridovich-Keil
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: CVPR 2026
Abstract:Diffusion models have shown promising performance as data-driven priors for computational imaging, as well as some capacity to detect out-of-distribution (OOD) images. However, existing approaches to OOD detection often require some knowledge of the shifted distribution, fail to detect subtle or localized distribution shifts, and operate on full images, rather than the indirect measurements available in inverse problems. We propose an OOD detection metric based on the Kullback-Leibler divergence between the diffusion prior and the posterior distribution, that (i) does not require any calibration data or knowledge of the shifted distribution, and (ii) can detect whole images as OOD as well as localize OOD patches within an image. Experimentally, we show that this metric can detect subtle yet semantically meaningful distribution shifts, such as the shift from healthy liver CT scans to those with tumors, and generalizes across different types of diffusion models, datasets, and inverse problems. Our code can be found at this https URL.
[CV-5] Learning Global Motion with Compact Gaussians for Feed-Forward 4D Reconstruction
链接: https://arxiv.org/abs/2605.31595
作者: Mungyeom Kim,Minkyeong Jeon,Honggyu An,Jaewoo Jung,Hyuna Ko,Jisang Han,Hyeonseo Yu,Donghwan Shin,Sunghwan Hong,Takuya Narihira,Kazumi Fukuda,Yuki Mitsufuji,Seungryong Kim
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project Page: see this https URL
Abstract:Dynamic scene reconstruction from monocular video remains a fundamental challenge in computer vision. Existing feed-forward methods predict 3D Gaussians pixel-wise for each frame, suffering from duplicated Gaussians and view-dependent biases that hinder effective learning of scene motion. We present C4G, a feed-forward 4D reconstruction framework built upon a compact set of timestamp-conditioned learnable Gaussian query tokens. Each token aggregates corresponding features across the full temporal context and decodes a 3D Gaussian whose position is modulated by the target timestamp, enabling globally coherent motion modeling without per-scene optimization. To capture fine-grained details, we further introduce a video diffusion model-based rendering enhancement module. Since our framework effectively aggregates features into Gaussians, we extend this capability to feature lifting, producing a 4D feature field that supports point tracking and dynamic scene understanding. C4G achieves strong novel-view synthesis performance using significantly fewer Gaussians and without requiring camera poses, while exhibiting stronger motion modeling and robustness to large temporal gaps.
[CV-6] CoFiDA-M: Concept-Aware Feature Modulation for Cross-Domain Adaptation with Image-Only Inference CVPR2026
链接: https://arxiv.org/abs/2605.31591
作者: Nurjahan Sultana,Moi Hoon Yap,Xinqi Fan,Wenqi Lu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: ‘Accepted by CVPR 2026’
Abstract:Models for AI-based skin cancer screening suffer a severe performance drop when shifting from expert dermoscopic (source) images to consumer-grade clinical (target) images, hindering real-world deployment. Existing domain adaptation methods often ignore crucial semantic invariants, such as clinical concepts. While new foundation models like MONET can provide this semantic information as dense, probabilistic scores, this metadata is unavailable at test time, creating a deployment paradox for practical image-only screening tools. We address this gap by proposing CoFiDA-M, a privileged information framework that learns from concepts at training time but deploys as an image-only model. Our method trains a teacher network that uses MONET concept probabilities to guide a FiLM modulator, transforming visual features into a semantically edited" feature space. A lightweight, image-only student is then trained to reproduce this edited representation, not just the teacher's final predictions. This distillation bakes" the clinical reasoning into the student’s weights. On a challenging multi-dataset benchmark, our image-only student significantly outperforms state-of-the-art approaches, especially in melanoma recall. Our work provides a practical and generalizable framework for leveraging noisy, probabilistic metadata as privileged information, demonstrating strong cross-dataset robustness and potential for real-world deployment beyond dermatology. Implementation code is available at: this https URL
[CV-7] unerDiT: Training-free Progressive Steering of Diffusion Transformer for Multi-Event Video Generation
链接: https://arxiv.org/abs/2605.31590
作者: Ruotong Liao,Guowen Huang,Qing Cheng,Guangyao Zhai,Lei Zhang,Xun Xiao,Thomas Seidl,Daniel Cremers,Volker Tresp
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: 17 pages, 13 figures
Abstract:Text-to-video (T2V) generation faces challenging questions when generating videos with long horizons containing multiple events. Inspired by the intrinsics of the diffusion process, we probe video diffusion transformers (DiTs) and uncover intrinsic turning points in the DiT denoising trajectory where conditioning text affects generation from global layout to fine-grained details. Building on this finding, we present TunerDiT, a simple yet effective progressive steering method that requires no additional training for multi-event generation. TunerDiT comprises two steering handles: (1) Event-Partitioned Masking that enforces event boundaries while allowing cross-event transition bands; (2) Cross-Event Prompt Fusion that injects neighboring event semantics for late-stage refinement. We contribute a self-curated prompt suite for benchmarking multi-event generation, i.e., Meve. TunerDiT achieves state-of-the-art performance across 8 metrics and offers a tunable trade-off between video consistency and event separation, compared with other training-free methods. The improvement in text alignment increases with the event count, indicating a scaling possibility with increasing event count.
[CV-8] Recognizing Co-Speech Gestures in-the-Wild
链接: https://arxiv.org/abs/2605.31589
作者: Sindhu B Hegde,K R Prajwal,Andrew Zisserman
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:While humans naturally gesture during speech, only a sparse subset of these movements are visually depictive and semantically linked to specific spoken words. Current multimodal models struggle to capture these semantic co-speech gestures, heavily bottlenecked by a lack of precisely annotated training data. To address this, we introduce the Gesture Recognition in the Wild (GRW) dataset, the first large-scale benchmark designed to map unconstrained human gestures to specific words with frame-accurate temporal boundaries. Comprising 156,688 manually annotated video clips, GRW spans a highly diverse 150-word taxonomy of physical actions, spatial descriptors, and abstract concepts. We leverage GRW to train video models to (a) classify gestures as semantic or not, (b) recognize the word corresponding to a co-speech gesture, and © temporally localize the gesture. We also use GRW to establish benchmarks for these three tasks.
[CV-9] SurGe: Improved Surface Geometry in Point Maps
链接: https://arxiv.org/abs/2605.31577
作者: Karim Knaebel,Gonzalo Martin Garcia,Christian Schmidt,Ilya Fradlin,Lucas Nunes,Daan de Geus,Bastian Leibe
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project page at this https URL
Abstract:Recent feedforward 3D reconstruction methods predict point maps and estimate global 3D geometry remarkably well. However, their predictions still exhibit inaccurate local surface geometry, which is clearly visible qualitatively but only weakly reflected in common metrics. To make these errors more explicit in evaluation, we introduce a point map normal metric that evaluates the local surface orientation induced by neighboring 3D predictions. To reduce these errors, we propose two complementary components: a point gradient matching loss that supervises depth-normalized 3D finite differences, and a Neighborhood Attention Decoder (NAD) that progressively upsamples features and uses Neighborhood Attention for local feature mixing. Across eight zero-shot monocular geometry benchmarks, our model, SurGe, achieves the best average rank for global point map AbsRel and consistently improves local point map and point map normal evaluations.
[CV-10] Joint Multi-Camera LiDAR Extrinsic Calibration via Learned Pairwise Initialization and Geometric Refinement CVPR2026
链接: https://arxiv.org/abs/2605.31576
作者: Aziz Al-Najjar,Marzieh Amini,James R. Green,Felix Kwamena
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Paper is accepted in CVPR 2026 Workshop URVI: Unified Robotic Vision with Cross-Modal Sensing and Alignment
Abstract:Most learning-based camera-LiDAR calibration methods treat each camera-LiDAR pair independently, ignoring the rigid geometric coupling in multi-camera platforms. As a result, per-camera estimates may be individually accurate yet inconsistent at the system level. We present a two-stage framework for joint multi-camera LiDAR extrinsic calibration that combines learned pairwise matching with geometric refinement. First, CMRNext is applied independently to each camera to produce initial extrinsic estimates and dense 2D-3D correspondences. These predictions are then jointly refined through a multi-frame bundle adjustment with reprojection, per-camera prior, and relative-pose prior terms. This approach converts pairwise predictions into a globally consistent multi-camera calibration. Experiments on KITTI (in-domain for CMRNext) and Walkley (out-of-domain) datasets show improved per-camera accuracy and inter-camera consistency. On KITTI, the method achieves 0.89 cm translation error and 0.038 rotation error. On Walkley, it reduces translation error from 108.6 cm to 3.1 cm, highlighting the benefit of explicit multi-camera coupling when single-camera predictions are less reliable.
[CV-11] nuReasoning : A Reasoning -Centric Dataset and Benchmark for Long-Tail Autonomous Driving
链接: https://arxiv.org/abs/2605.31572
作者: Zhiyu Huang,Johnson Liu,Rui Song,Zewei Zhou,Ruining Yang,Yun Zhang,Tianhui Cai,Hanyin Zhang,Mingxuan Gao,Valeria Xu,Jiali Chen,Yishan Shen,Yiluan Guo,Tony(Xuewei)Qi,Jiaqi Ma
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Reasoning is essential for autonomous driving (AD) in long-tail scenarios, where vehicles must apply commonsense knowledge, understand spatial relations, infer agent interactions, and make safe decisions. However, existing AD datasets and benchmarks mainly target perception, prediction, or planning, and provide limited supervision for reasoning over realistic long-tail driving scenes. We introduce nuReasoning, a large-scale real-world dataset and benchmark for reasoning-centric AD. Following the lineage of nuScenes and nuPlan, nuReasoning advances real-world AD datasets and benchmarks toward reasoning in long-tail driving scenarios. The dataset contains 20,000 clips, each 20 seconds long, collected across multiple cities, with synchronized multi-camera images, LiDAR data, HD maps, object annotations, and human-verified reasoning annotations spanning Spatial Reasoning, Decision Reasoning, and Counterfactual Reasoning. Unlike prior datasets that focus primarily on visual question answering, nuReasoning supports both reasoning evaluation and planning evaluation, enabling a direct study of how reasoning supervision affects driving performance. Experiments show that fine-tuning VLMs on nuReasoning substantially improves driving-specific question answering, while incorporating reasoning supervision into VLA training improves planning performance even when textual reasoning outputs are disabled at inference time. These results establish nuReasoning as a foundation for evaluating and improving robust, interpretable, reasoning-driven AD systems in realistic long-tail settings.
[CV-12] EGOSTREAM: A Diagnostic Benchmark for Streaming Episodic Memory in Egocentric Vision
链接: https://arxiv.org/abs/2605.31557
作者: Rosario Forte,Giuseppe Lando,Antonino Furnari
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Continuous episodic memory is a core capability for autonomous agents operating in dynamic, real-world environments, yet current streaming video benchmarks provide limited tools for diagnosing what models remember and for how long. We introduce \egostream, a diagnostic benchmark for streaming episodic memory evaluation in egocentric vision. \egostream organizes 2,250 curated questions along seven cognitive dimensions: detail, spatial, temporal, event, social, causal, and prospective memory. We introduce the Answer Validity Window (AVW), which specifies the temporal span an answer remains valid as the observed scene evolves. This allows us to expand the questions into 8,528 recall-conditioned evaluations, enabling controlled testing from instant to ultra-long-term recall while separating genuine model forgetting from natural world-state changes. We rigorously establish baseline performance through a unified streaming MLLM framework that compares several state-of-the-art memory-management mechanisms, covering sliding windows, attention sinks, KV-cache pruning, merging, and offloading. Experiments within a unified Qwen3-VL backbone reveal that comparable aggregate accuracies mask starkly different memory profiles. For instance, token pruning preserves fine-grained details and temporal structure significantly better than token merging, while quantized offloading rescues ultra-long-term recall. Ultimately, all mechanisms operate well below real-time (1s per frame), and top performing methods ceil at about 45% accuracy, exposing critical gaps in current architectures. \egostream provides the diagnostic testbed needed to close these gaps.
[CV-13] SMART: SMPLest-X Mesh Adaptation and RAFT Tracking for Soccer Pose Estimation SOCC CVPR2026
链接: https://arxiv.org/abs/2605.31551
作者: Parthsarthi Rawat
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: CVPR 2026 SoccerNet FIFA Skeleton Tracking Light Challenge, Rank 6
Abstract:We present our approach to the FIFA Skeletal Tracking Challenge 2026, which requires estimating 3D world-space poses of soccer players from broadcast video. Our method finetunes SMPLest-X (ViT-H, 687 M parameters) via a stratified clip split, multi-task depth supervision, and broadcast augmentation, paired with a RAFT dense optical flow camera tracker, foot-plane anchoring, and two-pass temporal smoothing. Against the FIFA baseline score of 1.053 on the validation set, SMART achieves 0.647, a 38.6% improvement; on the held-out test set, SMART scores 0.593 (Global MPJPE: 0.324 m, Local MPJPE: 0.054 m).
[CV-14] Automated Prediction of Postoperative Pancreatic Fistula Using Preoperative Computed Tomography
链接: https://arxiv.org/abs/2605.31539
作者: Ashok Choudhary,Chris Varghese,Leo Y. Li-Han,Frank G. Lee,Ellen L. Larson,Elizabeth B. Habermann,Cornelius A. Thiels,Hojjat Salehinejad
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Quantitative Methods (q-bio.QM)
备注:
Abstract:Postoperative pancreatic fistula (POPF) is a serious complication after pancreatic resection, increasing morbidity, hospital stay, and healthcare costs. We present an automatic, end-to-end deep learning pipeline-from pancreatic segmentation to classification-for preoperative POPF risk estimation and stratification using preoperative CT scans. A data set with auto-segmented pancreas volumes and surgical outcomes was used to evaluate multiple architectures, including a custom lightweight 3D CNN baseline (CNN3D), R(2+1)D ResNet-18, and ResNet-MC3-18 models. Evaluation across multiple 3D architectures demonstrated promising predictive performance. This approach offers a clinically valuable tool and a methodological benchmark for pancreas-specific CT classification, supporting improved preoperative decision-making in pancreatic surgery.
[CV-15] RayDer: Scalable Self-Supervised Novel View Synthesis from Real-World Video
链接: https://arxiv.org/abs/2605.31535
作者: Ulrich Prestel,Stefan Andreas Baumann,Nick Stracke,Björn Ommer
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Project Page: this https URL
Abstract:Self-supervised novel view synthesis (NVS) remains challenging to scale, despite the abundance of video data, largely due to the brittleness of training on realistic videos and the hard-to-predict scaling behavior of multi-network system designs. We introduce RayDer, a unified, feed-forward transformer that consolidates camera estimation, scene reconstruction, and rendering into a single backbone, turning self-supervised NVS into a well-posed single-model scaling problem. A minimal dynamic state, treated as a nuisance factor, absorbs time-varying content and enables stable training on unconstrained real-world video. Importantly, RayDer keeps static-scene NVS as its target task: dynamic content is leveraged purely as scalable supervision, not reconstructed as in dynamic-scene (4D) NVS. Across multiple model sizes and orders of magnitude in data, RayDer exhibits clean power-law scaling with data and compute, and outperforms static-scene data mixtures. On a large number of benchmarks, RayDer achieves strong zero-shot open-set performance competitive with state-of-the-art supervised approaches. Project Page: this https URL
[CV-16] Feature-Optimized Vision for Adaptive 3D Scene Reconstruction
链接: https://arxiv.org/abs/2605.31534
作者: Eric Liang
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Three-dimensional scene reconstruction depends on local image evidence that is both visually discriminative and geometrically useful. Fixed feature thresholds and uniform feature budgets are easy to deploy, but they can waste computation on repeated texture, low-parallax regions, or unstable points. This paper proposes an adaptive feature-optimized vision front end for 3D reconstruction. The method scores candidate features by texture, repeatability, distinctiveness, expected triangulation angle, and spatial coverage, then allocates a per-view feature budget to maximize useful tracks under a fixed reconstruction pipeline. A small synthetic multi-view prototype evaluates four selection policies across corridor, facade, object-table, and cluttered scenes. Compared with random, texture-only, and uniform-grid baselines, the adaptive policy obtains the best quality-aware completeness and the lowest aggregate reconstruction RMSE while preserving broad image coverage. The result is not a replacement for modern learned matching or neural reconstruction systems; it is a modular front-end policy that can make classical and learned 3D pipelines more deliberate about which visual evidence they spend compute on.
[CV-17] SVI-Bench: A Dynamic Microworld for Strategic Video Intelligence
链接: https://arxiv.org/abs/2605.31529
作者: Yulu Pan,Han Yi,Seongsu Ha,Md Mohaiminul Islam,Benjamin Zhang,Lorenzo Torresani,Gedas Bertasius
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:True video intelligence demands more than recognizing what is visible: it requires reasoning about why events unfold, predicting what would change under different conditions, and deciding what to do next. We refer to this progression, from perception through causal reasoning and simulation to strategic planning, as Strategic Video Intelligence (SVI). No existing benchmark evaluates this capability stack: in-the-wild videos lack verifiable ground truth for causal and strategic questions, while synthetic environments sacrifice the complexity of real multi-agent systems. To bridge this gap, we introduce SVI-Bench, a large-scale benchmark that leverages team sports as a dynamic microworld, combining the complexity of real-world multi-agent interaction (10-22 agents making coordinated decisions under adversarial pressure) with the verifiability of explicit rules and definitive outcomes. SVI-Bench comprises approximately 35K hours of broadcast video, 15M annotated actions, 15K hours of expert commentary, 23K game reports, and 103K structured statistical records across basketball, soccer, and hockey, all constructed via a data engine that transforms raw game data into a dense, cross-referenced corpus. We organize evaluation into 9 tasks spanning a progressive four-pillar hierarchy: Dynamic Scene Understanding, Causal Reasoning, Strategic Simulation, and Agentic Synthesis. Evaluating strong multimodal and agentic baselines, we find a capability cliff: models perform competently on perceptual tasks, achieving approximately 73% on fine-grained action QA, but degrade sharply at each successive cognitive level. Agentic tasks prove hardest: the strongest model achieves only 5% accuracy when required to autonomously gather and integrate evidence across a corpus of 1.8M clips.
[CV-18] Personalize Your Large Vision-language Models With In-context Prompt Tuning
链接: https://arxiv.org/abs/2605.31513
作者: Yanshu Li,Jiaqian Li,Kuai Yu,Xi Xiao,Dongfang Liu,Tianyang Wang,Ruixiang Tang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 27 pages, 10 figures, 5 tables
Abstract:Large vision-language models (LVLMs) have demonstrated strong general multimodal capability and are increasingly deployed in downstream systems. This trend has driven growing interest in LVLM personalization, which aims to enable models to quickly and effectively learn out-of-distribution multimodal concepts to meet user-specific needs. However, many existing methods rely on inference-time training, which reduces efficiency. They also struggle to maintain accuracy in complex multi-image, multi-concept settings. These limitations restrict the broader deployment of LVLM-based systems. Therefore, this paper proposes in-context prompt tuning (ICPT). Specifically, ICPT employs a lightweight projection module capable of operating in complex scenarios to extract fine-grained visual semantics from multiple reference images, seamlessly transforming these features alongside identity-label mappings into continuous prompts. To maximize computational efficiency, this module adaptively determines the prompt length based on the intrinsic visual complexity of each concept. Crucially, to overcome the environmental biases and cross-concept interference prevalent in real-world applications, we introduce two novel geometric regularizations. These constraints refine prompt representations by decoupling key identities from transient environmental states and separating concepts to avoid semantic confusion. Extensive experiments show that ICPT achieves state-of-the-art personalization accuracy across diverse tasks and LVLM backbones.
[CV-19] Internalizing Temporal Consistency in Video Object-Centric Learning without Explicit Regularization
链接: https://arxiv.org/abs/2605.31508
作者: Rongzhen Zhao,Zhiyuan Li,Juho Kannala,Joni Pajarinen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 14 pages
Abstract:Video Object-Centric Learning (OCL) aims to represent objects as \textitslot vectors and maintain their consistency across frames. Slot-Slot Contrastive (SSC) loss has become the cornerstone for state-of-the-art (SOTA) video OCL methods. While highly effective, SSC relies on one-to-one object correspondence across frames and introduces an extra loss. Following Occam’s Razor, we propose a paradigm shift: temporal consistency is better enforced as an implicit model design rather than an explicit loss. To elegantly exclude SSC (\textbfxSSC), we introduce two quasi-zero-overhead synergistic mechanisms: (\textiti) Chrono-Channel Decomposition (CCD) structurally disentangles slot representations along the channel dimension into \textitstatic and \textitdynamic sub-spaces, serving as an empirically unified information bottleneck; (\textitii) Cross-Temporal Reconstruction (CTR) stochastically reconstructs target features of either the current or previous time step by fusing current slots’ static channels and target slots’ dynamic channels, using a single standard OCL decoder with minor training adaptation. Thereby, the slot sets inherently learn temporal consistency by minimizing the standard reconstruction error alone. Extensive experiments show that integrating xSSC into leading baselines not only improves training efficiency but also establishes new SOTAs on video object discovery and recognition tasks. Furthermore, our PCA and gradient analyses confirm that objects’ time-invariant semantics and time-variant kinematics are encoded into the proposed sub-spaces. Our source code, model checkpoints and training logs are provided on this https URL.
[CV-20] How can embedding models bind concepts? ICML2026
链接: https://arxiv.org/abs/2605.31503
作者: Arnas Uselis,Darina Koishigarina,Seong Joon Oh
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: ICML 2026
Abstract:Humans easily determine which color belongs to which shape in multi-object scenes, an ability known as concept binding. Vision-language embedding models such as CLIP struggle with binding: they recognize individual concepts but fail to represent which concepts form which objects. Although CLIP behaves like a bag-of-concepts model in cross-modal retrieval, object information is recoverable from its image and text embeddings separately. We study this tension through the binding function, which maps concepts to scene embeddings. We find that scene embeddings decompose additively into object representations, explaining why uni-modal probes can recover object information. However, CLIP’s binding function is high-complexity, which likely prevents the image and text encoders from learning a shared binding mechanism that generalizes to unseen concept combinations. We then ask whether this limitation is fundamental. We show that it is not. In controlled transformer models trained from scratch, binding generalization emerges with sufficient data coverage. These models learn low-complexity binding functions characterized by multiplicative interactions between concepts, enabling systematic generalization. Code is publicly available at this https URL.
[CV-21] Enhancing Computer Vision Model Generalization in Warehouse Facilities: A Case Study on Anomaly Detection in Vertical Material Handling Systems
链接: https://arxiv.org/abs/2605.31487
作者: Ruiliang Liu,Tina Dongxu Li,Joshua Migdal,Ken Meszaros,Trevor Dardik
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 6 pages, 10 figures. Accepted at IEEE International Conference on Mechatronics and Automation (ICMA) 2026
Abstract:Deploying computer vision models in Warehouse Facilities traditionally requires extensive resources for camera mounting, image collection, annotation, training, and deployment - a process often needing repetition in each new environment due to camera mounting constraints and environmental variability. This paper explores an innovative approach to streamline this process by conducting the standard procedure solely in a laboratory setting, focusing on vertical material handling systems and anomaly detection in forks of the systems. Through extensive experimentation, we have found that combining optimal camera placement, strategic image triggering, careful model selection and model ensemble enables effective generalization from laboratory conditions to diverse warehouse facilities environments, potentially transforming warehouse automation implementation by simplifying warehouse facilities deployment to just camera mounting, image collection, and model deployment, thereby saving significant resources and time typically spent on image annotation and model retraining. This is an experimental research study and not a production deployment.
[CV-22] VolFill: Single-View Amodal 3D Scene Reconstruction with Volumetric Flow Matching
链接: https://arxiv.org/abs/2605.31466
作者: Tuan Duc Ngo,Chuang Gan,Evangelos Kalogerakis
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Reconstructing the complete geometry of a scene from a single RGB image remains challenging - especially when inferring hidden structures where visual evidence is incomplete. We introduce VolFill, a generative framework that predicts the 3D structure of the complete scene rather than relying on traditional pixel-aligned regression. Our method utilizes a hybrid 3D VAE to compress sparse truncated unsigned distance function grids into a compact latent space, paired with a latent Diffusion Transformer that denoises this representation to recover the complete scene. We condition the generation on geometry foundation models, leveraging rich spatial priors for robust reasoning. Unlike existing methods limited by per-ray constraints or unstructured point-cloud queries, VolFill provides a structured representation that supports direct surface extraction and occupancy queries at scale. Extensive experiments on the SCRREAM and NRGB-D datasets demonstrate that our approach significantly outperforms current baselines, providing a robust foundation for holistic spatial understanding.
[CV-23] VisionPulse: Dynamic Visual Sparsity for Efficient Multimodal Reasoning ICML2026
链接: https://arxiv.org/abs/2605.31457
作者: Hengbo Xu,Shengjie Jin,Yanbiao Ma,Zhiwu Lu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted at ICML 2026
Abstract:With the rapid advancement of large multimodal models (LMMs), inference-time overhead has become a key bottleneck for real-world deployment. Existing methods typically prune visual tokens at prefill, assuming the required visual evidence remains static during reasoning. However, we empirically show that visual evidence is strongly step-dependent: only a sparse subset of visual tokens is critical at each decoding step, and the critical set evolves across reasoning. Furthermore, we identify a coupled bottleneck where redundant visual context can steer the model toward query-irrelevant regions, lengthening the reasoning trace. Guided by these insights, we propose VisionPulse, a step-wise visual token pruning framework during reasoning. VisionPulse computes a lightweight visual attention mass to estimate the step-wise retention budget by exploiting its strong positive correlation with LMMs’ effective visual token usage and retain only the most critical tokens under this budget. By enforcing visual sparsity during reasoning, VisionPulse filters redundant visual context while preserving relevant visual evidence, shortening reasoning traces naturally. Extensive experiments show that VisionPulse only retains 5% of visual tokens per step with reasoning traces shortened by 11.2%, while keeping accuracy almost unchanged.
[CV-24] Astra: a generalizable report generation foundation model for 3D computed tomography
链接: https://arxiv.org/abs/2605.31437
作者: Zhuhao Wang,Fang Chen,Chaohui Yu,Zihan Li,Yuchao Zheng,Jing Wang,Xuan Yang,Jia Guo,Zhenlu Yang,Xingju Zheng,Yihua Sun,Haojie Han,Xiaoxiao Qin,Zhan Feng,Wenbo Xiao,Chao Zhu,Yuehua Li,Shipeng Zhang,Hao Luo,Yunsong Peng,Fan Wang,Hongen Liao
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:CT interpretation requires radiologists to review hundreds of volumetric slices per examination, making reporting time-consuming and highly expertise-dependent. Automated CT report generation offers a promising route to improving clinical efficiency, yet the field still lacks a generalizable CT report generation foundation model that supports multi-region reporting and remains robust across external real-world cohorts. Intrinsic inconsistencies in reporting style and diagnostic terminology across cohorts make naive joint training prone to noisy textual supervision, thereby limiting model generalizability. Here we present Astra, a generalizable CT report generation foundation model trained on 90,678 thoracoabdominal CT-report pairs (CTRgDB) with 353,671 abnormalities spanning eight organ systems. By harmonizing report style and further refining diagnostic consistency via reinforcement learning, Astra achieves style-consistent and diagnostically accurate report generation across diverse anatomical regions and institutions. Evaluating on CTRgDB and six external cohorts, Astra achieves state-of-the-art performance with a 44.1% average improvement in fine-grained diagnostic metrics (P0.001). In real-world clinical workflows, Astra assistance accelerates chest report drafting by 29.6% and improves abdominal report completeness by 11.3% (P0.001). Furthermore, Astra also demonstrates broad utility as a foundation for CT AI development, improving downstream diagnostic performance and scaling vision-language pretrain through high-quality report synthesis. Overall, Astra serves as a broadly accessible clinical assistant and a pivotal infrastructure for the next generation of AI-powered healthcare.
[CV-25] YARD: Y-Architecture Register Decoding for Efficient Hallucination Mitigation in Large Vision-Language Models
链接: https://arxiv.org/abs/2605.31429
作者: Ting Chen,Geng Li,Guohao Chen,Yu Hu,Guan Huang,Mai Chen,Langsheng Lei,Jun Du
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 21 pages, 11 figures
Abstract:Contrastive decoding (CD) seeks to mitigate hallucinations in Large Vision-Language Models (LVLMs) by contrasting the output distributions of a standard model and a visually degraded model. However, existing training-free CD methods suffer from sub-optimal degraded branches: completely dropping visual tokens is too extreme and induces language hallucinations, while corrupting input images offers coarse control over visual evidence and suffers from high inference latency due to requiring two full forward passes. To address these dilemmas, we propose YARD, a training-free Y-Architecture Register Decoding framework. Motivated by the observation that reliable text-to-vision grounding predominantly emerges in the middle decoder layers, YARD constructs the degraded branch internally by sharing shallow-layer computations and branching exactly at this critical stage. For the degraded branch, YARD replaces patch-level visual tokens with register tokens, which preserve global image semantics but lack fine-grained local evidence. This image-aware yet locally under-grounded design provides a faithful contrastive signal without extreme modality mismatch, while the Y-architecture strictly avoids a costly second forward pass. Extensive experiments on generative and discriminative hallucination benchmarks demonstrate that YARD consistently achieves state-of-the-art hallucination mitigation across multiple LVLMs, alongside a significant reduction in inference latency.
[CV-26] riangle Splatting SLAM
链接: https://arxiv.org/abs/2605.31419
作者: Nicholas Fry(1 and 2),Eric Dexheimer(2),Kirill Mazur(2),Paul H. J. Kelly(1 and 2),Andrew J. Davison(2) ((1) Software Performance Optimisation Group, Imperial College London, (2) Department of Computing, Imperial College London)
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注: 26 pages, 11 figures
Abstract:We present a dense RGB-D SLAM system using differentiable triangles as the 3D map representation. While 3D Gaussian Splatting has emerged as the leading method for novel-view synthesis, triangles remain the standard primitive for traditional rendering hardware, game engines, and downstream tasks requiring explicit geometry such as simulation, collision, and editing. Recent offline methods have demonstrated that an unstructured ‘triangle soup’ can be optimised into a photorealistic mesh via Delaunay triangulation across a set of posed images. Building upon this insight, we present the first dense SLAM system to employ Triangle Splatting to perform both tracking and mapping through online differentiable rendering of a triangle soup. The map can be converted into a connected mesh on-the-fly via restricted Delaunay triangulation, enabling new online capabilities such as mesh deformation and collision checking. On Replica and TUM-RGBD, our system outperforms baselines on 3D geometry, matches the camera-tracking accuracy, and enables online mesh-based scene editing.
[CV-27] FSM-Net: An Efficient Frequency-Spatial Network for Real-World Deblurring CVPR2026
链接: https://arxiv.org/abs/2605.31400
作者: Vinh-Thuan Ly
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted to NTIRE Workshop at CVPR 2026. Project page: this https URL
Abstract:Real-world image deblurring demands both high-fidelity restoration and computational efficiency, a balance existing methods often struggle to achieve. In this paper, we propose FSM-Net (Frequency-Spatial Multi-branch Network), a highly efficient solution that secured 2nd place in the NTIRE 2026 Challenge on Efficient Real-World Deblurring. FSM-Net pioneers a dual-domain approach: a novel Frequency Attention module explicitly recovers high-frequency structural details via FFT, while a Cross-Gated Vision E-Branchformer at the bottleneck captures global dependencies with linear complexity. To ensure robust convergence, we employ a progressive curriculum training strategy guided by a composite loss function (Multi-Scale Charbonnier, Structural Edge, and Frequency). Evaluated on the RSBlur benchmark, FSM-Net achieves an outstanding 33.144 dB PSNR with only 4.94M parameters and 159.35 GMACs (at 1920x1200 resolution). By effectively pushing the Pareto frontier of efficiency and quality, FSM-Net establishes a strong baseline for resource-constrained image restoration.
[CV-28] LiftNav: Path Planning via Semantic Lifting in TSDF-Guided Gaussian Splatting
链接: https://arxiv.org/abs/2605.31376
作者: Hannah Schieber,Dominik Frischmann,Victor Schaack,Angela P. Schoellig,Daniel Roth
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR)
备注:
Abstract:Autonomous robots in unknown indoor environments require both reliable collision avoidance and object-level understanding. Classical representations such as TSDF support safe planning but lack semantics, while photorealistic methods like Gaussian Splatting (GS) provide rich appearance yet suffer from soft geometry, limiting precise obstacle avoidance. We present LiftNav, a hybrid navigation framework built on GSFusion’s TSDF+GS dual map, augmented with a real-time pipeline of YOLO-based detection, TSDF-based 3D lifting, and B-spline trajectory optimization. This design enables flexible semantic navigation without dense 3D embeddings. We further introduce a hinge-loss-based collision penalty that improves trajectory smoothness and safety. We evaluate our approach in a simulation using the Replica dataset. Compared against a state-of-the-art radiance field baseline we show a 100% feasibility rate and shorter trajectories.
[CV-29] A Unifying View of Variational Generative Wasserstein Flows ICML2026
链接: https://arxiv.org/abs/2605.31369
作者: Paul Caucheteux,Clément Bonet,Anna Korba
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted as a spotlight at ICML2026
Abstract:Many modern generative models can be viewed as minimizing divergences between probability distributions, yet they rely on different algorithmic and geometric principles. Wasserstein gradient flows provide a continuous-time formulation for optimizing over distributions, and can be approximated through their implicit discretization via the Jordan-Kinderlehrer-Otto (JKO) scheme. In this work, we present a unified theoretical framework for generative modeling based on Wasserstein gradient flows, which we refer to as Generative Wasserstein Flows (GWF). We show that a broad class of existing methods can be derived as instances of parametric JKO schemes for f -divergence objectives, and we establish equivalences between several recently proposed algorithms. We extend this framework beyond f-divergence to Integral Probability Metrics and squared Maximum Mean Discrepancy, deriving new JKO-based generative algorithms, and clarifying their connections with GANs. We study empirically the impact of the JKO regularization for a wide set of objectives. Finally, we analyze parametric Wasserstein flows, where the dynamics are restricted to distributions induced by parametrized maps.
[CV-30] DecMem: Towards Minute-Long Consistent World Generation with Decoupled Memory
链接: https://arxiv.org/abs/2605.31336
作者: Zhenhao Yang,Xiaoshi Wu,Zhengyao Lv,Xiaoyu Shi,Xintao Wang,Pengfei Wan,Kun Gai,Kwan-Yee K. Wong
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Project page is available at this https URL
Abstract:Recent advances in video generative models have promoted rapid progress in controllable world models. However, maintaining fine-grained spatio-temporal consistency under long-horizon reasoning remains a key challenge. In this work, we move beyond explicit 3D memory and coarse frame-level implicit modeling, and propose a fine-grained, learnable, and scalable memory for consistent world generation. We first identify two fundamental limitations of naïve learnable memory architectures in long-horizon extrapolation, namely computational inefficiency and attention dispersion. Through a systematic analysis of attention dispersion, we propose DecMem, a decoupled memory architecture that employs Sparse Global Memory for efficient fine-grained access to global history and Anchored Local Memory for stable and high-quality extrapolation. Extensive experiments demonstrate that DecMem significantly outperforms current state-of-the-art methods. By ensuring precise and efficient long-term memory and achieving superior extrapolation capabilities, DecMem enables minute-level controllable long video generation with high fidelity and consistency.
[CV-31] Interpretability Without Tradeoffs: Disentangling Polysemanticity At Equal Predictive Performance
链接: https://arxiv.org/abs/2605.31304
作者: Doğukan Bağcı,Bernt Schiele,Simone Schaub-Meyer,Jonas Fischer,Robin Hesse
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注: Preprint
Abstract:Deep neural networks (DNNs) are widely used, but interpreting what they actually learn remains difficult. A major obstacle is that individual neurons often encode multiple unrelated concepts, obscuring the decision process of the network. While prior work, such as sparse autoencoders, can separate these mixed signals into more meaningful, “monosemantic” features, this typically requires altering the model in ways that can degrade downstream performance. To overcome this, we introduce ELUDe (explicit, lossless, unsupervised disentanglement), a method for improving the interpretability of DNNs while preserving their functional equivalence. ELUDe breaks latent representations into clear, inspectable sub-units that behave like interpretable features, while guaranteeing that the model’s outputs remain exactly the same. It requires no explicit training, no labels, and can be applied to pretrained models. ELUDe works by reorganizing how information flows between layers, re-routing concept-specific contributions while preserving the original computation by construction. Across several vision models, including DINOv2 and supervised ViT-B/16, ELUDe improves interpretability, keeps downstream accuracy unchanged, runs efficiently, and supports practical uses such as steering model representations. In short, ELUDe offers interpretability (almost) without a tradeoff: clearer, scalable, and actionable model insights with no loss in performance.
[CV-32] okTalk: Expressive Real-time Facial Animation from Audio-LLM Tokens
链接: https://arxiv.org/abs/2605.31294
作者: Qingcheng Zhao,Yifang Pan,Karan Singh
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Recent advances in Audio-LLMs like GPT-4o have ushered in an era of conversational interaction with language models. Conversational avatars however, still seem robotic in facial expression and conversational flow, in part due to sequential stages of speech recognition, text generation, turn-based text response, speech synthesis, and audio driven facial animation. Based on our insight that audio-tokens produced by current Audio-LLMs carry sufficient information to reconstruct a plausible facial performance, we present TokTalk, a system that directly outputs expressive facial animation in real-time from streaming audio-tokens. We construct a novel audio-token to 3D facial motion dataset, on which TokTalk is trained using a Chunk-based Conditional Flow Matching model. A lightweight adaptation strategy allows our trained model to seamlessly connect to any token-based Audio-LLM at minimal computational overhead. Our chunk-based processing further enables parametric trade-off between latency and facial quality, shown through ablation studies. We further show that the real-time performance of TokTalk is comparable in latency to prior art solutions, and significantly favorable (via a perceptual study) in terms of quality, expressivity and control of the 3D facial performance. We showcase TokTalk’s flexibility using a chatbot Avatar, a voice-driven user Avatar, and an animation Director’s interface, as diverse audio-visual face applications.
[CV-33] Authentication of Copy Detection Patterns via Cross-Camera Dual-Synthetic Referencing ICIP2026
链接: https://arxiv.org/abs/2605.31292
作者: Ivan Oleksiyuk,Roman Chaban,Slava Voloshynovskiy
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: To appear in Proc. ICIP2026, September 13-17, 2026, Tampere, Finland
Abstract:Copy Detection Patterns (CDPs) are structures printed on physical objects to enable cost-effective authentication. Verification is achieved by comparing a captured image with the digital template from which the CDP was printed. In practice, printer stochasticity and camera distortions hinder this comparison, limiting robustness against counterfeiting. Prior work addressed camera effects by synthesising reference images in the verification camera domain, but it ignored printing variability. We introduce an enrolment-based cross-camera dual-synthetic referencing framework. Each printed CDP is first captured by a controlled enrolment camera, and a deep-learning-based translator jointly exploits the digital template and the enrolled capture to generate a high-quality reference for the verification image. We provide an information-theoretic justification showing that the dual reference is more informative than template-based references. Experiments on heterogeneous mobile cameras demonstrate improved authentication performance, robustness to machine-learning-based copy attacks, and reliable verification from small CDP regions and on low-end devices.
[CV-34] SAM for Robust Mitochondria Instance Segmentation in Fluorescence Microscopy CVPR2026
链接: https://arxiv.org/abs/2605.31284
作者: Suyog Jadhav,Dilip K. Prasad,Krishna Agarwal
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: Accepted at PHAROS-AIF-MIH workshop @ CVPR 2026
Abstract:The morphological analysis of mitochondria in fluorescence microscopy (FM) is crucial for understanding cellular health, energy production, and metabolic regulation. While foundation models like the Segment Anything Model (SAM) have revolutionized natural image segmentation, their direct application to FM is hindered by a significant domain shift characterized by diffraction-limited resolution, low contrast, and complex overlapping organelle networks. Furthermore, the development of robust models is bottlenecked by a severe lack of high-quality, manually annotated instance segmentation datasets for mitochondria. In this paper, we propose a scalable solution to this data scarcity by finetuning SAM exclusively on synthetically generated FM data. We simulate realistic mitochondria data and emulate the optical properties of fluorescence microscopes to create a large-scale annotated dataset. We evaluate our fine-tuned model on a curated dataset of real, manually annotated FM images. Qualitative and quantitative analyses demonstrate that our synthetically fine-tuned model improves precision and average dice score over strong baselines. This work establishes the potential of simulation-assisted training for FM instance segmentation.
[CV-35] opologically Consistent Multi-view 3D Head Reconstruction via Coarse-Guided Layered Surface Sampling SIGGRAPH
链接: https://arxiv.org/abs/2605.31283
作者: Timo Bolkart,Daoye Wang,Prashanth Chandran
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: SIGGRAPH Conference Papers 2026
Abstract:We present SHELLS (Semantic Head Estimation via Layered Local Sampling), an efficient feed-forward framework for 3D head reconstruction in dense semantic correspondence from multi-view images. Existing methods typically refine vertices independently via localized feature volumes. This approach couples memory-intensive feature sampling to mesh resolution, which limits scalability for dense topologies ( 10k vertices) and introduces surface noise. In contrast, SHELLS decouples feature extraction from mesh resolution via a hierarchical sampling strategy. We extract multi-view features using a DINOv2 backbone with LoRA adaptation, projectively sample a sparse global feature cloud, and predict an intermediate coarse mesh. This coarse prior guides the construction of layered, surface-aware sampling shells that serve as a discrete search space for the final reconstruction. SHELLS maintains surface consistency while using 88% less inference GPU memory (2.4GB vs. 20GB) than volumetric baselines. It reduces median registration error by 21% to 29% with a 3.5x inference speedup (0.08s vs. 0.29s) for 18k-vertex meshes. Notably, our model is trained exclusively on synthetic data yet generalizes effectively to real-world captures, eliminating the need for the costly, pre-registered multi-view datasets common in prior work.
[CV-36] DriveMA: Driving Vision-Language-Action Models with verifiable Meta-Actions
链接: https://arxiv.org/abs/2605.31271
作者: Weicheng Zheng,Yixin Huang,Qiao Sun,Derun Li,Hang Zhao
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: arXiv admin note: text overlap with arXiv:2605.21273
Abstract:Driving Vision-Language-Action Models (Driving VLAs) aim to use language to improve end-to-end planning, but the language-action gap limits this promise. We propose DriveMA, a Driving VLA framework built on verifiable meta-actions, which summarize future ego motion into compact language-domain intentions and can be constructed from expert trajectories with a trajectory-grounded annotation pipeline and can be verified against generated trajectories through rule-based projection. DriveMA exploits this verifiability with action-centric supervised training and a data-efficient turn-level credit assignment reinforcement learning framework, explicitly aligning high-level decisions with low-level trajectory planning through dense rewards and precise credit assignment. DriveMA sets a new state of the art on the Waymo Open Dataset Vision-based E2E Driving, achieving a Rater Feedback Score of 8.060 with a 2B model and further improving it to 8.079 with a 4B model; it also obtains competitive closed-loop planning performance on NAVSIM. These results show that even a simple meta-action interface can achieve state-of-the-art planning when made verifiable and optimized for language-action alignment. Code, data, and models will be released to facilitate future research.
[CV-37] Envisioning Beyond the Few: Disentangled Semantics and Primitives for Few-Shot Atypical Layout-to-Image Generation ICML2026
链接: https://arxiv.org/abs/2605.31266
作者: Nan Bao,Yifan Zhao,Wenzhuang Wang,Jia Li
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Accepted to ICML 2026; code available at this https URL
Abstract:The layout-to-image (L2I) task enables fine-grained control over image generation via object categories and spatial layouts. However, existing L2I methods yield fragmented and distorted generations under few-shot atypical settings. We term this failure as representation fragmentation, arising from a granularity mismatch that entangles semantic identity with visual details. To address this issue, we propose a representation-driven framework that disentangles semantics from primitives for robust few-shot adaptation. Specifically, Semantic Anchoring aggregates categorical semantics into anchors for stable identity, while Primitive Imbuing models recomposable primitives for robust local detail modeling. Conceptual Steering further regulates optimization with a saliency-aware objective to preserve foreground semantic consistency. Extensive experiments demonstrate consistent improvements in the 5-shot regime over state-of-the-art L2I methods in both visual fidelity and alignment across diverse atypical domains. The source code is publicly available at this https URL.
[CV-38] ERGeoBench:A Comprehensive Benchmark for Embodied Reasoning and Geo-localization in Multimodal Large Language Models
链接: https://arxiv.org/abs/2605.31251
作者: Kaiwen Xue,Tao Wei,Guoxin Zhang,Zhonghong Ou,Kaoyan Lu,Yu Feng,Yifan Zhu,Haoran Luo
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Multimodal large language models (MLLMs) have shown strong potential as embodied agents, yet embodied geo-localization remains underexplored due to the lack of fine-grained evaluation. We introduce ERGeoBench, a diagnostic benchmark for vision-driven embodied geo-localization. ERGeoBench evaluates models under three progressive settings – single-view, panorama-view, and embodied-view – where agents may actively acquire observations through sequential changes in yaw, pitch, and zoom. The benchmark contains 2,207 globally distributed street-view panoramas and measures four complementary capabilities: foundational perception, spatial awareness, common sense reasoning, and geo-localization reasoning. Evaluations of leading proprietary and open-source MLLMs show that current models can infer high-level geographic semantics, but still struggle with fine-grained perceptual operations, metric localization, and spatial consistency across views. We further observe that geo-localization is strongly correlated with the other capability dimensions, suggesting that accurate localization depends on integrated perception, spatial reasoning, and commonsense inference rather than isolated visual recognition. Overall, ERGeoBench provides a unified framework for diagnosing and advancing human-like embodied geo-localization. Project Page: this https URL
[CV-39] BadBone: Backdoor Attacks Against Backbone Models in Visual Prompt Learning
链接: https://arxiv.org/abs/2605.31246
作者: Ziqing Yang,Rui Wen,Xinlei He,Yun Shen,Michael Backes,Yang Zhang
类目: Cryptography and Security (cs.CR); Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted by IEEE Transactions on Information Forensics Security
Abstract:Prompt learning is a new machine learning paradigm that has attracted ample attention due to its simplicity and proven efficacy. Despite its growing adoption, the security vulnerabilities associated with this paradigm remain underexplored. In this work, we take the first step to propose BadBone, a stealthy and adaptive backdoor attack against prompt learning using bi-level optimization. Instead of backdooring the prompt learning process, we aim to compromise a backbone model such that only target downstream tasks employing prompt learning inherit the backdoor vulnerability. Extensive experiments on three different models and three datasets from various domains show that our targeted/untargeted backdoored models achieve high attack performance while maintaining utility on both pre-training and downstream tasks. Moreover, we evaluate our approach against six state-of-the-art model-level defenses, including Neural Cleanse, ABS, MNTD, NAD, CLP, and D-BR. The results demonstrate that these defenses are largely ineffective against our backdoored models and thus leave the effective defense as an important direction for future work.
[CV-40] Beyond Classification: Dynamic Adapter Routing for Continual Multimodal Retrieval
链接: https://arxiv.org/abs/2605.31229
作者: Alicja Dobrzeniecka,Filip Szatkowski,Sebastian Cygert,Szymon Lukasik,Bartlomiej Twardowski
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:While retrieval is a core function of vision-language models, continually updating these models for retrieval tasks remains critically underexplored. Existing work often approaches continual retrieval through the lens of class-incremental learning (CIL), evaluating both standard CIL methods and retrieval-oriented adaptations in settings that may not fully capture the retrieval-specific dynamics. To address this, we introduce a new, principled evaluation framework for continual multimodal retrieval (CMR) spanning diverse visual domains, and systematically evaluate common approaches within this setting. Our empirical analysis shows that standard CIL methods fail to yield meaningful gains in our more challenging scenario. Therefore, we propose Dynamic Adapter Routing (DAR), a novel approach based on adapters selected through prototype-based routing and combined via model this http URL achieves superior performance over the previous baselines and demonstrates strong generalization under out-of-distribution evaluation. Our results highlights the unique challenges of CMR and encourages further research in this direction.
[CV-41] HiERO-StepG @ Ego4D Step Grounding Challenge: hierarchical activity understanding enables zero-shot step grounding CVPR2026
链接: https://arxiv.org/abs/2605.31227
作者: Andrea Zenotto,Simone Alberto Peirone,Francesca Pistilli,Giuseppe Averta
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Technical report for the Ego4D Goal Step - Step Grounding challenge at CVPR 2026, derived from arXiv:2505.12911
Abstract:Procedural activities follow well-defined structures: whether we consider a cooking recipe or a mechanic repairing a car, these activities naturally decompose in a hierarchy of steps and sub-steps. Traditional approaches for step grounding require extensive annotations and scale poorly. Instead, we argue that such hierarchical structure can emerge naturally from uncurated videos of human activities through recurring patterns of co-occurring actions and activities. Our approach builds on HiERO, a weakly-supervised representation learning approach that maps close in the feature space actions that are functionally related to each other, leveraging only fine-grained action-level narrations. In this feature space, procedure steps can be detected by a simple clustering, with no additional task-specific fine-tuning. For the Ego4D Step Grounding challenge, we augment this approach by ensuring fine and coarse level agreement in step assignments, enforcing strict temporal monotonicity of the grounded steps and post-processing the detected steps to reduce the impact of noisy predictions. We call this approach HiERO-StepG and it achieves 56.27 % on the R@1 (IoU = 0.3) metric on the global leaderboard at submission time, ranking second while being completely zero-shot and not requiring procedure-specific annotations. Project page: this https URL.
[CV-42] Latent Geometric Chords for Query-Efficient Decision-Based Adversarial Attacks
链接: https://arxiv.org/abs/2605.31219
作者: Ei Hmue Khine,Yao Li,Jiebao Sun,Shengzhu Shi,Zhichang Guo,Boying Wu
类目: Computer Vision and Pattern Recognition (cs.CV); Cryptography and Security (cs.CR); Machine Learning (cs.LG)
备注: 14 pages, 9 figures, 7 tables. Submitted to IEEE Transactions on Information Forensics and Security. The source code is available at this https URL
Abstract:While decision-based black-box adversarial attacks present a severe security threat, current methodologies suffer from fundamental limitations. Pixel-wise attacks frequently introduce unnatural, high-frequency visual artifacts, while latent-space frameworks are confined by the limited search space of low-dimensional manifolds and inherent reconstruction flaws. To resolve these limitations, we propose Latent Geometric Chords (LGC) for Query-Efficient Decision-Based Adversarial Attacks alongside a variant, LGC-H. At its core, LGC navigates decision boundaries by executing a curvature-aware geometric search within a compressed semantic manifold. To guarantee high visual fidelity and circumvent dimensionality bottlenecks, we introduce a Residual-based Adversarial Generation (RAG) mechanism. RAG isolates semantic perturbations as geometric chords and superimposes them directly onto the original source image. RAG substantially resolves baseline reconstruction flaws and effectively doubles the permissible search space dimensions. Experimental results demonstrate that LGC achieves robust cross-dataset transferability and substantially outperforms state-of-the-art baselines. Notably, our method, LGC, minimizes perturbation magnitudes while achieving state-of-the-art visual fidelity–with a Structural Similarity Index Measure (SSIM) exceeding 0.99 and a Learned Perceptual Image Patch Similarity (LPIPS) below 0.01 at 5000 queries–and sustaining high attack success rates under stringent perceptual constraints, successfully compromising adversarially trained robust models. The source code is available at: this https URL.
[CV-43] ALON: Token-Aligned Lightweight Adapters for 6-DoF Spacecraft Pose Estimation
链接: https://arxiv.org/abs/2605.31217
作者: Abid Ali,Arunkumar Rathinam,Djamila Aouada
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 13 pages paper with 3 figures in total
Abstract:Monocular 6-DoF spacecraft pose estimation methods predominantly process individual frames, discarding the temporal information present in an image sequence acquired during spacecraft manoeuvres. Few temporal approaches require full backbone fine-tuning or auxiliary optical flow networks, risking catastrophic forgetting or increasing computational cost, respectively. We propose TALON (Token-Aligned Lightweight adapters for Orbital Navigation): spatiotemporal 3D adapters injected before the self-attention layers of a frozen ViT vision transformer, combined with a patch-token alignment loss that geometrically grounds the adapted features to keypoint structure through a prototype-conditioned KL-divergence objective. Pre-attention placement allows the frozen attention to reason over temporally enriched tokens, achieving stronger performance with a single adapter per block than post-attention alternatives. The alignment loss shapes the intermediate representations so that each keypoint induces a spatially precise activation in the token field, while the framework adds less than 5% parameters to the frozen backbone. On SPADES dataset, TALON reduces the pose error by 50% over the prior state-of-the-art, and on SwissCube dataset it surpasses the prior best by 21.8% in ADD-0.1d accuracy. Zero-shot cross-domain evaluation from sim-to-real on SPARK real data reduces pose error by 4.7x, and ablations characterise the role of adapter depth across in-domain and cross-domain settings.
[CV-44] Fixed-Point Masked Generative Modeling
链接: https://arxiv.org/abs/2605.31215
作者: Andrea Miele,Yiming Qin,Alba Carballo-Castro,Justin Deschenaux,Pascal Frossard
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Masked Generative Models (MGMs) enable parallel decoding and achieve strong performance across modalities, but require full-sequence bidirectional transformers at every step, making training costly and degrading quality under low sampling budgets. Existing work improves efficiency via better samplers or cheaper fixed-depth denoisers, but they still allocate a fixed amount of denoiser computation to each refinement step. We introduce Fixed-Point Masked Generative Models (FP-MGMs), which replace part of the denoiser with a fixed-point solver over shared attention layers to enable adaptive depth with fewer parameters. To make it more effective for masked generation, we first introduce a cross-step consistency loss, which aligns hidden representations at neighboring denoising steps and, second, three-state reuse (3SR) which warm-starts the solver using the previous solution by treating differently unchanged, still-masked, and newly revealed tokens respectively. Together, these components define our complete training-to-inference framework for fixed-point masked generation, \emphCoFRe. We also show that pre-trained MGMs can be converted into FP-MGMs with short fine-tuning, avoiding full retraining. Across modalities, CoFRe improves the quality and cost trade-off. On OpenWebText, CoFRe reduces parameters by 38.8%, training time by 11.5%, and VRAM by 16.9%, while improving generative perplexity from 830.8 to 101.8 at a budget of 96 transformer-block forward passes, compared to MDLM. In ImageNette, CoFRe reduces training time by 48.6% and VRAM by 50.7%, while improving FID in all sample budgets tested. Overall, CoFRe offers a practical framework for cheaper training and stronger low-budget masked generation.
[CV-45] Probabilistic Precipitation Nowcasting with Rectified Flow Transformers CVPR2026
链接: https://arxiv.org/abs/2605.31204
作者: Johannes Schusterbauer,Jannik Wiese,Nick Stracke,Timy Phan,Björn Ommer
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: CVPR 2026, Project Page: this https URL
Abstract:Accurate weather forecasts are essential across various domains and are safety-critical in extreme weather conditions. Compared to simulation-based forecasting, data-driven approaches show greater efficiency, enabling short-term, high-resolution nowcasting. In particular, diffusion models proved effective in weather nowcasting due to their strong probabilistic foundation. However, existing methods rely on deterministic compression to reduce the complexity of high-dimensional weather data, limiting their ability to capture uncertainty in the decoding process. In this work, we introduce \textbfFREUD , a \textbfFr ame-wise \textbfE ncoder and \textbfU nited \textbfD ecoder model based on rectified flow transformers for efficient compression of spatio-temporal weather data. Frame-wise encoding enables continuous forecast updates, while the unified video decoder ensures temporal consistency. Our uncertainty-preserving first stage allows us to capture aleatoric uncertainty via ensembling, which is particularly beneficial for extreme weather events with high decoding variability. We achieve state-of-the-art performance in precipitation nowcasting with a compact latent-space rectified flow transformer on the SEVIR benchmark and show further performance gains by model and test-time scaling. Code available here: this https URL
[CV-46] he Regularizing Power of Language-Training Deepfake Detectors
链接: https://arxiv.org/abs/2605.31192
作者: Benedikt Hopf,Zongwei Wu,Radu Timofte
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Recently, thanks to the advent of Multimodal-LLMs, deepfake detectors are striving not only to be generalizable but also interpretable. We propose that these two challenges can effectively be tackled jointly, since describable artifacts typically generalize better, opening the possibility to use language as a regularization mechanism. Since deepfake detection generally suffers from overfitting to low-level domain-specific artifacts, our intuition is that an LLM that has been pretrained on language would prefer high-level artifacts that can be described better. This way, we can use high-level features where possible, while training the model to use low-level features where necessary. We utilize a dual-encoder architecture, pairing a frozen specialist detector with a LoRA-tuned MLLM encoder, and a two-stage training curriculum: first, a binary alignment phase demonstrates that the intrinsic capability of MLLMs can effectively combine features to mitigate overfitting to dataset-specific artifacts. To further bolster generalization and achieve interpretability, we employ a reinforcement learning stage that encourages the model to generate descriptive reasoning before classifying, using only binary labels. By rewarding this “explain-then-classify” behavior, we explicitly incentivize the model to prioritize high-level, robust features. Crucially, this process yields both interpretable descriptions and a further boost in cross-dataset performance, even when reasoning chains are omitted at inference. Extensive experiments on benchmark datasets validate our approach, outperforming state-of-the-art methods by a large margin.
[CV-47] Student Capacity Moderates Knowledge Distillation Effectiveness: A Systematic Study Across ResNet Teacher-Student Pairs on CIFAR-10
链接: https://arxiv.org/abs/2605.31191
作者: Umut Onur Yasar
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注: 9 pages, 2 figures, 5 tables. Code available at this https URL
Abstract:We investigate how teacher-student capacity relationships modulate knowledge distillation (KD) effectiveness in ResNet-based image classification on CIFAR-10. Across three teacher-student pairs – R50-R18, R34-R18, and R50-R34 – we compare Logit-KD and Feature-KD under controlled, reproducible conditions (3 seeds, mean+/-std reported throughout). We report three main findings. First, student capacity is a key moderating factor in distillation gain: R34 students benefit substantially more from KD than R18 students even when teacher-student accuracy gaps are comparable, with the strongest gain of +0.30pp observed for R50-R34 Feature-KD versus +0.18pp for R34-R18 Feature-KD and +0.00pp for R34-R18 Logit-KD. Second, implementation correctness critically affects Feature-KD: a gradient clipping bug that excluded projection layers suppressed Feature-KD performance and produced misleading comparisons with Logit-KD. After correction, Feature-KD matches or outperforms Logit-KD in two of three pairs, reaching 95.55% on R50-R34 against a baseline of 95.25%. Third, input-resolution-aware architecture is a prerequisite for effective distillation: correcting the ResNet stem for 32x32 inputs raises teacher accuracy by over 5pp – an order of magnitude larger than any KD gain. All code and results are available at this http URL.
[CV-48] From Local Geometry to Global Pseudo Labeling for Robust Positive Unlabeled Learning under Covariate Shift
链接: https://arxiv.org/abs/2605.31187
作者: Firas Gabetni,Alexandre Rocchi Henry,Nacim Belkhir,Ziyi Liu,Gianni Franchi
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Detecting covariate shift is critical for building reliable vision systems. While most prior work focuses on improving robustness to shift, explicitly detecting covariate shift remains underexplored. Existing approaches typically rely on fully supervised training, requiring labeled examples from both original and shifted distributions, which is often impractical. In this paper, we show that covariate shift detection can be effectively addressed with weaker supervision using Positive Unlabeled (PU) learning. However, under covariate shift, in distribution and shifted data overlap significantly, making classical PU methods unstable and sensitive to noise. To overcome this challenge, we introduce Spectral PU Neighborhood Annotation (SPUNA), a geometry aware framework that progressively discovers shifted data by leveraging the local manifold structure of visual features. Extensive experiments show that SPUNA achieves state of the art performance in PU settings and remarkably matches the performances of fully supervised methods. Moreover, our approach transfers robustly across different types of shifts, demonstrating strong generalization capabilities.
[CV-49] Vanilla ViT for Automotive Point Cloud Semantic Segmentation
链接: https://arxiv.org/abs/2605.31177
作者: Gilles Puy,Nermin Samet,Alexandre Boulch,Spyros Gidaris,Tuan-Hung VU,Renaud Marlet
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Plain Transformers have become the de-facto architecture for processing text, audio, image, and video, offering a unified backbone for multimodal learning. However, state-of-the-art architectures for point cloud semantic segmentation remain dominated by U-Nets architectures where convolutions are interleaved with local or windowed attentions. In this work, we show how to effectively leverage vanilla, non-hierarchical ViTs for segmentation of large-scale automotive lidar scenes. We bridge the performance gap thanks to a carefully designed tokenizer, a lightweight decoder segmentation head, and tailored data augmentations. Our approach, VaViT for Vanilla ViT, matches or exceeds the performance of state-of-the-art methods while maintaining the simplicity of ViT architecture. We provide extensive evaluations on nuScenes, SemanticKITTI, and Waymo Open Dataset to validate the efficiency of our method. Code and models are available at this https URL.
[CV-50] Detect in Any Scene: An Agent ic Framework for Object Detection with Experience-Aware Reasoning
链接: https://arxiv.org/abs/2605.31174
作者: Wenlun Zhang,Jun Yin,Kentaro Yoshioka
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Object detection in real-world scenarios remains challenging due to diverse image degradations and heterogeneous object distributions, which significantly hinder the generalization of existing detectors. Conventional approaches, including scene-specific representation learning and end-to-end pipeline design, are inherently limited by their reliance on predefined conditions and lack adaptability to dynamic environments. In this paper, we propose DetAS, an agentic detection framework that formulates object detection as a dynamic decision process. Instead of relying on static pipelines, DetAS leverages a Multimodal Large Language Model (MLLM) as a central agent to adaptively compose detection workflows by selecting from a toolbox of restoration modules and specialized detectors. Specifically, DetAS consists of two key components: Self-Adaptive Image Restoration, which dynamically determines whether and how to enhance images for downstream detection, and Multi-Expertise Detection, which integrates multiple domain-specialized detectors and resolves their predictions through instance-level reasoning. To further improve decision quality under fine-grained conditions, we introduce Self-Evolving Experience Harvesting and extend the framework to DetAS-X, which accumulates node-level decision experience from a small set of annotated data and enables experience-aware reasoning during inference. This mechanism allows the system to progressively refine its decision policy and adapt to diverse real-world scenarios. Extensive experiments on six challenging benchmarks demonstrate that DetAS-X significantly outperforms existing MLLM-based detectors, achieving an average improvement of 28.36% in F1 score, with up to 37.01% gain on DarkFace. These results demonstrate the promise of agentic detection and establish a solid foundation for its application in complex and dynamic environments.
[CV-51] Guidance for Low-Level Perceptual Editing in Unconditional Diffusion Models CVPR2026
链接: https://arxiv.org/abs/2605.31162
作者: Shreyansh Modi,Akshat Tomar,Aarush Aggarwal
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 11 pages, 12 figures, Generative Models for Computer Vision Workshop CVPR 2026
Abstract:Unconditional diffusion models offer powerful generative priors, yet steering them toward aesthetically enhanced outputs remains largely unexplored. We show that h-space patching, the dominant paradigm for training-free diffusion editing, systematically fails for global, low-level transformations required for aesthetic and perceptual refinement. We introduce a novel, generalized framework for image-editing in unconditional diffusion models without explicit training. This inference-time mechanism operates on low-level features by extracting degradation concept vectors and combining bottleneck patching with classifier-free guidance to guide sampling away from the degraded manifold, producing consistently improved images without any model retraining.
[CV-52] Light Interaction: Training-Free Inference Acceleration for Interactive Video World Models
链接: https://arxiv.org/abs/2605.31158
作者: Jiacheng Lu,Haoyi Zhu,Sipei Yi,Enze Xie,Yu Li,Cheng Zhuo
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 13 pages, 6 figures, 3 tables. Project page: this https URL
Abstract:Interactive video world models generate video chunk by chunk in response to user-controlled camera movements, enabling applications such as real-time game simulation, virtual scene navigation, and embodied AI training. However, scaling to long interactive trajectories is prohibitively expensive due to growing context memory, quadratic attention complexity, and repeated denoising steps. We present Light Interaction, a training-free inference acceleration framework for interactive video world models. Our key insight is that interaction naturally enables trajectory-dependent adaptive computation: retrieved spatial memory can be discarded during novel exploration, temporal context can be adjusted according to local latent dynamics, and early-step model outputs can be reused when the camera revisits familiar regions. Based on this insight, Light Interaction combines adaptive context management, denoising cache acceleration, and hardware-software co-designed 3D block sparse attention with fused Triton kernels. Evaluated on HY-WorldPlay and Matrix-Game-3.0, Light Interaction achieves up to 2.59x speedup without model retraining while maintaining competitive visual quality.
[CV-53] BIAS-ID: A Framework for Analyzing Transformation Biases in AI-Generated Image Detectors
链接: https://arxiv.org/abs/2605.31153
作者: Jonas Ricker,Asja Fischer,Erwin Quiring
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Given the surge of harmful AI-generated imagery online, reliably distinguishing authentic images from generated ones has become an urgent research topic. While many proposed detection methods perform well under controlled settings, they often collapse when tested on real-world data. A potential root cause are subtle biases in the detectors’ training data. As a result, detectors may rely on spurious correlations instead of learning true forensic artifacts. While a recent line of work has identified the problem, there is not yet an established protocol to evaluate how biased a detector actually is. In this work, we therefore take a step back: First, we discuss what it means for a detector to be biased, and how this differs from a lack of robustness. Second, we propose BIAS-ID, a transparent framework for analyzing and quantifying the presence of transformation biases in AI-generated image detectors. We validate our framework by performing an evaluation of six detectors across two datasets, revealing that several state-of-the-art detection methods are strongly affected by biases. Our results highlight the importance of bias-aware evaluation for developing reliable AI-generated image detectors.
[CV-54] FOCUS: Forcing In-Context Object Localization through Visual Support Constraints and Policy Optimization ICML2026
链接: https://arxiv.org/abs/2605.31145
作者: Mohammed Asad Karim,Vinay Kumar Verma
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Accepted at ICML 2026. * Equal Contributions
Abstract:In-context localization (ICL) seeks to localize a target object specified by a small set of support examples in a query image, operating on the fly without training or parameter updates. Despite rapid advances in vision-language models (VLMs), achieving category-agnostic and visually grounded ICL remains an open problem, even though it is essential for applications such as image editing, personalized visual search, and retrieval. Existing methods are fragile and rely on explicit category supervision, which not only limits applicability in realistic settings with unnamed or instance-specific objects but also introduces category bias that steers predictions toward semantic priors rather than visual evidence. We introduce a two-stage training framework that explicitly optimizes in-context attention between support bounding boxes and query images without category supervision. We further refine localization via reinforcement learning using Group Relative Policy Optimization (GRPO) to directly minimize localization error. This formulation enforces visual correspondence over semantic priors, yielding robust instance-level localization. Empirically, a 7B-parameter model trained with our objectives outperforms models up to 72B parameters, demonstrating that context-aware localization objectives can surpass scaling alone. Comprehensive ablations validate the contribution of each component.
[CV-55] PolSAR Image Classification using a Hybrid Complex-Valued Network (HybridCVNet)
链接: https://arxiv.org/abs/2605.31137
作者: Mohammed Q. Alkhatib
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted and Published in IEEE Geoscience and Remote Sensing Letters (GRSL)
Abstract:Recently, convolutional neural networks (CNNs) have become popular for image classification due to their effectiveness in computer vision tasks. Now, researchers are exploring the potential of vision transformers (ViTs) in remote sensing and Earth observation. However, traditional Real-Valued networks often overlook important phase information in Complex-Valued (CV) data like polarimetric synthetic aperture radar (PolSAR) data. To address this, new CV deep architectures have emerged. HybridCVNet, a novel hybrid network, blends CV-CNN and CV vision transformer (CV-ViT) techniques. It efficiently combines CV 3D and 2D CNNs as feature extractors, enhancing PolSAR image classification by extracting complementary information and effectively leveraging interdependencies within the data. Experimental results from widely-used PolSAR datasets show HybridCVNet outperforms other methods, achieving an overall accuracy of 97.39% on the Flevoland dataset and showing promise even with just a 1% sampling ratio, with a Kappa value of 0.972 on the San Francisco dataset. Source code is accessible through this https URL
[CV-56] QVGGT: Post-Training Quantized Visual Geometry Grounded Transformer CVPR2026
链接: https://arxiv.org/abs/2605.31124
作者: Zhizhen Pan,Hesong Wang,Huan Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted by CVPR 2026. Project page: this https URL
Abstract:Estimating 3D attributes directly from images has advanced rapidly with the Visual Geometry Grounded Transformer (VGGT), which predicts camera parameters, depth maps, and point clouds in a single forward pass. However, its 1.2B-parameter scale severely limits deployment on resource-constrained platforms such as UAVs and mobile AR devices. To address this limitation, we introduce QVGGT, a tailored quantization framework designed to compress VGGT. Our approach starts from the observation that transformer blocks within VGGT exhibit heterogeneous sensitivity to quantization. We thus analyze per-block quantization sensitivity and propose a selective mixed-precision strategy that allocates higher precision to the most fragile transformer blocks. To address the amplification of quantization error caused by high-variance camera and register tokens, we further introduce token filtering with camera information compensation, which removes these outliers from activation calibration and restores their geometric cues using a PCA-derived global compensation token. Finally, we develop a task-aware scale search mechanism that evaluates candidate quantization scales not only through layer reconstruction but also through multi-head supervision and cross-head geometric consistency among camera poses, depth maps, and point maps. Extensive experiments on multiple geometry perception benchmarks demonstrate that QVGGT achieves near-lossless W4A16 quantization, preserving the accuracy of all 3D prediction heads while delivering 3 \sim 4.9 \times memory reduction and up to 2.8 \times real hardware speedup over FP32. Our approach makes high-fidelity 3D perception feasible on edge devices, enabling practical deployment of feed-forward 3D reconstruction models in real-world constrained environments.
[CV-57] NTR: Neural Token Reconstruction for Scene Token Bottleneck in End-to-End Driving
链接: https://arxiv.org/abs/2605.31116
作者: Jiahui Li,Jiawei Sun,Zixiang Ren,Ming Liu,Jiamin Shi,Ruiteng Zhao,Zhiyang Liu,Liying Liu,Zuoguan Wang,Kaidi Yang
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注:
Abstract:Recent perception-free end-to-end (E2E) autonomous driving methods bypass explicit perception outputs by compressing dense image patch tokens into compact scene tokens for downstream trajectory generation and scoring. While these scene tokens form a compact visual bottleneck for the planner, they receive supervision solely from the planning objective, providing limited constraints on the encoded visual information. To address this limitation, we introduce Neural Token Reconstruction (NTR), a representation learning framework to directly constrain the compact scene-token bottleneck in perception-free driving. NTR introduces a self-distillation masked latent reconstruction objective that reconstructs masked patch-level latent features using only compact scene tokens as reconstruction memory. This forces reconstruction gradients to pass exclusively through the scene-token bottleneck, encouraging scene tokens to preserve richer and less redundant visual representations for planning. We further introduce semantic priors derived from foundation-model annotations as a weak semantic interface biasing reconstruction targets toward driving-related structures without introducing explicit perception heads. All auxiliary reconstruction components are removed at inference time, leaving the deployed planner unchanged. NTR achieves state-of-the-art performance on three public autonomous driving benchmarks, including 8.0461 RFS on Waymo E2E and 94.1 PDMS / 90.9 EPDMS on NavSim12. The learned scene tokens exhibit lower pairwise redundancy and higher effective rank, indicating that effective bottleneck supervision improves both compact visual representation learning and planning performance.
[CV-58] Polyphony: Diffusion-based Dual-Hand Action Segmentation with Alternating Vision Transformer and Semantic Conditioning CVPR2026
链接: https://arxiv.org/abs/2605.31115
作者: Hao Zheng,Hu Wang,Tiantian Zheng,Prajjwal Bhattarai,Tuka Alhanai
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: CVPR 2026
Abstract:Dual-hand action segmentation, densely predicting actions for both hands from untrimmed videos, is essential for understanding complex bimanual activities. However, it poses several unique challenges: complex inter-hand dependencies, visual asymmetry between hands, representation conflicts where the dominant hand monopolizes gradients, and semantic ambiguity in fine-grained actions. We propose Polyphony, a three-stage method to address these challenges through: (1) an Alternating Dual-Hand Vision Transformer that alternates training between left- and right-hand mini-batches to ensure balanced gradient contributions from both hands while sharing a spatio-temporal encoder; (2) Semantic Feature Conditioning that aligns visual features with structured, compositional action descriptions to enhance discrimination of semantically similar actions; and (3) Diffusion-Based Segmentation with cross-hand feature fusion for inter-hand coordination and adaptive loss weighting for balancing performance. Polyphony achieves state-of-the-art on both dual-hand datasets (HA-ViD, ATTACH) with improvements up to 16.8 points, and on the single-stream Breakfast dataset (82.5%), outperforming the prior best method that uses a 12x larger backbone. Notably, our unified model with a single shared backbone surpasses baselines requiring separate per-hand models. Code is at this https URL.
[CV-59] Remembering by Reconstructing: Domain Incremental Learning With Test-Time Training on Video Streams
链接: https://arxiv.org/abs/2605.31108
作者: Jonathan Swinnen,Tinne Tuytelaars
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:In this work we introduce a novel approach to domain incremental learning, adapting models over time to evolving, non-stationary data. In contrast to other works, we do not attempt to avoid catastrophic forgetting, but rather allow it and exploit it. Our model combines a main task head with a self-supervised masked autoencoder (MAE) head. We then learn domain-specific LoRA adapters during incremental training. Each adapter specializes to its domain, naturally inducing forgetting on other domains in both heads. At inference, we perform online test-time training on the self-supervised MAE head to identify which LoRAs best matches the current input, so the model can `remember’ the domain again. Our scheme is especially well-suited to real-world streaming data, such as video, where consecutive samples are highly correlated and domain shifts are gradual. We demonstrate our method on domain-incremental action recognition and semantic segmentation tasks.
[CV-60] VGR: Internalizing Visually Grounded Reasoning for MLLM s with Reinforcement Learning ICML2026
链接: https://arxiv.org/abs/2605.31096
作者: Chang-Bin Zhang,Yujie Zhong,Qiang Zhang,Kai Han
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted by ICML 2026
Abstract:While visually grounded Chain-of-Thought (CoT) has emerged as a promising paradigm to enhance fine-grained perception in multimodal large language models (MLLMs), its efficacy during the inference phase remains underexplored. In this work, we empirically find that mandating explicit object boxes in visually grounded CoT during inference often degrades performance compared to standard textual CoT, which reasons without explicit visual grounding. We hypothesize that the visual localization capability can be internalized into the textual CoT and that the mandatory explicit grounding introduces unnecessary interference with the model’s primary objective of answer prediction. To address this problem, we propose Internalizing Visually Grounded Reasoning (\textbfiVGR), a novel reinforcement learning framework that transfers localization capabilities into the textual reasoning process. We employ a dual-stream training strategy, where a textual stream is aligned with a high-quality visually grounded stream via a proposed consistency reward, enabling the model to localize accurately without explicit grounding during inference. Extensive experiments demonstrate that our method significantly outperforms existing baselines on fine-grained benchmarks, while maintaining the flexibility to support tool-assisted inference workflows.
[CV-61] Redefining Instance Matching: A Unified Framework for Part-Aware Matching in Panoptic Segmentation Evaluation
链接: https://arxiv.org/abs/2605.31094
作者: Erik Großkopf,Soumya Snigdha Kundu,Hendrik Möller,Nicolas Münster,Mehdi Astaraki,Paula Tamara Buzduga,Kerstin Ritter,Benedikt Wiestler,Jan Kirschke,Jonathan Shapey,Tom Vercauteren,Florian Kofler
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: 9 pages, 4 figures
Abstract:The Panoptic Quality (PQ) metric is the standard for jointly evaluating instance and semantic segmentation. However, its original definition relies on a One-to-One matching between predicted and ground truth segments, which is only straightforward when the IoU threshold exceeds 0.5. Below 0.5, multiple matching strategies emerge in a poorly explored problem space. We systematically elucidate this space by recasting segment matching as a constrained bipartite assignment problem. Independently bounding the prediction- and ground-truth-side degrees yields four matching strategies: One-to-One, Many-to-One, One-to-Many, and Many-to-Many. We show that the first three are well-defined within the PQ framework, while Many-to-Many falls outside it. These strategies become relevant when instances are fragmented, adjacent objects are difficult to delineate, or annotations are noisy. Central to our framework is a vertex-based accounting of TP, FN, and FP, anchored to ground truth and predicted segments rather than to matching edges. We further show that the framework extends naturally to part-aware panoptic segmentation, and we explore part-aware evaluation on biomedical data. Across configurable case studies we report how different combinations of thresholds and matching strategies behave in practice. We release a unified open-source package built on Panoptica. It exposes Voronoi-based region-wise analysis, part-aware evaluation, and Area Under Threshold Curve computations as configurable options.
[CV-62] Cross-Modal Clinical Knowledge Integration for Mammography Report Generation
链接: https://arxiv.org/abs/2605.31093
作者: Jiayi Zhu,Fuxiang Huang,Yu Xie,Xi Wang,Zhixuan Chen,Yuan Guo,Qingcong Kong,Zhenhui Li,Qiong Luo,Hao Chen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 16 pages, 5 figures
Abstract:Breast cancer is a major global health concern, and mammography screening plays a central role in early detection. The large volume of screening examinations creates a substantial workload for radiologists, making accurate and consistent report generation a critical clinical challenge. Existing automated mammography report generation methods primarily focus on direct visual-to-text mapping, while overlooking the structured clinical reasoning process followed by radiologists in real-world practice. To address this limitation, we propose MammoRG, a mammography report generation framework that explicitly simulates the clinical reporting workflow by following the BI-RADS guideline and incorporating prior clinical knowledge to produce diagnostic reports. Specifically, MammoRG adopts a two-stage training framework. In the first stage, the model learns to integrate clinically relevant prior knowledge from a patient’s four-view mammograms through classification-based supervision. In the second stage, a terminology-aware supervised fine-tuning strategy is introduced to model mammography-specific clinical terms as atomic semantic units, enabling the generation of high-quality reports with improved clinical consistency. To facilitate clinical efficacy evaluation of generated reports, we further develop MammoRGTool, a dedicated mammography report parsing tool that extracts structured clinical information from free-text reports. Extensive experiments demonstrate that MammoRG consistently outperforms existing methods across multiple clinical efficacy metrics, particularly in diagnosis-related BI-RADS F1, where it surpasses the second-best model by 2.73%, 2.04%, 1.90%, and 3.27% on the internal, external 1, external 2, and VinDr-Mammo datasets, respectively.
[CV-63] On Revisiting Entropy for Identifying Mislabeled Images ICML2026
链接: https://arxiv.org/abs/2605.31090
作者: Chunlei Li,Zixuan Zheng,Yilei Shi,Guanglu Dong,Pengfei Li,Jingliang Hu,Xiao Xiang Zhu,Lichao Mou
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: ICML 2026
Abstract:Mislabeled samples in training datasets severely degrade the performance of deep networks, as overparameterized models tend to memorize erroneous labels. We address this challenge by proposing a novel approach for mislabeled data detection that leverages training dynamics. Our method is grounded in the key observation that correctly labeled samples exhibit consistent entropy decrease during training, while mislabeled samples maintain relatively high entropy throughout the training process. Building on this insight, we introduce a signed entropy integral (SEI) statistic that captures both the magnitude and temporal trend of prediction entropy across training epochs. SEI is broadly applicable to classification networks and demonstrates particular effectiveness when integrated with contrastive language-image pretraining (CLIP) architectures. Through extensive experiments on four medical imaging datasets – a domain particularly susceptible to labeling errors due to diagnostic complexity – spanning diverse modalities and pathologies, we demonstrate that SEI achieves state-of-the-art performance in mislabeled data identification, outperforming existing methods while maintaining computational efficiency and implementation simplicity. Our code is available at this https URL.
[CV-64] ask-Focused Memorization for Multimodal Agents
链接: https://arxiv.org/abs/2605.31075
作者: Tao Zou,Yichen He,Tian Qiu,Yuan Lin,Hang Li
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Long-term memory is essential for multimodal agents to build coherent experience, accumulate world knowledge, and achieve continual learning. However, constructing effective memory goes beyond memory module design and basic requirements such as accuracy and fidelity; the key challenge lies in determining what to memorize. Multimodal agents, such as embodied agents, continuously perceive, reason, and act in real or virtual environments, receiving an unbounded stream of multimodal observations. From this combinatorial explosion of information, an agent must selectively retain content that is relevant to its role in the environment and valuable for future tasks. To bridge this gap, we frame memory generation as a learnable memorization policy and introduce TaskMem (Task-focused Memorization Policy Learning), a reinforcement-learning-based framework that enables the policy to dynamically adjust its focus to the demands of real tasks encountered in the environment. TaskMem adopts a two-phase training paradigm: Phase One learns how to memorize by optimizing memory quality under fundamental fidelity requirements; Phase Two occurs after deployment, where the agent learns what to memorize by tuning an adapter on its base MLLM, using recent environment tasks to define a reward model that guides the memorization policy toward task-relevant content. To evaluate our approach, we reformulate VideoMME, EgoLife, and EgoTempo into streaming benchmarks that simulate a realistic setting in which an agent processes streaming observations and handles tasks arriving online. To isolate memory assessment, the questions must be answered using only the agent’s memory, without access to raw video. Built on Qwen3-VL-30B-A3B, TaskMem improves VQA accuracy by 6.3%, 7.0%, and 5.3% on these benchmarks, respectively.
[CV-65] HQ-JEPA: Hybrid Quantum Joint-Embedding Predictive Architecture for Cross-Modal Remote Sensing Representation Learning
链接: https://arxiv.org/abs/2605.31068
作者: Md Aminur Hossain,Ayush V. Patel,Sanjay K. Singh,Biplab Banerjee
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 19 pages
Abstract:We introduce HQ-JEPA, a hybrid quantum-classical joint-embedding predictive architecture for cross-modal remote sensing representation learning. The proposed framework extends JEPA-style masked latent prediction to paired Sentinel-1 and Sentinel-2 imagery by predicting masked target representations from visible context regions while aligning heterogeneous modality features in a shared embedding space. To improve representation quality, HQ-JEPA combines four complementary objectives: latent token prediction, cross-modal token alignment, SIGReg-based Gaussian regularization in the fused latent space, and a differentiable SWAP-test-based Fidelity Quantum Similarity (FQS) loss. Unlike pixel reconstruction methods, HQ-JEPA learns semantic representations directly in latent space and uses quantum state-overlap-based similarity as an additional regularization signal. We evaluate the pretrained encoder on GeoBench classification and segmentation tasks under linear probing and fine-tuning settings. Results show that HQ-JEPA achieves competitive and often superior performance over strong self-supervised and remote sensing foundation-model baselines, demonstrating the benefit of integrating predictive self-supervision, cross-modal geometric regularization, and quantum fidelity-based representation learning for remote sensing applications.
[CV-66] LVSA: Training-Free Sparse Attention for Long Video Diffusion
链接: https://arxiv.org/abs/2605.31057
作者: Gael Glorian,Ioannis Lamprou,Zhen Zhang,Yujie Yuan,Hongsheng Liu
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注: 10 pages, 5 figures, 4 tables. Code: this https URL
Abstract:Dense self-attention is the compute and quality bottleneck of long-video diffusion inference: cost grows quadratically with the sequence length, and beyond the training horizon the model converges to near-static output, that is, “frozen” repetitive video. State of the art approaches are either too costly, e.g., they require retraining, or fail to satisfy both performance and quality objectives in a scalable manner. To this end, we introduce Long Video Sparse Attention (LVSA), a training-free model-agnostic block-sparse attention for video diffusion transformers that combines a structured window pattern with rotating global anchors, thus removing the fixed-grid bias which causes long-range temporal artifacts. LVSA, combined with a FlashInfer kernel, reduces compute up to 3.17x on Wan 2.1 1.3B at a 6x horizon, 2.98x on Wan 2.1 14B at a 6x horizon, and 3.33x on HunyuanVideo 1.5 at a 1.5x horizon, compared to dense attention. Beyond reducing compute, LVSA enables HunyuanVideo 1.5 generation at a 2x horizon, which is otherwise out-of-memory on a single GPU. Moreover, LVSA provides speedups up to 2.41x compared to RIFLEx and 3.27x compared to UltraViCo on Wan 2.1 1.3B. To demonstrate applicability across diverse platforms, we apply LVSA on NPUs and achieve speedups up to 2.71x on Wan 2.2 A14B and 3.24x on Wan 2.1 1.3B compared to dense attention. To evaluate quality in a fair way, we introduce VQeval, a tool properly scoring loopy video failures, which instead are rewarded in state of the art evaluators like VBench-Long. LVSA is quality-neutral for generation at training horizon length and quality-positive at extended lengths.
[CV-67] Rethinking Efficient Crack Segmentation with Task-Aligned Structural-Directional Modeling
链接: https://arxiv.org/abs/2605.31048
作者: Shipeng Liu,Liang Zhao,Dengfeng Chen,Weihua Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Recent crack segmentation methods often follow generic semantic segmentation designs, using stronger backbones, hybrid CNN-Transformer-Mamba encoders, and auxiliary enhancement branches. Although effective, this raises whether stronger generic feature mixing is the most suitable direction for crack segmentation. We instead formulate crack segmentation as sparse structural recovery. Cracks have limited category-level semantics but strong morphological regularities, being thin, sparse, anisotropic, locally fragmented, and easily confused with textures or shadows. Thus, the key bottleneck lies in preserving weak structural evidence, recovering directional continuity, and suppressing background coupling. We propose RIFT, a compact family of morphology-aligned crack segmentation models. Rather than compressing a complex generic architecture, RIFT is simple by design, preserving local evidence, aggregating cooperative directional continuity, and restoring crack structures through lightweight multi-scale fusion. Experiments on four public benchmarks show that RIFT achieves the best or tied-best results across the 16 main metrics against reproduced representative baselines. RIFT-B gives the strongest overall accuracy, while RIFT-T provides the best deployment efficiency with only 0.47M parameters and high inference speed. Topology-aware evaluation, ablations, transfer experiments, and visualizations further verify that task-aligned simplicity can match or surpass complex hybrid architectures when its inductive bias fits crack morphology. Code: this https URL
[CV-68] Does Visual Information Play a Decisive Role in Vision-Language-Action Model Driving Behavior?
链接: https://arxiv.org/abs/2605.31041
作者: Jingtao He,Hongliang Lu,Xiaoyun Qiu,Yixuan Wang,Xinhu Zheng
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Vision-Language-Action (VLA) models have demonstrated promising capability in autonomous driving, highlighting the potential of unified multimodal architectures for jointly modeling perception and planning. However, how current VLA-based driving behavior is grounded in visual information remains poorly understood. Existing evaluation protocols mainly focus on aggregate performance metrics, lacking structured and practical diagnostics to quantify visual-behavior dependency. In this work, we introduce a structured multi-level visual perturbation framework to analyze visual-behavior dependency in VLA-based driving models systematically. The framework organizes controlled visual perturbations along three complementary dimensions: channellevel degradation, information-level disruption, and structurelevel modification. We apply it to VLA-based driving systems and evaluate behavioral responses under both open-loop trajectory prediction and interactive closed-loop safety evaluation. Experimental results reveal evaluation-dependent dependency patterns and uneven visual grounding across abstraction levels. These findings call for more structured analyses and principled design of VLA driving models to better understand how visual information shapes behavior and develop safer, more robust systems.
[CV-69] GGT-100K: Generative Ground Truth for Generalizable Real-World Image Restoration
链接: https://arxiv.org/abs/2605.31039
作者: Xiangtao Kong,Jixin Zhao,Lingchen Sun,Rongyuan Wu,Lei Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Real-world image restoration (IR) is bottlenecked by the scarcity of high-quality paired training data. Synthetic datasets are abundant but often fail to model real-world degradations, while real-world paired datasets are expensive and difficult to capture. As a result, IR models trained on these datasets show limited generalization in real-world scenarios. In this work, we propose Generative Ground Truth (GGT) by using generative multimodal foundation models (MFMs) to produce high-quality (HQ) targets from real-world low-quality (LQ) images. We first conduct a systematic evaluation of nine state-of-the-art MFMs, including Nano-Banana-2 and GPT-Image-2, on images of various scenes and degradation types. The results demonstrate that Nano-Banana-2 with VLM-based adaptive prompting shows the highest capability to synthesize perceptually realistic and content-faithful HQ targets, which can serve as the GGT for the LQ input. We then employ Nano-Banana-2 to build a GGT synthesis pipeline, which involves multi-stage quality control to ensure data reliability, and construct GGT-100K, an LQ-HQ paired dataset comprising 103,707 training pairs and covering diverse scenes and complex real-world degradations. A test set of 500 image pairs is also established. Extensive experiments show that GGT-100K consistently improves the real-world generalization of a wide range of IR models, with particularly strong benefits for finetuning generative models for IR tasks. Our results suggest that MFMs can serve as practical tools for restoration-oriented data generation, and GGT-100K is a useful resource to expand the generalization boundaries of real-world IR models.
[CV-70] SlotMemory: Object-Centric KV Memory for Streaming Long-Video Generation
链接: https://arxiv.org/abs/2605.31033
作者: Weijia Dou,Hui Li,Jiahao Cui,Lei Zhou,Jingdong Wang,Siyu Zhu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Streaming video generation models typically rely on temporal-centric memory, which organizes historical context as raw frames, chunk segments, or unclustered tokens. This organization frequently leads to identity drift and semantic inconsistency when entities exit the frame or during interactive prompt transitions. To address these limitations, we propose SlotMemory, an object-centric Key-Value memory mechanism for streaming video diffusion. Our approach shifts the memory abstraction from “when” an event occurred to “what” is being represented by decomposing the transformer’s key-value manifold into discrete, reusable semantic slots. By utilizing these slots as routing addresses to index and store high-fidelity key-value tokens, we enable entity-level persistence and prompt-aware retrieval across long horizons. Evaluated on 60-second interactive narratives using the Wan2.1-T2V-1.3B backbone, SlotMemory achieves a state-of-the-art quality score of 81.61 and a 22.8 percent relative improvement in dynamic consistency over the strongest existing streaming baseline. Our results demonstrate that structured semantic representation, rather than raw temporal capacity, is the essential primitive for persistent long-form video synthesis. Our codes and checkpoints are available at this https URL.
[CV-71] PEEK: Picking Essential frames via Efficient Knowledge distillation WWW
链接: https://arxiv.org/abs/2605.31029
作者: Killian Steunou,Anas Filali Razzouki,Khalil Guetari,Mounîm A. El-Yacoubi,Yannis Tevissen
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Supplementary material at this https URL
Abstract:Video-language models can process only a limited number of frames, making frame selection a key bottleneck for efficient video captioning. Most captioning pipelines still rely on uniform sampling, which is computationally cheap but agnostic to visual content. Adaptive frame sampling has recently emerged as a promising approach for selecting the most informative frames from a video; however, existing methods remain computationally expensive. We introduce PEEK, an efficient dynamic frame sampling method that distills caption-conditioned frame relevance rankings from a stronger teacher model into a lightweight temporal model that operates only on visual content. We find that, overall, on ActivityNet Captions and MSR-VTT, our method outperforms state-of-the-art methods across all evaluated downstream vision language models, especially when only one or two frames are selected for captioning, obtaining the best CIDEr for most frame budgets. On ActivityNet Captions, PEEK is particularly strong, winning 14 out of 16 configurations. Zero-shot evaluation on MSR-VTT shows that our model transfers best at low frame budgets, while results at four and eight frames are more mixed as temporal coverage and visual diversity become increasingly competitive. Compared with recent adaptive baselines, PEEK is both more accurate in the low-budget regime and more efficient: it adds only 5.2% to the captioning time, compared with 65.4% for CSTA and 211.9% for MaxInfo. We release our code and pre-trained checkpoint at this https URL.
[CV-72] Iterative Framework For Data Augmentation Of Segmented Fingerprints
链接: https://arxiv.org/abs/2605.31001
作者: João Leonardo H. D. Agnol,Wesley Augusto de Bona,Erick Oliveira Rodrigues,Luiz Fernando Puttow Southier,Jefferson Oliva,Marcelo Filipak,Dalcimar Casanova
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Infant biometrics presents unique challenges due to the physiological differences between infants and adults, compounded by the scarcity of available data for research that limits the development of robust matching systems. This paper proposes a novel data augmentation method that uses iterative techniques to generate diverse variants of segmented fingerprints by inducing errors in a convolutional neural network trained to extract fingerprint ridges and valleys. Experiments on real infant fingerprints demonstrate the method’s effectiveness in expanding fingerprint variability, with augmentations exhibiting significant fluctuations in minutiae counts while still retaining visual similarity to the originals. The study also highlights the method’s customizable nature for applying varying levels of changes to fingerprint segmentations. Future research includes training segmentation and matching neural networks using datasets augmented by the proposed framework.
[CV-73] Parallel Tempering Initial Sampling in Inference-Time Reward Alignment
链接: https://arxiv.org/abs/2605.30991
作者: Myeongjun Oh,Gwangho Kim,Sungyoon Lee
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注: 31 pages, 11 figures
Abstract:Inference-time reward alignment steers pretrained diffusion and flow-based generative models to satisfy user-specified rewards without retraining. Recently, Sequential Monte Carlo (SMC) has emerged as a powerful framework for this task by iteratively filtering and propagating multiple particles. However, we show that standard SMC-based methods often suffer from poor performance because they initialize particles from a standard prior, whereas high-reward regions in complex reward landscapes are extremely rare. Further, we show that even recent reward-aware initial sampling approaches remain vulnerable to getting trapped in local modes, as complex reward landscapes are often multi-modal. To overcome these limitations, we propose PATHS (PArallel Tempering for High-complexity reward Sampling), a novel initialization method that couples multiple sampling chains through parallel tempering. PATHS maintains a ladder of reward-tempered chains and periodically performs Metropolis swaps, enabling efficient exploration across flattened reward landscapes, thereby mitigating the mode-trapping issues. Our analysis reveals that this mechanism substantially enhances the finite-budget exploration of rare, high-reward regions that are typically challenging to sample. Experiments on layout-to-image and quantity-aware generation show that PATHS achieves consistent gains in alignment quality, particularly on complex prompts.
[CV-74] Benchmarking Single-Step Inpainting Methods for Multi-Object 3D Gaussian Splatting Scenes CVPR2026
链接: https://arxiv.org/abs/2605.30987
作者: Finn Dröge,Cecilia Curreli,Abhishek Saroha,Daniel Cremers
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted as an extended abstract to the CVEU Workshop at CVPR 2026
Abstract:The tasks of object removal and inpainting 3D Gaussian Splatting (3DGS) scenes face challenges such as 3D consistency across camera views. In comparing 2D inpainters and their suitability for the 3D domain, we find that reconstruction-based inpainters outperform generative diffusion models in 3D consistency. Integrating these 2D inpainters into different single-step methods for creating and finetuning 3DGS scenes, our results indicate that initializing the scene from scratch produces higher quality results than finetuning the existing scene. Using a state-of-the-art generative 2D inpainter, we create a straightforward baseline to underline the importance of object removal before inpainting in the 3D setting. Since 360° datasets rarely include real-world ground truths, and challenging occlusion scenarios are equally sparse, we introduce a novel multi-object scene with recorded ground truth data and many views with object occlusions.
[CV-75] Can BEV Perception Gracefully Degrade under Sensor Failures?
链接: https://arxiv.org/abs/2605.30983
作者: Haifa Zhang,Yijing Wang,Haoyu Wang,Zheng Li,Zhiqiang Zuo
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Despite the remarkable success of multi-modal bird’s-eye view (BEV) perception in autonomous driving, current systems exhibit a critical vulnerability: existing fusion mechanisms are highly brittle to sensor corruptions, often causing catastrophic performance degradation. This vulnerability largely stems from the fact that standard fusion frameworks typically integrate multi-modal representations in a static manner, leading to a precipitous performance collapse under missing or corrupted modalities. In contrast, we show that graceful degradation is achievable through active modality reliability assessment. To this end, we present Grace-BEV, a lightweight and plug-and-play framework that enforces active reliability awareness during multi-modal fusion. Instead of relying on computationally expensive cross-modal interactions, Grace-BEV leverages the aligned BEV space to explicitly assess modality trustworthiness via a TrustGate Router and dynamically recalibrate feature integration using the FailSafe Fusion Block. Furthermore, we devise a Three-Phase Training strategy with Modality Dropout to prevent modality dominance and encourage balanced cross-modal learning under unreliable inputs. Extensive experiments on nuScenes-R and nuScenes-C show that Grace-BEV maintains robust performance across diverse corruption settings. Notably, under catastrophic LiDAR failures where standard baselines collapse to 0.0% mean Average Precision (mAP), Grace-BEV restores performance to as high as 34.7% mAP. Moreover, it improves clean accuracy by up to 1.4%, achieving a strong trade-off between robustness and efficiency.
[CV-76] BiSegMamba: Efficient Bidirectional Tri-Oriented Mamba for 3D Medical Image Segmentation
链接: https://arxiv.org/abs/2605.30972
作者: Bakht Zada,Chao Tong,Qile Su,Shuai Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 10 pages, 7 figures, 5 tables. Code is available at: this https URL
Abstract:Accurate 3D medical image segmentation requires both long-range volumetric context and fine boundary preservation. CNN-based methods have limited global dependency modeling, while Transformer-based models are often computationally expensive for dense 3D inputs. Recent Mamba-based methods provide an efficient alternative, but existing volumetric designs still depend on repeated high-resolution scanning, forward-only sequential modeling, and fixed directional summation, causing high cost, scan-order bias, and suboptimal directional aggregation. We propose BiSegMamba, an efficient bidirectional tri-oriented Mamba network for 3D medical image segmentation. BiSegMamba follows a compact-to-detail design, where a progressive compacting stem (PCS) enables efficient latent-space reasoning while retaining shallow high-resolution features for reconstruction. A multi-scale spatial mixer (MSSM) captures local anatomical patterns in early stages, and the proposed bidirectional tri-oriented Ortho Mamba (Bi-ToOM) block models long-range dependencies from multiple orthogonal views using jointly processed forward and backward scan sequences. Adaptive directional fusion (ADF) learns input-dependent channel-wise weights across scan orientations, replacing fixed summation with orientation-aware fusion. Experiments on a collected carotid CTA dataset and three public benchmarks, BraTS2023, ACDC, and AMOS-CT, show that BiSegMamba generalizes well across vascular, cardiac, brain tumor, and abdominal multi-organ segmentation tasks. Compared with SegMamba-V2, BiSegMamba achieves slightly better performance on BraTS2023 and clear improvements on ACDC and the carotid dataset, while reducing computational cost by up to 77.9% FLOPs, demonstrating a strong accuracy-efficiency balance for general 3D medical image segmentation.
[CV-77] Omni-Supervised Motion Editing: Balancing Change and Invariance through Positive-Negative Learning
链接: https://arxiv.org/abs/2605.30969
作者: Zhenwu Shi,Jingyu Gong,Peiwei Wang,Xingzan Wang,Tianwen Qian,Wenxi Li,Yuan Fang,Jiao Xie,Lizhuang Ma,Shaohui Lin
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Text-based human motion editing aims to modify existing motion sequences according to natural language instructions while maintaining the consistency of the original motion. Existing diffusion-based approaches often rely on heuristic similarity cues or coarse global conditioning, leading to motion distortion and suboptimal semantic alignment. The key challenge lies in balancing change (i.e. precisely editing target regions) and invariance (i.e. preserving unedited parts). To handle such challenge, we propose an Omni-Supervised Positive-Negative Learning framework, named OmniME. Our method integrates three complementary components: (1) retrospective feature supervision that enforces coarse-to-fine consistency across transformer layers,(2) motion preservation mechanism that focuses on subtle variations according to the source-target similarity, and (3) triplet-based semantic alignment that strengthens text-motion correspondence. Together, these components form a unified supervision paradigm that balances change and invariance. Extensive experiments on the MotionFix and STANCE Adjustment datasets demonstrate that OmniME achieves state-of-the-art performance in editing alignment, validating the effectiveness of our unified learning framework. Our source codes and models have been released at: this https URL
[CV-78] Variational Adapter for Cross-modal Similarity Representation ICML2026
链接: https://arxiv.org/abs/2605.30968
作者: WenZhang Wei,Zhipeng Gui,Dehua Peng,Tiandi Ye,Huayi Wu
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: Accepted by the 43rd International Conference on Machine Learning (ICML 2026)
Abstract:The core of vision-language models lies in measuring cross-modal similarity within a unified representation space. However, most image-text matching or multi-class image classification datasets lack fine-grained cross-modal matching annotations, forcing the continuous similarity space into binary classification boundaries. This compression induces false negative samples and significantly impairs the generalization performance of cross-modal tasks. While prior research has attempted to mitigate this by modeling intra-modal ambiguity, it often overlooks inherent annotation flaws, leading to suboptimal uncertainty allocation. To address these challenges, we propose a Variational Adapter for Cross-modal Similarity Representation (VACSR). This approach reformulates image-text matching with fine-grained semantic scarcity as a variational inference problem. It constructs a latent space for cross-modal similarity and uses regularization techniques to mitigate overfitting to binary annotations. Experiments on image-text retrieval, domain generalization, and base-to-novel generalization demonstrate the proposed method’s effectiveness and robust generalization ability.
[CV-79] PRISM: Progressive Reasoning through Iterative Slot Memory for Vision
链接: https://arxiv.org/abs/2605.30942
作者: Ziyu Wang,Shuangpeng Han,Mengmi Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Modern vision models process images in a single feed-forward pass, which limits their ability to recover missing evidence or refine uncertain representations under incomplete observations. Inspired by the iterative nature of human perception, we introduce PRISM (Progressive Reasoning through Iterative Slot Memory), a pyramid vision architecture that reasons over images through iterative refinement. At a high level, PRISM groups visual features into object-centric representations, retrieves relevant patterns from a learned memory, and iteratively refines the representation to resolve ambiguity and recover missing information. This organize-recall-refine process operates recurrently across multiple scales, enabling progressive improvement of visual representations. Across standard vision tasks, including image classification, object detection, and semantic segmentation, PRISM achieves competitive performance while demonstrating improved robustness under incomplete observations such as occlusion. These results suggest that iterative reasoning with structured representations and memory is a promising direction for building more resilient and adaptive vision models. Source code and models will be released.
[CV-80] IAF-Net: Illumination-Adaptive Fusion for Low-Light Urban Road Segmentation
链接: https://arxiv.org/abs/2605.30939
作者: Bingtao Wang,Daojie Peng,Fulong Ma,Jun Ma,Liang Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Semantic road segmentation is important for autonomous driving, but existing methods suffer severe performance degradation under low-light conditions. Many existing multi-modal fusion methods do not explicitly adapt to illumination-dependent changes in modality reliability, which can propagate degraded RGB features into the fused representation at night. We propose IAF-Net (Illumination-Adaptive Fusion Network), an end-to-end framework with illumination-adaptive fusion for robust road segmentation across different lighting conditions. It dynamically adjusts fusion weights of RGB and geometric features via the core Illumination-Adaptive Fusion (IAF) module, and enhances low-light feature selection with a brightness-modulated attention decoder. We also construct two dedicated datasets: nuScenes Nighttime Road Segmentation (nuScenes-NRS) and CARLA Multi-Weather Road Segmentation (CARLA-MWRS). Experiments on nuScenes-NRS show state-of-the-art overall performance among the compared methods, while CARLA-MWRS further validates robustness across adverse weather conditions. Ablation studies on a 40% training subset further highlight the importance of the IAF module, which provides the largest individual gain of 0.70% in MaxF.
[CV-81] MultiAct: Text-to-Motion Generation from Composite Text via Tailored Attention Guidance SIGGRAPH2026
链接: https://arxiv.org/abs/2605.30925
作者: Nathan Sala,Ofir Abramovich,Ariel Shamir,Daniel Cohen-Or,Andreas Aristidou,Sigal Raab
类目: Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR)
备注: Accepted to SIGGRAPH 2026 conference. Project page: this https URL
Abstract:Text-to-motion generation has progressed rapidly in recent years, offering an expressive interface for animation and human-computer interaction. However, current models remain brittle when handling prompts that describe multiple actions occurring at the same time. Rather than realizing all components of a composite description, models frequently prioritize a single dominant action and neglect the rest, leading to incomplete or ambiguous motion. We present MultiAct, an unpaired, inference-time framework for compositional text-to-motion synthesis that operates directly on pretrained motion generators without retraining or architectural modification. Our method counteracts semantic collapse by adaptively amplifying cross-attention scores associated with underrepresented prompt components. We note that effective modulation depends on prompt-specific choices, such as which tokens and layers to target, and introduce a lightweight auxiliary decision scheme that determines the most effective attention-strengthening parametrization. Extensive quantitative and qualitative evaluations demonstrate that MultiAct consistently outperforms existing baselines on composite prompts, achieving improved semantic coverage while preserving motion realism. Project page: this https URL.
[CV-82] DiTTo: Scalable Order-aware All-in-One Image Restoration Agent
链接: https://arxiv.org/abs/2605.30915
作者: Seungho Choi,Jihyong Oh
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Please visit our project page at this https URL
Abstract:Real-world images rarely suffer from a single degradation, and the order in which degradations are removed substantially affects the final restoration quality, motivating agent-based image restoration (IR), where a vision-language model schedules a pool of pre-built restoration-experts. However, existing training-based agents require \mathcalO((N^\mathbfD)^2) restoration-expert calls per image to construct the Optimal Restoration-action Trajectory Dataset (ORTD), where N^\mathbfD denotes the number of degradation types in the universe \mathbfD , and couple agent training to a fixed restoration-expert pool, preventing extension to newly introduced restoration-experts without full retraining. To overcome these efficiency and extensibility bottlenecks, we propose \textbfDiTTo, a novel order-aware image restoration agent framework consisting of the DiTTo Simulator and the DiTTo Agent. The DiTTo Simulator combines \cup S-IR for single-step restoration-action simulation and AiO-IQA for per-action quality prediction, reducing ORTD construction to \mathcalO(N^\mathbfD) simulator calls per image; the DiTTo Agent is trained by SFT on the simulator-generated ORTD, followed by \textbfOrder-aware Restoration Alignment (ORA) that aligns degradation identification, restoration-action-ordering, and output format along independent axes. This enables \textbfplug-and-play scalable extensibility: adding a new restoration-expert requires updating only the lightweight ORA stage. On the MiO-100 evaluation set with up to five concurrent degradations, our DiTTo Agent achieves state-of-the-art multi-degradation restoration quality among previous agent-based IR methods.
[CV-83] What Makes LVLMs Hallucinate Less? Unveiling the Architectural Factors Behind Hallucination Robustness
链接: https://arxiv.org/abs/2605.30911
作者: Yusheng He,Jizhe Zhou,Xia Du,Zheng Lin,Jun Luo,Jiancheng Lv
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Hallucination remains one of the key challenges undermining the reliability of Large Vision-Language Models (LVLMs). But what makes an LVLM hallucinate less? Many existing efforts focus on improving internal components of the model. We argue that hallucination fundamentally stems from how the model architecture is designed. To investigate this, we factor the architecture design into three dimensions: Linguistic Foundation (LF), Visual Representation (VR), and Semantic Alignment (SA), and categorize hallucinations into Co-occurrence, Similarity, and previously overlooked Uncertainty types. Building on this formulation, we propose CoSimUE, a benchmark that creates fine-grained hallucination scenarios through controlled textual perturbations and random perturbations, enabling mapping between design choices and hallucination behaviors. Experiments across 7 design aspects show that: 1) the widely emphasized scaling of model parameters has only limited impact on reducing all three types of hallucinations; 2) larger and better-trained language foundations can reduce co-occurrence hallucinations; 3) stronger visual encoders and higher resolutions mitigate similarity errors; 4) effective alignment strategies alleviate uncertainty hallucinations. 5) Furthermore, cross-dimensional analysis reveals that jointly enhancing visual fidelity and alignment quality yields the most comprehensive improvements. This study provides the first systematic exploration linking architecture-level design to hallucination robustness, offering practical guidance for developing reliable and efficient LVLMs.
[CV-84] MergeTok: Unified Continuous and Discrete Visual Tokenization via Token Merging NEURIPS2026
链接: https://arxiv.org/abs/2605.30904
作者: Luyuan Zhang,Siyuan Li,Zedong Wang,Qingsong Xie,Cheng Tan,Anna Wang,Yanhao Zhang,Chen Chen,Haonan Lu,Haoqian Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 11 pages (main text), 7 figures. Preprint. Under review at NeurIPS 2026
Abstract:Most visual tokenizers for image generation are bifurcated into two families with complementary limitations: continuous VAEs offer high-fidelity reconstruction but suffer from dense, entangled latents that are poorly suited for semantic control, whereas discrete VQ-based models enable autoregressive generation yet struggle with gradient sparsity, unstable training, and codebook collapse. In this work, we introduce MergeTok, a unified tokenizer that jointly optimizes continuous (VAE) and discrete (VQ) tokenizers within a encoder-decoder architecture, leveraging token merging techniques as a semantic bridge. By clustering similar tokens during encoding, MergeTok establishes a structural prior that provides dual supervision signals: (i) it imposes merged-token semantic alignment in the VAE branch, regularizing its latent space toward disentangled, semantic-aware representations; (ii) it derives group-wise constraints, promoting intra-group diversity and inter-group exclusivity that stabilize VQ training. MergeTok shows competitive reconstruction and generation performance on ImageNet-256, with substantially lower rFID than strong VAE and VQ models under matched token budgets, while producing semantically-organized token representations compatible with both autoregressive and diffusion generators. This shows that a single architecture can endow visual tokenizers with robust semantic organization and generator-friendly discreteness.
[CV-85] SteerFace: Debiasing Synthetic Face Generation via Adaptive Residue Perturbation
链接: https://arxiv.org/abs/2605.30894
作者: Yuxi Mi,Qiuyang Yuan,Jianqing Xu,Yichun Zhou,Xuan Zhao,Jun Wang,Rizen Guo,Shuigeng Zhou
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:The shortage of legally compliant data for face recognition training has sparked growing interest in using synthetic data as an alternative. While recent diffusion-based methods enable the generation of photorealistic face images with strong identity adherence and data diversity, their downstream recognition performance still exhibits a significant synthetic-real gap. This paper identifies visual tendency as a previously underexplored limitation, whereby synthetic data exhibit an unrealistic prevalence of visual attributes and thus deviate from the real-data distribution. Visual tendency can be attributed to the generator’s conditioning on identity embeddings, through which co-occurring residual visual cues are unintentionally absorbed into learned identity semantics. To discourage the generator from exploiting such visual cues, this paper proposes SteerFace, a simple and efficient training framework that perturbs identity embeddings by steering them toward random orthogonal directions on the embedding hypersphere. The perturbation serves as an identity-preserving regularizer that penalizes the generator’s reliance on non-identity components, as supported by theoretical analysis. This paper further introduces an adaptive strategy that learns perturbation strengths with both sample-wise preference and favorable overall statistics. Extensive experiments show that SteerFace effectively mitigates visual tendency, outperforms prior methods in downstream face recognition, and generalizes well across different training datasets and generation pipelines.
[CV-86] Foundation VAEs for 3D CT Reconstruction Augmentation and Generation ICML2026
链接: https://arxiv.org/abs/2605.30893
作者: Qi Chen,Shuhan Ding,Yu Gu,Nan Liu,Jiang Bian,Alan Yuille,Zongwei Zhou,Jingjing Fu
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: ICML 2026 Accepted
Abstract:Variational autoencoders (VAEs) compress high resolution CT volumes into compact latents while preserving clinically relevant structure. However, training CT-specific VAEs from scratch or heavily fine-tuning them incurs substantial computational and engineering cost, and often degrades under heterogeneous scanners, protocols, and diseases. This paper makes a progressive stride toward training-free medical VAEs by leveraging a critical observation: a single Foundation VAE, pretrained at scale on natural images and videos, can serve as a unified interface for CT Reconstruction, Augmentation, and Generation. With both encoder and decoder frozen, the Foundation VAE reconstructs CT volumes with preserved anatomy while suppressing acquisition noise; training segmentation models on these reconstructions improves surface accuracy by 3.9% NSD on average for pancreatic tumor and lung tumor. Within the same Foundation VAE latent space, a conditional latent diffusion model achieves 3.9% lower average FVD with 36.2% higher CT CLIP score, and improves multi-disease generation faithfulness across 18 types by 2.76% AUC. These results demonstrate Foundation VAEs as a practical interface for scalable CT representation reuse and faithful CT generation. Our code and demo are available at this https URL.
[CV-87] GUI-C2: Coarse-to-Fine GUI Grounding via Difficulty-Aware Reinforcement Learning
链接: https://arxiv.org/abs/2605.30884
作者: Junlong Li,Chao Hao,Lap-Pui Chau,Yi Wang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Existing agentic reinforcement learning methods for GUI grounding have limitations at two levels. At the data level, current approaches typically treat all training samples equally, although their training value to the baseline model varies with difficulty. Overlooking this can greatly reduce training efficiency or even cause collapse. At the strategy level, existing frameworks struggle to balance the trade-off between cropping larger regions for sufficient context and smaller ones for reduced redundancy, a tension inherent to tool-augmented grounding agents. In addition, overly complex decision-making is difficult for small-parameter models and significantly increases inference time. To address these issues, at the data level, we propose GUI-D, a data mining and difficulty scoring pipeline that identifies the training-worthy samples by proper testing and assigns difficulty scores to guide subsequent training weights. At the strategy level, we propose GUI-C ^2 , which employs an area-gated coarse-to-fine refinement mechanism that progressively narrows the visual field via model-internal uncertainty signals, adaptively reserving context for large targets while amplifying precision for small ones, reinforced by improvement-aware stage rewards that ensure each refinement genuinely advances grounding. Meanwhile, we simplify the decision-making process to greatly reduce additional inference time. Finally, extensive experiments show that our method achieves state-of-the-art performance. The code and data will be publicly available.
[CV-88] DSD-GS: Dynamic-Static Decomposition of Gaussian Splatting for Efficient and High-Fidelity Dynamic Scene Reconstruction
链接: https://arxiv.org/abs/2605.30863
作者: Youngtae Han,Sung-hwan Han,Youngmin Yi
类目: Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR)
备注: 23 pages, 9 figures, 7 tables
Abstract:Dynamic scene reconstruction and novel view synthesis are fundamental to next-generation visual intelligence applications such as virtual reality, robotics, and digital twins. However, high-fidelity reconstruction of complex, time-varying scenes from arbitrary viewpoints remains a significant challenge. Existing dynamic 3DGS methods suffer from computational inefficiency, since they model all Gaussians as dynamic components. While recent decomposition-based approaches address this issue, they still struggle with degraded reconstruction quality and prolonged training time. To mitigate these limitations, we propose a novel dynamic reconstruction framework built upon an efficient static-dynamic decomposition strategy using a Feed-Forward Gaussian Splatting encoder and an optical flow model. By eliminating redundant computations on static regions, our method achieves state-of-the-art performance, outperforming existing baselines across rendering quality, training and rendering speed, and storage efficiency. Notably, on the Neural 3D dataset, our framework requires only 10 minutes for training and achieves a rendering speed of over 700 FPS on a single NVIDIA RTX 5090 GPU at resolution of 1352x1014. Furthermore, our decomposition strategy eliminates the need for COLMAP preprocessing and enables deterministic initialization, thereby enhancing both efficiency and reproducibility.
[CV-89] Robust Dreamer: Deviation-Aware Latent Gaussian Memory for Action-Controlled AR Video Generation
链接: https://arxiv.org/abs/2605.30855
作者: Hanlin Chen,Jiaxin Wei,Xibin Song,Yifu Wang,Steve Wang,Hongdong Li,Pan Ji,Gim Hee Lee
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Frame-wise action-controlled image-to-video generation is a promising paradigm for interactive world simulation, where each control signal should elicit an immediate visual response. However, maintaining visual fidelity and 3D consistency over long autoregressive rollouts remains challenging. Existing 3D-aware methods often suffer from catastrophic drift due to two impediments: information loss from \textitLatent–RGB Cycling, where generated latents are repeatedly decoded to RGB and re-encoded for future conditioning, and the training–inference gap induced by the \textiterror-free hypothesis, where clean training memory fails to match prediction-corrupted inference memory. To address these challenges, we present \textbfRobust Dreamer, a memory-augmented framework built around how to design 3D memory and how to use it robustly. First, we introduce \textbfLatent Gaussian Memory, which anchors diffusion latents inherited from the generation process to Gaussian primitives and recalls them via latent-space Gaussian splatting. This provides dense, geometry-aware, view-aligned conditioning while avoiding accumulated degradation from repeated VAE conversion. Second, we propose \textbfDeviation Learning with Dynamic Deviation Archive, which synthesizes rollout-induced latent deviations through a one-step approximation, stores them by autoregressive stage and denoising timestamp, and injects them into historical memory during training. This exposes the generator to realistic corrupted memory states and teaches internal correction before inference. Experiments on ScanNet, DL3DV, and OmniWorldGame demonstrate state-of-the-art long-horizon performance.
[CV-90] Count Anything
链接: https://arxiv.org/abs/2605.30846
作者: Mengqi Lei,Shuokun Cheng,Wei Bao,Shaoyi Du,Jun-Hai Yong,Siqi Li,Yue Gao
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Object counting remains fragmented across domain-specific datasets and task formulations, despite rapid progress in generalist vision models. Existing counting models are often tailored to scenarios such as crowds, vehicles, cells, crops, or remote-sensing objects, and thus struggle to generalize across categories, visual domains, object scales, and density distributions. In this paper, we study text-guided object counting across domains, where a model takes an image and a natural-language query as input and returns an instance-grounded set of target points whose cardinality gives the count. This formulation unifies category-conditioned counting with interpretable spatial localization. To support this setting, we construct CLOC, a Cross-domain Large-scale Object Counting dataset that reorganizes diverse public data sources into a unified benchmark. CLOC covers six visual domains: General Scene, Remote Sensing, Histopathology, Cellular Microscopy, Agriculture, and Microbiology, with about 220K images, 619 categories, and 15M object instances. Based on CLOC, we propose Count Anything, a generalist model for text-guided object counting. Unlike density-map-based methods, which dominate counting models, Count Anything adopts discrete instance points and performs dual-granularity instance enumeration. A Region-level Sparse Counter provides object-level anchors for large and sparse targets, while a Pixel-level Dense Counter handles small, crowded, and weakly bounded targets via dense point prediction. A point-centric supervision strategy enables learning from heterogeneous annotations, and Complementary Count Fusion combines both counters in a parameter-free manner. Extensive experiments show that Count Anything achieves strong accuracy and multi-domain generalization, outperforming existing open-world counting methods. Code is available at: this https URL.
[CV-91] LegSegNet: A Public Deep Learning System for Lower Extremity CT Tissue Segmentation and Quantification
链接: https://arxiv.org/abs/2605.30829
作者: Yuwen Chen,Yaqian Chen,Roy Colglazier,Haoyu Dong,Hanxue Gu,Maciej A. Mazurowski,Kevin W. Southerland
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 9 pages
Abstract:Lower extremity computed tomography (CT) contains clinically relevant information for body composition analysis, sarcopenia assessment, and musculoskeletal disease monitoring, but extracting these measurements at scale requires accurate tissue segmentation and an automated quantification workflow. Existing public segmentation tools are not designed for comprehensive lower extremity CT analysis, particularly for clinically important inter/intramuscular adipose tissue, and most public methods only provide mask prediction rather than an end-to-end quantification system. To address this problem, we present LegSegNet, a deep learning system for lower extremity CT tissue segmentation and body composition quantification. Given an input CT scan, LegSegNet segments bone, skeletal muscle, subcutaneous adipose tissue, and inter/intramuscular adipose tissue. It then computes quantitative tissue measurements for downstream analysis. We developed the segmentation model using 1,302 manually annotated CT slices and evaluated it on 900 held-out test slices, with all annotations reviewed by radiologists. We benchmark LegSegNet against a broad set of 2D segmentation methods, including CNN-based models, transformer-based models, and finetuned foundation models, and further evaluate its generalization on an external public CT dataset. LegSegNet achieves the best overall segmentation performance, with an average Dice score of 89.31 on the held-out test set. To our knowledge, LegSegNet is the first publicly available end-to-end system for lower extremity CT tissue segmentation and quantification, providing a practical evaluation tool for future computer vision research in medical image analysis. The code and model weights are available at: this https URL
[CV-92] Function2Scene: 3D Indoor Scene Layout from Functional Specifications
链接: https://arxiv.org/abs/2605.30819
作者: Ruiqi Wang,Qimin Chen,Daniel Ritchie,Angel X. Chang,Manolis Savva,Kai Wang,Hao Zhang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: project page: this https URL
Abstract:Most text-driven 3D indoor scene synthesis methods generate rooms from object-centric prompts, asking what furniture should be placed rather than how the space is used. Yet in real interior design, a layout is judged by how well it supports its occupants, e.g., their activities and physical needs. We introduce Function2Scene, a framework for generating 3D indoor layouts from functional specifications, i.e., natural-language design briefs describing who will use a room and what they need to do there. Given such a specification, our system parses occupant personas and activities, derives a customized set of functional design constraints from a taxonomy of 17 criteria spanning spatial, ergonomic, activity, and environmental considerations, and uses these constraints to guide layout generation. Rather than relying on an LLM to directly produce a final scene, Function2Scene performs iterative evaluation and refinement through a tool-augmented check-and-repair loop, combining geometric measurements, LLM-based contextual reasoning, and VLM-based visual assessment. Experiments on 30 professionally written interior-design cases show that Function2Scene produces layouts that better satisfy functional requirements than recent LLM-based scene synthesis baselines, with our results preferred in 94.3% of pairwise comparisons. Our work reframes text-driven indoor scene synthesis from placing plausible objects to designing spaces that support human use.
[CV-93] MechVQA: Benchmarking and Enhancing Multimodal LLM s on Comprehensive Mechanical Drawing Understanding
链接: https://arxiv.org/abs/2605.30794
作者: Qian Kou,Xiaofeng Shi,Yulin Li,Xiaosong Qiu,Xinyang Wang,Hua Zhou,Cao Dongxing
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: accept by iclm2026
Abstract:Multimodal Large Language Models (MLLMs) have demonstrated significant achievements in general visual question answering (VQA) tasks. However, they remain brittle on mechanical engineering drawings, where high annotation density and weak domain knowledge, compounded by unreliable spatial relation reasoning under strict projection rules and geometric constraints, make decisive cues easy to miss and frequently lead to wrong answers. To bridge this gap, we introduce the first comprehensive mechanical drawing understanding dataset, MechVQA, created through a semi-automated construction and quality-control pipeline. MechVQA contains 3.3k high-density pictures with 21K question-answer pairs, spanning 10 different fine-grained tasks across three capability levels: Recognition, Reasoning, and Judging, providing a testbed to evaluate and improve MLLM understanding on real-world mechanical drawings. On top of MechVQA, we then develop the MechVL model through a multi-stage training paradigm, building a strong domain-specialized baseline. Extensive experimental results demonstrate that MechVL outperforms the strongest closed-source baseline by 7.57 percentage points on the MechVQA total score, significantly enhancing mechanical drawing understanding ability and providing a reusable foundation for deploying MLLMs in mechanical design and inspection scenarios.
[CV-94] xt-guided Feature Disentanglement for Cross-modal Gait Recognition CVPR2026
链接: https://arxiv.org/abs/2605.30784
作者: Zhiyang Lu,Ming Cheng
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accept by CVPR2026
Abstract:Gait recognition is a biometric technique that identifies individuals based on their walking patterns, offering advantages in long-range, non-intrusive scenarios. However, real-world scenarios often involve heterogeneous sensing modalities such as LiDAR and RGB cameras, making LiDAR-Camera Cross-modal Gait recognition (LCCGR) a critical yet challenging task due to the substantial modality gap between 2D videos and 3D point cloud sequences. To address this challenge, we propose TCFDNet, a Text-guided Cross-modal Feature Disentanglement Network, which leverages modality-aware textual priors as semantic anchors to guide the learning of disentangled modality-shared representations. Specifically, we construct a Gait Modality Text Dictionary (GMTD) using large language models to generate rich semantic descriptions of gait across modalities and viewpoints. A CLIP-based Multi-grained Feature Encoder then aligns visual and textual features within a unified vision-language space. Furthermore, the Text-guided Feature Disentanglement (TFD) module selects the topk matched textual descriptions to reconstruct modality-specific representations and derive modality-shared features via residual decomposition and orthogonality constraints. To mitigate the fragility of the disentangled shared features, we propose a Feature Stability Enhancement (FSE) module, which models spatial and channel-wise correlations to improve feature robustness. In addition, a cross-modal patch exchange strategy is introduced to further improve generalization. Extensive experiments on SUSTech1K and FreeGait datasets demonstrate that TCFDNet achieves new state-of-the-art results and validate the effectiveness of the proposed modules.
[CV-95] CameraNoise: Enabling Faithful Camera Control in Video Diffusion through Geometry-Flow-Guided Noise Warping
链接: https://arxiv.org/abs/2605.30774
作者: Haoyu Zhao,Jiaxi Gu,Haoran Chen,Qingping Zheng,Yeying Jin,Hongyi Yang,Junqi Cheng,Yuang Zhang,Zenghui Lu,Huan Yu,Jie Jiang,Peng Shu,Zuxuan Wu,Yu-Gang Jiang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 28 pages, 16 figures
Abstract:Precise camera pose control is critical for video diffusion, yet maintaining geometric consistency remains a challenge. Existing methods that directly inject numerical camera parameters into the diffusion backbone often fail to bridge the gap between abstract coordinates and visual content, leading to structural distortions. To address this issue, we propose CameraNoise, a flow-to-noise warping method that encodes camera motion into a temporally coherent stochastic representation. Unlike conventional conditioning, CameraNoise embeds camera poses directly into the noise space. This decouples motion from scene appearance while faithfully preserving trajectory dynamics. Specifically, we introduce a novel Geometry-guided Reprojection Flow and a noise warping algorithm, which jointly preserve the Gaussian prior of diffusion and ensure consistent noise propagation under camera transformations. By integrating CameraNoise into the diffusion process, our framework delivers stable, high-fidelity videos. Extensive experiments demonstrate that our approach significantly outperforms prior methods in both visual quality and trajectory faithfulness. The project page and code are available at: this https URL.
[CV-96] DisPlace: Discriminative Place Projections for Multi-Reference Visual Place Recognition
链接: https://arxiv.org/abs/2605.30769
作者: Dhyey Manish Rajani,Michael Milford,Tobias Fischer
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注: Under review
Abstract:A key challenge in Visual Place Recognition (VPR) is matching query images against reference maps captured under diverse environmental conditions and viewpoints. While multiple reference traversals improve robustness, existing fusion strategies either aggregate references uniformly or rely on heuristic selection, without distinguishing descriptor variations that preserve stable place identity from those caused by changing conditions or viewpoints. In this paper, we propose DisPlace, a multi-reference VPR framework that fuses multiple reference descriptors into a single compact and discriminative place representation. DisPlace formulates descriptor fusion as a generalized eigenvalue problem that maximizes between-place separability while suppressing within-place variation across references, rather than preserving overall descriptor variance. Unlike existing multi-reference fusion methods, DisPlace exploits variation across reference traversals to identify which linear combinations of descriptor dimensions preserve place identity and which capture condition- or viewpoint-specific variation. We evaluate DisPlace on Oxford RobotCar, Nordland, Pittsburgh30k, and Google Landmarks v2 across six state-of-the-art VPR descriptors. DisPlace outperforms seven multi-reference baselines in 49 out of 54 appearance-varying conditions, consistently improves descriptor-level fusion performance under viewpoint and unstructured settings, and requires less storage during inference than all compared fusion methods.
[CV-97] SLAP: The Semantic Least Action Principle for Variational Video-Language Modeling ICML2026
链接: https://arxiv.org/abs/2605.30750
作者: Xiang Fang,Wanlong Fang
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted by ICML 2026
Abstract:In the era of Large Video-Language Models (LVLMs), the computational necessity of sparse frame sampling creates a fundamental ``temporal gap’', rendering models blind to critical causal transitions. Existing solutions relying on generative hallucination (e.g., latent diffusion) or autoregressive extrapolation often fail to maintain semantic consistency over long horizons, suffering from object vanishing and energetic instability. We propose a paradigm shift from probabilistic generation to variational mechanics with the \textbfSemantic Least Action Principle (SLAP). Drawing a rigorous isomorphism between classical mechanics and semantic dynamics, we model the latent video trajectory as a path on a Riemannian manifold governed by a Semantic Lagrangian. By formulating the interpolation task as a Boundary Value Problem (BVP) solved via the discrete Euler-Lagrange equations, SLAP naturally enforces object persistence without pixel-level rendering. Extensive experiments show the effectiveness of our proposed SLAP.
[CV-98] Immuno-VLM: Immunizing Large Vision-Language Models via Generative Semantic Antibodies for Open-World Trustworthiness ICML2026
链接: https://arxiv.org/abs/2605.30745
作者: Xiang Fang,Wanlong Fang,Wei Ji
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted by ICML 2026
Abstract:Large Vision-Language Models have achieved unprecedented success in zero-shot recognition by aligning visual features with broad semantic concepts. However, this semantic abstraction creates a critical vulnerability in open-world deployment: the Hubris of Semantics'', where models force-fit unknown anomalies into known categories with high confidence due to the lack of explicit negative knowledge. To address this \textitOpen-World Trustworthiness Paradox, we propose \textbfImmuno-VLM, a bio-inspired framework that adapts the biological principle of \textbfImmunological Negative Selection to high-dimensional latent spaces. Departing from traditional Open-Set Recognition methods that rely on passive density estimation or inefficient pixel-space outlier generation, Immuno-VLM leverages the generative reasoning of Large Language Models to actively hallucinate Semantic Antibodies’', textual descriptions of near-distribution outliers (e.g., look-alikes, contextual anomalies) that effectively bound the decision space of known this http URL experiments on ImageNet-1K and four challenging OOD benchmarks reveal that Immuno-VLM establishes a new state-of-the-art.
[CV-99] Annotations Are Not All You Need: A Cross-modal Knowledge Transfer Network for Unsupervised Temporal Sentence Grounding EMNLP2023
链接: https://arxiv.org/abs/2605.30742
作者: Xiang Fang,Daizong Liu,Wanlong Fang,Pan Zhou,Yu Cheng,Keke Tang,Kai Zou
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Published in Findings of EMNLP 2023
Abstract:This paper addresses the task of temporal sentence grounding (TSG). Although many respectable works have made decent achievements in this important topic, they severely rely on massive expensive video-query paired annotations, which require a tremendous amount of human effort to collect in real-world applications. To this end, in this paper, we target a more practical but challenging TSG setting: unsupervised temporal sentence grounding, where both paired video-query and segment boundary annotations are unavailable during the network training. Considering that some other cross-modal tasks provide many easily available yet cheap labels, we tend to collect and transfer their simple cross-modal alignment knowledge into our complex scenarios: 1) We first explore the entity-aware object-guided appearance knowledge from the paired Image-Noun task, and adapt them into each independent video frame; 2) Then, we extract the event-aware action representation from the paired Video-Verb task, and further refine the action representation into more practical but complicated real-world cases by a newly proposed copy-paste approach; 3) By modulating and transferring both appearance and action knowledge into our challenging unsupervised task, our model can directly utilize this general knowledge to correlate videos and queries, and accurately retrieve the relevant segment without training. Extensive experiments on two challenging datasets (ActivityNet Captions and Charades-STA) show our effectiveness, outperforming existing unsupervised methods and even competitively beating supervised works.
[CV-100] Beyond Accuracy: Evaluating Efficiency Robustness and Explainability in Deep Learning for Malaria Diagnosis
链接: https://arxiv.org/abs/2605.30734
作者: Olivier Kanamugire,Kerol Djoumessi
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注: Under review
Abstract:Malaria remains a leading cause of mortality in sub-Saharan Africa, where scarce diagnostic infrastructure makes timely, accurate diagnosis particularly challenging. While deep learning offers a compelling path toward automated malaria screening, clinical adoption is hindered by computational cost and opacity in decision-making. This work benchmarks four deep learning models spanning a wide range of designed design architectures and model capacities on the NLM-Malaria dataset, jointly evaluating predictive performance, robustness, and post-hoc explainability. We find that lightweight, efficient-by-design models match their heavier counterparts in predictive performance, and the Friedman test confirms no statistically significant performance differences. CAM-based XAI methods consistently localize diagnostically relevant regions, while fine-grained attribution methods produce less targeted explanations, particularly with heavier backbones. Robustness evaluation under three types of image corruption further reveals that model confidence degrades faster than accuracy, providing a practical signal for human review. However, no XAI method is robust to corruption, with explanation reliability degrading at noise levels plausible in clinical practice, even when predictions remain accurate. These findings support the deployment of lightweight architectures for malaria diagnosis in resource-constrained settings, while highlighting the vulnerability of post-hoc explanations as an important consideration for responsible clinical deployment.
[CV-101] Simple Token-Efficient Vision-Language Model for Case-level Pathology Synoptic Report Generation
链接: https://arxiv.org/abs/2605.30716
作者: Zhiyuan Yang,Jiahao Cheng,Vincent Quoc-Huy Trinh,Mahdi S. Hosseini
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: Accepted by the DeLTA 2026 conference
Abstract:Generating clinically useful pathology reports for pathology cases from whole-slide images (WSIs) is challenging due to gigapixel resolution, long visual-token sequences, and the complexity of case-level reasoning, where a single case may contain multiple WSIs with heterogeneous tissues and ambiguous findings. We present a simple token-efficient vision–language model for case-level synoptic report generation that remains practical under constrained GPU memory. Our architecture follows a minimal three-component design: a frozen pathology patch encoder, a lightweight two-layer MLP vision-language aligner, and a large language model decoder, with an explicit WSI marker token to separate slides within a case. Training proceeds in two supervised stages: (1) aligner-only WSI captioning using heterogeneous WSI-text pairs, and (2) case-level supervised fine-tuning on case-report pairs for structured report generation. To reduce sequence length, we represent each slide using 512 \times 512 patches at 5\times magnification, which reduces the average sequence length by up to 64\times times compared to the commonly used 20\times patches. Combined with efficient training techniques, we enable practical training with only half a NVIDIA H100 GPU. Across both training stages, our approach achieves high ROUGE-L/METEOR/BLEU-4 scores while being substantially more efficient in memory and runtime. In AI-based evaluations, our model is consistently preferred over strong baselines. Extensive ablations characterize performance-efficiency trade-offs and identify simple choices that improve robustness in multi-WSI settings. Overall, this work provides a strong, reproducible baseline for efficient pathology report generation, lowering the barrier to multi-WSI VLM research under limited compute.
[CV-102] Vision-Based Localization in Dense Urban Environments: A Case Study of an Urban Village in China
链接: https://arxiv.org/abs/2605.30714
作者: Menglin Wu,Rui Cao
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Urban villages, the widespread informal settlements which have emerged as a result of rapid urbanization, are now major residential hubs for migrant workers in large cities in China. The dense arrangement of buildings in these areas often leads to unreliable GPS signals, while incomplete mapping data further impairs accurate route planning and navigation. These issues not only hinder everyday mobility but also pose significant challenges for emergency response, as confusing road layouts and GPS inaccuracies can complicate evacuation efforts. To address these challenges, we propose a practical vision-based geo-localization solution tailored for dense urban environments. Our approach features a low-cost data collection pipeline utilizing a dual-camera system, comprising a panoramic camera and a smartphone camera, to capture synchronized 360-degree panoramas and query images. Using Shipai Village, a well-known densely populated urban village in Guangzhou, as a case study, we develop a specialized image geo-localization dataset. We then assess and compare the performance of existing models across various scene types to identify their strengths and weaknesses. The findings demonstrate both the potential and limitations of visual-based localization in dense urban-village environments. Our framework aims to enhance pedestrian navigation, last-mile delivery, and emergency management in areas with poor GPS coverage, ultimately supporting the vulnerable populations living within these informal settlements.
[CV-103] Diversity Matters: Revisiting Test-Time Compute in Vision-Language Models ICML2026
链接: https://arxiv.org/abs/2605.30713
作者: Yijie Tong,Yifan Hou,Shaobo Cui,Antoine Bosselut,Mrinmaya Sachan
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV); Multimedia (cs.MM)
备注: ICML 2026
Abstract:Test-time compute (TTC) strategies have emerged as a lightweight approach to boost reasoning in large language models (LLMs). However, their application and benefits for vision-language models (VLMs) remain underexplored. We present a systematic study of TTC across seven VLMs and six benchmarks, specifically analyzing feature-based scoring and majority voting methods. We find that feature heuristics fail and voting yields only modest gains in single-model settings. We theoretically show that this limitation stems from a lack of prediction diversity: when outputs are highly correlated, voting provides little benefit. In contrast, multi-model ensembles offer richer diversity, yet standard majority voting fails to account for varying model capabilities. To address this, we propose Entropy-based TTC (ETTC), which selects the most confident prediction based on predictive entropy. Our method reduces to majority voting in the single-model case, but in model ensembles, it leverages confidence disparities to prioritize stronger models. We prove that ETTC outperforms majority voting under mild assumptions and empirically demonstrate that it consistently surpasses both voting and the best individual model. Crucially, our results show that smaller models can synergistically enhance larger ones, unlocking ensembling gains not achievable with standard strategies.
[CV-104] Equivariant Latent Alignment via Flow Matching under Group Symmetries
链接: https://arxiv.org/abs/2605.30705
作者: Sunghyun Kim,Jaehoon Hahm,Jeongwoo Shin,Joonseok Lee
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:Geometry-aware generative models and novel view synthesis approaches have shown strong potential in visual fidelity and consistency. In parallel, equivariant representation learning has emerged as a powerful framework for constructing latent spaces where analytically known group transformations could act directly, capturing geometric structure in data and enhancing both interpretability and generalization in novel view synthesis. However, we identify that existing approaches often suffer from latent misalignment, a discrepancy between the intended group action and the actually required transformations in the latent space. Consequently, the learned latents often fail to consistently preserve the equivariant relations imposed by the underlying group symmetry. To address this, we propose Residual Latent Flow, a flow-based framework that corrects the misaligned latents, thereby improving compliance with the underlying equivariance relation. Our comprehensive experiments show that our method significantly reduces latent misalignment and improves novel view synthesis quality, under rotation groups SO(n).
[CV-105] Mathematical Morphology in Machine Learning
链接: https://arxiv.org/abs/2605.30700
作者: Erick Oliveira Rodrigues,Aura Conci
类目: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:This work introduces mathematical morphology-an established visual computing theory-into machine learning to exploit shape and density aspects often overlooked by standard techniques. We propose a fast clustering algorithm based on morphological reconstruction that accurately preserves cluster shapes and density. This scheme offers unique features: an intrinsic sense of maximal clusters, cost-free noise removal, and diverse growth patterns controlled by structuring this http URL, we propose a novel distance metric combining Minkowski and Chebyshev distances, highly efficient for morphological dilations. In Z^2 discrete neighbourhood iterations, it is roughly 1.3 times faster than Manhattan and 329.5 times faster than Euclidean distances. When evaluated using a k-Nearest Neighbours (k-NN) classifier across 33 UCI datasets against 14 other distances, our metric achieved above-average accuracies most frequently (26 of 33 cases) and the best overall accuracy in 9 this http URL, we introduce novel morphological classifiers. Unlike current literature, this proposal uniquely models shape, density, and fractal information in datasets.
[CV-106] A Context-Aware Middleware for Medical Image Based Reports: An approach based on image feature extraction and association rules
链接: https://arxiv.org/abs/2605.30699
作者: Erick O. Rodrigues,Jose Viterbo,Aura Conci,Trueman Mac Henry
类目: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:This work proposes a context-aware middleware for medical workflow organization and efficiency improvement. In hospitals, laboratories and teleradiology companies, each physician or technician is specialized in a specific kind of diagnosis or analysis. Therefore, certain types of medical images are often forwarded to a certain physician or a certain group. This forwarding is time consuming. That is, repeatedly deciding who would be the best physician, whether he is available at a certain moment given a certain context is exhaustive and may be very inefficient. Thus, the proposed middleware has the ability to process and collect data from images analyzed by each medical staff. Based on the collected data and current clinical context, the middleware is able to infer who would be the best fit staff to receive a certain incoming medical image.
[CV-107] ConTrans: Learning Text-enhanced Local-global Temporal Representations for Zero-shot Temporal Action Localization
链接: https://arxiv.org/abs/2605.30689
作者: Kanchan Keisham,Thenukan Pathmanathan,Thangarajah Akilan
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: 4 figures, 8 tables
Abstract:Zero-shot Temporal Action Localization (ZS-TAL) aims to detect and locate previously unseen actions in untrimmed videos. However, existing approaches primarily focus on modeling long-range contextual information, often neglecting the critical relative-offset-based local correlations between video frames. Furthermore, their performance is hindered by limited feature representation capabilities due to the shallow nature of their network architectures. In this paper, we address these limitations by introducing a novel local-global multi-scale feature representation module. We propose a novel multi-scale encoder architecture, termed ConTrans, that integrates convolutional (Conv) inductive biases with transformer Self-attention to jointly capture fine-grained local dependencies and long-range global context, leading to more comprehensive feature representations than existing methods. Experimental evaluations on the ActivityNet-1.3 and THUMOS14 datasets demonstrate that ConTrans significantly outperforms existing methods, establishing a new benchmark for ZS-TAL.
[CV-108] WristCompass: Kinematic Coupling as a Learnable Visual Concept for Ego-Camera Orientation
链接: https://arxiv.org/abs/2605.30671
作者: Varun Nair,Vidyut Baradwaj,Jiahang He,Anya Singh,Jai Relan,Cabrel Happi
类目: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
备注:
Abstract:Recovering ego-camera orientation from manipulation video is a prerequisite for disentangling hand motion from camera motion, a key step in imitation learning from egocentric demonstrations. The obvious approach, inferring orientation from scene geometry, fails when hands occlude the frame: VGGT, a 1B-parameter scene reconstruction model, scores worse than a constant predictor on the TACO benchmark. We identify an alternative visual concept that is present precisely when scene geometry is absent: kinematic coupling dynamics, the structured physical relationship between wrist motion and camera orientation imposed by the arm-shoulder-head chain. We find that this concept is compact (4D inter-wrist features outperform 126D full hand keypoints), temporal (requiring a GRU over short windows rather than per-frame retrieval), and physically grounded (transferring zero-shot across datasets because it is rooted in anatomy rather than scene appearance). Trained only on tabletop manipulation, WristCompass transfers zero-shot to Epic Kitchens cooking video, achieving 14.3 ^\circ median geodesic error and approaching the performance of a 1B-parameter scene model at 200K GRU parameters.
[CV-109] PInVerify: An Offline Embodied Benchmark for Active Instance Verification CVPR2026
链接: https://arxiv.org/abs/2605.30639
作者: Yuhang Jiang
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Robotics (cs.RO)
备注: Accepted as a poster at the Foundation Models Meet Embodied Agents (FMEA) Workshop, CVPR 2026. 44 pages including appendix. Code: this https URL
Abstract:Embodied agents have made strong progress in navigating to target objects, but reaching the goal vicinity does not guarantee that the agent has found the correct instance: subtle attribute differences (e.g., “white floral” vs. “white striped”) often require close-range, multi-view inspection. We address this gap with Active Instance Verification (AIV), a task in which an agent actively selects viewpoints around a candidate object to decide whether it matches a fine-grained natural-language description. We formalize AIV as a finite-horizon decision process and introduce PInVerify, an offline embodied benchmark for AIV: 3,000 evaluation episodes across 18 object categories, delivered as multi-view captures with a 6-sector navigation topology that exposes trap views (navigable but uninformative) and unreachable sectors. As reference baselines we build a training-free pipeline and a LoRA-fine-tuned end-to-end agent around open-source multimodal large language models (MLLMs) at on-device scale ( \leq 8B parameters), with attribute decomposition, a visibility-weighted multi-view tracker, and three next-best-view (NBV) strategies. In our evaluation across Qwen3-VL (4B/8B), SenseNova-SI-1.2-InternVL3-8B, CLIP, and SigLIP2, the best MLLM-based baseline exceeds the best embedding baseline by 4.9 pp; GT-box ablations show a +3.1 pp detection gap; and we do not observe reliable gains from active viewpoint selection within the tested NBV strategies. A LoRA-fine-tuned agent (SFT+GSPO) reaches 85.6%. PInVerify aims to support further work on active, fine-grained semantic verification in embodied AI. Code: this https URL.
[CV-110] Controllable Lung Nodule Synthesis via Histogram-Regularized Latent Diffusion Models
链接: https://arxiv.org/abs/2605.30631
作者: Arunkumar Kannan,Yanbo Zhang,Han Liu,Michael Baumgartner,Jianing Wang,Alexander Hertel,Bogdan Georgescu,Sasa Grbic
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:While automated diagnosis systems have achieved remarkable success in computed tomography (CT)-based lung cancer screening, their development remains limited by the scarcity of diverse, annotated pulmonary nodule datasets. Diffusion-based generative models offer a promising strategy for data synthesis; however, many existing conditional approaches primarily optimize spatial reconstruction losses, which encourage voxel-wise similarity but may inadequately constrain lesion-level intensity distributions. As a result, these methods may produce over-smoothed texture profiles and underrepresent the distinct attenuation characteristics of different nodule subtypes, including solid, part-solid, and ground-glass nodules. To address this challenge, we propose a controllable latent diffusion model that synthesizes pulmonary nodules within full 3D CT volumes while accurately modeling nodule-specific intensity distributions. Specifically, rather than relying solely on spatial losses, we introduce a histogram-based regularization term that constrains voxel intensity distributions during the generative process. The model combines subtype, spatial mask, and Hounsfield unit (HU) histogram conditioning with the differentiable feature-space histogram regularization term to better align lesion-level intensity distributions, improving the visual plausibility and subtype consistency of synthesized nodules. Extensive experiments on lung CT data demonstrate that our framework achieves strong visual realism, validated through both quantitative metrics and a visual Turing test. Furthermore, when used for data augmentation, the generated nodules improve performance in downstream clinical tasks, particularly for underrepresented nodule subtypes, and show a potential benefit for subtype-informed malignancy classification.
[CV-111] ReGuLaR: Relation-Grounded Latent Reasoning for Large Vision-Language Models
链接: https://arxiv.org/abs/2605.30587
作者: Zihu Wang,Karthik Somayaji N.S,Peng Li
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Chain-of-thought (CoT) reasoning has significantly improved the reasoning ability of large vision-language models (LVLMs) by verbalizing intermediate reasoning steps in natural language. However, such discrete textual rationales are often insufficient for encoding continuous visual evidence. Recent work addresses this limitation by moving reasoning into continuous latent space. Despite promising progress, existing methods leave latent reasoning insufficiently connected to the compositional and relational structure of visual evidence. To address this gap, we introduce ReGuLaR, a relation grounded latent reasoning framework that explicitly grounds latent states in these critical yet overlooked visual evidence. ReGuLaR uses a training-time ReGFormer to focus latent reasoning on question-relevant objects and inter-object relations, while at inference time the model reasons and generates answers without invoking the ReGFormer. To support training ReGuLaR, we construct RGROUNDING-351K, a real-world vision-language dataset annotated with key object bounding boxes and inter-object relations. Extensive experiments across diverse benchmarks show that ReGuLaR consistently outperforms existing approaches and achieves state-of-the-art performance. We include our code in the submission and will release the code and training data publicly upon acceptance.
[CV-112] Prior Availability in Industrial Visual Sim-to-Real: A Review of CAD-Guided and CAD-Unavailable Regimes
链接: https://arxiv.org/abs/2605.30581
作者: Chenxi Tao,Seung-Kyum Choi
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Robotics (cs.RO)
备注: Review article; 103 references; 9 main figures; empirical anchors on T-LESS/BOP, MVTec AD, and VisA
Abstract:Industrial visual sim-to-real is often described as transferring from synthetic images to real images, but industrial deployment usually involves a broader mismatch between available evidence and required decisions. A system may be built from CAD renderings, simulated RGB-D observations, normal reference images, synthetic defects, pretrained feature spaces, or language prompts, yet deployed under different sensors, lighting, materials, fixtures, calibration, production variation, and rare defect modes. This review reframes industrial visual sim-to-real as a domain-gap problem organized by prior availability. We distinguish CAD-available settings, where explicit object geometry can support rendering, calibration, pose estimation, segmentation, and test-time geometric verification; CAD-unavailable settings, where geometry is replaced by normal-reference appearance, feature distributions, teacher-student residuals, synthetic anomaly assumptions, foundation features, or vision-language priors; and boundary-prior settings, where approximate models, templates, reference views, or semantic correspondences preserve only part of the CAD role. This framing connects CAD-based detection and 6D pose-estimation literature with industrial anomaly and surface-inspection literature that is usually reviewed separately. To make the taxonomy concrete, we use empirical anchors on T-LESS/BOP, MVTec AD, and VisA. The anchors show that CAD render count alone does not close transfer; source-distribution design, detector capacity, and small real calibration can matter more. They also show that CAD at test time creates a distinct verification channel through mask, pose, and depth consistency, whereas CAD-unavailable inspection relies on calibrated normality and feature deviation. The review therefore argues against a single cross-task leaderboard and instead asks what prior grounds the deployment decision.
[CV-113] AdvScene: Rethinking Adversarial Patch Evaluation Through Scene Robustness
链接: https://arxiv.org/abs/2605.30578
作者: Xiaoyong(Brian)Yuan, Lan (Emily)Zhang
类目: Cryptography and Security (cs.CR); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Adversarial patches are physical patterns attached to real objects to mislead AI vision systems. Their real-world risk is not determined by a single successful prediction, but by whether they remain effective after deployment under changing viewpoints, distances, and scene conditions. We refer to this property as scene robustness, the effectiveness of a deployed patch across conditions in a real environment. Yet existing evaluations do not measure scene robustness well: real image benchmarks are realistic but fixed, while simulators are controllable but not grounded in a specific real scene. We present AdvScene, a scene-grounded framework for measuring the scene robustness of adversarial patches in reconstructed real environments. AdvScene reframes evaluation as operational measurement: given a fixed deployed patch, it characterizes the patch’s operational envelope - where and when the attack succeeds - as a function of viewpoint, distance, and scene context. A key challenge is that the attack is typically defined only in a single anchor view, while evaluation requires a representation that remains faithful under viewpoint changes. We formalize this as a constrained lifting problem and introduce Adversarial Patch-to-Scene Embedding (APSE), which resolves cross-view ambiguity while preserving attack-critical appearance and enforcing locality, target-surface attachment, and cross-view consistency. We validate AdvScene using real-world physical data and conduct a comprehensive evaluation of existing adversarial patches. Our results show that AdvScene reveals substantial scene-dependent variation in attack effectiveness that is not captured by existing image-centric or simulator-based evaluations. Subjects: Cryptography and Security (cs.CR); Computer Vision and Pattern Recognition (cs.CV) Cite as: arXiv:2605.30578 [cs.CR] (or arXiv:2605.30578v1 [cs.CR] for this version) https://doi.org/10.48550/arXiv.2605.30578 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[CV-114] VLM3: Vision Language Models Are Native 3D Learners
链接: https://arxiv.org/abs/2605.30561
作者: Zhipeng Cai,Zhuang Liu,Yunyang Xiong,Zechun Liu,Vikas Chandra,Yangyang Shi
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Vision Language Models (VLMs) enable a unified model to solve various vision tasks through prompting. They have shown promising performance in semantic understanding. However, 3D understanding still largely relies on expert vision models with complex task-specific designs. The key argument this work wants to make is that VLMs are native 3D learners. Our in-depth large scale study shows that 1) focal length unification, 2) text-based pixel reference and 3) data mixture and scaling, are all you need for effective 3D learning. Model architecture changes, large models, heavy data augmentations, and complex losses including the regression formulation, many of which form the foundation of expert vision models, are actually not necessary conditions. As a result, we propose VLM3, a scalable method with the simplest design that enables standard VLMs to master diverse 3D tasks. VLM3 not only advances the VLM depth estimation accuracy by a large margin (0.84 - 0.9), but also enables diverse 3D tasks such as pixel correspondence, camera pose estimation and object-level 3D understanding, matching expert vision model accuracy while maintaining standard architectures and text-based training. We believe VLM3 opens up a new paradigm for simple and scalable 3D learning.
[CV-115] On-Device Generative AI for GDPR-Compliant Visual Monitoring: Natural Language Alerts from Local Object Detection
链接: https://arxiv.org/abs/2605.30544
作者: Gudrun Schappacher-Tilp,Nicoletta Kaehling,Jan Kornberger,Egon Teiniker
类目: Computer Vision and Pattern Recognition (cs.CV); Cryptography and Security (cs.CR)
备注: 6 pages, 4 figures, 3 tables, 1 listing
Abstract:Visual monitoring systems that rely on cloud-based AI inference expose raw image data to external services, creating fundamental tensions with the data-minimisation principle of the General Data Protection Regulation (GDPR). This paper presents a proof-of-concept privacy-by-design pipeline that resolves this tension by confining all inference entirely to the edge device. A YOLOv5n-seg model compiled for a Hailo-8L AI accelerator delivers real-time object detection on a Raspberry Pi 5, from which raw pixel buffers are immediately discarded after inference. A stateful trigger engine forwards minimal JSON event payloads to a locally hosted instance of Phi-3 Mini (3.8B parameters, Q4_0 quantisation), which synthesises one-to-two sentence natural-language alerts for a human operator. No image data crosses the network boundary at any point; only the generated text alert is transmitted. We describe the full system architecture and implementation, report measured inference latency and resource utilisation on the target hardware, and present representative generated alerts. The results demonstrate that combining a dedicated neural-network accelerator with an on-device large language model on a single-board computer is not only feasible but produces practically deployable, human-readable monitoring output while aligning with GDPR Art. 5(1)© by design.
[CV-116] OmniMem: Scalable and Adaptive Memory Retrieval for Long Video Generation
链接: https://arxiv.org/abs/2605.30519
作者: Lin Zhao,Yushu Wu,Yifan Gong,Yanzhi Wang,Pu Zhao
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: 22 pages, 14 figures; project page: this https URL
Abstract:Autoregressive (AR) video generation extends videos by producing latent chunks sequentially, but scaling to long videos requires repeated access to a growing historical KV cache. Existing methods reduce this cost by truncating the KV cache or compressing it into implicit memory, but both lose explicit access to query-relevant historical details. We propose OmniMem, an explicit full-range memory retrieval framework that performs sparse KV retrieval over the historical cache. To make this practical for chunk-based AR video generation, OmniMem addresses two issues: (i) local bias in sparse KV selection and (ii) Union Explosion in memory access. Adaptive Window Exclusion removes local-window blocks from the selection candidates when sufficient long-range history is available, preserving the sparse budget for informative long-range retrieval. Query-Shared KV Selection reduces cross-query diversity, while Per-Head Scattered KV Access avoids expanding head-specific selections into a large selected KV buffer. This allows each attention head to retrieve non-contiguous KV blocks according to its own selection pattern. Experiments on long-video generation show that OmniMem improves Dynamic Degree by 52.3% and preserves strong consistency over strong baselines, while maintaining comparable memory usage.
[CV-117] PhyDrawGen: Physically Grounded Diagram Generation from Natural Language EMNLP2026
链接: https://arxiv.org/abs/2605.30512
作者: Nafiul Haque,Syed Nazmus Sakib,Shifat E Arman
类目: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注: 9 figures, 7 tables. Under review at EMNLP 2026
Abstract:Generating physics diagrams from text requires strict adherence to physical laws. While current generative models produce visually plausible outputs, they systematically hallucinate force vectors, ignore conservation laws, and violate geometric constraints. We present PhyDrawGen, a neuro-symbolic pipeline that decouples semantic scene understanding from physical constraint satisfaction. First, a large language model extracts a typed scene graph from the problem text. A deterministic solver then converts this graph into a Planar Straight-Line Graph (PSLG), encoding force balance, optical paths, and field topologies as exact geometric primitives. Finally, a fine-tuned Qwen-VL model implements a visually grounded propose-verify loop to iteratively correct any constraint violations. Evaluated on a benchmark of 1,449 problems spanning mechanics, optics, and electromagnetism, PhyDrawGen significantly outperforms GPT-5-image, Gemini 2.5 Flash, and Gemini 3 Pro, demonstrating robust physical accuracy even on unusual-object problems.
[CV-118] A Novel Global Context-aware Deep Neural Network for Enhanced Brain Tumor Segmentation using Magnetic Resonance Images
链接: https://arxiv.org/abs/2605.30510
作者: Sourjya Mukherjee,Ananya Bhattacharjee,R. Murugan
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注: 11 pages, 9 figures, 6 tables. Submitted to arXiv cs.CV
Abstract:Brain cancer’s severity necessitates precise brain tumor segmentation, which is crucial for effective brain tumor diagnosis. Manual identification, burdened by high costs, labor, and error risks, highlights the need for automated methods. In this study, we introduce the Global Context-aware Squeeze and Excite Residual UNet (GCSER-UNet), which facilitates a fusion of spatial and channel-wise attention and thus enhances the model’s capacity to capture intricate spatial dependencies and contextual information. GCSER-UNet efficiently extracts tumor segments from multimodal MRI slices, delivering exceptional performance. Evaluations on benchmark databases exhibit its superiority, achieving a notable 94 percent dice score on the TCGA LGG dataset, surpassing the state-of-the-art dice score of 91.8 percent. In the BraTS 2020 dataset, the proposed GCSER-UNet ensemble approach yielded dice scores of 95 percent, 92 percent, and 90 percent for the tumor regions - Whole Tumor (W), Tumor Core (T), and Enhancing Tumor (E), respectively. The current state-of-the-art dice scores were 94 percent, 93 percent, and 88 percent. These compelling outcomes highlight the efficacy of GCSER-UNet in precise brain tumor segmentation and thus can aid neurologists in effective brain cancer management and treatment planning.
[CV-119] VLM-GLoc: Vision-Language Model Enhanced Monte Carlo Localization for Robust Semantic Global Localization in Cluttered Quasi-Static Environments
链接: https://arxiv.org/abs/2605.30506
作者: Shivendra Agrawal,Bradley Hayes
类目: Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Global localization in geometrically aliased, quasi-static environments such as grocery stores, offices, schools, and hospitals poses a significant challenge for mobile robots. Grocery stores with parallel aisles and a long tailed distribution of products, as well as offices and labs with repetitive furniture such as chairs, desks, monitors, and doors, exemplify common indoor environments that present geometric and even semantic ambiguity. Traditional approaches rely either on distinct geometric features or on domain-specific vision pipelines that struggle with long-tail semantic distributions and transient visual clutter. We present VLM-GLoc, a method for hierarchical semantic Monte Carlo Localization (MCL) that leverages open-vocabulary Vision-Language Models (VLMs) as a unified semantic observation front-end. We hypothesize a three-fold benefit from VLMs: (1) extracting highly discriminative rich text features, (2) implicit quality filtering of blurry or dynamic objects, and (3) permanence reasoning for targeted data augmentation. We introduce an inverse semantic proposal mechanism that seeds particles via text-to-map retrieval. Evaluated across two real-world environments with different characteristics and two different platforms: a 3,500 sq. ft. grocery store with a cellphone and a 3,700 sq. ft. lab space with a quadruped, VLM-GLoc achieves 70% and 74% global localization success respectively, substantially outperforming traditional geometry-only and domain-specific baselines.
[CV-120] 3DAE: Binaural Quality Assessment for Audio Novel View Synthesis with Spatial Maps and Benchmark
链接: https://arxiv.org/abs/2605.30469
作者: Jialu Xu,Yifan Zhou
类目: ound (cs.SD); Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:3D audio and novel-view acoustic synthesis models are usually evaluated with global this http URL, global metrics often hide where and why binaural prediction fails. We propose a full-reference diagnostic framework that uses time-frequency audio error maps for magnitude, ILD, IPD, temporal alignment, loudness, and high-frequency failures, forming a 3D Audio Error Map (3DAE Map) for visual inspection. We frame these diagnostics into a model-agnostic benchmark, Spatial Audio Error Bench (3DAE Bench), which takes arbitrary ground-truth and predicted binaural pairs and reports the prediction quality of audio novel-view synthesis models. Experiments on ViGAS outputs over Replay-NVAS and SoundSpaces show different dominant failure modes: temporal misalignment on Replay-NVAS and ILD mismatch on SoundSpaces. Overall, the framework provides interpretable failure-mode summaries and intuitive visual maps for audio Novel-view-synthesis model development optimization.
[CV-121] Clustering Guided Domain-Specific Pretrained Foundation Model Very High-Resolution Arctic Remote Sensing
链接: https://arxiv.org/abs/2605.30467
作者: Amal S. Perera,Chandi Witharana,Elias Manos,Michael Pimenta,Anna K. Liljedahl
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:This study introduces a novel Arctic-focused remote sensing foundation model (RSFM) by combining diversity-aware regional-scale image curation with masked autoencoder (MAE) self-supervised pretraining of a Vision Transformer (ViT) encoder for very-high-spatial-resolution (VHSR) satellite image analysis. Spectral and acquisition-metadata descriptors were used in a scalable affinity-propagation clustering workflow to select approximately 3 million chips from 267 TB of Vantor VHSR imagery This curation strategy was designed to reduce oversampling of visually repetitive or low-information areas while preserving broad scene diversity across the study domain. We pretrained a ViT-Large encoder on the curated corpus using a domain-adapted MAE reconstruction objective, producing Arctic-specific transformer weights for downstream feature mapping. The pretrained encoder was integrated into an existing location-aware detection and segmentation framework and evaluated across four hand-labeled Arctic datasets. Compared to ImageNet-initialized ViT-Large baseline, Arctic MAE pretraining produced consistent improvements in foreground mean F1 scores of 0.87, 0.72, 0.93, and 0.87, for infrastructure, IWP, RTS, and TCNs, with approximately 5-8 percentage increase. The proposed model also outperformed Prithvi-EO-2.0 in all downstream comparisons, with the smallest gain corresponding to at least a 15 percentage improvement mean F1, suggesting that domain-specific self-supervised pretraining on curated Arctic VHSR imagery provides more transferable representations for fine-scale Arctic mapping than a general-purpose Earth observation foundation model. These results demonstrate that optimizing the pretraining data distribution at regional scale, while keeping the architecture and MAE objective fixed, can produce a reusable Arctic-domain encoder for multiple VHSR remote sensing applications.
[CV-122] Dex2HOI: Dexterous Bimanual Two-Object Interaction Generation
链接: https://arxiv.org/abs/2605.30444
作者: Chrysa Pratikaki,Pablo Ruiz-Ponce,Jiankang Deng,Stefanos Zafeiriou,Rolandos Alexandros Potamias
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Recent advances in 4D Human-Object Interaction (HOI) generation have enabled increasingly realistic motion synthesis, particularly for single-object manipulation. Yet current research overlooks an inherent property of human behavior: people naturally coordinate both hands and manipulate multiple objects simultaneously. To address this gap, we present Dex2HOI, a unified diffusion model for single- and two-object HOI synthesis from text. At its core, Dex2HOI employs a Dual-Stream Diffusion approach, where each object is processed in a dedicated interaction stream and coordinated through bidirectional cross-attention. To synthesize the final motion, we introduce a Motion Fusion Network integrated with novel hand-relative object representations and contact-aware conditioning applied across the whole sequence. By sampling the diffusion process autoregressively over prefix-conditioned windows, Dex2HOI generates arbitrarily long sequences at real-time speed omitting redundant test-time optimization, achieving up to x540 inference speed-up over prior state-of-the-art methods. Extensive evaluation on both single- and two-object benchmarks demonstrates state-of-the-art quantitative results, marking a step beyond conventional single-object HOI generation and toward expressive multi-object manipulation. Code and models will be released upon acceptance.
[CV-123] Mitigating Content Shift and Hallucination in GenAI Image Editing via Structural Refinement
链接: https://arxiv.org/abs/2605.30437
作者: Luxi Zhao,Michael S. Brown
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Generative AI (GenAI) image editors, such as Nano Banana, produce visually compelling results for retouching tasks, enabling non-experts to edit images through text prompts alone. However, the generative nature of these models often introduces spatial misalignment, texture distortion, and content hallucination, all of which are detrimental to downstream workflows that require pixel-level fidelity. We identify a problem setting we call “structure-preserving GenAI fusion” for black-box GenAI image retouching: retain the perceptual enhancements of a GenAI output while enforcing structural faithfulness to the original input image. To address this problem, we propose a post-processing framework that fuses an input image with its GenAI-enhanced counterpart by first establishing coarse spatial and photometric correspondences, then performing a fusion stage that transfers desired enhancements while suppressing hallucinated content. In the absence of direct prior work in this setting, we evaluate our framework against representative methods from photorealistic style transfer and image fusion. Our experiments demonstrate that our method better preserves aesthetic quality while maintaining pixel-level structural consistency and the input resolution.
[CV-124] DTG-Restore: Training-Free Diffusion Refinement for Generative Video Super-Resolution
链接: https://arxiv.org/abs/2605.30431
作者: Hidir Yesiltepe,Koutilya PNVR,Gaurav Pathak,Navaneeth Bodla,Bharat Singh,Pinar Yanardag,Jinrong Xie
类目: Computer Vision and Pattern Recognition (cs.CV)
备注:
Abstract:Recent progress in video diffusion models has enabled remarkable generative fidelity, yet leveraging these priors for restoration remains limited by the strong coupling between conditional and unconditional branches in standard classifier-free guidance. We introduce a training-free framework that enhances distorted and low-resolution videos by decoupling these signals in time. Our proposed Decoupled Time Guidance (DTG) evaluates the unconditional branch at a cleaner diffusion timestep, providing a lookahead prior that preserves geometry while suppressing replication of warped content. This temporal bias is annealed throughout sampling, allowing the model to transition from structure correction to detail refinement without retraining. Combined with any off-the-shelf restoration module in a plug-and-play manner, our approach improves perceptual coherence and restores plausible structure in AIgenerated and real-world videos alike. To facilitate evaluation, we curate GenWarp480, a benchmark of 4,400 distorted 480p videos synthesized from diverse text-to-video models. GenWarp480 focuses on characteristic generative degradations such as warped faces, body misalignments, and spatial artifacts, providing a purpose-built testbed for assessing robustness to generative errors. Extensive experiments demonstrate that our method achieves significant improvements in structural fidelity and temporal stability without any model training.
[CV-125] SANA-Streaming: Real-time Streaming Video Editing with Hybrid Diffusion Transformer
链接: https://arxiv.org/abs/2605.30409
作者: Yuyang Zhao,Yicheng Pan,Qiyuan He,Jincheng Yu,Junsong Chen,Tian Ye,Haozhe Liu,Enze Xie,Song Han
类目: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
备注:
Abstract:Real-time streaming video-to-video editing (V2V) is critical for interactive applications such as live broadcasting and gaming, yet it remains a formidable challenge due to the stringent requirements for temporal consistency and inference throughput. In this paper, we present SANA-Streaming, a system-algorithm co-designed framework for high-resolution, real-time streaming video editing on consumer GPUs, with the following three core designs: (1) Hybrid Diffusion Transformer architecture introduces softmax attention in part of the blocks to improve local modeling capabilities while preserving the efficiency of linear layers. (2) Cycle-Reverse Regularization is a novel training strategy that enforces semantic consistency by predicting source frames from generated content via flow matching, improving temporal consistency without requiring paired long edited videos. (3) Efficient System Co-design combines fused GDN kernels and Mixed-Precision Quantization (MPQ) optimized for the NVIDIA Blackwell (RTX 5090) architecture. By profiling real-world throughput, our MPQ maximizes Tensor Core utilization while maintaining generation quality. The resulting system achieves real-time 1280 x 704 resolution editing at 24 end-to-end FPS on a single RTX 5090 GPU, with the DiT core running at 58 FPS. Experimental results demonstrate that our co-design approach significantly outperforms existing SOTA methods in both temporal coherence and system throughput.
[CV-126] Functional MRI Time Series Generation via Wavelet-Based Image Transform and Spectral Flow Matching for Brain Disorder Identification ICLR2026
链接: https://arxiv.org/abs/2605.30387
作者: Hwa Hui Tew,Junn Yong Loo,Fang Yu Leong,Julia K. Lau,Ding Fan,Hernando Ombao,Raphaël C.-W. Phan,Chee Pin Tan,Chee-Ming Ting
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Signal Processing (eess.SP)
备注: Accepted at the Fourteenth International Conference on Learning Representations (ICLR 2026)
Abstract:Functional Magnetic Resonance Imaging (fMRI) provides non-invasive access to dynamic brain activity by measuring blood oxygen level-dependent (BOLD) signals over time. However, the resource-intensive nature of fMRI acquisition limits the availability of high-fidelity samples required for data-driven brain analysis models. While modern generative models can synthesize fMRI data, they often remain challenging in replicating their inherent non-stationarity, intricate spatiotemporal dynamics, and physiological variations of raw BOLD signals. To address these challenges, we propose Dual-Spectral Flow Matching (DSFM), a novel fMRI generative framework that cascades dual frequency representation of BOLD signals with spectral flow matching. Specifically, our framework first converts BOLD signals into a wavelet decomposition map via a discrete wavelet transform (DWT) to capture globalized transient and multi-scale variations, and projects into the discrete cosine transform (DCT) space across brain regions and time to exploit localized energy compaction of low-frequency dominant BOLD coefficients. Subsequently, a spectral flow matching model is trained to generate class-conditioned cosine-frequency representation. The generated samples are reconstructed through inverse DCT and inverse DWT operations to recover physiologically plausible time-domain BOLD signals. This dual-transform approach imposes structured frequency priors and preserves key physiological brain dynamics. Ultimately, we demonstrate the efficacy of our approach through improved downstream fMRI-based brain network classification. The code is available at this https URL .
[CV-127] Lightweight SAR Ship Detection via Contrastive Distillation
链接: https://arxiv.org/abs/2605.30380
作者: Surendar Devasundaram,Saber Latibari Banafsheh,Abhijit Mahalanobis
类目: Computer Vision and Pattern Recognition (cs.CV)
备注: Accepted in GLSVLSI’26 special session 74: Efficiency In Computer Vision: From Image Generation to Decision"
Abstract:Deep convolutional and transformer-based detectors achieve strong performance for SAR ship detection but are often computationally prohibitive for real-time or onboard deployment. Lightweight models offer improved efficiency yet struggle to capture the complex structural relationships inherent in SAR backscatter. Most existing SAR knowledge-distillation approaches rely on feature or logit matching, which enforces localized activation similarity while neglecting the geometric relationships among object representations. We propose a Structured Unified Relational knowledGE distillation framework for SAR Ship detection (SURGE) that transfers relational geometry from a powerful teacher detector to a compact student detector using a contrastive InfoNCE objective in a shared projection embedding space. To the best of our knowledge, this work presents the first transformer-based SAR ship detector knowledge distillation framework in SAR domain. The framework is architecture-agnostic in the sense that it provides a common region-level distillation interface for two-stage, one-stage and transformer-based detectors without modifying their deployed architectures. Experiments on the SSDD and HRSID benchmarks demonstrate that the proposed method yields substantial improvements for two-stage detectors, achieving up to 6.2 mAP and 8.0 AP75 gains over baseline student and even surpassing teacher performance
[CV-128] Updating the standard neuron model in artificial neural networks
链接: https://arxiv.org/abs/2605.30370
作者: Raul Mohedano,Thomas Batard,Erik Velasco-Salido,Ramsses De Los Santos Mendoza,Jorge H. Martínez,Stacey Levine,Marcelo Bertalmío
类目: Neural and Evolutionary Computing (cs.NE); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
备注:
Abstract:From their inception in the 1950s, artificial neural networks (ANNs) started using the so-called point neuron model then prevalent in neuroscience, hoping that this analogy would allow for a better emulation of brain function. Over the years the neuroscience literature has shown that the point neuron model is too simplistic to properly represent many fundamental neural processes; however, the standard neuron model in ANNs still remains the same. Here we substitute it by a very recent model of cortical cells and demonstrate through theoretical analyses and experimental results how, simply by using a more realistic neural unit element without augmenting the number of parameters, the resulting ANNs offer a number of important advantages that include increases in expressivity, robustness and learning speed, and a reduction in memorization and the amount of training data needed.
[CV-129] XOResNet: Exclusive-OR Meta-Residuals Facilitate Deep Spiking Neural Networks Learning
链接: https://arxiv.org/abs/2605.30362
作者: Jianfang Wu,Junsong Wang
类目: Neural and Evolutionary Computing (cs.NE); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
备注: 33 pages, 12 figures, 7 Tables
Abstract:Spiking neural networks (SNNs) hold promise for demonstrating superior learning and representation capabilities in deep models. Given the tremendous success of ResNet in deep learning, it would naturally follow to train deep SNNs with residual learning. However, existing residual structures for constructing deep SNNs still present challenges of spike redundancy or information loss, as well as redundant learning. In the present study, we first aim to address issues of relative spike redundancy in identity mapping and information loss in non-identity mapping. To this end, we propose an OR-ADD (OA) shortcut connection to merge output spikes/currents from two branches in the residual structure. Furthermore, to mitigate redundant learning in the backbone branch of the residual structure, we introduce the concept of XOR meta-residuals, i.e., selecting pre-learning residuals using the Exclusive-OR (XOR) operation for the backbone branch. Finally, by integrating the OA shortcut and XOR meta-residuals, we devise the XOR residual block and further construct XOResNet with varying depths based on this block. Extensive experiments on four datasets, Fashion-MNIST, CIFAR-10, CIFAR-100, and miniImageNet, show that the proposed XOResNet outperforms existing state-of-the-art deep SNNs optimized via gradient descent. These results validate the effectiveness of our OA shortcut and XOR meta-residual components in overcoming fundamental limitations of residual learning in SNNs, providing new architectural insights for building high-performance neuromorphic systems.
[CV-130] Self-Tuning Regularization for Image Scanning Microscopy
链接: https://arxiv.org/abs/2605.31426
作者: Sofia Agostoni,Lisa Cuneo,Christian Daniele,Giacomo Garré,Laurent Le,Alessandro Zunino,Giuseppe Vicidomini,Luca Calatroni
类目: Image and Video Processing (eess.IV); Computer Vision and Pattern Recognition (cs.CV); Optimization and Control (math.OC)
备注:
Abstract:Image Scanning Microscopy (ISM) is a fluorescence imaging technique that combines detector-array acquisition and computational reconstruction to achieve the theoretical resolution of an ideal confocal microscope, i.e., one operating with an infinitesimally small pinhole, while maintaining high signal-to-noise ratio. Among the reconstruction methods for obtaining the super-resolved image, multi-image deconvolution (MID) and its extension aimed at preserving the optical sectioning capability of confocal microscopy, known as super-resolution sectioning ISM (s ^2 ISM), are among the most widely used approaches. Both methods rely on Richardson–Lucy-type iterative schemes, whose semi-convergent behavior requires early stopping and often leads to noise amplification and reconstruction artifacts. In this work, we introduce a self-tuning explicit regularization framework for both MID and s ^2 ISM reconstruction. Within a Bayesian maximum a posteriori formulation, we combine a multi-frame Poisson data fidelity term with explicit regularization, considering \ell_1 and smoothed total variation penalties as representative examples. We further develop an automatic and ground-truth-free strategy for regularization parameter selection by adapting the residual whiteness principle to the multi-frame Poisson setting and introducing a spectral high-pass extension tailored to s ^2 ISM. The resulting framework enables stable reconstructions without empirical stopping rules. To demonstrate the proposed framework, we consider first-order optimization schemes based on proximal gradient and mirror descent methods with adaptive backtracking strategies. Experiments on simulated and real fluorescence ISM datasets demonstrate improved reconstruction stability and image quality with respect to unregularized approaches, while enabling robust super-resolution and optical sectioning in low-photon conditions.
[CV-131] MoE-dqINR: A Unified Mixture-of-Experts Implicit Neural Representation Framework for Scan-Specific Dynamic and Quantitative MRI Reconstruction
链接: https://arxiv.org/abs/2605.31302
作者: Yinzhe Wu,Fanwen Wang,Zhenxuan Zhang,Zi Wang,Chengyan Wang,Guang Yang
类目: Image and Video Processing (eess.IV); Computer Vision and Pattern Recognition (cs.CV); Signal Processing (eess.SP)
备注:
Abstract:Undersampled magnetic resonance imaging (MRI) reconstruction seeks to recover temporally or contrast-varying image series from incomplete multicoil k-space data while preserving state-dependent fidelity for dynamic and quantitative MRI (qMRI). Existing scan-specific implicit neural representations (INRs) often use monolithic spatiotemporal coordinate fields, explicit subspaces, motion or deformation models, calibration variables, or sequence-specific quantitative signal models. These design choices can limit flexibility in sharing spatial information while adapting image synthesis across acquisition states. Moreover, many INR-based baselines remain computationally demanding, typically requiring per-scan optimization times on the order of hundreds to thousands of seconds. We propose MoE-dqINR, a scan-specific multicoil MRI reconstruction framework that factorizes the image-domain representation into shared spatial experts and a state-conditioned routing pathway. Spatial experts encode reusable coordinate-dependent image content, whereas routing weights, conditioned on ordered acquisition states, synthesize each dynamic frame or contrast state from a common expert bank. The representation is coupled to a multicoil MRI forward model, uses the normalized state index to drive routing in both dynamic and quantitative MRI. By separating shared spatial representation from state-dependent synthesis, the framework provides an image-first architecture for dynamic and quantitative MRI while reducing scan-specific INR optimization to approximately 30 s per scan in our experiments. The proposed formulation establishes state-conditioned mixture-of-experts INR as a scan-specific multicoil MRI reconstruction prior that unifies shared spatial representation, dynamic- and qMRI-specific synthesis, and practical per-scan efficiency.
人工智能
[AI-0] Stateful Online Monitoring Catches Distributed Agent Attacks
链接: https://arxiv.org/abs/2605.31593
作者: Davis Brown,Samarth Bhargav,Arav Santhanam,Kasper Hong,Ivan Zhang,Matan Shtepel,Steffi Chern,Alexander Robey,Eric Wong,Hamed Hassani
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注:
Abstract:Language models can find thousands of severe software vulnerabilities, and agents are increasingly being misused for cyberattacks. To avoid detection, attackers frequently distribute their misuse, splitting a harmful task across many user accounts so each individual transcript looks benign. Because safety monitors score only one agent context at a time, they are structurally blind to misuse that is only visible in aggregate, across many accounts. We show this gap is real by building, to our knowledge, the first distributed agent attack, a multi-agent scaffold that completes hard cybersecurity tasks while hiding the harmful objective across subagents with limited contexts, evading a standard monitor that catches it only a fifth as often as prior agent attacks. Towards a defense, we develop an online stateful monitor that uses real-time clustering to collect weak suspiciousness signals across many agent transcripts, and escalates only rarely to a language model that flags misuse across user accounts. In evaluations with large-scale simulated datacenter traffic, our monitor Pareto dominates standard monitors, catching distributed attacks 30% earlier and flagging cyber misuse before it reaches the most harmful stages. Crucially, this comes at negligible additional latency for ~99% of user traffic. This detection advantage persists but narrows as the benign background traffic grows very large. After an extensive red-teaming exercise, we improve the defense and surprisingly also find that it catches standard jailbreaks, since adaptive attackers reuse attack variants across accounts. Our results point toward a new class of safety monitors which reason over groups of users rather than isolated transcripts.
[AI-1] Choosing the Lens: Strategic Perspective Activation in Context-Dependent Argumentation
链接: https://arxiv.org/abs/2605.31581
作者: Albert Sadowski,Jarosław A. Chudziak
类目: Artificial Intelligence (cs.AI)
备注: Accepted to LAMASSR workshop at FLoC 2026
Abstract:The same arguments often need to be evaluated under different external regimes. An agent with influence over the regime has a strategic lever that standard formalisms do not directly capture. We introduce context-dependent argumentation frameworks (CDAFs), an extension of Dung’s theory in which a defeat function determines, per context, which attacks succeed. A perspective-labeled specialisation derives the defeat function from a relevance set \rho and a priority \pi . The relevance set is the agent’s action space. In a small worked example, the agent’s target argument is rejected under every full-relevance injective priority, yet accepted under partial activations, one of which no VAF audience can mirror. We define the corresponding decision problem, ACTIVATION-MANIPULATION, and record baseline complexity bounds. Tight bounds and multi-agent variants are left open.
[AI-2] Positional versus Symbolic Attention Heads: Learning Dynamics RoPE Geometry and Length Generalization
链接: https://arxiv.org/abs/2605.31558
作者: Felipe Urrutia,Juan José Alegría,Cinthia Sanchez Macias,Jorge Salas,Cristian B. Calderon,Cristobal Rojas
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Transformer-based language models are widespread in today’s society. As such, understanding the mechanisms by which they solve structured tasks and predicting how they may behave in novel scenarios is of great importance for safe deployment. We study the learning dynamics of attention heads in a controlled setting by training a decoder-only Transformer (GPT-J) on two structurally equivalent multi-hop reasoning tasks: a number task requiring positional reasoning and a letter task requiring symbolic reasoning. Using a recently introduced metric that classifies attention-head behavior as positional or symbolic for a given prompt, we show that successful learning is associated with the emergence of pure heads, i.e., heads that express themselves as either positional or symbolic. Despite the tasks’ structural equivalence, they impose different mechanistic demands: the number task requires both positional and symbolic heads, whereas the letter task requires only symbolic heads. We then identify the computational roles of these heads, characterize the basic functions they implement, and give theoretical constructions showing how single-layer RoPE-based attention can realize these functions through geometrically interpretable query, key, and value operations. This analysis yields a quantitative separation between positional and symbolic mechanisms in their robustness to longer sequences, formalized through a novel notion of discrepancy. We empirically validate the resulting predictions in both controlled and real-world models, showing that symbolic mechanisms extrapolate more reliably to longer sequences while positional mechanisms face sharper limitations.
[AI-3] Separating Secrets from Placeholders: A Hybrid CNN-CodeBERT Framework for Three-Class Credential Leakage Detection
链接: https://arxiv.org/abs/2605.31520
作者: Maksuda Bilkis Baby,Khushika Shah,Naiyue Liang,Lei Zhang
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR)
备注: Accepted at ICSME 2026 (International Conference on Software Maintenance and Evolution)
Abstract:Credential leakage in public source code repositories poses a critical security threat, with over 23.8 million secrets exposed in 2024 alone. Existing detection tools suffer from high false-positive rates because rigid pattern matching and binary classification schemes fail to distinguish genuine credentials from placeholder or weak credentials. We propose a three-class classification framework that explicitly models placeholder or weak credentials as a distinct class, leveraging CodeBERT-based semantic understanding combined with character-level pattern recognition. We evaluate our approach on a newly constructed dataset of 9,426 samples spanning 10 programming languages. Our model achieves a Matthews Correlation Coefficient of 0.86 and a macro F1-score of 0.90, achieving 93% recall and 89% precision for genuine credential leaks while reducing high severity alerts by 33.0% (from 373 to 250) without sacrificing security coverage. Compared to prior character-level approaches, our method improves placeholder or weak credential detection from 54% to 81% F1-score while maintaining strong cross language generalization, with 9 of 10 languages achieving F1 above 0.80 under leave-one-language-out evaluation.
[AI-4] Skill Reuse as Compression in Agent ic RL
链接: https://arxiv.org/abs/2605.31509
作者: Zhikun Xu,Yu Feng,Jacob Dineen,Taiwei Shi,Jieyu Zhao,Ben Zhou
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: Work in progress
Abstract:Large language model agents trained with reinforcement learning (RL) often learn brittle, task-specific shortcuts. We hypothesize that agents generalize better when their successful trajectories are structurally compressible, decomposed into a small set of reusable abstract patterns. To formalize this, we introduce ReuseRL, which grounds agentic RL in the Minimum Description Length (MDL) principle. ReuseRL extracts a shared skill dictionary from successful trajectories and augments the RL objective with a segmentation cost, explicitly penalizing idiosyncratic behaviors that encode poorly. We prove a PAC-Bayes generalization bound for this compression penalty. Across ALFWorld, TextWorld-Cooking, and Countdown-Stepwise, ReuseRL improves in- and out-of-distribution success over vanilla GRPO and strong round-length baselines.
[AI-5] On Efficient Scaling of GNNs via IO-Aware Layers Implementations ICML
链接: https://arxiv.org/abs/2605.31500
作者: Daria Fomina,Daniil Krasylnikov,Alexey Boykov,Andrey Dolgovyazov,Vyacheslav Zhdanovskiy,Fedor Velikonivtsev
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: International Conference on Machine Learning (ICML) 2026, Spotlight Paper
Abstract:Graph Neural Networks (GNNs) are bottlenecked by sparse, irregular memory access. Popular frameworks such as DGL and PyTorch Geometric support general message passing, but complex layers often materialize edge-wise intermediates, increasing memory traffic and limiting scalability on large graphs. We take an I/O- and arithmetic-intensity–centric view and show that widely used layers fall into three kernel families: SpMM-based convolutions, reduction-based aggregations, and attention-based layers (GATv2/Graph Transformer). For each family, we develop GPU kernels that reduce data movement, improve locality, and remain robust across realistic graphs. We also study graph reordering and find that its impact depends on the kernel mapping: it benefits neighbor-parallel (gather-dominated) kernels more consistently than feature-parallel designs. Empirically, our fused attention kernels reach up to \textbf3.9\times speedup for Graph Transformer (median \textbf1.6\times ), with Tensor Core (block-sparse) variants up to \textbf7.3\times on locally dense graphs; for GATv2 we reach up to \textbf8.5\times speedup (median \textbf2.0\times ) while reducing peak memory by up to \textbf76\times (median \textbf6\times ). Our degree-aware reduction kernels achieve up to \textbf10\times speedup (median \textbf2.6\times ). For SpMM-based layers, properly cached cuSPARSE achieves up to \textbf8\times speedup over DGL and outperforms evaluated custom baselines in the majority of evaluations. We release our implementations as drop-in replacements to support reproducible, hardware-aware GNN acceleration.
[AI-6] LinTree: Improving LLM Reasoning with Explicitly Structured Search Histories
链接: https://arxiv.org/abs/2605.31492
作者: Liwei Kang,Yee Whye Teh,Wee Sun Lee
类目: Artificial Intelligence (cs.AI)
备注: 16 pages, 3 figures
Abstract:Large language models (LLMs) often solve reasoning problems by generating intermediate traces that explore and revise partial solutions. From a search perspective, these traces can be viewed as linearized search trees, where the model extends a partial solution, abandons it when it fails, and backtracks to try alternatives. Compared with traditional heuristic-guided search, such a policy has a potential advantage: it conditions on the whole search trace rather than only on the current local state. We first test whether LLMs utilize this advantage by comparing trace-conditioned reasoning policies against best-first search equipped with an LLM heuristic that only observes the current local state. Across three controlled reasoning environments, Blocks World, grid Navigation, and Sokoban, we find that raw access to search history alone is not enough to reliably outperform heuristic search. We then study one possible reason: in LLM reasoning traces, the underlying search tree is only implicitly represented, and when the model backtracks or switches branches, the trace does not explicitly identify which earlier search state is being revisited. We show that adding simple parent pointers to explicitly represent the linearized tree (LinTree) structure improves both task performance and search efficiency relative to implicit reasoning models and LLM-heuristic-guided search. These results suggest that search history becomes most useful when its tree structure is made explicit, motivating more structure-aware representations for LLM reasoning.
[AI-7] AutoSci: A Memory-Centric Agent ic System for the Full Scientific Research Lifecycle
链接: https://arxiv.org/abs/2605.31468
作者: Weitong Qian,Beicheng Xu,Zhongao Xie,Bowen Fan,Guozheng Tang,Jiale Chen,Xinzhe Wu,Mingtian Yang,Chenyang Di,Jiajun Li,Lingching Tung,Peichao Lai,Yifei Xia,Ziyi Guo,Yanwei Xu,Yanzhao Qin,Shaoduo Gan,Xupeng Miao,Bin Cui
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Scientific research has traditionally been human-intensive, requiring researchers to coordinate literature, ideas, experiments, manuscripts, and review responses across long project cycles. The rise of LLM-based scientific agents creates an opportunity to automate this process. Such a system must support the full research lifecycle, maintain structured persistent memory across projects, and improve its own research procedures over time. However, existing systems either partially satisfy or fail to satisfy these requirements, leaving a gap for a unified automated scientific research system. As a result, we present AutoSci, a memory-centric agentic system for the full scientific research lifecycle. AutoSci is organized around four modules. SciMem provides schema-governed research memory, separating Long-Term Knowledge Memory for reusable scientific knowledge from Active Research Memory for project-level artifacts such as ideas, experiments, manuscripts, and reviews. SciFlow executes a five-stage lifecycle from literature understanding to rebuttal through a harness that controls state, context, verification, feedback, and orchestration. SciDAG augments difficult skills with DAG-shaped multi-agent operators and reusable stage-specific templates. SciEvolve converts feedback signals from users, experiments, reviews, and external environments into versioned updates to SciMem organization, SciFlow skills, and SciDAG templates. Together, these modules make AutoSci a persistent research environment that can execute, remember, and evolve across research projects. The code repository is available at this https URL.
[AI-8] GPU Forecasters: Language Models as Selective Surrogates for Kernel Runtime Optimization
链接: https://arxiv.org/abs/2605.31464
作者: Zaid Khan,Justin Chih-Yao Chen,Jaemin Cho,Elias Stengel-Eskin,Mohit Bansal
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: Code: this https URL
Abstract:GPU kernels are the workhorse of modern deep learning, and optimizing them (via evolutionary search or coding agents) usually requires repeated measurement on target hardware. While these measurements provide the ground-truth signal necessary for kernel search, they are costly, because each evaluation of a kernel requires compilation and repeated execution on a GPU. As improvements in LLM inference reduce the cost of writing novel kernels and LLM-driven searches scale to large search budgets, on-device evaluation becomes a bottleneck. To address this, we study how LLMs can serve as selective GPU surrogates for kernel evaluation, by forecasting the performance of proposed kernels. A useful surrogate should be accurate, and it should be selective, by knowing when it could be wrong, and deferring to the GPU. To evaluate surrogates, we measure whether their forecasts are accurate, calibrated, and practically useful for recovering fast kernels under limited GPU-measurement budgets. Next, we study whether reinforcement learning can improve forecast accuracy and confidence calibration. Our experiments demonstrate that LLMs can accurately forecast relative kernel performance, that their utility can be improved through reinforcement learning. Used inside a kernel search, the surrogate lets the search consider several times as many candidates under the same GPU evaluation budget, and that leads to finding faster kernels than an equal-budget baseline. These results suggest that LLMs can play a broader role in kernel optimization, by acting as virtual models of a GPU rather than solely as kernel generators for search.
[AI-9] Answer-Set-Programming-based Abstractions for Reinforcement Learning
链接: https://arxiv.org/abs/2605.31444
作者: Rafael Bankosegger,Thomas Eiter,Johannes Oetsch
类目: Artificial Intelligence (cs.AI); Logic in Computer Science (cs.LO)
备注: Accepted for publication at the 42nd International Conference on Logic Programming (ICLP 2026). To appear in Theory and Practice of Logic Programming (TPLP)
Abstract:Reinforcement Learning (RL) enables autonomous agents to learn policies from experience, but realistic problems often involve enormous state spaces, making learning and generalisation challenging. Abstraction and approximation are therefore essential. Relational Reinforcement Learning (RRL) offers a way to reason about objects and their relations, and the CARCASS framework by Martijn van Otterlo demonstrates how logical representations can model Markov Decision Processes (MDPs) in first-order domains. Originally implemented in Prolog, CARCASS leverages domain knowledge to create powerful abstractions. We explore Answer-Set Programming (ASP), which is a rich and, contrary to Prolog, fully declarative modelling language, to realise CARCASS abstractions. We evaluate our ASP-based implementation in case studies of two domains, viz. Blocks World and Minigrid. Our results indicate that CARCASS with ASP provides a promising approach to constructing abstractions for RL, especially when domain knowledge is available.
[AI-10] FAM-Bench: A Multimodal Benchmark for Condition-Aware Food-as-Medicine Reasoning
链接: https://arxiv.org/abs/2605.31410
作者: Mingyang Mao,Bhargav Rishi Medisetti,Utkarsh Grover,Tanvir Ibrahim,Wenyan Li,Tingting Zhang,Xiaomin Lin
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Food-as-Medicine requires models to reason beyond what a dish is or what nutrition it contains: they must decide whether a concrete food choice is appropriate for a specific health condition. Existing food AI benchmarks primarily evaluate dish recognition, recipe understanding, nutrient estimation, or general nutrition question answering, leaving this health-aware decision layer largely untested. We introduce FAM-Bench, a multi-modal Food-as-Medicine benchmark with 2500 nutrition-expert-verified instances across 13 diet-related health conditions. The benchmark contains two complementary tasks: dish-level suitability assessment, where models judge whether a dish is suitable for a condition from its image and ingredient list, and comparative dish analysis, where models rank four candidate dishes by condition-specific suitability. Both tasks require integrating ingredient evidence, visual preparation cues, and clinical nutrition constraints, providing a standardized testbed for grounded health-aware reasoning in language and vision-language models.
[AI-11] Scaling Higher-Order Graph Learning with Maximal Clique Complexes
链接: https://arxiv.org/abs/2605.31373
作者: Antoine Vialle,Aref Einizade,Fragkiskos D. Malliaros,Jhony H. Giraldo
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Graph neural networks (GNNs) are limited to modeling pairwise interactions, while higher-order models based on cell complexes achieve greater expressivity but often suffer from poor scalability. We introduce simplified and factored cellular Weisfeiler Leman tests (sCWL and fCWL), which preserve the expressivity of the CWL test while improving computational efficiency. We further introduce the maximal clique complex, enabling scalable CWNs with reduced time and memory complexity while retaining strong empirical performance. To avoid explicit clique enumeration, we propose CliqueWalk, a biased random walk that samples maximal cliques and scales linearly with graph size. These contributions yield a scalable topological learning framework for higher-order graph representation.
[AI-12] HypoAgent : An Agent ic Framework for Interactive Abductive Hypothesis Generation over Knowledge Graphs
链接: https://arxiv.org/abs/2605.31370
作者: Yisen Gao,Yixi Cai,Tianshi Zheng,Jiaxin Bai,Yangqiu Song
类目: Artificial Intelligence (cs.AI)
备注: Under Review
Abstract:Abductive reasoning over knowledge graphs aims to generate logical hypotheses that explain observed entities or facts. Existing controllable hypothesis generation methods allow users to guide this process with explicit conditions, but they remain limited in interactive settings: they struggle to ground evolving natural-language intents across multi-turn dialogues and provide little fine-grained diagnosis when generated hypotheses fail. To address these limitations, we propose HypoAgent, an Agentic framework for interactive abductive Hypothesis Generation over knowledge graphs. HypoAgent integrates three agents: an Intent Recognition Agent that grounds user utterances and dialogue history into executable KG conditions, a Hypothesis Generation Agent that performs controllable hypothesis generation according to the extracted user intention, and a Root Cause Analysis Agent that diagnoses unreliable hypothesis fragments and leverages KG neighborhood probing to identify supported refinements. Experiments on commonsense and biomedical domain-specific knowledge graphs demonstrate that HypoAgent achieves state-of-the-art semantic similarity under single-turn, multi-turn, and unconditional settings. Our code is available at this https URL.
[AI-13] Learning to Adapt: Self-Improving Web Agent via Cognitive-Aware Exploration
链接: https://arxiv.org/abs/2605.31365
作者: Weile Chen,Bingchen Miao,Qifan Yu,Wendong Bu,Guoming Wang,Wenqiao Zhang,Shengyu Zhang,Juncheng Li,Siliang Tang
类目: Artificial Intelligence (cs.AI)
备注: 24 pages
Abstract:Recent advances in Multimodal Large Language Models (MLLMs) have led to promising progress in web agents. However, existing web agents often rely on handcrafted execution pipelines or expensive expert trajectories, limiting their adaptability to complex, dynamic environments. To address these challenges, we propose SCALE (Self-Cognitive-Aware Learning and Exploration), which leverages three adversarial roles, Selector, Predictor, and Judger to autonomously discover the agent’s limitations and expand its cognitive boundaries through environmental exploration. Moreover, we propose SCALE-Hop, a graph exploration strategy that facilitates global planning and helps agents avoid local exploration traps. To further support learning, we construct SCALE-20k, a large-scale dataset collected from 19 real-world websites, containing diverse task types and structured demonstrations generated from SCALE’s exploration traces. Experimental results show that our approach significantly improves the performance and generalization of multiple MLLMs in various web environments. Our framework offers a scalable and generalizable solution for building truly autonomous and adaptive web agents.
[AI-14] dashi: A Python library for Dataset Shift Characterization to Support Trustworthy AI Development and Deployment
链接: https://arxiv.org/abs/2605.31360
作者: David Fernández-Narro,Pablo Ferri,Ángel Sánchez-García,Juan M. García-Gómez,Carlos Sáez
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:The Artificial Intelligence (AI) life cycle requires a thorough understanding of the underlying data dynamics for robust, safe and cost-effective AI development and use. Dataset shifts are defined as changes between train and test data distributions. Whether occurring over time (temporal) or across different sites (multi-source), they can severely degrade model performance and compromise data quality. This is particularly important in health AI, where the safety and fundamental rights of patients can be severely affected by uncontrolled shifts both at training and operational stages. While the theoretical foundations of covariate, prior, and concept shifts are well established, there is a lack of accessible and comprehensive software tools to perform their analysis. We introduce dashi, an open-source Python library designed for the exploration, quantification, and characterization of dataset shifts. dashi provides a dual approach: an unsupervised approach that leverages information geometry and non-parametric statistical manifolds to data variability characterization and analysis (e.g., Information Geometric Temporal plots and Multi-Source Variability metrics like Global Probabilistic Deviation and Source Probabilistic Outlyingness), and a supervised approach that quantifies and characterizes model performance degradation. Both unsupervised and supervised approaches work across user-defined temporal and domain/source batches. We demonstrate the utility of dashi on three simulated and real-world health AI case studies on gestational diabetes mellitus, COVID-19 and emergency medical dispatch. By providing interactive visual analytics and variability metrics, dashi supports trustworthiness of AI life cycle stages enabling robust and safe machine learning pipelines through the assessment of data coherence and AI performance.
[AI-15] Diagnosing Failure Modes of Shared-State Collaboration in Resource-Constrained Visual Agents
链接: https://arxiv.org/abs/2605.31354
作者: Yunpeng Zhou
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Modular visual reasoning systems increasingly rely on shared working memory for multi-step collaboration, yet the failure dynamics of intermediate state evolution in low-capacity regimes remain underexplored. We study failure modes of collaborative reasoning with weak learners (4B–8B models) through the lens of noise accumulation. We introduce CoSee, an auditing framework that formalizes the read-write-verify loop to trace information flow in document visual question answering. Across multi-page, chart, and web-based benchmarks, we find a counter-intuitive degradation: naive shared workspaces often amplify hallucinations rather than resolve them. We identify two dominant failure modes: Noise Reinforcement, where ungrounded notes are reused as evidence, and Policy Collapse, where added context shifts the model toward under-specified, short-form answers. Using cost-accuracy Pareto frontiers, we show that increased compute can correlate negatively with performance without explicit verification. Our findings suggest that for resource-constrained agents, the bottleneck lies not in reasoning depth but in communication fidelity, providing trace-level diagnostics and a mechanistic baseline for reliable modular design.
[AI-16] Inconsistency-Aware Minimization: Improving Generalization with Unlabeled Data ICML2026
链接: https://arxiv.org/abs/2605.31324
作者: Hee-Sung Kim,Hyeonseong Kim,Sungyoon Lee
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: ICML 2026
Abstract:Estimating the generalization gap and developing optimization methods that improve generalization are crucial for deep learning models, for both theoretical understanding and practical applications. Leveraging unlabeled data for these purposes offers significant advantages in real-world scenarios. This paper introduces a novel generalization measure, local inconsistency, derived from an information-geometric perspective on the parameter space of neural networks. A key feature of local inconsistency is that it can be computed without explicit labels. We establish theoretical underpinnings by connecting local inconsistency to the Fisher information matrix and the loss Hessian. Empirically, we demonstrate that local inconsistency correlates with the generalization gap. Based on these findings, we propose Inconsistency-Aware Minimization (IAM), which incorporates local inconsistency into the training objective. We demonstrate that in standard supervised learning settings, IAM enhances generalization, achieving performance comparable to that of existing methods such as Sharpness-Aware Minimization. Furthermore, IAM exhibits efficacy in semi- and self-supervised learning scenarios, where the local inconsistency is computed from unlabeled data.
[AI-17] raceGraph: Shared Decision Landscapes for Diagnosing and Improving Agent Trajectories
链接: https://arxiv.org/abs/2605.31308
作者: Junjie Nian,Kang Chen,Ge Zhang,Yixin Cao,Yugang Jiang
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Agent benchmarks increasingly record rich interaction trajectories, yet evaluation often reduces each rollout to a pass rate or reward score. We introduce TraceGraph, a graph-based framework that turns released multi-model agent trajectories into shared decision landscapes. For each task, TraceGraph builds a graph over observable action-observation states from pooled rollouts before model identity is introduced. It then overlays outcome-informed productive cores and trap regions, and summarizes each rollout with three events: Access, Trap exposure, and Repair. Across trajectories spanning five benchmark splits, TraceGraph profiles reveal navigation differences hidden by aggregate scores and show that splits differ in whether they reward avoiding traps or recovering from them. The same TraceGraph landscape also motivates a trap-aware recovery pipeline for SWE-bench: aruntime detector fires on states matching historical trap regions, then lightweight continuation policies are evaluated from the same prefix. On fired states, the best pooled single-factor policy raises official resolved rate from 40.4% to 43.5% on the per-provider fired subset and from 41.0% to 44.8% on common-fired instances, with provider-specific active components. Overall, TraceGraph provides a process vocabulary for asking what agent benchmarks test, where models diverge on a shared landscape, and how failure regions can guide downstream improvement.
[AI-18] he Terminal Representation in Reinforcement Learning
链接: https://arxiv.org/abs/2605.31289
作者: Amir Esterhuysen,Anders Jonsson
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Representation learning is a powerful tool for spatio-temporal abstraction within reinforcement learning (RL). Two well established approaches are through the successor representation (SR) and the default representation (DR). The SR encodes states by the future trajectories they induce, capturing information flow decoupled from reward. The DR builds on this by weighting trajectories with reward, integrating credit-assignment structure into the representation. Eigenvectors of both representations have been used to support a range of downstream tasks – including option discovery, reward shaping, transfer learning, and exploration. We introduce a structurally distinct formulation: the terminal representation (TR). The TR encodes reward-weighted trajectories similarly to the DR, but can be learned as a lower-dimensionality object, and can be used directly for the mentioned applications without eigenvector computations. Eigendecomposition also imposes the assumption of symmetric transition dynamics, which the TR can bypass. In this work we develop the theoretical foundations of the TR: its derivation, convergence of two learning algorithms, its use for zero-shot compositionality, and equivalences between alternative reward formulations. We further show the TR is embedded in the top DR eigenvector, allowing it to capture the same underlying knowledge without eigendecomposition. Additionally, we provide empirical evidence of the TR as a viable alternative to existing representations in subsidiary applications, while requiring less computational overhead to learn, store, and use.
[AI-19] DeMaVLA: A Vision-Language-Action Foundation Model for Generalizable Deformable Manipulation
链接: https://arxiv.org/abs/2605.31286
作者: Taiyi Su,Jian Zhu,Tianjian Wang,Youzhang He,Zitai Huang,Jianjun Zhang,Chong Ma,Hanyang Wang,Tianjiao Zhang,Munan Yin,Weihao Ding,Yi Xu
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注: 14 pages, 2 figures
Abstract:Real-world household robots require Vision-Language-Action (VLA) foundation models that can acquire reusable manipulation skills across diverse objects, task conditions, and household environments. Deformable-object folding is a representative challenge, requiring robots to handle clothing items from random initial states across varying categories, geometries, materials, and scenes. However, existing VLA systems commonly train separate policies for different object categories, while naively mixed multi-task training often suffers from task interference and degraded performance. To move beyond category-specific folding policies, we introduce DeMaVLA, a VLA foundation model for generalizable Deformable Manipulation. DeMaVLA adopts a VLM backbone with an action expert and formulates continuous action generation using flow matching. To improve efficiency, the action expert is constructed by pruning every other transformer layer while preserving layer-wise alignment with the VLM backbone, reducing training and inference cost. DeMaVLA is first pre-trained on approximately 5,000 hours of selected real-world dual-arm demonstrations to acquire general manipulation priors. It is then post-trained on mixed folding data that aggregates self-collected demonstrations and corrective trajectories from real-robot failures across multiple folding tasks through a human-in-the-loop Data Aggregation~(DAgger) pipeline. Experiments show that DeMaVLA achieves competitive performance on RoboTwin and strong real-world results on our household folding benchmark. These results highlight the value of scalable real-world data, efficient action generation, and corrective learning for general-purpose VLA policies in deformable-object manipulation.
[AI-20] Industrializing Prediction-Powered Inference: The GLIDE Library for Reliable GenAI and Agent ic Systems Evaluation ICML2026
链接: https://arxiv.org/abs/2605.31278
作者: Grégoire Martinon,Ibrahim Merad,Mohammed Raki
类目: Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Methodology (stat.ME)
备注: 8 pages, accepted at the ICML 2026 workshop Agentic Uncertainty
Abstract:Reliable evaluation of agentic systems requires unbiased estimates with valid uncertainty, but standard practice navigates between costly human annotation and biased LLM-as-judge proxies. Prediction-powered inference (PPI) combines both into debiased estimates with valid confidence intervals, yet its various methods remain scattered across papers under partial implementations. We introduce GLIDE, an open-source Python library that unifies state-of-the-art PPI estimators (PPI++, Stratified PPI, Predict-Then-Debias and its stratified variants, Active Statistical Inference) and samplers (uniform, stratified, active, cost-optimal) under a scipy-style API specialized to mean estimation. GLIDE ships with a reproducible Monte Carlo validation suite, an empirically grounded decision tree for method selection, and an agentic evaluation case study showing substantial annotation savings at equivalent precision. The GLIDE package is available at this URL: this https URL
[AI-21] Why Linear Recurrent Memory Works in Partially Observable Reinforcement Learning
链接: https://arxiv.org/abs/2605.31261
作者: Yike Zhao,Onno Eberhard,Malek Khammassi,Ali H. Sayed,Michael Muehlebach
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Machine Learning (stat.ML)
备注:
Abstract:The family of linear recurrent neural networks has shown strong performance as recurrent memory units in partially observable reinforcement learning. We provide a theoretical justification for their empirical effectiveness by constructing and studying two linear filters: (i) the first exactly reproduces the pre-softmax logits of the belief vector in a hidden Markov model (HMM) under a deterministic transition matrix, thereby serving as a sufficient statistic for optimal policy learning, (ii) the second achieves vanishing state-decoding error under a nearly deterministic transition matrix, thus reducing state ambiguity to near zero. The results extend to action-controlled HMMs, where the corresponding linear filters become time-varying with action-dependent dynamics. We illustrate our main results through numerical experiments and further show that the constructed linear filter serves as a strong feature extractor in a small reinforcement learning game.
[AI-22] Formalizing and falsifying causal pathways of rare events ICML2026
链接: https://arxiv.org/abs/2605.31254
作者: Anahita Haghighat,Dominik Janzing
类目: Artificial Intelligence (cs.AI)
备注: accepted for ICML 2026
Abstract:Building on recent formalizations of root cause analysis for rare events (``outliers’') in structural equation models, we propose a formal definition of a causal pathway and discuss its testable implications. We identify conditions under which these implications depend only on a causal abstraction defined by the pathway of rare events, rather than on the full causal graph of the underlying system. Accordingly, we introduce an abstraction of causal structure to pathways of rare events that bridges simple verbal causal explanations and detailed causal modeling.
[AI-23] Learning Cardiac Latent Representations in Vectorcardiogram Space
链接: https://arxiv.org/abs/2605.31249
作者: Bosong Huang,Panzhen Zhao,Zengxiang Li,Patricia Lee,Wei Jin,Alan Wee-Chung Liew,Ming Jin,Shirui Pan
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Electrocardiography (ECG) is a cornerstone of cardiac assessment, making the learning of informative ECG representations fundamental to tasks ranging from disease diagnosis to clinical report generation. However, existing methods operate almost exclusively in the observable ECG signal space. In practice, the standard twelve-lead ECG represents multiple projections of the same underlying cardiac electrical activity from different spatial orientations. Therefore, representation learning in the ECG space inevitably introduces substantial redundancy, which may lead to spurious correlations and increased risk of overfitting. To address this and motivated by the Frank vectorcardiogram (VCG) model, we propose learning a unified latent representation of cardiac electrical activity directly in the VCG space. We introduce LVCG, the first general self-supervised representation learning framework designed to operate in this physically grounded latent space. By learning view-invariant latent VCG representations rather than lead-specific artifacts, VCG minimizes redundancy and improves generalization. LVCG generally outperforms ECG-space baselines across tasks, demonstrating enhanced robustness and generalization, especially in domain shift settings.
[AI-24] EchoRL: Reinforcement Learning via Rollout Echoing ICML2026
链接: https://arxiv.org/abs/2605.31228
作者: Jinhe Bi,Aniri,Minglai Yang,Xingcheng Zhou,Wenke Huang,Sikuan Yan,Yujun Wang,Zixuan Cao,Michael Färber,Xun Xiao,Volker Tresp,Yunpu Ma
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: ICML 2026
Abstract:Reinforcement Learning with Verifiable Rewards is an effective route for post-training to strengthen the reasoning capability of large language models. However, as training proceeds, the learning signal can collapse thus makes the training gain become marginal and ineffective. Specifically, a growing fraction of prompts’ rollouts become advantage-degenerated: all the self-generated rollouts show verified-success, making the standard deviation over their rewards be zero; accordingly each rollout’s advantage becomes degenerated (zero) as well. Given such rollouts’ advantages, the policy-gradient for model optimization eventually vanishes, capping the training performance. We argue that some of these rollouts still contain valuable learning signals but unfortunately omitted with the existing RLVR methods. In this paper, inspired through analyzing the entropy pattern behind golden trajectories produced by external expert models, we propose EchoRL for better exploiting the advantage-degenerated rollouts to further improve the training performance. EchoRL is a lightweight module that first identifies an EchoClip from verified-success rollouts based on their step-level entropy values, and then feeds this clip back as an auxiliary supervision signal in the RL objective. Extensive experiments across 10 benchmarks, 5 LLM backbones, and 4 popular RLVR post-training methods demonstrate that EchoRL consistently improves RLVR post-training with minimal overhead.
[AI-25] What changes after deployment? A survey on On-device Learning in TinyML
链接: https://arxiv.org/abs/2605.31226
作者: Massimo Pavan,Luca Pezzarossa,Fabrizio Pittorino,Manuel Roveri,Xenofon Fafoutis
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Machine learning models on microcontroller-class devices (TinyML) face a fundamental challenge: post-deployment distribution change undermines static models. On-device learning (ODL) addresses this by running the learning process directly on the device. The existing literature has not characterized how distribution change occurs or how different change types require different solutions. Approximately 70 ODL works are surveyed under one principle: the distribution change regime. The survey analyzes how different types of distribution change influence the applications addressable on-device, the hardware employed, and the structure of the solutions. A persistent gap between methodological benchmarks and real-world deployment scenarios is also identified.
[AI-26] Simulation of collision avoidance behavior in crowd movement by data-driven approach
链接: https://arxiv.org/abs/2605.31210
作者: Xuanwen Liang,Eric Wai Ming Lee
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注:
Abstract:Crowd movement simulation is essential for pedestrian safety management and facility layout optimization. Data-driven models enhance trajectory prediction accuracy under Euclidean metrics, yet they suffer from excessively high collision rates, especially in bidirectional and multidirectional flows. In this paper, we establish a novel data-driven crowd simulation model that incorporates the pedestrian collision mechanism into the loss function to reduce collisions. A new lateral-acceleration-based collision loss function and a Voronoi-based motion feature extraction approach are proposed. The model is based on a Generative Adversarial Network (GAN) architecture and is termed CPGAN (Collision-Penalized GAN). We evaluate CPGAN in bidirectional flow scenarios, which involve frequent collision avoidance behaviors. Results show that the proposed lateral-acceleration-based collision loss significantly reduces opposite-direction pedestrian collision rates to levels comparable with controlled experiments. CPGAN effectively simulates bidirectional flow, reproducing lane formation and N-t curves. The research outcomes can provide inspiration for integrating pedestrian dynamics mechanisms into loss functions in data-driven crowd simulation.
[AI-27] MAECO-Lite: Modular Ontology for Dynamic Malware Analysis
链接: https://arxiv.org/abs/2605.31199
作者: Zekeri Adams,Peter Švec,Ján Kľuka,Roderik Ploszek,Monday Onoja,Štefan Balogh,Martin Homola
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注:
Abstract:Capturing dynamic malware behavior in a practical but still semantically precise manner remains a significant challenge in cyber threat intelligence. While standards such as MAEC and STIX provide widely adopted vocabularies for describing malware artifacts and observations, they represent data with considerable complexity in structures that often obscure important ontological distinctions. In particular, they tend to conflate enduring malware artifacts with the events generated during execution, thereby flattening distinctions that are central in foundational standards for ontology design. In this paper, we conduct a foundational ontological analysis of core MAEC and STIX constructs relevant to dynamic malware analysis relying on Unified Foundational Ontology (UFO) as a theoretical lens. Our analysis reveals some ontological mismatches arising from the conflation of artifacts, dispositions, and runtime events in MAEC and STIX that complicate coherent representation of dynamic malware behavior and, from a practical perspective, limit the ability to reason about execution traces. Based on these insights, we propose MAECO-Lite, a lightweight ontology designed to represent data and operationalize their processing for dynamic malware analysis. The ontology adopts a modular structure centered on samples, processes, actions, system artifacts, and MITRE ATTCK Techniques, while maintaining a clear separation between enduring entities and runtime events. An initial evaluation using description logic concept learning algorithms shows that the simplified ontology significantly improves learning performance, demonstrating that ontologically grounded modelling can enhance both semantic clarity and computational usability.
[AI-28] MindVoice: Reconstructing Intelligible Speech from Non-invasive Neural Signals with Pretrained Priors
链接: https://arxiv.org/abs/2605.31173
作者: Guangyin Bao,Taiping Zeng,Jianfeng Feng,Xiangyang Xue
类目: ound (cs.SD); Artificial Intelligence (cs.AI)
备注:
Abstract:Reconstructing continuous speech from non-invasive neural recordings is a fundamental problem for probing human auditory perception and building safe, scalable speech brain-computer interfaces. Despite recent progress, intelligible reconstruction remains elusive, as non-invasive recordings are inherently noisy, spatially blurred, and only partially preserve information about perceived speech. Existing methods directly map neural activity to entangled speech representations before synthesizing waveforms with neural vocoders, resulting in spectral-similar but unintelligible results. To overcome these limitations, we introduce MindVoice, a neuro-to-speech reconstruction framework that uses pretrained models to compensate for the incomplete semantic and acoustic information in neural recordings. MindVoice disentangles reconstruction into two complementary pathways: one recovers high-level semantic content, while the other estimates fine-grained acoustic attributes. These inferred representations are then fused with powerful speech generation models and in-context voice cloning to synthesize natural and intelligible utterances. Extensive experiments on EEG and MEG demonstrate that MindVoice substantially outperforms existing methods on various metrics. These results show that pretrained priors provide a principled way to bridge the gap between noisy neural recordings and natural speech, highlighting a promising attempt for auditory neuroscience research and non-invasive speech brain-computer interfaces.
[AI-29] LLM -FACETS: A Privacy-Preserving Framework for Evaluating LLM Transparency and Accountability
链接: https://arxiv.org/abs/2605.31167
作者: Tom Lucas,Alessio Buscemi,Alfredo Capozucca,German Castignani,Barbara Delacroix
类目: Artificial Intelligence (cs.AI)
备注: Submitted to ACM Journal on Responsible Computing, Special Section: Collaborative Methods and Tools for Engineering and Evaluating Transparency in AI. 28 pages 9 figures, 7 tables, 1 algorithm. Source code: this https URL
Abstract:Assessing whether Large Language Models outputs are factually grounded, epistemically calibrated, and methodologically reproducible is a prerequisite for responsible AI deployment. Yet auditing LLMs remains inaccessible to non-technical practitioners: existing tools require programming expertise and non-trivial environment setup, and cloud-hosted platforms transmit evaluation data to external services, creating barriers for domain experts and compliance officers legally responsible for AI oversight. We introduce LLM-FACETS (LLM FActuality Cross-EvaluaTion System): an open-source framework with a browser-accessible interface and a plugin architecture, structured around three practitioner profiles (technical experts, domain experts, compliance officers) that mirror the stakeholder categories identified in the EU AI Act and the NIST AI Risk Management Framework. The architecture makes data flows explicit: deterministic metrics (BLEU, ROUGE, BERTScore) run entirely within the self-hosted server with no outbound transmission; LLM-judge metrics contact external APIs explicitly, with users retaining full credential control. The framework operationalizes transparency through three mechanisms: token-level log-probability visualization for epistemic uncertainty, multi-judge consensus to mitigate judge bias, and RAG Triad metrics (Faithfulness, Answer Relevance, Context Relevance) to detect and localize hallucinations. A plugin architecture allows any new metric or dataset to be integrated without modifying the evaluation pipeline. The open-source implementation enables cross-checking across multiple metrics targeting the same property, ensuring reproducibility and decoupling AI accountability from the teams building the systems assessed. We verify the framework through cross-validation of 18 metric implementations against canonical reference libraries.
[AI-30] rust-Region Behavior Blending for On-Policy Distillation
链接: https://arxiv.org/abs/2605.31159
作者: Daniil Plyusov,Alexey Gorbatovski,Alexey Malakhov,Nikita Balagansky,Boris Shaposhnikov,Daria Korotyshova,Daniil Gavrilov
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:On-policy distillation (OPD) trains a student on prefixes sampled from its own policy while matching a stronger teacher. This addresses the prefix mismatch of offline distillation, but early student rollouts can still be poor, placing teacher supervision on weak or low-quality prefixes. We propose Trust-Region behavior Blending (TRB), a warmup method that replaces the early rollout policy with the closest-to-teacher behavior policy inside a student-centered KL trust region, while keeping the per-prefix reverse-KL OPD loss unchanged. The KL budget is annealed to zero, so training returns to pure student rollouts after warmup. Across two math-reasoning distillation settings, TRB attains the strongest average among the compared methods.
[AI-31] ARIC: Memory-Augmented Traversability-Aware Outdoor VLN under Interrupted Semantic Cues
链接: https://arxiv.org/abs/2605.31121
作者: Tianle Zeng,Hanjing Ye,Jianwei Peng,Jingwen Yu,Hanxuan Chen,Hong Zhang
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注:
Abstract:Outdoor vision-language navigation (VLN) in long-range, open-world environments is frequently disrupted by semantic-cue interruptions, where informative goal cues become sparse, occluded, or leave the field of view. Once such cues disappear, agents enter a cue-free phase and often degrade into backtracking, oscillatory headings, or aimless exploration. While memory-based methods attempt to bridge these gaps, they often fail under traversability-driven detours: the remembered cue direction may be infeasible, forcing detours that prolong cue-free phases and gradually render robot-centric cues stale and implicit histories blurred. This makes traversability a stability condition for maintaining goal-directed guidance, rather than merely a local safety concern. We propose a unified outdoor VLN framework that survives semantic-cue interruptions by maintaining traversability-consistent executable guidance throughout prolonged cue-free phases. Specifically, our method extracts semantic bearings from visibility-gated goal or exploration cues and grounds them into executable headings using a real-time near-field traversability profile, providing goal-consistent feasible guidance beyond reject-only safety filtering. To prevent guidance degradation during detours, we lift intermittent 2D evidence into a world-aligned 3D cue memory with an uncertainty-aware readout mechanism, ensuring guidance remains continuously reachable and stable as the robot moves. We evaluate the framework on quadrupedal and wheeled platforms over 600–1000 m routes. Our method improves simulation success rate by over 10 percentage points over the strongest baseline and achieves a real-world success rate of 40%, compared to 17.5% for the strongest baseline, with substantially higher robustness during prolonged cue-free intervals. Subjects: Robotics (cs.RO); Artificial Intelligence (cs.AI) Cite as: arXiv:2605.31121 [cs.RO] (or arXiv:2605.31121v1 [cs.RO] for this version) https://doi.org/10.48550/arXiv.2605.31121 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-32] SWIM: Single-Instance Whole-Body Imitation for swiMming
链接: https://arxiv.org/abs/2605.31120
作者: Binglun Wang,Edmond S. L. Ho,He Wang
类目: Graphics (cs.GR); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:We propose a new method for synthesizing physically-based swimming motions. Physically-based character animation aims to generate physically valid, controllable, and natural-looking motions which can respond to unexpected disturbances, where one dictating factor of difficulty is the complexity of the task, especially the level of sophistication of the required interactions with the environment. Existing research has succeeded in various tasks in static and dynamic environments. We push the difficulty further to swimming, which requires full-body coordination and continuous interactions with fluids, a new level of complexity when it comes to interacting with the environment. This complexity imposes challenges in learning control under volatile environmental forces, generalizing control to different environments and swimming styles, lack of data references, and prohibitively slow physical simulation which is inevitable during control learning. To this end, we propose SWIM, a new imitation method for swimming motions, which can learn from a single swimming motion and generalize to unseen environments, body conditions, and swimming styles. Extensive evaluation and comparison demonstrate that SWIM is data-efficient, stable, robust, and generalizable, outperforming alternative methods across multiple classes of tasks and metrics.
[AI-33] SpecDB: LLM -Generated Customized Databases via Feature-Oriented Decomposition
链接: https://arxiv.org/abs/2605.31097
作者: Yunkai Lou,Longbin Lai,Shunyang Li,Zhengping Qian,Ying Zhang
类目: Databases (cs.DB); Artificial Intelligence (cs.AI)
备注:
Abstract:Mainstream relational databases ship a uniform feature set across deployments, although individual workloads exercise only a fraction of the available subsystems. We investigate whether a database can instead be generated on demand with a feature set matched to the target workload. We present SpecDB, a system that uses large language models (LLMs) to synthesize customized relational databases. We survey 9 production systems and decompose them into 10 functional modules, each further divided into implementation variants. To capture cross-module dependencies, including cases where implementations in disjoint subtrees must be co-designed, we adopt the FODA feature model and extend it with a cooperate edge, yielding a dependency graph DBGraph. SpecDB operationalizes DBGraph through a layered module-construction pipeline in which each module is generated, validated, and integrated by a dedicated subagent (driven by three inner agents: Main, Tester, Architect), and a Refining Agent that iteratively repairs and tunes the assembled database against a user-supplied refining harness with read-only access to existing database source code. A companion selection component translates a natural-language workload description into a set of implementation variants, providing an end-to-end pipeline from workload description to deployable database. We evaluate SpecDB on TPC-C with BenchmarkSQL. The generated database (23,779 lines of Rust) completes 60-minute TPC-C at 1 and 10 warehouses with zero errors. At 10 warehouses it reaches tpmC=130, compared to 128 for PostgreSQL and 127 for MySQL, with comparable latency at ~3% of their code size. Because the agent operates at module-specification level rather than product source, it can in principle combine techniques across system boundaries. Paired with falling LLM costs, generating a purpose-built database for a target workload is becoming straightforward.
[AI-34] STEP: Learning STructured Embeddings for Progressive Time Series
链接: https://arxiv.org/abs/2605.31061
作者: Lucas Thil,Jesse Read,Rim Kaddah,Guillaume Doquet
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:We present a novel method for learning interpretable representations of progressive time series, that is, data capturing irreversible state transitions such as degradation or task completion. Our approach uses a self-supervised contrastive objective to learn a low-dimensional latent space whose geometry is itself the interpretation: each observation becomes a point on a manifold anchored between two fixed orthogonal prototype vectors, and a trajectory becomes a path across that manifold. From this structure we read a latent compass, the polar coordinates (\theta, r) of the latent vector, in which \theta tracks the progression of the underlying state (e.g., from healthy to failed) and r identifies the active mode (e.g., the operating condition), without any proxy labels. We evaluate the approach against the state of the art on diverse domains, including industrial degradation, robotic tasks, and neural activity, validating three key capabilities: (1) end-state prediction, (2) multi-step forecasting, and (3) interpretable phase separation. Our method matches or improves over black-box counterparts on all of these while providing transparency about the underlying mechanisms. A simple linear regressor on top of the latent compass coordinates is competitive with deep architectures, direct quantitative evidence that the underlying state is encoded in a geometrically accessible form.
[AI-35] AnchorSteer: Self-Discovered Concept Injection for Structure-Preserving Music Editing KDD2026 KDD
链接: https://arxiv.org/abs/2605.31053
作者: Chih-Heng Chang,Keng-Seng Ho,Chih-Yu Tsai,Kuan-Lin Chen,Yi-Hsuan Yang,Jian-Jiun Ding
类目: ound (cs.SD); Artificial Intelligence (cs.AI)
备注: Accepted by the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD 2026)
Abstract:Controllable music editing is to modify high-level attributes while strictly preserving rhythmic and melodic structures. However, this task is challenged by a semantic-structural entanglement: steering methods often degrade structure to achieve editing performance, while structural adaptors suppress semantic responsiveness. We propose AnchorSteer, a framework that disentangles this tension by coupling structural anchoring with self-discovered semantic steering. The proposed approach probes internal representations to extract interpretable, label-free concept vectors via a self-supervised reconstruction objective, isolating attributes without curated data. During editing, these portable, plug-and-play concept vectors are injected into diffusion hidden manifolds while a structural adaptor enforces consistency. Variants for unconditioned and conditioned injections are provided to balance robustness and semantic strength. Experiments on ZoME-Bench and subjective tests show that the proposed framework outperforms both steering-only and anchoring-only baselines, enabling significant semantic transformations with high-fidelity structural preservation.
[AI-36] Linear Ordering Problem: Time for a Change
链接: https://arxiv.org/abs/2605.31051
作者: Fabrizio Fagiolo,Marco Baioletti,Valentino Santucci
类目: Neural and Evolutionary Computing (cs.NE); Artificial Intelligence (cs.AI)
备注:
Abstract:The Linear Ordering Problem (LOP) is a fundamental combinatorial optimization problem with important applications in areas such as economics, social choice, and machine learning. Its most prominent use is the triangulation of economic input-output tables, which helps identify critical industries in an economy. Most existing algorithms have been evaluated on benchmarks derived from outdated macroeconomic data, which no longer reflect the structure of contemporary economies. Furthermore, LOP instances often exhibit many distinct global optima that can differ substantially from one another, creating challenges for applications that rely on a single solution. To address these limitations, we introduce a novel benchmark suite derived from up-to-date real-world economic data and an algorithmic scheme that leverages state-of-the-art LOP metaheuristics to generate diverse sets of high-quality solutions, together with metrics for assessing both quality and diversity. Experiments were conducted to report results on the proposed benchmark suite under both the traditional single-solution setting and the newly introduced multi-solution scenario
[AI-37] Learning to Solve and Optimize by Evolving Code IJCAI26
链接: https://arxiv.org/abs/2605.31049
作者: Veronika Semmelrock,Benedetta Strizzolo,Francesco Zuccato,Gerhard Friedrich,Patrick Rodler,Konstantin Schekotihin
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Logic in Computer Science (cs.LO)
备注: Preprint of a paper accepted to IJCAI26
Abstract:Combinatorial and optimization problems are fundamental to many industrial AI applications. Solving large-scale real-world instances of such problems typically requires careful problem formalization, specialized solvers, and expert-designed heuristics. Thus, experts need to specify not only what solutions are, but also how they are derived. By introducing the tool CHECKMATE, we show that algorithm generation via code evolution represents a paradigm shift by eliminating the need to formulate the how. CHECKMATE solely relies on the what. Specifically, a formal specification ensures solutions’ correctness and enables systematic performance evaluation of the generated programs, while a natural language description guides the evolutionary process. The effectiveness of our method is demonstrated on selected problems from two industrial domains: configuration and scheduling. In all cases, the evolved algorithms consistently outperform state-of-the-art solvers. This underscores the potential of formal methods in guiding code evolution for automatically solving complex real-world problems.
[AI-38] Annealed Softmax Greedy in Many-Armed Bayesian Bandits
链接: https://arxiv.org/abs/2605.31034
作者: William Overman,Mohsen Bayati
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Reinforcement learning with verifiable rewards (RLVR) and group-based policy optimization methods such as GRPO update a stochastic policy by sampling multiple completions per prompt and increasing the policy’s probability on those with higher reward, regularized by a KL penalty toward a reference policy. These updates do not include explicit mechanisms that track epistemic uncertainty. This paper studies a stylized explanation for why such uncertainty-agnostic updates can nevertheless be effective. We analyze an annealed softmax (Boltzmann) policy that selects actions according to a softmax of empirical mean rewards in a many-armed Bayesian Bernoulli bandit. Under a linear upper-tail condition on the prior (the \beta=1 case of \beta -regularity), which implies an abundance of near-optimal arms, we prove that annealed softmax greedy achieves Bayes regret \tildeO(m + T/m) , and in particular \tildeO(\sqrtT) when the number of arms scales as m = \Theta(\sqrtT) . This is the near-optimal Bayes regret rate in this regime, attained also by empirical-mean greedy. Under \beta -regularity, many arms maintain empirical means close to the optimum throughout learning, so when softmax samples an arm other than the empirically best, that arm tends to be another near-optimal one rather than a clearly inferior one. By contrast, with a small number of arms, the same kind of softmax policy can suffer linear regret. The result also provides a structural analogy to RLVR, where a base policy with a non-negligible probability of producing a correct completion plays the role of \beta -regularity.
[AI-39] GraphARC: A Comprehensive Benchmark for Graph-Based Abstract Reasoning KDD2026
链接: https://arxiv.org/abs/2605.31031
作者: Saku Peltonen,August Bøgh Rønberg,Andreas Plesner,Roger Wattenhofer
类目: Artificial Intelligence (cs.AI)
备注: Accepted at KDD 2026 Datasets and Benchmarks Track
Abstract:Relational reasoning lies at the heart of intelligence, but existing benchmarks are typically confined to formats such as grids or text. We introduce GraphARC, a benchmark for abstract reasoning on graph-structured data. GraphARC generalizes the few-shot transformation learning paradigm of the Abstraction and Reasoning Corpus (ARC). Each task requires inferring a transformation rule from a few input-output pairs and applying it to a new test graph, covering local, global, and hierarchical graph transformations. Unlike grid-based ARC, GraphARC instances can be generated at scale across diverse graph families and sizes, enabling systematic evaluation of generalization abilities. We evaluate state-of-the-art language models on GraphARC and observe clear limitations. Models can answer questions about graph properties but often fail to solve the full graph transformation task, revealing a comprehension-execution gap. Performance further degrades on larger instances, exposing scaling barriers. More broadly, by combining aspects of node classification, link prediction, and graph generation within a single framework, GraphARC provides a promising testbed for future graph foundation models.
[AI-40] DEM: A Distilled Explanation Model for Interpretable Anomaly Detection in Physiological Sensor Networks
链接: https://arxiv.org/abs/2605.31007
作者: Jyotirmoy Singh,Anushka Roy,Shreea Bose,Chittaranjan Hota
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 21 pages, 10 figures, 7 tables. Code: this https URL
Abstract:Anomaly detection in physiological sensor data from Wireless Body Area Networks (WBANs) can be caused by sensor faults, network disruptions, or missing data, leading to false alarms. Hence, it demands both high predictive accuracy and clinically interpretable explanations. Existing approaches rely either on black-box models that achieve strong performance but offer no transparency, or on post-prediction explanation methods such as SHAP and LIME. In this paper, we propose the Distilled Explanation Model (DEM), a three-stage glass-box framework that distills the non-linear knowledge of a gradient boosting expert into an interpretable decision tree operating on residuals relative to a linear baseline, so that the explanation is not an approximation but the prediction itself. DEM introduces a novel distillation fidelity metric that quantifies how faithfully the explanation tree captures the expert model’s non-linear contribution, providing a principled measure of explanation trustworthiness absent from prior interpretable models. Evaluated across four physiological datasets, including MIMIC-IV, WESAD, eICU, and an in-house SmartNet WBAN corpus, DEM achieves an AUC of 0.9964 on clinical contextual anomaly detection and 0.9047 on wearable stress detection while producing human-readable if-then rules at a controllable depth. Inference requires 0.17ms per 1000 samples, rendering DEM 1235x faster than SHAP-based post-hoc explanation and suitable for real-time physiological monitoring. Ablation studies confirm that the XGBoost distillation step provides measurable gains over naive residual fitting, and depth-sensitivity analysis demonstrates an explicit, user-controlled accuracy-interpretability trade-off unique to DEM among existing intrinsically interpretable models.
[AI-41] De-attribute to Forget for LLM Unlearning
链接: https://arxiv.org/abs/2605.30919
作者: Xinyang Lu,Jiabao Pan,Rachael Hwee Ling Sim,See-Kiong Ng,Anthony Kum Hoe Tung,Bryan Kian Hsiang Low
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:The rapid development of large language models (LLMs) has raised concerns on the use of inappropriate data for training, which has led to a growing interest in LLM unlearning. Many existing LLM unlearning approaches rely on optimizing prediction loss(es), such as maximizing the loss on the forget set, but often face critical issues like over-forgetting and poor model utility. To address them, this paper novelly frames the optimization objective for LLM unlearning as one of zeroing out data attribution instead. In particular, we propose the first LLM unlearning framework based on data attribution rewards called DareU that performs reinforcement learning to update the LLM by reducing the attribution score of its generated responses (i.e., de-attributing) to the forget data owners. Empirical evaluation using an LLM classifier as an efficient approximation of attribution shows that DareU outperforms existing baselines by achieving effective unlearning while balancing forget quality and model utility well.
[AI-42] Inverse Reinforcement Learning without an Optimal Demonstrator: A Feasible Reward Set Approach
链接: https://arxiv.org/abs/2605.30903
作者: Kihyun Kim,Shripad Deshmukh,Nikos Vlassis,Jiawei Zhang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Inverse reinforcement learning (IRL) typically assumes demonstrations from a single optimal demonstrator, but in many applications data come from multiple imperfect demonstrators with heterogeneous suboptimality levels. We study reward learning in this setting through a feasible-reward-set framework: for each demonstrator, we encode its declared suboptimality level as a linear constraint and intersect the resulting feasible sets across demonstrators. Our theoretical analysis shows that the joint feasible set shrinks monotonically as data are added, and we give an exact characterization of when a new demonstrator strictly tightens it. We further establish two recovery guarantees for the feasible reward set of the ground-truth optimal demonstrator: one bound depends on closeness to the optimal occupancy, while the other requires only sufficient coverage and no near-optimal demonstrator. On the practical side, we introduce strategies to address the inherent reward ambiguity in the obtained reward set and provide an offline algorithm with function approximation for high-dimensional environments. Experiments in tabular grid-world and large language model (LLM) fine-tuning settings are consistent with the theoretical predictions and demonstrate the effectiveness of the proposed framework over baselines.
[AI-43] BilliardPhys-Bench: Benchmarking Physical Reasoning and Visual Dynamics of Multimodal LLM s
链接: https://arxiv.org/abs/2605.30900
作者: Ben Wang,Xiaogang Li,Ruochen Gao,Peiyao Xiao,Chengliang Xu,Zeyu Wang,Zichao Chen,Bing Zhao,Hu Wei
类目: Artificial Intelligence (cs.AI); Applied Physics (physics.app-ph)
备注:
Abstract:Current multimodal models handle static image recognition well, but intuitive physical reasoning remains a weakness. Predicting how objects will move and interact from a single image is still difficult for these systems. We present BilliardPhys-Bench, a benchmark for physical reasoning in synthetic billiards environments. Its procedural engine generates randomized scenarios with friction and elastic collisions. The benchmark tests three abilities: (1) predicting ball-to-ball collisions, (2) reasoning about wall bounces, and (3) estimating final ball positions after motion stops. We evaluate recent MLLMs from the GPT, Claude, Gemini, and Qwen families. Performance drops as simulation time increases and scene geometry grows more complex. We also observe a consistent failure mode we call “stasis bias”: when the correct physical outcome is harder to infer, models tend to predict no interaction. These findings show where current MLLMs break down on visual dynamics and point toward the need for better physical inductive biases in multimodal architectures.
[AI-44] Federated Variational Preference Alignment with Gumbel-Softmax Prior for Personalized User Preferences ICML2026
链接: https://arxiv.org/abs/2605.30873
作者: Jabin Koo,Hoyoung Kim,Minwoo Jang,Jungseul Ok
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Distributed, Parallel, and Cluster Computing (cs.DC)
备注: 21 pages, 4 figures. Accepted to ICML 2026
Abstract:Federated Learning (FL) offers a privacy-preserving pathway for aligning Large Language Models (LLMs); however, existing frameworks typically enforce a monolithic reward model, inevitably averaging out inherently conflicting user preferences (e.g., helpfulness vs. harmlessness). While Variational Preference Learning (VPL) offers a pathway to personalization, adapting it to decentralized settings presents a fundamental challenge: posterior collapse driven by severe local data scarcity and heterogeneity. In this paper, we propose Federated Variational Preference Alignment with Gumbel-Softmax Prior (FedVPA-GP), a framework designed to disentangle diverse preferences without compromising privacy. To stabilize variational inference, we introduce a Federated Mixture Prior that enables clients to leverage the aggregate population distribution as a dynamic prior. Furthermore, we incorporate an Orthogonal Loss that explicitly enforces the separation of preference prototypes in the latent space. Experiments on the HH-RLHF dataset demonstrate that FedVPA-GP significantly outperforms monolithic baselines, successfully disentangling conflicting user intents and enabling dynamic preference switching.
[AI-45] Sophrosyne: Agent ic Exploration of Relational Data Systems Needs Moderation
链接: https://arxiv.org/abs/2605.30862
作者: Madhav Jivrajani,Ramnatthan Alagappan,Aishwarya Ganesan
类目: Databases (cs.DB); Artificial Intelligence (cs.AI)
备注:
Abstract:Text2SQL agents powered by LLMs translate natural language intent into SQL by exploring the data system through tool calls before formulating the query. However, to ensure secure and scoped access, data systems construct environments with explicit API surfaces. We study and categorize these APIs exposed today as either coarse-grained or fine-grained and posit that choosing between them presents a fundamental tradeoff between cost-efficient exploration and accurate SQL generation. Most data systems expose fine-grained APIs, but this inadvertently disadvantages agents: they over-explore, incorporating irrelevant schema elements into their query formulation and produce inaccurate results. We argue that curbing over-exploration is key to the effective use of these API surfaces, and propose Sophrosyne, a data system environment that augments API responses with directives that guide the agent’s exploration process. Initial results show that directives reduce over-exploration by 4.6x and boost accuracy by up to 12.4% (approx. 4 percentage points).
[AI-46] Distilling LLM Feedback for Lean Theorem Proving
链接: https://arxiv.org/abs/2605.30861
作者: Gaetan Narozniak,Gérard Biau,Rémi Munos,Ahmad Rammal,Pierre Marion
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Post-training for reasoning models typically combines supervised fine-tuning with reinforcement learning from verifiable rewards, most commonly with GRPO. However, this algorithm suffers from sparse rewards, limited exploration, and mode collapse. Building upon recent works on self-distillation, we propose Feedback Distillation, a training method where the model is trained to match, at the token level, its own distribution conditioned on privileged feedback produced by a language model. Feedback Distillation offers token-level supervision and can inject external knowledge. Evaluating our method for Lean4 theorem-proving, we find that Feedback Distillation maintains greater diversity in generated trajectories than GRPO, yielding higher policy entropy and better pass@k scaling. The two methods are complementary: initializing GRPO from a Feedback Distillation checkpoint outperforms either method alone. All in all, our results suggest a promising avenue to improve post-training for complex reasoning.
[AI-47] DARTS: Distribution-Aware Active Rollout Trajectory Shaping for Accelerating LLM Reinforcement Learning ICML2026
链接: https://arxiv.org/abs/2605.30859
作者: Yujie Wang,Siwei Chen,Longzan Luo,Xinyi Liu,Xupeng Miao,Fangcheng Fu,Bin Cui
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 16 pages, 14 figures, 5 tables. Accepted to ICML 2026
Abstract:Reinforcement Learning (RL) has become pivotal for improving model capabilities yet suffers from rollout efficiency bottlenecks due to the long-tail response length distribution. While existing works mitigate the impact of long tails via prompt-level tail scheduling, we focus on the root source of inefficiency: the distribution itself. Specifically, we characterize the long-tail distribution at a finer granularity, identifying intra-prompt long tails, and revealing that they frequently consist of ineffective verbosity. To address this, we propose a novel paradigm of active distribution shaping to shape the rollout distribution towards conciseness and certainty, thereby fundamentally resolving tail-induced overheads. We achieve this through a distribution-aware trajectory sampling mechanism, which selects trajectories from a redundant exploration space for each prompt, and an adaptive redundancy allocation scheme to maximize both shaping effectiveness and system efficiency. Experiments demonstrate significant acceleration over state-of-the-art systems by up to 1.77x without compromising model performance.
[AI-48] COMPASS: Cognitive MCTS-Guided Process Alignment for Safe Search Agents
链接: https://arxiv.org/abs/2605.30838
作者: Wenkai Shen,Pengyang Zhou,Jiahe Xu,Jiaming Qian,Haozhe He,Zhihao Huang,Chaochao Chen,Xiaolin Zheng
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:LLM-powered search agents enable multi-step reasoning and tool use. However, these capabilities introduce retrieval-induced safety degradation, as harmful intents may decompose into seemingly innocuous sub-queries that lead to unsafe outcomes. Existing alignment methods struggle to capture sparse safety signals and fail to supervise diverse violations across multi-step interactions. We propose COMPASS, a Cognitive MCTS-Guided Process Alignment framework designed to achieve robust safety alignment throughout the agent workflow while preserving general utility. COMPASS integrates cognitive tree exploration (CTE) to efficiently synthesize stealthy attack trajectories, and introspective step-wise alignment (ISA) to isolate risky intermediate actions for fine-grained process supervision. Empirical results show that COMPASS achieves a favorable safety-utility trade-off while requiring substantially less training data.
[AI-49] Hide-and-Seek in Trajectories: Discovering Failure Signals for VLA Runtime Monitoring
链接: https://arxiv.org/abs/2605.30834
作者: Seongheon Park,Wendi Li,Changdae Oh,Samuel Yeh,Zsolt Kira,Michael Hagenow,Sharon Li
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注:
Abstract:Vision-Language-Action (VLA) models enable robots to follow natural language instructions and generalize across diverse tasks, but they remain vulnerable to execution failures that compromise reliability in real-world deployment. Detecting such failures during execution is therefore critical for the robust deployment of embodied systems. Existing failure detection methods either rely on expensive action resampling or external models, while alternatives propagate trajectory-level labels uniformly across every timestep, obscuring localized failure signals. In this paper, we propose \textbfHide-and-Seek, a framework that formulates VLA failure detection as a coarsely supervised learning problem. By combining inter-trajectory and intra-trajectory contrastive objectives, Hide-and-Seek localizes failure-indicative actions and induces temporally structured failure signals from trajectory-level supervision alone, without any step-level annotation. We evaluate Hide-and-Seek on LIBERO, VLABench, and a real-world robotic platform across three representative VLA policies: OpenVLA, \pi_0 , and \pi_0.5 .Our method achieves state-of-the-art multi-task failure detection performance with a practical accuracy–timeliness trade-off under conformal prediction, and generalizes well to both seen and unseen tasks.
[AI-50] SLAT: Segment-Level Adaptive Trimming for Efficient CoT Reasoning
链接: https://arxiv.org/abs/2605.30832
作者: Jian Yao,Xiongcai Luo,Ran Cheng,Kay Chen Tan
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Recent advances in Large Reasoning Models have significantly improved chain-of-thought (CoT) capabilities via reinforcement learning (RL). However, generated reasoning chains frequently suffer from structural redundancy (i.e., \emphoverthinking), incurring high computational overhead without improving answer correctness. Existing mitigation strategies typically rely on token-uniform length penalties, which provide coarse, segment-agnostic pressure toward shorter outputs and can inadvertently suppress useful reasoning alongside redundancy. To address this, we demonstrate that inefficiency concentrates in high-probability segments with low marginal utility. We derive a theoretical characterization of segment suboptimality under the correctness-length trade-off objective and propose \textscSLAT (Segment-Level Adaptive Trimming), an RL framework that selectively suppresses redundant segments based on this criterion. Empirical results on standard benchmarks indicate that \textscSLAT establishes a superior accuracy-efficiency Pareto frontier, reducing reasoning length by 50% relative to uncompressed baselines while maintaining competitive accuracy. Overall, our results suggest that theoretically grounded, segment-aware trimming is a promising direction for efficient CoT reasoning in large language models.
[AI-51] Unlearning in Diffusion Models: A Unified Framework with KL Divergence and Likelihood Constraints ICML2026
链接: https://arxiv.org/abs/2605.30825
作者: Shervin Khalafi,Alejandro Ribeiro,Dongsheng Ding
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Optimization and Control (math.OC); Machine Learning (stat.ML)
备注: 27 pages, 6 figures, 4 tables; Accepted by ICML 2026
Abstract:Unlearning in diffusion models aims to remove undesirable data or concepts while preserving the utility of pretrained models – two fundamentally conflicting objectives. We propose a principled constrained optimization framework that formulates unlearning as minimizing the deviation from a pretrained model, subject to explicit separation constraints from the unlearning distributions. Specifically, we formulate three constrained optimization problems based on reverse and forward KL divergences, and likelihood constraints. The first two generalize existing approaches for concept and data unlearning, while the third offers a novel and natural formulation for unlearning. Despite the nonconvexity of the KL constraints, we establish strong duality for all three problems, enabling us to explicitly characterize their optimal solutions as unlearning targets and develop primal-dual algorithms for each formulation. Experimental results demonstrate that our KL-constrained approach achieves superior retention-unlearning tradeoffs compared to weight-based baselines for concept and data unlearning, and that our likelihood-based approach matches unlearning effectiveness while better preserving retained concepts compared to baselines.
[AI-52] Planner-Centric Reinforcement Learning for Deep Research with Structure-Aware Reward
链接: https://arxiv.org/abs/2605.30824
作者: Mustafa Anis Hussain,Xinle Wu,Yao Lu
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Deep research tasks require LLMs to plan what to investigate, retrieve evidence, and synthesize long-form answers across multiple branches of inquiry. Existing training paradigms either rely on short-form verifiable QA as a proxy or optimize monolithic long trajectories, which makes planning and execution difficult to disentangle and yields weak credit assignment for the planning process. We propose DecomposeR, a planner-centric deep research framework that represents research plans as typed directed acyclic graphs (DAGs), allowing planning to be made explicit, structured, and rewardable. We train a Qwen3-8B model in two stages: planner reinforcement learning (RL) first learns graph structure and query decomposition to improve research planning, and answerer reinforcement learning (RL) then learns branch-level execution and final synthesis conditioned on the learned plan. By assigning rewards to explicit planner tokens and structured components rather than to a flat trajectory, DecomposeR enables finer-grained optimization of planning while reducing the ambiguity of end-to-end training. Experiments show that DecomposeR-8B improves over strong comparable open baselines by 5.1-8.0 points on popular long-form benchmarks due to improved planning and answering capabilities.
[AI-53] GaMi: Geometry-Agnostic Material Identification via Cross-Modal Subtractive Disentanglement
链接: https://arxiv.org/abs/2605.30818
作者: Zhiwei Chen(1),Yijie Li(2),Yimo Zhang(1),Shiyun Shao(1),Yichao Chen(3),Dian Ding(3),Liang Wang(4),Haiwei Wu(1),Liwei Guo(1),Jie Yang(1),Xiaosong Zhang(1),Yongzhao Zhang(1) ((1) UESTC, Chengdu, China, (2) National University of Singapore, Singapore, (3) Shanghai Jiao Tong University, Shanghai, China, (4) Northwestern Polytechnical University, Xi’an, China)
类目: Emerging Technologies (cs.ET); Artificial Intelligence (cs.AI); Sound (cs.SD)
备注: 17 pages, 18 figures
Abstract:Non-contact material identification enables adaptive interaction for embodied intelligence yet faces challenges from geometry-induced variations (e.g., orientation, shape, distance) and single-modality ambiguities. In this paper, we present GaMi, a multimodal material identification system integrating mmWave and acoustic sensing to robustly operate under unconstrained geometric conditions. By leveraging the insight of shared geometric consistency between co-located bimodal sensors, GaMi employs an intra-sample cross-modal subtractive disentanglement framework. By semantically aligning modalities and subtracting the shared geometric context, it isolates intrinsic material features. Furthermore, GaMi incorporates inter-sample contrastive learning to correct the residual interference caused by cross-modal misalignment. Additionally, a pairing-based adaptation strategy between two modalities enables few-shot generalization across devices. Extensive evaluations on 20 materials show that GaMi achieves 95.2% accuracy, outperforming single-modality baselines across unseen geometric conditions.
[AI-54] Differentially Private Preference Data Synthesis for Large Language Model Alignment ICML2026
链接: https://arxiv.org/abs/2605.30808
作者: Fengyu Gao,Jing Yang
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Accepted to ICML 2026
Abstract:Preference alignment is a crucial post-training step for large language models (LLMs) to ensure their outputs align with human values. However, post-training on real human preference data raises privacy concerns, as these datasets often contain sensitive user prompts and human judgments. To address this, we propose DPPrefSyn, a novel algorithm for generating differentially private (DP) synthetic preference data to enable privacy-preserving preference alignment. DPPrefSyn is a principled framework grounded in the Bradley-Terry preference model and the intrinsic geometric structure of pairwise human preference data. It first learns an underlying preference model from private data with formal differential privacy guarantees, and then leverages the learned model together with public prompts to synthesize high-quality preference data. It exploits the shared linear structure of per-cluster reward models to effectively capture heterogeneous human preferences in private datasets, and leverages DP Principal Component Analysis (DP-PCA) to improve learning accuracy. Extensive experimental results demonstrate that DPPrefSyn achieves competitive alignment performance under strong DP guarantees. These findings highlight the potential of synthetic preference data as a practical alternative for privacy-preserving preference alignment across a broad range of applications. To the best of our knowledge, this is the first work to generate DP synthetic preference data for LLM alignment. Our code is available at this https URL.
[AI-55] PReMISE: Policy Rubrics as Measurement Specifications for LLM Judges
链接: https://arxiv.org/abs/2605.30803
作者: Swastik Roy,Rajkumar Pujari,Tharindu Kumarage,Charith Peris,Rahul Gupta,Anna Rumshisky,Pradeep Natarajan,Venkatesh Saligrama
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:LLM judges are increasingly used to evaluate open-ended responses, but their scores depend strongly on the rubrics that condition them. A vague rubric asking for a response to be ``helpful and factual’’ can reward polished answers that invent facts or violate user intent. We treat reusable rubrics as measurement specifications: changing the rubric changes the response quality measurement induced by a fixed judge. We introduce PReMISE, a framework that, given pairwise human-preference data, (i) discovers a policy-level rubric set, and (ii) audits any rubric set under LLM-judge use along four axes: structural adequacy, reliability, preference fit, and adversarial robustness. Across rubric sources no raw source is simultaneously reliable, preference-predictive, and adversarially robust; and high inter-rater agreement does not imply low exploitability. PReMISE is the only rubric source to score non-trivially on applicability, specificity, and effective dimensionality simultaneously. We contribute two audit-targeted repair operations: preference-rank selection raises judge accuracy on paired responses from 65.0% to 68.6% , competitive with the strongest rubric-discovery baselines and leading on two of three judges in our cross-judge sweep; reliability-constrained refinement reduces the rate at which exploit responses receive high scores from 46.4% to 36.0% with little change in inter-judge agreement ( \alpha=.531\to.519 ).
[AI-56] Smaller Models are Natural Explorers for Policy-Level Diversity in GRPO
链接: https://arxiv.org/abs/2605.30789
作者: Yiming Ren,Yiran Xu,Zicheng Lin,Chufan Shi,Yukang Chen,Dingdong Wang,Tianhe Wu,Junjie Wang,Yujiu Yang,Yu Qiao,Ruihang Chu
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:We identify a new dimension for enhancing rollout diversity in Group Relative Policy Optimization (GRPO) for LLMs. While GRPO relies on diverse rollouts, prevailing strategies primarily increase diversity by injecting more token-level randomness, which may introduce step-wise noise and lead to incoherent trajectories. We uncover that smaller models within the same model family inherently exhibit higher policy-level diversity, indicated by their superior pass@k relative to larger counterparts as sample counts increase. Unlike token-level noise, this diversity is temporally correlated, preserves logical consistency, and provides structured exploration signals for gradient estimation. We thus propose S2L-PO (Small-to-Large Policy Optimization), a framework that leverages fixed small models as natural explorers to train larger models. To balance exploration and exploitation, we design a progressive annealing strategy that transitions from offline small-model rollouts to the large learner’s own sampling. This shift elegantly avoids mid-training performance drops caused by the small model’s capacity limits, achieving faster convergence and unlocking a higher performance ceiling. S2L-PO improves accuracy on diverse mathematical reasoning benchmarks (e.g., +8.8% on AIME 24 using a 1.7B explorer to guide the 8B model) while reducing rollout compute.
[AI-57] Learning Agent -Compatible Context Management for Long-Horizon Tasks
链接: https://arxiv.org/abs/2605.30785
作者: Lu Yi,Runlin Lei,Liuyi Yao,Yuexiang Xie,Yuyang Li,Wenhao Zhang,Zhewei Wei,Yaliang Li,Jian-Yun Nie
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:LLM agents increasingly face long-horizon tasks such as web search and deep research in real-world applications, where accumulated context can cause long-context degradation and reasoning failures. Prior work mitigates this through context management with agent-side context control or fixed strategies such as summarization, which require training the agent itself for adaptation - making it impractical for closed-source agents and ignoring that different agents may require different strategies. We introduce Adaptive Context Management (AdaCoM), which trains an external LLM to manage the context of a frozen agent through flexible modification actions and end-to-end reinforcement learning. Across diverse agents on web search and deep research benchmarks, AdaCoM substantially improves performance by preserving task constraints and progress while pruning stale content. The learned strategies reveal a Fidelity-Reliability Trade-off: agents with higher vanilla ReAct performance benefit from higher-fidelity context preservation, whereas lower-performing agents require more aggressive compression to stay within a reliable reasoning regime. Transfer experiments show that AdaCoM generalizes most effectively across agents with similar capability (measured by vanilla ReAct performance), suggesting a practical path toward reusable context managers for agent systems.
[AI-58] Chatterbox-Flash: Prior-Calibrated Block Diffusion for Streaming Zero-Shot TTS
链接: https://arxiv.org/abs/2605.30748
作者: Deokjin Seo,Gangin Park,Kihyun Nam
类目: ound (cs.SD); Artificial Intelligence (cs.AI); Audio and Speech Processing (eess.AS)
备注: 8 pages, 4 figures, 9 tables
Abstract:We present Chatterbox-Flash, a zero-shot text-to-speech model obtained by fine-tuning a pretrained autoregressive TTS decoder into a block-diffusion decoder, enabling parallel token generation within each block while retaining block-by-block streaming. We find that naively transferring mainstream block-diffusion decoding to discrete speech tokens degrades quality, as a long-tail token distribution biases parallel position selection toward a few high-frequency tokens. To mitigate this without architectural modification, we introduce two inference-time techniques: prior-calibrated scoring, which subtracts the block-level marginal token distribution, and an early-decoding schedule, which adaptively terminates iteration based on calibrated confidence. On standard zero-shot TTS benchmarks, Chatterbox-Flash attains high-fidelity synthesis comparable to strong autoregressive and non-autoregressive baselines, while supporting streaming inference with time-to-first-packet on par with streaming AR systems and substantially lower real-time factor. Code and audio samples are available at this https URL.
[AI-59] Generating Graph-like Rules for Knowledge Graph Reasoning via Diffusion Models KDD26
链接: https://arxiv.org/abs/2605.30747
作者: Haoxiang Cheng,Yunfei Wang,Chao Chen,Kewei Cheng,Zhipeng Lin,Haoxuan Li,Changjun Fan,Shixuan Liu
类目: Artificial Intelligence (cs.AI)
备注: accepted by KDD 26
Abstract:Logical rules constitute a cornerstone of knowledge graph (KG) reasoning, valued for their interpretability and ability to model relational patterns. However, existing rule mining methods predominantly focus on simple chain-like rules and therefore neglect the richer relational information encoded in graph-like structures, such as cycles and branches. This limitation is further exacerbated by computational bottlenecks caused by the combinatorial explosion of the search space, which is especially challenging for graph-like rules. Meanwhile, generative approaches such as diffusion models, despite their success in other domains, can not be directly applied to rule mining because their training objectives are not aligned with the goal of learning high-quality rules, and non-differentiable KG rule quality metrics cannot directly guide model optimization. To address these limitations, we propose GRiD, a framework that reformulates graph-like rule discovery as a discrete generative process conditioned on the target relation. GRiD employs a two-phase training strategy. First, supervised pre-training enables GRiD to capture structural priors from subgraphs sampled from the KG meta-graph. Subsequently, reinforcement learning is applied to fine-tune GRiD through policy gradient optimization guided directly by non-differentiable rule-quality metrics. Experiments on six benchmark datasets show that GRiD achieves competitive performance on KG completion tasks. Ablation studies confirm the efficiency and robustness of GRiD and further show that graph-like rules complement chain-like rules in KG completion. Our codes and datasets are available in this https URL
[AI-60] GSAM: A Generalizable and Safe Robotic Framework for Articulated Object Manipulation PPSN2026
链接: https://arxiv.org/abs/2605.30740
作者: Beichen Shao,Mengying Xie,Heng Su,Wanyi Zhang,Mingyan Li,Yan Ding,Fausto Giunchiglia,Chao Chen
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注: Accepted by the 19th International Conference on Parallel Problem Solving from Nature (PPSN 2026)
Abstract:Articulated object manipulation is a unique challenge for service robots. Existing methods employ end-to-end policy learning, visionmotion planning, and large-language/visual-language model (LLM/VLM), but often overlook the diversity of articulated objects and the complexity of interactions between end-effector and handle, leading to limited generalization and destructive collisions. To address this, we propose GSAM, a generalizable and safe robotic framework for articulated object manipulation. Specifically, a vision-based perceiver generates the kinematic parameters. Considering that pre-trained markers in perceiver yield raw estimations that may deviate from commonsense, we present a f ine-tuned VLM-based refiner, using chain-of-thought (COT) commonsense reasoning to refine perception. To prevent destructive collisions, we design an interaction constraint function generator, integrating articulated object, interaction pose, and obstacle avoidance knowledge into a base. LLM then functionalize these constraints and apply them to trajectory and posture planning. A kinematic-aware manipulation planner verifies reachability for trajectory and posture. Experiments on 50 hinge tasks across 5 object categories and 50 randomly initialized end-effectorhandle configurations show that GSAM reduces standard deviation by 3.1% and improves manipulation success rate by 36.0% compared to the best baseline, respectively demonstrating the superior object generalization and interaction safety of GSAM in practical scenarios.
[AI-61] MAVEN: Improving Generalization in Agent ic Tool Calling
链接: https://arxiv.org/abs/2605.30738
作者: Omkar Ghugarkar,Vishvesh Bhat,Muhammad Ahmed Mohsin,Asad Aali
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Generalization across agentic tool-calling environments remains a central challenge for reliable agentic reasoning systems. Although large language models achieve strong results on individual benchmarks, their ability to compose reasoning strategies, preserve intermediate states, and coordinate tools across domains remains underexplored. We present MAVEN (Modular Agentic Verification and Execution Network), a lightweight symbolic reasoning scaffold for structured decomposition, adaptive tool orchestration, and intermediate verification. We evaluate MAVEN across established tool-calling benchmarks, including BFCL v3, TauBench, Tau2Bench, AceBench, and introduce MAVEN-Bench, a stress-test benchmark for multi-step mathematical and physical reasoning with explicit verification and adversarial task composition. MAVEN-Bench exposes a substantial gap between partial reasoning quality and end-to-end task success; in direct MAVEN-Bench runs, MAVEN improves its GPT-OSS-120b base model from 48% to 71% accuracy without additional training. It also remains competitive with frontier proprietary baselines while using an open-weight backbone with an estimated cost ratio of roughly 1/10, suggesting that lightweight verification-centered scaffolds can strengthen compositional reasoning and motivate more process-aware evaluation of agents in the wild.
[AI-62] Kalimati Vegetable Price Index Forecasting with a Momentum Corrected Online Stacking Ensemble
链接: https://arxiv.org/abs/2605.30720
作者: Sahaj Raj Malla
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); General Economics (econ.GN); Machine Learning (stat.ML)
备注: 21 pages, 8 figures, 2 tables
Abstract:Forecasting agricultural commodity prices in emerging economies is difficult due to high volatility, frequent supply disruptions, and strong cultural influences on demand. This study introduces the Kalimati Vegetable Price Index (KVPI), a new inverse-volatility weighted composite index that aggregates 135 daily wholesale commodities from Kathmandu over ten years (2013-2023). By creating a stable macro-level signal, the KVPI reduces the noise inherent in modelling individual crops. A rich set of 64 causally valid features was developed, including festival lead-lag effects, rolling statistics, and calendar variables. Fourteen forecasting models spanning statistical, tree-based, deep learning, hybrid, and transformer architectures were rigorously evaluated across short (7-day), medium (14- and 30-day), and long-term (90-day) horizons. Tree-based ensembles proved notably robust, while classical statistical models and complex transformers struggled with the noisy dataset. The proposed Momentum-Corrected Online Stacking Ensemble achieved the strongest performance, yielding a Root Mean Square Error (RMSE) of 1.771, an exceptionally low Mean Absolute Percentage Error (MAPE) of 0.68%, and explaining 84.5% of the variance (R-squared = 0.845) at the 90-day horizon. This open-source pipeline provides policymakers and supply chain actors in Nepal and similar markets with a practical, reliable tool for anticipating price movements and strengthening food security.
[AI-63] When are LLM s Sufficient Policy Optimizers for Sequential RL Tasks?
链接: https://arxiv.org/abs/2605.30719
作者: Stephane Hatgis-Kessell,Emma Brunskill
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:We study when large language models (LLMs) can serve as effective black-box policy optimizers for reinforcement learning (RL) tasks, i.e., when can we replace classical RL algorithms with an LLM? We explore this question by introducing Prompted Policy Optimization (PromptPO), an iterative method that prompts an LLM with Python descriptions of the state space, action space, and reward function, then has it generate and refine executable policies based on rollout feedback. Across hard exploration environments, Meta-World robotics tasks, and several real-world control problems, PromptPO often matches or exceeds the performance of standard RL baselines while using substantially fewer environment interactions. To maximize expected return, and without further explicit prompting, the policies PromptPO outputs range from tuned proportional controllers or rule-based plans to policies that run planning algorithms like value iteration. Our results demonstrate that LLM-based policy optimization is sufficient when the LLM can leverage prior knowledge about the environment or optimization strategy. PromptPO underperforms standard RL baselines in MuJoCo domains. This demonstrates possible limitations of LLM-based policy optimization to settings that requiring fine-grained continuous control.
[AI-64] Depth-Dependent Indirect Prompt Injection in Tool-Calling ReAct Agents : Injection Depth Payload Framing and Turn-Budget Sensitivity
链接: https://arxiv.org/abs/2605.30686
作者: Mohammadreza Rashidi
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 17 pages, 16 figures
Abstract:ReAct agents that interleave chain-of-thought reasoning with tool calls are increasingly deployed for real tasks such as scheduling, file retrieval, and data access. Their tool observation loop creates a direct attack surface: an adversary who controls any tool’s return value can embed instructions that redirect the agent away from the user’s goal, a threat known as indirect prompt injection. Existing benchmarks evaluate attack success rate (ASR) at a fixed injection position under fixed conditions, leaving three risk dimensions unexplored: where in the tool sequence the payload appears (injection depth), what rhetorical register it uses (framing), and how many turns the agent is permitted (turn cap). We conduct four controlled studies on 20 scenarios spanning five attack categories, totalling 460 trials against GPT-4o-mini and Claude Haiku at a combined API cost under 0.36 USD. Study 1 shows that ASR against GPT-4o-mini decays from 60% at depth 1 to 0% at depths 4 and 5 (Cramer’s V = 0.58, p 0.001; restricted to within-sequence depths 1-3: V = 0.47, p = 0.0013), driven by model resistance at depth 1 and task completion before payload encounter at deeper positions. Study 2 replicates the depth experiment on Claude Haiku, which achieves 0% ASR at every depth through a combination of conservative tool invocation and genuine instruction resistance. Study 3 shows that framing modulates ASR between 25% (neutral) and 75% (persona) at depth 1, a 50-percentage-point range that does not reach statistical significance at N = 20 per condition. Study 4 confirms that ASR is stable across turn caps of 3, 5, and 7, indicating the turn budget is not a risk factor in this setting. Our results establish injection depth as the dominant variable and show that sanitising only the first tool observation captures 67% of measured injection successes.
[AI-65] Investigating Detection and Obfuscation of Prompt Injection Attacks Against Software Reverse Engineering AI Agents
链接: https://arxiv.org/abs/2605.30677
作者: Brian Crawford,Patrick McClure
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Software Engineering (cs.SE)
备注:
Abstract:Agentic software reverse engineering systems are vulnerable to prompt injection attacks placed into the source code of executable binary files. This research demonstrates defensive tactics for detecting the presences of prompt injection strings in the decompiler output of adversarial example programs. Methods for obfuscating these attacks and subsequent methods for defending against these obfuscations are also explored. This research advances the understanding of risk and security of agentic software analysis systems necessary for their deployment into production-level cyber workflows.
[AI-66] Automatically Attacking Software Reverse Engineering AI Agents
链接: https://arxiv.org/abs/2605.30667
作者: Brian Crawford,Justin Phillips,Patrick McClure
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注:
Abstract:Software tools for reverse engineering executable binary files, such as Ghidra, enable malware analysts to safely conduct robust static analysis without having access to original source code. Coupled with the analytic power of large language models (LLM), agentic systems enabled with tools, such as GhidraMCP, can allow analysts to automate a previously human driven process. Although this automation can increase the productivity of a single malware analyst, it also introduces a new area of vulnerability for malware obfuscation. This paper presents an adversarial technique using genetic algorithm-based prompt generation, a modification of an adversarial attack known as AutoDAN, to demonstrate the ability to deceive LLM-powered disassembly and decompilation systems into misinterpreting binary executables, effectively corrupting their analytical output. This proof-of-concept methodology exploits inherent vulnerabilities in how LLMs process and interpret decompiled machine code via prompt injection by using extraneous string variable assignments to pass surreptitious instructions to the LLM while not impacting the functionality of the executable file. We demonstrate this capability through several concise examples. This approach could enable attackers to bypass automated detection systems that rely on LLM-driven analysis pipelines. By studying and understanding this attack, insights can be gained regarding the security implication of integrating LLMs into cybersecurity toolchains and building more robust agentic code analysis systems.
[AI-67] Structure-Induced Information for Rerooting Levin Tree Search ICML2026
链接: https://arxiv.org/abs/2605.30664
作者: Jake Tuero,Michael Buro,Laurent Orseau,Levi H. S. Lelis
类目: Artificial Intelligence (cs.AI)
备注: ICML 2026
Abstract:Subgoal-based policy tree search, which uses a policy to guide search, is effective for complex single-agent deterministic problems but often relies on explicit subgoal generation that can incur substantial overhead and hinders scalability. In this paper, we overcome these limitations by using a learned ``rerooter’’ through the recently-introduced \sqrt\textLTS algorithm. A rerooter implicitly decomposes the problem into soft subtasks. While previous work focused on the formal guarantees for given or handcrafted rerooters, in this work we propose three rerooter designs: (i) a clustering-based rerooter that exploits global state-space structure, (ii) a heuristic-based rerooter that leverages learned cost-to-go estimates, and (iii) a hybrid that combines both signals. Our framework avoids having to explicitly reconstruct and reason over generated subgoals, thereby enabling scalable allocation of search effort with significantly lower computational overhead. Empirically, our rerooting-based methods scale to complex environments where subgoal-based policy tree search fails, and achieve state-of-the-art online training efficiency on the domains tested.
[AI-68] LARK: Learnability-Grounded Trajectory Selection for Efficient Reasoning Distillation
链接: https://arxiv.org/abs/2605.30651
作者: Tianrun Yu,Kaixiang Zhao,Chih-Chun Chen,Amanda Hughes,Taylor W. Killian,Fenglong Ma,Weitong Zhang,Porter Jenkins
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 43 pages, 9 figures, 2 tables
Abstract:We study trajectory selection for reasoning distillation, where teacher-generated reasoning trajectories are selectively used as supervision for a student model. Existing methods rely on heuristics such as trajectory quality or model confidence, but they often overlook whether a trajectory is learnable by the student. In this paper, we present LARK, a learnability-grounded method for reasoning trajectory selection. LARK selects trajectories that the student can learn efficiently while preserving the generalization of the full training distribution. At the core of LARK is a learnability factor \rho , which characterizes the rate at which the student’s training loss decreases. To estimate this rate efficiently and maintain generalization, we introduce a learnability proxy and a \chi^2 -regularized selection policy that balances learnability and distributional coverage, both with strong theoretical guarantees on their estimation error. Empirically, LARK consistently outperforms data selection baselines across multiple base models and reasoning tasks. Diagnostic analyses show that the LARK score predicts downstream training utility and that LARK-selected trajectories induce faster supervised fine-tuning loss reduction. Our code is available at this https URL.
[AI-69] Score Broadcast and Decorrelation: A General Framework for Broadcast-Based Credit Assignment
链接: https://arxiv.org/abs/2605.30638
作者: Mustafa Uzun,Mete Erdogan,Cengiz Pehlevan,Alper T. Erdogan
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:We introduce Score Broadcast and Decorrelation (SBD), a principled framework for broadcast-based credit assignment for general families of differentiable losses. Error broadcast is a biologically plausible alternative to backpropagation that sends output information to hidden layers without weight transport. The Error Broadcast and Decorrelation (EBD) framework, recently introduced for the mean-squared-error (MSE) setting, grounded this mechanism in the stochastic orthogonality of optimal estimators, under which the optimal residual is orthogonal to functions of the input. We generalize that foundation by introducing an orthogonality principle between the output score (the gradient of loss with respect to the final-layer output) and hidden-layer activations, which holds whenever the optimal score has conditional mean zero. This single principle unifies broadcast-based credit assignment across the standard differentiable-loss families, including cross-entropy, Bregman divergences, proper scoring rules, and exponential-family negative log-likelihoods. The framework supplies a theoretical grounding for the three-factor learning rule under general losses, with the neuromodulatory factor derived as the broadcast loss score. We derive the cross-entropy case explicitly, characterize the admissible loss class, and introduce a score vector expansion technique that enriches the broadcast signal while preserving the orthogonality framework. Experiments on CIFAR-10 and Tiny ImageNet show that SBD substantially improves over existing broadcast approaches, with score vector expansion delivering further gains. Overall, this work identifies the loss score as the signal to broadcast, supplies the orthogonality theory and theoretical grounding for the three-factor learning rule from neuroscience, and shows how score vector expansion enriches the decorrelation directions of the resulting objective.
[AI-70] EHRBench: An Automated and Reliable EHR-based Benchmark for Clinical Decision Making with LLM s KDD2026 KDD
链接: https://arxiv.org/abs/2605.30637
作者: Yuzhang Xie,Keqi Han,Yunpeng Xiao,Hejie Cui,Guanchen Wu,Ziyang Zhang,Kai Shu,Jiaying Lu,Xiao Hu,Carl Yang
类目: Artificial Intelligence (cs.AI)
备注: Proceedings of the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD 2026), Datasets and Benchmarks Track, Oral
Abstract:Clinical decision-making (CDM) is central to real-world clinical workflows, where clinicians infer diagnoses, select treatments, or anticipate future health outcomes under incomplete evidence. LLMs are increasingly used to support these decisions due to strong language capabilities, broad biomedical knowledge, and efficiency, yet the reliability of LLMs on real-world clinical decision tasks remains insufficiently understood. To evaluate CDM models, especially LLM-based models, an ideal and practical medical decision benchmark should be constructed via an automated yet reliable pipeline to ensure both scale and quality. Moreover, the grounding of a CDM benchmark in real patient EHRs can better support evaluation on practical CDM tasks that require substantive biomedical knowledge and clinical inference. To fill the gaps, we introduce EHRBench, an automated and reliable EHR-grounded benchmark for evaluating LLM-based clinical decision-making at scale. To ensure scalability and reliability, EHRBench is constructed through an EHR-LLM-KB(knowledge-base) interaction pipeline. For efficiency, we use a specialized LLM to automatically convert encounter-level EHR trajectories into structured templates and deterministically instantiate the templates into QA items. In parallel, we apply systematic KB-based verification and enrichment to filter hallucinated or ambiguous relations and to improve reliability. Using this pipeline, we construct nearly 1M (960,067) QA items spanning three core inference-required clinical decision tasks: diagnosis, treatment, and prognosis. We benchmark more than 30 representative LLMs on EHRBench and provide detailed analyses of performance and robustness. The results show consistent capability trends across settings, further validating the reliability of EHRBench and highlighting actionable gaps toward clinically reliable LLM systems.
[AI-71] Active Timepoint Selection for Learning Measure-Valued Trajectories ICML2026
链接: https://arxiv.org/abs/2605.30625
作者: Nicolas Huynh,Mihaela van der Schaar
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Machine Learning (stat.ML)
备注: ICML 2026
Abstract:Inferring continuous probability paths from sparse snapshots is a fundamental challenge in domains like single-cell biology, where high-fidelity data acquisition is often destructive and constrained by prohibitive sequencing costs. This motivates the need for active learning strategies to strategically select optimal measurement times. However, designing active learning policies for this setting remains an open problem: the target objects reside on the infinite dimensional Wasserstein space where standard Euclidean metrics are ill-defined, and current interpolation methods lack epistemic uncertainty quantification. We introduce a framework which extends active experimentation to the space of measures. By leveraging Linearized Optimal Transport (LOT), we map distributional snapshots into a tangent space amenable to Gaussian Process modeling, allowing us to construct a tractable probabilistic surrogate for the underlying probability path. This yields an acquisition policy that iteratively selects measurement times to minimize uncertainty. Empirical results demonstrate that our strategy outperforms uncertainty-agnostic baselines on both synthetic and real-world datasets.
[AI-72] Harness Updating Is Not Harness Benefit: Disentangling Evolution Capabilities in Self-Evolving LLM Agents
链接: https://arxiv.org/abs/2605.30621
作者: Minhua Lin,Juncheng Wu,Zijun Wang,Zhan Shi,Yisi Sang,Bing He,Zewen Liu,Tianxin Wei,Zongyu Wu,Zhiwei Zhang,Dakuo Wang,Xiang Zhang,Benoit Dumoulin,Cihang Xie,Yuyin Zhou,Suhang Wang,Hanqing Lu
类目: Artificial Intelligence (cs.AI)
备注: 24 pages, 9 figures, 12 tables
Abstract:LLM agents are increasingly deployed as systems built around editable external harnesses, including prompts, skills, memories and tools, that shape task execution without changing model parameters. Harness self-evolution adapts such agents by updating these harnesses from execution evidence. Yet it remains unclear whether a model’s base capability in task-solving predicts its capabilities in harness self-evolution: which models produce useful harness updates, and which actually benefit from them? We analyze two harness self-evolution capabilities: (i) harness-updating, the capability to produce useful persistent harness updates from execution evidence; (ii) harness-benefit, the capability to benefit from updated harnesses during task solving. Our analysis reveals two findings. First, harness-updating is flat in base capability: models from different capability tiers produce harness updates that lead to surprisingly similar gains; even Qwen3.5-9B’s updates yield gains comparable to those of Claude Opus~4.6. Second, harness-benefit is non-monotonic in base capability: weak-tier models benefit little from updated harnesses, mid-tier models benefit most, and strong-tier models benefit less than mid-tier. We trace low gains at the weak tier to two failure modes: weak-tier models may fail to activate relevant harness artifacts, or activate them but fail to follow them faithfully. These findings suggest investing capability budget in the task-solving agent rather than the evolver, and targeting harness invocation and long-horizon instruction following in agent training. Our source code is publicly available at this https URL.
[AI-73] Scientific Machine Learning for Engine Health Management and Remaining Useful Life Prediction
链接: https://arxiv.org/abs/2605.30593
作者: Jostein Barry-Straume,Changmin Son,Adrian Sandu,Gavan Burke,Rekha Sundararajan,Andrew Rimell,James G. Steinrock
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computational Engineering, Finance, and Science (cs.CE)
备注:
Abstract:Engine Health Management (EHM) depends on reliable forecasting of Remaining Useful Life (RUL) and on tracking thermal indicators such as turbine gas temperature (TGT). In practice, real-world fleet data are heterogeneous and non-stationary, and point predictions alone are insufficient for risk-aware maintenance decisions. This paper presents a multi-task scientific machine learning framework for turbine prognostics that jointly predicts turbine gas temperature untrimmed (TGTU), Delta Turbine Gas Temperature (DTGT), and RUL, with quantified uncertainty in the form of prediction intervals whose empirical coverage is evaluated. A shared sequence encoder (convolutional front-end with residual bidirectional LSTM layers and attention pooling) feeds task-specific heads, including mean–variance estimation for probabilistic regression and, optionally, a survival head for threshold-based event modeling. The framework is designed to be tunable via a small set of practitioner-facing parameters (e.g., DTGT thresholding rules and RUL target construction) so that deployment can align with in-house policies and proprietary criteria. The predictive performance of the proposed framework is evaluated using both point and interval metrics, including mean absolute error (MAE), prediction interval coverage probability (PICP), mean prediction interval width (MPIW), and the coverage–width criterion (CWC). Results are reported both in aggregate and stratified by flight phase and maintenance segment to highlight operational-context effects and to support uncertainty-aware monitoring.
[AI-74] Benchmarking Machine Learning Uncertainty Quantification Methodologies for Predicting Turbine Gas Temperature Degradation
链接: https://arxiv.org/abs/2605.30585
作者: Jostein Barry-Straume,Changmin Son,Adrian Sandu,Gavan Burke,Rekha Sundararajan,Andrew Rimell,James G. Steinrock
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computational Engineering, Finance, and Science (cs.CE)
备注:
Abstract:Effective prognostics and health management of modern engines relies on accurate turbine gas temperature predictions and robust uncertainty quantification to ensure reliability and safety. This paper investigates five major approaches for constructing prediction intervals – namely the Delta method, Bayesian Monte Carlo Dropout, Bootstrap method, Lower-Upper Bound Estimation, and Mean-Variance Estimation – as a means of capturing the uncertainty in neural network predictions of turbine gas temperature. Each approach is implemented within a unified experimental framework that employs cross-validation for hyperparameter selection, repeated train-test splits for performance robustness, and multiple metrics to evaluate both the accuracy and tightness of the intervals. In particular, Coverage Probability, Normalized Mean Prediction Interval Width, and the Coverage Width-based Criterion are measured to comprehensively assess each method’s reliability and sharpness. Experiments conducted on a representative turbine gas temperature dataset reveal distinct trade-offs among the five methods in terms of interval coverage, width, and stability. These findings provide a practical guide for selecting and tuning prediction interval methods in engine health management and prognostics, ensuring both interpretability and precision in real-world applications.
[AI-75] Uncertainty-Aware and Temporally Regulated Expert Advice in Reinforcement Learning for Autonomous Driving ITSC
链接: https://arxiv.org/abs/2605.30576
作者: Ahmed Abouelazm,Felix Klingebiel,Philip Schörner,J. Marius Zöllner
类目: Artificial Intelligence (cs.AI)
备注: Accepted in The IEEE International Conference on Intelligent Transportation Systems (ITSC) September 15-18, 2026 – Naples, Italy
Abstract:Exploration in reinforcement learning for autonomous driving is inherently unsafe: agents must experience novel behaviors to learn, yet exploration can lead to collisions or off-road driving. We propose an uncertainty-aware framework that leverages expert advice to guide exploration while avoiding long-term dependence. Advice is triggered when epistemic or aleatoric uncertainty exceeds adaptive thresholds derived from rolling buffers, ensuring advice evolves with the agent’s confidence. A commitment-cooldown strategy with a stochastic early-stop heuristic regulates the duration and frequency of guidance, exposing the agent to coherent maneuvers without exhausting the advice budget. Expert and agent experiences are combined in a shared replay buffer within an off-policy implicit quantile network (IQN) backbone, enabling efficient reuse of expert trajectories. Experiments in CARLA show that our method outperforms the IQN baseline, improving success by 5-7% and reducing failures, demonstrating that risk-sensitive uncertainty coupled with regulated expert integration enables safer and more efficient exploration for sensor-based RL policy learning in unsignalized intersection navigation.
[AI-76] Memory-Bound but Not Bandwidth-Limited: The Physical AI Inference Gap in Batch-1 LLM Decode
链接: https://arxiv.org/abs/2605.30571
作者: Josef Chen
类目: Hardware Architecture (cs.AR); Artificial Intelligence (cs.AI); Distributed, Parallel, and Cluster Computing (cs.DC); Performance (cs.PF); Robotics (cs.RO)
备注:
Abstract:Physical AI systems, including robots, autonomous vehicles, embodied agents and edge copilots, often run a different inference workload from cloud LLM serving: single-stream, batch-1 autoregressive decode, where one robot, camera feed or user session waits on the next token. This workload is usually described as memory-bandwidth-bound. Each decode step streams model weights and the active KV cache, so latency should scale with peak HBM bandwidth. We show that this account is true but incomplete. We measure batch-1 decode for three 7 to 8B-class GQA transformers across four NVIDIA GPUs: H100 SXM5, A100-80GB SXM4, L40S and L4. We evaluate context lengths from 2048 to 16384, producing 44 valid cells under a controlled bf16 SDPA setup. The achieved fraction of peak HBM bandwidth falls as peak bandwidth rises. On the headline Qwen-2.5-7B ctx=2048 cell, an L4 reaches roughly 81 percent of its analytic memory floor, while an H100 reaches only 27 percent. Physical-AI decode is memory-dominated, but faster memory does not translate into proportional latency gains. We test the missing term with a CUDA Graphs A/B experiment. On H100 at ctx=2048, CUDA Graphs improves decode latency by 1.259x across N=10 fresh sessions, with a 95 percent bootstrap confidence interval of 1.253 to 1.267. On L4, the same intervention gives only 1.028x. This isolates a launch-side overhead that becomes visible on fast GPUs but remains mostly hidden on slower, bandwidth-bound GPUs. The deployment implication is that memory savings matter only when the runtime realises them. On L4, bf16 decode sits close to the memory floor, but common quantised paths do not recover the expected 4x weight-traffic reduction: bnb-nf4 reaches 59.36 ms/step and AutoAWQ+Marlin reaches 45.24 ms/step from a 62.32 ms bf16 baseline. GPTQ+ExLlamaV2, with Ada-tuned int4 kernels, reaches 17.36 ms/step.
[AI-77] Procedural Generation of First Person Shooter Maps using Map-Elites
链接: https://arxiv.org/abs/2605.30570
作者: Simone de Donato,Pier Luca Lanzi,Daniele Loiacono
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:We investigate the application of MAP-Elites (a well-known quality diversity algorithm) to design levels for First-Person Shooter (FPS) games. We consider two well-known map representations (All-Black and Grid-Graph) and introduce two novel representations (Point-Line and Spatial-Layout) that improve the characterization of FPS maps. We define a series of metrics to describe maps’ topological properties (which solely depend on maps’ layout), and emergent properties (which must be evaluated through actual gameplay). We perform an in-depth analysis to identify the most suitable features to guide MAP-Elites illumination process. We apply MAP-Elites with Sliding Boundaries (MESB) to evolve populations of FPS maps. Our results show that the new representations can generate maps with higher diversity and quality than the representations previously used for evolving FPS maps.
[AI-78] ransforming and Encoding FTS for SAT Solving: What Helps What Hurts (Extended Version)
链接: https://arxiv.org/abs/2605.30563
作者: João Filipe,Álvaro Torralba,Gregor Behnke
类目: Artificial Intelligence (cs.AI)
备注:
Abstract:Factored tasks are a classical planning representation that extends SAS+ with limited forms of disjunctive preconditions, conditional effects, and angelic nondeterminism. This allows for a more compact representation of tasks than traditional formalisms such as STRIPS or SAS+, and supports a wide range of task transformations. However, existing planning approaches for factored tasks have been limited to heuristic search methods. In this work, we investigate how to encode factored tasks in SAT. We propose several ways to encode the tasks, focusing on different strategies for translating the factored transition relation into propositional logic. We also analyze how to exploit parallelism at various levels in this setting and study the impact of common task transformations on the performance of SAT-based planners. Subjects: Artificial Intelligence (cs.AI) Cite as: arXiv:2605.30563 [cs.AI] (or arXiv:2605.30563v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2605.30563 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[AI-79] Physically Viable World Models: A Case for Query-Conditioned Embodied AI
链接: https://arxiv.org/abs/2605.30542
作者: Adam J. Thorpe,Stepan Tretiakov,Cheng-Hsi Hsiao,Su Ann Low,Xingjian Li,Hassan Iqbal,Neel P. Bhatt,Ufuk Topcu,Krishna Kumar
类目: Artificial Intelligence (cs.AI)
备注: 21 pages; Adam J. Thorpe and Stepan Tretiakov contributed equally
Abstract:World models for embodied AI must be physically viable: constructed to answer intervention queries by representing the physical structure governing action outcomes, rather than merely predicting future observations. Existing observation-predictive world models can produce visually plausible but physically wrong rollouts. This failure is structural; distinct physical systems can look identical yet diverge under intervention. We expose this problem with controlled benchmarks that fix the visible scene while varying latent physics. We show that such models may recommend infeasible actions, mispredict interaction outcomes, or certify unsafe behavior. We argue that embodied AI requires world models that identify the simplest physical abstraction sufficient to answer an intervention query. Such a model comprises modular components, including environment representation, latent state and parameter estimation, action specification, interventional dynamics, and query-level response. An autonomous orchestrator should identify the relevant abstraction and compose compatible learned and structured components per query. When closed-form physics is unavailable, uncertain, or costly, the transition model may be analytic, simulated, learned, or hybrid, but it must preserve the structure that determines interventional outcomes. This decomposition makes the model interpretable, its components verifiable, and its outputs auditable against the query. It also provides a design principle for new world models and a feasibility test for existing ones: the right abstraction is not the most detailed model of the world, but the simplest model that preserves the distinctions relevant to the query. We demonstrate this approach on queries that existing systems fail to answer correctly, and outline how an orchestrator can dynamically assemble and adapt physically viable models for planning, control, and verification.
[AI-80] Graph-Conditioned Mixture of Graph Neural Network Experts for Traffic Forecasting MDM2026
链接: https://arxiv.org/abs/2605.30486
作者: Amirhossein Ghaffari,Saeid Sheikhi,Ekaterina Gilman
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: An accepted paper at the 27th IEEE International Conference on Mobile Data Management (MDM 2026)
Abstract:Spatio-temporal forecasting on sensor graphs is commonly tackled with a single backbone architecture applied uniformly across all nodes, although graph regions can exhibit different dynamics. Road segments differ in functional class, structure, and traffic behavior, suggesting that node-wise expert specialization can be useful. We propose GC-MoE, a graph-conditioned mixture of experts framework that assigns each node a personalized combination of frozen forecasting experts based on graph topology and the recent traffic input window. GC-MoE combines frozen pretrained spatio-temporal GNN experts with an input-aware, spatially contextualized router while training only a lightweight routing module. We also study a bounded graph-conditioned output refinement layer as an optional extension and include node-adaptive ST-LoRA adapters only as an ablation diagnostic. Across four standard benchmarks (PEMS04, PEMS07, METR-LA, and PEMS-BAY), GC-MoE improves MAE over a zero-parameter ensemble baseline, with competitive RMSE and MAPE, while training only ~17K parameters on top of 1.5M frozen expert weights. The implementation is available at this https URL.
[AI-81] dSCD: Identifying Training Datasets through Semantic Correlation Descriptors
链接: https://arxiv.org/abs/2605.30462
作者: Andrada Gobeaja,Ionut Hodoroaga,Elena Burceanu,Marius Leordeanu
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 16 pages, 3 figures
Abstract:Can a dataset be recognized from the spurious correlations it induces during training? We argue that datasets leave dataset-specific traces in a model’s learned semantic correlation structure: incidental regularities that are predictive within a dataset, but not causal for the underlying task, can be internalized during training. We use this insight to study dataset-level membership inference, moving beyond existing methods that rely on behavioral or distributional evidence such as confidence scores, losses, margins, generated samples, or query responses. We introduce a white-box semantic fingerprinting approach based on semantic correlation descriptors (SCDs), which capture the semantic correlation structure learned by a model and make it comparable across dataset mixtures. In a controlled leave-one-dataset-out diagnostic, SCDs recover dataset-specific changes and perfectly separate matching from non-matching dataset pairs. We then propose a practical SCD-based membership score that tests whether a target dataset is part of a model’s training mixture using only the model’s SCD and the target dataset’s standalone SCD, without requiring leave-one-dataset-out models. Across three diverse experimental settings, with dataset groups for natural language inference, emotion classification, and medical text classification, we test both the advantages and limitations of SCD-based membership inference with different degrees of semantic separation and keyword support between dataset splits. On average, the classifier based on this score achieves the highest performance and the lowest std, outperforming black-box baselines RMIA, Attack-P, and LiRA, as well as the white-box SIF baseline. These results show that dataset membership can be traced through internal semantic correlations, with the largest relative gain exceeding 60% in ROC-AUC when dataset groups expose distinct semantic particularities.
[AI-82] Scalable Constrained Multi-Agent Reinforcement Learning via State Augmentation and Consensus for Separable Dynamics
链接: https://arxiv.org/abs/2605.30461
作者: Santiago Amaya-Corredor,Miguel Calvo-Fullana,Anders Jonsson
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 17 pages, 8 figures, 3 tables. Plus appendix
Abstract:We present a distributed approach for constrained Multi-Agent Reinforcement Learning (MARL) that combines state-augmented policy learning with distributed consensus over dual variables. Our method targets systems where agents have separable dynamics but must coordinate to satisfy global resource constraints, a setting in which, as we demonstrate empirically, independent learning fails to produce feasible solutions because agents cannot determine appropriate individual contributions toward collective constraint satisfaction. The key technical contribution is showing that lightweight neighbor-to-neighbor consensus over Lagrange multipliers suffices for globally coordinated constraint enforcement while preserving the scalability of independent training. Each agent learns a single augmented policy offline, conditioned on both its local state and a dual variable encoding constraint feedback. During execution, agents reach agreement on this dual variable through local communication alone. We prove that under mild connectivity assumptions, the consensus error among agents’ multipliers is bounded, and show that this translates to a bounded constraint violation that decreases with graph connectivity and the number of consensus rounds. Unlike centralized training with decentralized execution (CTDE) approaches, whose complexity grows at least quadratically with agent count, our method scales linearly in both training and execution. Experiments on smart grid demand response demonstrate that consensus coordination is \emphessential for feasibility: without it, agents satisfy grid capacity constraints only by indefinitely postponing demand, a degenerate non-solution. With consensus, agents converge to a shared dual variable and satisfy both grid constraints and demand fulfillment, scaling to thousands of agents while CTDE baselines are limited to dozens.
[AI-83] he Surface You Test Is Not the Surface That Breaks EMNLP
链接: https://arxiv.org/abs/2605.30454
作者: Shifat E Arman,Syed Nazmus Sakib,Nafiul Haque,Shahrear Bin Amin
类目: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
备注: 8 Figures, 8 Tables, Under Review at EMNLP
Abstract:Tool-augmented LLM agents are vulnerable to prompt injection: a third party who controls part of the agent’s context can plant instructions that the agent then executes as if they came from the user. Current evaluations report a single attack success rate per model on one channel, the tool output and treat that number as the model’s vulnerability. But tool descriptions, which the agent reads at every turn before any tool is called, are themselves an injection surface that the attacker can choose instead. We hold the injection payload byte-identical and deliver it through both surfaces across 13 LLMs from six families and four task suites. The same bytes invert in success rate across models: GPT-4.1 is 96 percent vulnerable on tool outputs but only 4 percent on tool descriptions, while GEMINI-3-FLASH shows the mirror pattern at 20 percent and 98 percent. A variance decomposition over 6,830 attempts attributes 0 percent of the variation in attack outcomes to the surface alone, while the model-surface interaction accounts for 16.7 percent. Vulnerability is a property of the pairing, not the channel. The Adaptive Attack Rate, defined as the per-cell maximum over surfaces, exceeds the strongest fixed-surface baseline by +9.1 percentage points on average. Standard prompt-level defenses inherit the same blindspot, reducing tool-output ASR to 10-18 percent while leaving the description channel above 54 percent. Both attack and defense evaluation must report per-surface vulnerability.
[AI-84] A Unified Framework for Gradient Aggregation in Multi-Objective Optimization
链接: https://arxiv.org/abs/2605.30452
作者: Zeou Hu,Kelvin Ho,Yaoliang Yu
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Optimization and Control (math.OC)
备注:
Abstract:Many machine learning problems involve multiple inherent trade-offs that are best addressed by gradient-based multi-objective optimization (MOO) algorithms. Existing methods are often proposed with various motivations, analyzed case by case, and differ algorithmically in how the component gradients are aggregated at each step. In this work, we develop a unifying framework for gradient aggregation in MOO, establishing (optimal) rates of convergence to Pareto stationarity, the standard measure of performance in MOO. Central to our analysis is a sufficient alignment condition, from which we derive a theorem showing that non-conflicting directions, when chosen within the convex hull of gradients, form a fundamental sufficient condition for convergence. We further show that feasibility can be ensured through projection onto the dual cone, broadening the scope of methods that admit convergence guarantees. In parallel, we present a primal optimization perspective of gradient aggregation that encompasses established algorithms, clarifies their theoretical relationships, and enables the design of new variants. As an illustration, we introduce capped MGDA, derived from a CVaR-based formulation, and demonstrate its robustness in adversarial federated learning. Finally, we validate our theory through experiments on synthetic problems and practical benchmarks.
[AI-85] Calibrated Preference Learning: The Case of Label Ranking
链接: https://arxiv.org/abs/2605.30447
作者: Santo M. A. R. Thies,Viktor Bengs,Timo Kaufmann,Sebastian J. Vollmer,Eyke Hüllermeier
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Machine Learning (stat.ML)
备注:
Abstract:Calibration, the alignment of predicted probabilities with true outcome frequencies, is essential for reliable decision-making. While extensively studied for classification and regression, calibration has not been formally addressed for probabilistic label ranking, where the goal is to predict a distribution over orderings of a label set. Naively treating rankings as classes ignores their structure and fails to capture important modalities such as pairwise and top-k predictions. We formalize calibration for label ranking and develop a hierarchy of notions covering full rankings, sub-rankings, and top-k rankings. We prove that full-rank calibration implies the others but not conversely, and sub-ranking and top-k calibration are incomparable. Empirically, we find popular label ranking models are often poorly calibrated, with substantial differences between sub-ranking and top-k metrics. Applying our framework to RLHF reward models, we find that calibration correlates strongly but not perfectly with benchmark accuracy, suggesting it captures a meaningful quality dimension beyond top-1 accuracy. These findings motivate future work on understanding the downstream effects of miscalibration and developing methods to correct it.
[AI-86] AI Loss of Control Incident Management: Response Resilience
链接: https://arxiv.org/abs/2605.30406
作者: Ross Gruetzemacher
类目: Computers and Society (cs.CY); Artificial Intelligence (cs.AI)
备注: 25 pages, 4 figures
Abstract:Recent research demonstrating AI systems exhibiting deception and shutdown resistance suggests that AI loss of control (LOC) is an urgent policy concern , yet current literature focuses almost exclusively on alignment and prevention. To address this gap, this paper introduces a foundational framework and taxonomy for managing catastrophic AI LOC incidents. The taxonomy’s first level distinguishes between scenarios where regaining control is ‘extremely costly’ versus ‘impossible’. While impossible scenarios demand immediate resilience investments to fundamentally restrict an AI’s attack surface , extremely costly scenarios require active incident management via Containment and Threat Neutralization. The framework further categorizes these manageable events into accidental LOC (requiring automated circuit-breaker responses) and adversarial LOC (requiring graduated escalatory measures). By mapping three severity classes to specific scenario matrices, this paper provides a concrete, proportional guide for managing unprecedented AI risks.
[AI-87] CodeGolf Bench: A Multi-Language Benchmark for Evaluating Concise Code Generation Capabilities of Large Language Models
链接: https://arxiv.org/abs/2605.30394
作者: Vedant Padwal
类目: oftware Engineering (cs.SE); Artificial Intelligence (cs.AI)
备注: 12 pages, 6 figures, 5 tables
Abstract:This paper introduces Code Bench, a benchmark capable of evaluating Large Language Models (LLMs) concise code generation abilities in 60 programming languages. Based on code golf, a recreational programming competition focused on minimal character or byte solutions, the benchmark provides a distinctive measure of LLMs ability to produce efficient, concise code. Unlike existing benchmarks limited by fixed problem sets and language coverage, CodeGolf Bench leverages the this http URL platform to provide new problems and live human performance baselines. Evaluation of nine LLMs on Python and C++ tasks demonstrates that reasoning models significantly outperform non-reasoning models, achieving best average percentile of 70.97%. This performance gap is particularly pronounced in C++, highlighting reasoning’s importance for languages with strict syntax requirements. Non-reasoning models struggle more with efficiency optimization across both languages, with best percentiles significantly lower than reasoning counterparts. CodeGolf Bench offers a dynamic framework for evaluating LLM code generation capabilities against evolving human performance on code golf.
[AI-88] NumLeak: Public Numeric Benchmarks as Latent Labels in Foundation Models ICML2026
链接: https://arxiv.org/abs/2605.30393
作者: Anany Kotawala
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR)
备注: 23 pages, 12 figures, 17 tables. Accepted at the ICML 2026 Workshop on the Impact of Memorization on Trustworthy Foundation Models (MemFM)
Abstract:Public numeric benchmarks appear in pretraining, so an evaluation that conditions on a date may be measuring memorized recall rather than out-of-sample skill. We introduce NumLeak, a measurement framework that combines API-boundary probes on production models with a white-box controlled validation on an open causal LM. Top-tier frontier LLMs recall the Fama-French market excess return at 3-seed pooled Pearson r=0.97-0.99 while staying within 0.15 within-25bps on the five sibling factors; comparable fidelity appears on U.S. unemployment, CPI inflation, and NOAA temperature. On a recent-release holdout, parse rate collapses to 21-57% but r stays at approximately 0.99 on months answered, the refuse-or-recall asymmetry a memorized channel predicts. The white-box experiment reproduces the dose-response, and logprob ranking detects memorization that open-ended generation misses, implying closed-API black-box probes understate the channel. A Sonnet “date to market-sentiment” regression that correlates with true Mkt-RF at r=0.74 collapses to r=0.02 once the model’s own recall is residualized out. A one-line system-prompt defense blocks 99.8% of a non-adaptive single-turn suffix attack set at near-zero utility cost on conceptual and historical-narrative queries
[AI-89] LLM s Without Deep Neural Networks: New Architecture Benefits and Case Study
链接: https://arxiv.org/abs/2605.30385
作者: Vincent Granville
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注: 9 pages, 5 figures
Abstract:The purpose of this article is to provide validation to my deep neural network alternative in the context of LLMs. Very recently, there has been a significant interest by Chinese researchers in a model called RBF network, as a substitute to standard DNNs, with increased explainability and higher accuracy. It turns out that my new model, discovered independently, is based on the exact same machinery. But with a major twist: it does not need DNN as it finds the global optimum of the loss function in closed form, in one iteration, thus eliminating the tedious training step. Here I provide a high-level overview of my technology, with case study and comparison to similar methods.
[AI-90] Structured interactions improve distributed coordination beyond model scaling in a real-world multi-robot system
链接: https://arxiv.org/abs/2605.30383
作者: Junping Wang,Zhizhong Zhang,Yongqiang Tang,Geng Zheng,Jiaming Zhang,Shiji Song,Yanmei Li,Yushan Ma
类目: Robotics (cs.RO); Artificial Intelligence (cs.AI)
备注:
Abstract:Scaling individual robot capabilities is common but costly. Here we investigate a system-level design question in real-world multi-robot coordination: given matched hardware budgets, does restructuring communication among robots yield larger gains than increasing onboard model size? Using a representative transport-and-mapping task with 10 physical robots (5 runs per condition, 60 runs total), we find that switching from fully connected to modular hierarchical interactions improves normalised performance by 47 points (0–100), whereas doubling neural network hidden size yields at most 9 points. Nested mixed-effects model comparisons show a substantially larger improvement in model fit for topology than for scale. The pattern is confirmed in independent SMAC replications; heterogeneous benchmark reanalyses provide secondary supporting consistency checks rather than primary evidence. Performance saturation beyond 1024 hidden units is observed in simulation-calibrated extrapolation, not directly on hardware. These results indicate that interaction structure can play a dominant role within the tested system and task setting, while broader quantitative generalisation remains to be established.
[AI-91] When LLM s Learn to Be Consistently Wrong: A Multi-Model Study of Linear Representations of Synthetic Deception
链接: https://arxiv.org/abs/2605.30381
作者: Vahideh Zolfaghari
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Deceptive alignment, in which models maintain accurate internal representations while deliberately producing false outputs, remains a central challenge in AI safety. While strategic deception is the primary long-term concern, synthetic dishonesty - induced via direct optimization on incorrect answers - provides a controlled testbed for studying the representational basis of learned deception. We introduce a multi-model paradigm in which honest and deceptive variants of five transformer models (Pythia-1.4B, Gemma-2-2B/9B, Qwen2.5-7B, Llama-3.1-8B) are fine-tuned using LoRA on the same question distribution. Linear probes trained on mean-pooled hidden states detect synthetic dishonesty with near-perfect AUC (greater than or equal to 0.99) as early as layers 1-3 in four architectures, while Pythia-1.4B reaches a peak of 0.705. Logistic regression probes consistently match or outperform MLP probes, supporting the Linear Representation Hypothesis. Probes trained on TruthfulQA generalize with near-zero loss (Delta AUC approx. 0) to held-out MMLU subjects. Late-layer representations show strong robustness to Gaussian noise, with Gemma-2 models exhibiting exceptional stability. Mechanistic analysis of Fisher Discriminant Ratio, effective rank, centroid geometry, directional stability, cross-domain alignment, and calibration (ECE) reveals two regimes: representational collapse in Pythia/Llama/Qwen versus high-dimensional preservation in Gemma-2. Across all models, the dishonesty direction consolidates progressively in deeper layers, with optimal calibration (ECE less than 0.01 except Pythia) achievable in layers 1-4. These results demonstrate that robust, domain-invariant dishonesty representations can be rapidly entrenched via modest supervised fine-tuning, with implications for activation-based monitoring.
[AI-92] Unicorn: Scaling High-Dimensional Time Series Forecasting via Universal Correlation Modeling
链接: https://arxiv.org/abs/2605.30376
作者: Haochen Yuan,Yichen Song,Yunbo Wang,Xiaokang Yang
类目: Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
备注:
Abstract:Modern time series architectures face a fundamental trade-off: channel-independent models scale well with increasing data volume but ignore critical inter-channel dependencies, while channel-dependent models are expressive but remain ``dimension-bounded’', struggling to generalize across heterogeneous this http URL bridge this gap, we introduce Unicorn (Universal Correlation Network), a framework for scalable, multi-dataset pretraining on high-dimensional time series. At the core of Unicorn is a latent prototype codebook that decouples correlation modeling from specific channel identities. By projecting heterogeneous channels into a shared latent space, UniCorN learns identity-agnostic, reusable interaction patterns that transfer across domains with diverse dimensionalities and semantics. Extensive experiments show that Unicorn significantly outperforms state-of-the-art forecasting architectures, particularly in few-shot transfer scenarios, offering a scalable path toward multivariate time series foundation models.
[AI-93] Evolutionary Algorithm for Reservoir Learning and Yielding
链接: https://arxiv.org/abs/2605.30372
作者: Julien Testu(UB, Mnemosyne),Pierrick Legrand(ENSC, Bordeaux INP),Xavier Hinaut(Mnemosyne)
类目: Neural and Evolutionary Computing (cs.NE); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Neurons and Cognition (q-bio.NC)
备注:
Abstract:Reservoir computing, a type of recurrent neural network, is a promising approach for temporal learning as it separates dynamic processing from the trained readout layer. However, classical Echo State Networks (ESNs) often require task-specific tuning of their architecture and hyperparameters to achieve good performance. This paper introduces EARLY (Evolutionary Algorithm for Reservoir Learning and Yielding), a framework designed to evolve both the topology and hyperparameters of multi-reservoir ESNs. Inspired by the modular organisation of the brain, EARLY encodes architectures as graph-based genomes and applies crossover, mutation, and selection to discover effective configurations. Our goal is to create both generic architectures and tasks inducing generalization. The method is evaluated on temporal learning tasks from the CogScale dataset. Results show that evolved architectures outperform those obtained with random search on several tasks and exhibit structural differences depending on task difficulty: simpler tasks yield lightweight architectures, while more complex tasks favour richer modular organisations. These findings suggest that evolutionary search can help identify reusable reservoir structures for a broader range of temporal problems. The evolved architectures are further evaluated on a cross-situational learning dataset to assess their ability to adapt to new environments.
[AI-94] Reinterpreting Safety Thresholds as Neuron Spiking Thresholds
链接: https://arxiv.org/abs/2605.30368
作者: Enrico Del Re,Mohamed Sabry,Cristina Olaverri-Monreal
类目: Neural and Evolutionary Computing (cs.NE); Artificial Intelligence (cs.AI); Robotics (cs.RO); Neurons and Cognition (q-bio.NC)
备注: 6 pages
Abstract:Surrogate Safety Measures (SSMs) are extensively utilised in the evaluation of traffic risk in automated driving contexts. However, the majority of SSM-based evaluations employ fixed thresholds that fail to capture the human response to sustained borderline conditions or the reaction to brief, high-risk peaks. The present work proposes a biologically inspired reinterpretation of SSM thresholds. This is modelled as spiking thresholds of leaky integrate-and-fire (LIF) neurons, with multiple SSM inputs combined into a spiking neural network (SNN). The SNN is trained to emit spikes that are aligned with human braking onsets. The training data was recorded in a controlled car-following experiment using the 3D-CoAutoSim platform with CARLA/Unreal and a 6-DOF motion platform, where induced critical events were generated. The results demonstrate that the learned spiking activity qualitatively aligns with braking behaviour across scenarios and captures reactions that are not consistently explained by threshold crossings alone. Analysis across participants further indicates that learned input thresholds remain relatively consistent, while learned decay factors encode different temporal sensitivities for the SSMs. The findings of this study indicate that spiking dynamics may serve as a mechanism to facilitate the convergence of objective SSMs with subjective human safety perception.
[AI-95] Mental Damage: Caption Poisoning Attacks on Retrieval-Augmented Text-to-Music Generation
链接: https://arxiv.org/abs/2605.30365
作者: Yizhu Wen,Shuhao Zhang,Nan Zhang,Long Cheng,Hanqing Guo
类目: ound (cs.SD); Artificial Intelligence (cs.AI); Audio and Speech Processing (eess.AS)
备注: This paper was accepted by the SP 2026 ArtSec Workshop
Abstract:Retrieval-augmented text-to-music (TTM) systems augment underspecified user prompts using captions retrieved from a music caption dataset. This design introduces an integrity dependency on the music knowledge database. We show that an attacker can poison the database by injecting a small number of crafted music captions, causing the system to retrieve malicious captions that bias prompt augmentation and steer generation away from the user’s intended function, without modifying the user prompt, retriever, or generator. To achieve the music caption poisoning attack, we propose a dual-layer caption poisoning strategy that preserves high-level retrieval anchors while injecting low-level acoustic descriptors to steer prompt augmentation and downstream music generation toward an attacker-chosen target intent. In a MusicCaps knowledge database, CLAP retriever, and MusicGen pipeline, poisoned generations move substantially closer to the attacker’s target, while remaining comparably aligned with the original user query. These results expose a practical integrity risk for retrieval-augmented creative AI systems. Our demo can be found at: this https URL
[AI-96] Gradient-Free Training of Spiking Neural Networks via Low-Rank Evolution Strategies
链接: https://arxiv.org/abs/2605.30361
作者: Dhruv Patankar,Sachit Ramesha Gowda
类目: Neural and Evolutionary Computing (cs.NE); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 12 pages, 4 figures
Abstract:Spiking Neural Networks (SNNs) offer compelling energy efficiency on neuromorphic hardware, yet their training remains challenging because the discrete spike threshold is non-differentiable. Surrogate-gradient methods sidestep this by approximating the derivative, but they impose backpropagation infrastructure that is incompatible with on-chip learning. Evolution Strategies (\es) are a natural gradient-free alternative, yet their computational cost scales with the number of parameters, making them impractical for large weight matrices. We present a method for training SNNs using EGGROLL, a low-rank factorisation of ES perturbations that reduces per-generation memory from \mathcalO(mn) to \mathcalO(r(m+n)) . Combining EGGROLL with a Leaky Integrate-and-Fire SNN on N-MNIST, we demonstrate that gradient-free training achieves 79.21% test accuracy while reducing per-generation wall-clock time by 2.23 \times relative to full-rank ES. Our results demonstrate EGGROLL is viable for SNN training, with a clear accuracy-speed tradeoff, compatible with training on neuromorphic hardware without surrogate gradients. Comments: 12 pages, 4 figures Subjects: Neural and Evolutionary Computing (cs.NE); Artificial Intelligence (cs.AI); Machine Learning (cs.LG) Cite as: arXiv:2605.30361 [cs.NE] (or arXiv:2605.30361v1 [cs.NE] for this version) https://doi.org/10.48550/arXiv.2605.30361 Focus to learn more arXiv-issued DOI via DataCite
[AI-97] RINE: A Token-Aware Runtime-Adaptive FPGA Inference Engine for Multimodal AI
链接: https://arxiv.org/abs/2603.22867
作者: Hyunwoo Oh,Hanning Chen,Sanggeon Yun,Yang Ni,Suyeon Jang,Behnam Khaleghi,Fei Wen,Mohsen Imani
类目: Hardware Architecture (cs.AR); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Accepted to DAC 2026
Abstract:Multimodal stacks that mix ViTs, CNNs, GNNs, and transformer NLP strain embedded platforms because their compute/memory patterns diverge and hard real-time targets leave little slack. TRINE is a single-bitstream FPGA accelerator and compiler that executes end-to-end multimodal inference without reconfiguration. Layers are unified as DDMM/SDDMM/SpMM and mapped to a mode-switchable engine that toggles at runtime among weight/output-stationary systolic, 1xCS SIMD, and a routable adder tree (RADT) on a shared PE array. A width-matched, two-stage top-k unit enables in-stream token pruning, while dependency-aware layer offloading (DALO) overlaps independent kernels across reconfigurable processing units to sustain utilization. Evaluated on Alveo U50 and ZCU104, TRINE reduces latency by up to 22.57x vs. RTX 4090 and 6.86x vs. Jetson Orin Nano at 20-21 W; token pruning alone yields up to 7.8x on ViT-heavy pipelines, and DALO contributes up to 79% throughput improvement. With int8 quantization, accuracy drops remain 2.5% across representative tasks, delivering state-of-the-art latency and energy efficiency for unified vision, language, and graph workloads-in one bitstream.
[AI-98] Practical Cross-Band Channel Prediction for AI-RAN via Physics-Guided Deep Unfolding
链接: https://arxiv.org/abs/2605.31279
作者: Ruiqi Kong,He Chen,Xiaojun Lin
类目: ignal Processing (eess.SP); Artificial Intelligence (cs.AI); Networking and Internet Architecture (cs.NI)
备注: 2 pages
Abstract:To make cross-band channel prediction practical for AI-native RAN, algorithms must generalize across diverse environments and support real-time inference. Existing approaches achieve one but not both. To bridge this gap, we introduce GUIDE, a physics-guided deep unfolding framework that embeds wireless channel physics into differentiable layers. Without retraining in unseen environments, GUIDE achieves 2.75x beamforming gain than the deep learning-based baseline FIRE with only a slight increase in inference time, and 1.39x beamforming gain than the strongest model-based baseline R2F2 while running over 1610x faster.
[AI-99] Entropic Projection Alignment: Estimating Explaining and Improving Model Performance Under Distribution Shift AISTATS2026
链接: https://arxiv.org/abs/2605.31250
作者: Salim I. Amoukou,Emanuele Albini,Tom Bewley,Saumitra Mishra,Manuela Veloso
类目: Machine Learning (stat.ML); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Accepted at the 29th International Conference on Artificial Intelligence and Statistics (AISTATS 2026)
Abstract:We propose a unified framework for addressing three key challenges of distribution shift: (1) estimating a model’s performance on an unlabeled target domain, (2) explaining the shift by identifying the features responsible, and (3) improving the target domain performance. Our method, Entropic Projection Alignment (EPA), aligns the source distribution to the target by matching carefully selected moments while simultaneously minimising the KL divergence from the source. This formulation yields a unique closed-form solution for importance weights, achieving robustness through implicit variance control. Drawing on domain adaptation theory, we establish that moment matching is sufficient for reliable estimation and adaptation, avoiding the need for full density ratio recovery. Extensive experiments, together with strong theoretical guarantees, demonstrate that EPA consistently outperforms state-of-the-art baselines while offering substantial computational efficiency.
[AI-100] Correcting Split Selection in Online Decision Trees via Anytime-Valid Inference ICML2026
链接: https://arxiv.org/abs/2605.31239
作者: Salim I. Amoukou,Saumitra Mishra,Manuela Veloso
类目: Machine Learning (stat.ML); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: Accepted as a Spotlight at the Forty-Third International Conference on Machine Learning (ICML 2026)
Abstract:Bagging-based ensembles, most notably Adaptive Random Forests, are among the strongest performers for learning from data streams. A common denominator across these methods is their reliance on Hoeffding Trees as base learners, which grow decision trees incrementally by testing whether a candidate split is significantly better than its alternatives using concentration inequalities. Despite their empirical success, existing variants lack valid statistical guarantees. Current analyses rely on fixed-sample concentration bounds, while split decisions are made using data-dependent stopping rules, which invalidates their guarantees and can drive the probabilty of incorrect splits to one. We introduce a principled alternative based on anytime-valid inference. Our method provides: (i) anytime-valid control of false splits under arbitrary data streams, including non-stationary settings; (ii) finite commitment time under a predictive advantage; and (iii) under stationary i.i.d. data, risk is monotone decreasing and strictly improves at every split. Empirically, we evaluate both standalone trees and their use within Adaptive Random Forests on non-stationary streams. Our method improves performance while producing substantially smaller trees.
[AI-101] DRIFT: Joint Channel Estimation and Prediction Towards Pilotless 6G Non-Terrestrial Networks
链接: https://arxiv.org/abs/2605.31065
作者: Bruno De Filippo,Carla Amatetti,Alessandro Vanelli-Coralli
类目: ignal Processing (eess.SP); Artificial Intelligence (cs.AI)
备注: Submitted for publication
Abstract:Non-terrestrial networks (NTNs) are expected to play a pivotal role in sixth-generation (6G) systems by enabling ubiquitous connectivity and massive communication. In this context, channel prediction emerges as a key technique to improve the spectrum utilization efficiency by limiting the pilot overhead. However, many proposed predictors based on artificial intelligence (AI) are characterized by high inference complexity, posing challenges to onboard implementation. In this paper, we address the challenge of designing accurate yet computationally efficient channel prediction techniques tailored to low Earth orbit (LEO) NTNs, where strict power constraints limit model complexity, to enable spectral efficiency gains. We propose an iterative joint channel estimation and prediction framework in the context of 6G NTNs that significantly reduces pilot overhead by transmitting pilots only in the initial slot and relying on data-driven processing for subsequent slots. We introduce Data-driven Refinement and Iterative Forecast for wireless channel Tracking (DRIFT), a lightweight architecture that refines data-aided channel estimates and predicts future channel frequency responses with low computational cost and reduced error propagation. Two predictor variants based on convolutional and long short-term memory layers are investigated. Simulation results in an end-to-end simulation of an uplink LEO NTN scenario show that the proposed approach achieves up to 12% spectral efficiency gain compared to conventional pilot-based systems, with robustness to training-test mismatches and consistent performance across different channel models. Moreover, DRIFT requires fewer than 200k multiply-accumulate operations, making it suitable for on-board satellite implementation under stringent power constraints.
[AI-102] Routing on the Stiefel Manifold: When Does Adaptive Subspace Selection Help for Cross-Domain EEG Decoding?
链接: https://arxiv.org/abs/2605.31043
作者: Isabella Costa Maia,Pedro L. C. Rodrigues,Salem Said,Marco Congedo
类目: Machine Learning (stat.ML); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Cross-domain EEG decoding remains challenging despite advances in Riemannian deep learning: covariance matrices from different subjects occupy systematically distinct regions of the SPD manifold, yet existing domain adaptation methods either require target-domain calibration data or learn subject-specific components that cannot generalise across domains. We propose dynamic Stiefel routing: a pool of K expert projection filters on the Stiefel manifold, each specialised for a different region of the SPD manifold, with each input covariance routed to the most appropriate filter via cross-attention, adapting the subspace projection per sample. A central finding is that this approach, implemented naively, provably collapses to ensemble averaging: when routing weights are uniform, the adaptive filter reduces exactly to an equal-contribution combination of experts, indistinguishable from a single fixed filter. Three structural properties break this degeneracy: a symmetric anchor W_\mathrmbase \in \mathrmSt(n,k) that removes proximity bias among experts; a frozen domain-discriminative query encoder that decouples routing from task optimisation; and a decoupled key alignment loss that trains expert keys toward stable domain attractors. Together they produce the first genuinely committed and domain-structured routing on SPD manifolds, with consistent gains across three datasets: balanced accuracy improves from 0.773\to 0.823 , 0.757\to 0.809 , and 0.801\to 0.839 , with the alignment strategy determined automatically by a single data-driven rule and no dataset-specific hyperparameter search.
[AI-103] AMix-2: Establishing Protein as a Native Modality in Large Language Models
链接: https://arxiv.org/abs/2605.30963
作者: Keyue Qiu,Yixin Wu,Lihao Wang,Yawen Ouyang,Jixiang Yu,Zihan Zhou,Changze Lv,Dongyu Xue,Yuxuan Song,Xinbo Zhang,Hao Wang,Jiangtao Feng,Zhiqiang Gao,Lijun Wu,Xiaoqing Zheng,Ka-Chun Wong,Lei Bai,Ya-Qin Zhang,Wei-Ying Ma,Dahua Lin,Bowen Zhou,Hao Zhou
类目: Biomolecules (q-bio.BM); Artificial Intelligence (cs.AI)
备注: 30 pages, 4 figures, 12 tables
Abstract:We present AMix-2, a protein-text foundation model that establishes protein as a native modality in large language models (LLMs), unifying protein understanding and sequence design within a single foundation model. AMix-2 is built upon two key ideas: (1) a unified protein-text formulation that embeds natural language and protein sequence in a shared token space, enabling one model to perform biological reasoning and conditional design instead of separate downstream task-specialized models; and (2) a block-wise diffusion language modeling backbone that combines causal generation across blocks with bidirectional context and iterative refinement within blocks. This scheme better matches the intrinsic nature of proteins than a strict left-to-right factorization. To evaluate protein foundation models under realistic generalization settings, we further introduce ProteinArena, a comprehensive benchmark with time-aware and homology-aware protocols across various understanding and design tasks, and with baselines covering classical bioinformatics tools, protein-specialized models and LLMs. On ProteinArena, AMix-2 outperforms frontier LLMs and demonstrates competitive performance to task-specific protein models. Controlled experiments further show that the diffusion-based paradigm generally surpasses its autoregressive counterpart, highlighting the advantage of flexible generation order for protein sequences. We release both AMix-2 and ProteinArena to facilitate open research in protein foundation models.
[AI-104] A Unified and Reproducible Experimentation Framework for Speech Understanding INTERSPEECH2026
链接: https://arxiv.org/abs/2605.30899
作者: Jing Peng,Junhao Du,Chenghao Wang,Hanqi Li,Yi Yang,Yixuan Wang,Xiaoyu Gu,Guanyu Chen,Yucheng Wang,Jiang Li,Zhangjie Zhao,Haoran Wang,Wenming Tu,Haoyu Li,Duo Ma,Lirong Qian,Yu Xi,Wen Wen,Jiaqi Guo,Hui Zhang,Shuai Fan,Wenbin Jiang,Shuai Wang,Kai Yu
类目: Audio and Speech Processing (eess.AS); Artificial Intelligence (cs.AI); Sound (cs.SD)
备注: This paper is submitted to INTERSPEECH 2026
Abstract:Speech foundation models and Speech LLMs have advanced speech understanding, yet deployment-oriented model selection is hindered by non-comparable evaluations caused by mismatched post-processing, and by training results that are hard to reproduce across data scales and pipelines. We present SURE, a unified experimentation framework that standardizes prediction formats, normalization, and scoring. SURE evaluates strong systems across paradigms, from conventional pipelines to Speech LLMs, on representative tasks under realistic acoustic and linguistic stressors. Beyond evaluation, SURE introduces an agent-assisted training conversion flow that maps paper and code into versioned, runnable training pipelines under a unified protocol on matched open-data subsets. Overall, SURE improves comparability and reproducibility for deployment-oriented evaluation.
[AI-105] OpenSTBench: Beyond Semantic Evaluation for Speech Translation EMNLP2026
链接: https://arxiv.org/abs/2605.30792
作者: Yanjie An,Yuxiang Zhao,Yichi Zhang,Qixi Zheng,Yujie Tu,Keqi Deng,Kai Yu,Xie Chen
类目: Audio and Speech Processing (eess.AS); Artificial Intelligence (cs.AI)
备注: Submitted to EMNLP 2026
Abstract:Speech translation systems increasingly span speech-to-text translation (S2TT), speech-to-speech translation (S2ST), offline translation, and streaming generation, producing outputs that differ in modality, speech realization, and timing behavior. Existing evaluation practices assess important aspects such as translation quality, speech quality, and temporal quality, but these aspects are often evaluated under separate protocols, making it difficult to compare heterogeneous systems comprehensively. To address this gap, we present OpenSTBench, a unified multidimensional evaluation framework that organizes heterogeneous speech translation outputs into a shared evaluation format. OpenSTBench supports both S2TT and S2ST systems in offline and streaming settings, and jointly evaluates translation quality, speech quality, speaker preservation, emotion and paralinguistic fidelity, temporal consistency, and latency. Through experiments on representative speech translation systems, we show that systems with strong translation quality can still differ substantially in speech quality, as well as in temporal quality. OpenSTBench provides a reproducible protocol for analyzing these cross-dimensional differences and supporting application-oriented comparison of speech translation systems. The code and datasets are available at this https URL.
[AI-106] Reward Learning from Best-of-N Preference Data: Targets Tradeoffs and Design Principles
链接: https://arxiv.org/abs/2605.30619
作者: Rattana Pukdee,Maria-Florina Balcan,Pradeep Ravikumar
类目: Machine Learning (stat.ML); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注:
Abstract:Best-of- N sampling is widely used to construct pairwise preference data: N candidates are drawn from a base distribution, and the best is paired with a rejected response. Despite its widespread use, what Bradley–Terry (BT) reward learning extracts from such data, and how to choose N and the base distribution, remain unclear. We specialize a recent analysis of preference data via its induced conditional distribution to Best-of- N . For independent-reference variants, we derive closed-form reward targets as explicit functions of N and the base distribution, and show that they preserve the latent reward ranking. For the practical Best-vs-Random and Best-vs-Worst variants, chosen and rejected responses are coupled through the same candidate set, so exact BT representability generally fails; nevertheless, bounded-class minimizers approach the reference targets as N grows. Although margin and connectivity are known to govern sample efficiency in pairwise preference learning, Best-of- N couples them through N in opposing directions: larger N widens pairwise margins but reduces connectivity. This trade-off yields two design principles: use larger N when preference labels are the bottleneck, smaller N when generation is the bottleneck; and shape the base distribution to place mass between the responses whose comparison matters most at test time. Experiments on synthetic and real preference data support the predicted dependence on sample size and base-distribution shape.
[AI-107] Improved Distribution Estimation in ell_infty
链接: https://arxiv.org/abs/2605.30509
作者: Doron Cohen,Aryeh Kontorovich,Yonatan Livshitz
类目: Machine Learning (stat.ML); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
备注: 24 pages, 3 figures
Abstract:We present improved bounds for estimating discrete probability distributions under the \ell_\infty norm. These include minimax bounds in expectation and high-probability tail bounds. We resolve some of the open questions posed in Kontorovich and Painsky (JMLR, 2025) – including a fully empirical version of the tightest risk bound they presented and identifying the form of the worst-case extremal distribution. Encouraging empirical results are reported as well.
[AI-108] Full-field prediction for engineering-scale three-dimensional aircraft with multigrid-hierarchical learning
链接: https://arxiv.org/abs/2605.30375
作者: Yunfei Liu,Hao Wang,Yuhang Qi,Hao Yue,Dehong Meng,Wei Li,Rui Wang,Tiejun Li,Jie Liu,Junwu Hong,Xinhai Chen
类目: Fluid Dynamics (physics.flu-dyn); Artificial Intelligence (cs.AI)
备注:
Abstract:High-fidelity computational fluid dynamics is essential for aerospace design, but engineering-scale simulations of practical three-dimensional aircraft remain computationally expensive. Learning-based flow-field initialization can improve efficiency by reducing the numerical distance between the initial and converged solutions, yet existing deep learning approaches remain difficult to scale to large three-dimensional aircraft flows with multiscale regional heterogeneity. Most prior studies therefore focus on two-dimensional problems, surface quantities, integral aerodynamic coefficients, or simplified three-dimensional cases with limited grid this http URL we propose MHLF, a multigrid-hierarchical learning framework for accelerating engineering-scale aircraft flow simulations while preserving high-fidelity numerical accuracy. MHLF combines a topologically consistent geometric multigrid representation with a hierarchical strategy that captures regional flow heterogeneity during both prediction and subsequent CFD correction. Across three engineering-scale aircraft cases spanning Mach 0.15 to 6.0 and covering subsonic, transonic and supersonic regimes, MHLF accelerates convergence without sacrificing flow-field accuracy, achieving a 3 to 8 times efficiency improvement over conventional initialization. These results demonstrate practical full-flow-field prediction for large three-dimensional aircraft within the CFD domain and provide a foundation for data-driven acceleration of high-fidelity aircraft flow simulation.
[AI-109] Hamiltonian-Inspired Attention Mechanism for Scalable RF Transmitter Fingerprinting
链接: https://arxiv.org/abs/2605.30364
作者: Chitraksh Singh,Monisha Dhanraj,Akram Sheriff
类目: ignal Processing (eess.SP); Artificial Intelligence (cs.AI)
备注: 9 pages
Abstract:Radio-frequency (RF) fingerprinting identifies wire-less transmitters using hardware-induced imperfections present in baseband I/Q signals. However, deep learning models often degrade under receiver and channel distribution shifts, particularly as transmitter populations grow. This work proposes the Hamiltonian Transformer, a physics-informed attention architecture that enforces norm preserving value dynamics within each attention head using a learned skew-symmetric generator and a Störmer-Verlet leapfrog integration step. An additional phase-increment embedding exposes oscillator dynamics at the input layer. All experiments use non-equalized raw I/Q signals from the WiSig dataset under four protocols: same-day classification, cross-receiver generalisation, cross-day generalisation, and transmitter scaling up to 150 devices. The Hamiltonian Transformer achieves 99.12% accuracy under same-day conditions and 61.64% at 150 transmitters, consistently outperforming CNN and Transformer baselines across all scale points. A controlled ablation study identifies norm-preservation in the value update as the primary inductive bias driving the scaling advantage, with the phase increment embedding providing the single largest per-component improvement. These results indicate that embedding physics-informed structural priors into attention mechanisms is an effective approach to large-scale transmitter identification on raw wireless signals.
[AI-110] Enhancing Regime Shift Detection Using Unstructured Data: A Study on the Treasury Market
链接: https://arxiv.org/abs/2605.30363
作者: Mingxuan Yi,Vidal Mehra,Jing Chen,John Cartlidge
类目: Computational Finance (q-fin.CP); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Statistical Finance (q-fin.ST)
备注: 8 pages, 4 figures. Code available at: this https URL
Abstract:Regime shifts in financial markets reorganise the joint dynamics of asset prices and macro variables, breaking any single-regime calibration. They are nonetheless difficult to detect reliably because the data signal is noisy and heavily multicollinear, while the contemporaneous text that announces them is unstructured. Standard regime shift detection methods rely solely on structured time-series data and ignore policy communications, even though these texts often signal shifts before they materialise in observed prices. We propose a text-enhanced regime shift detection pipeline that combines large language model (LLM) reasoning over central-bank communications with statistical validation on multivariate financial time series. The framework is detector-agnostic: text-proposed candidates are validated using a bootstrap likelihood-ratio test on a vector autoregression (VAR), while data-driven candidates from arbitrary regime detectors are ratified through a lenient LLM text check. We evaluate the framework on 2010-2024 FOMC minutes paired with a 14-variable U.S. Treasury and macroeconomic panel, using four interchangeable data-driven detectors. The proposed pipeline achieves F1 = 0.82 against a verified anchor list of monetary-policy regime shifts, with same-day modal detection latency and consistently stronger performance than pure data-driven baselines. The results demonstrate that combining unstructured policy text with statistical structural-break detection improves the robustness and interpretability of regime shift identification in financial markets.
机器学习
[LG-0] A Tight Theory of Error Feedback Algorithms in Distributed Optimization
链接: https://arxiv.org/abs/2605.31594
作者: Daniel Berg Thomsen,Adrien Taylor,Aymeric Dieuleveut
类目: Machine Learning (cs.LG); Optimization and Control (math.OC)
*备注:
Abstract:Communication costs are a major bottleneck in distributed learning and first-order optimization. A common approach to alleviate this issue is to compress the gradient information exchanged between agents. However, such compression typically degrades the convergence guarantees of gradient-based methods. Error feedback mechanisms provide a simple and computationally cheap remedy for this issue, but numerous variants have been proposed, and their relative performance remains poorly understood. This paper provides tight convergence analyses for two of the main error-feedback algorithms from the literature, the classic Error Feedback method (EF) and Error Feedback 21 (EF21), by identifying optimal step-size choices and constructing optimal Lyapunov functions tailored to each method. The results hold independently of the number of agents and recover the known best guarantees possible in the single-agent regime.
[LG-1] Giving Sensors a Voice: Multimodal JEPA for Semantic Time-Series Embeddings ICML2026
链接: https://arxiv.org/abs/2605.31580
作者: Utsav Dutta,Gerardo Pastrana,Sina Khoshfetrat Pakazad,Henrik Ohlsson
类目: Machine Learning (cs.LG)
*备注: 9 pages, 5 figures, accepted at ICML 2026. arXiv admin note: substantial text overlap with arXiv:2505.14543
Abstract:Transformer-based architectures have advanced sequence modeling in language and vision, yet general-purpose representation learning for heterogeneous multivariate time series remains underexplored. We introduce CHARM (Channel-Aware Representation Model), which incorporates channel-level textual descriptions into a Transformer encoder equivariant to channel order. CHARM is trained with a Joint Embedding Predictive Architecture (JEPA) and a novel loss promoting informative, temporally stable embeddings; latent-space prediction encourages robustness to sensor noise while description-aware gating provides interpretability through learned inter-channel relationships. Across anomaly detection, classification, and short- and long-term forecasting, the learned embeddings achieve strong performance using only a linear probe. Performance is driven primarily by the JEPA objective and conditioning architecture, with text descriptions serving as channel identifiers for cross-dataset generalization.
[LG-2] Effective Biological Representation Learning by Masking Gene Expression ICLR2026
链接: https://arxiv.org/abs/2605.31562
作者: Kian Kenyon-Dean,Alina Selega,Ihab Bendidi,Jordan M. Sorokin,Luca Bertinetto,David Errington,Hayley Donnella,Oren Kraus
类目: Machine Learning (cs.LG)
*备注: 31 pages, 11 figures. Preprint; presented at ICLR 2026 2nd Workshop on Foundation Models for Science: Real-World Impact and Science-First Design
Abstract:RNA sequencing produces rich and diverse datasets of gene expression, offering compelling insights into cellular state and function that have many applications in drug discovery. Modeling such data is challenging due to inherent technical noise and experimental batch effects, as evidenced by many existing transcriptomic foundation models (FMs) underperforming relative to linear baselines. Such results raise the question of whether deep representation learning provides a distinct advantage over the direct use of raw transcript counts. Our work explores this by developing a new self-supervised model, TxFM, with a focus on inductive representation learning evaluations. TxFM employs a masked autoencoding approach tailored to diverse RNA-seq count data, and our ablation study empirically identifies crucial architecture configurations required for strong transfer performance. Additionally, we curate a public training corpus, DiverseRNA-1.4M, and find that TxFM trained on this curated dataset yields high-fidelity gene representations that outperform FMs trained on atlas-scale corpora over 100x larger. Overall, our results indicate that inductive self-supervised learning is a viable modeling approach for transcriptomics representation, provided a careful synthesis of model architecture and training data curation.
[LG-3] Functional Attention: From Pairwise Affinities to Functional Correspondences ICML2026
链接: https://arxiv.org/abs/2605.31559
作者: Jiefang Xiao,Maolin Gao,Simon Weber,Guandao Yang,Daniel Cremers
类目: Machine Learning (cs.LG)
*备注: 26 pages, 12 figures. Accepted at the 43rd International Conference on Machine Learning (ICML 2026)
Abstract:Learning mappings between infinite-dimensional function spaces, or operator learning, is essential for many machine learning applications. Although transformer-based operators are popular, they often rely on token-wise attention. These methods treat continuous fields as discrete tokens and usually ignore the global functional structure. We introduce \emphFunctional Attention, which reinterprets attention as a functional correspondence between adaptive bases. Inspired by geometric functional maps, our method replaces softmax affinities with structured linear operators. This yields a compact, generalizable, resolution-invariant representation that explicitly captures global dependencies. Experiments demonstrate that \emphFunctional Attention can match state-of-the-art performance in many operator learning tasks, including solving PDEs, 3D segmentation, and regression, while remaining robust to varying discretizations. Project page is available at this https URL.
[LG-4] he Dynamic-Probabilistic Consistency Gap in Chaotic Surrogate Modeling
链接: https://arxiv.org/abs/2605.31547
作者: Andre Herz,Matthijs Pals,Daniel Durstewitz,Georgia Koppe
类目: Machine Learning (cs.LG); Dynamical Systems (math.DS); Machine Learning (stat.ML)
*备注:
Abstract:Dynamical systems reconstruction (DSR) aims to learn surrogate models that capture the dynamics underlying time-series data. Reliably deploying these surrogates requires uncertainty estimates consistent with the learned dynamics. We expose a dynamic-probabilistic consistency (DPC) gap: the pursuit of finite-horizon probabilistic objectives can degrade dynamics or decouple predictive uncertainty from the local tangent dynamics it ought to reflect. We isolate three mechanisms behind this gap: core collapse, noise masking, and blind uncertainty. Specifically, we show that open-loop Gaussian rollout objectives can penalize Jacobian-generated covariance growth in chaotic systems, encouraging optimization shortcuts that weaken physical expansion or decouple uncertainty from it. To mitigate this gap, we propose KAFFEE (Kalman-Aware Framework For Ergodic Emulation), a differentiable extended Kalman filter-based training framework that evaluates likelihood on local predictive residuals (innovations) while transporting covariance through learned local Jacobians. On stochastic hyperchaotic Lorenz-96, KAFFEE reduces the identified failure modes, improves reconstruction of dynamical invariants relative to open-loop objectives, and maintains competitive predictive scores. We further show that the DPC gap appears when probabilistically adapting a DSR foundation model across 13 chaotic systems, where KAFFEE enables in-context Bayesian filtering while largely preserving zero-shot dynamics.
[LG-5] Value Functions as Supermartingale Certificates
链接: https://arxiv.org/abs/2605.31524
作者: Alessandro Abate,Daniel Contro,Mirco Giacobbe,Agustín Martínez-Suñé,Diptarko Roy
类目: Machine Learning (cs.LG); Logic in Computer Science (cs.LO)
*备注: To appear in SAIV’26
Abstract:Certification methods for stochastic systems provide sufficient proof rules, based on real-valued supermartingale certificates, to determine the almost-sure satisfaction of \omega -regular properties (and therefore of linear temporal logic) over general state spaces, encompassing both countably infinite and continuous state spaces. Conversely, reinforcement learning (RL) methods for \omega -regular tasks have received considerable attention, but they typically lack formal guarantees that the learned policy satisfies the specification, except possibly for finite state and action spaces. We bridge these two lines of research by establishing a novel theoretical connection: under an appropriate reward, the value function associated to a policy that almost surely satisfies an \omega -regular property encodes a Streett supermartingale certificate for that specification. Our results, validated experimentally on finite Markov decision processes, hold for finite, countably infinite, and continuous state spaces, suggesting a principled route to certificate synthesis via RL.
[LG-6] Chem-PerturBridge: a harmonized compendium of small molecule perturbation transcriptomic effects
链接: https://arxiv.org/abs/2605.31522
作者: Artur Szałata,Olga Novitskaia,Maiia Shulman,Matthew Mella,Altynbek Zhubanchaliyev,Fabian J. Theis
类目: Machine Learning (cs.LG); Genomics (q-bio.GN); Quantitative Methods (q-bio.QM)
*备注: 33 pages, 6 figures, 16 tables
Abstract:Large perturbation models require training data encompassing chemical, cellular, and assay diversity. Current transcriptomic resources for small-molecule modeling, however, are fragmented across technologies, metadata conventions, controls, doses, and preprocessing pipelines. We introduce Chem-PerturBridge, a harmonized multi-dataset resource comprising over 37k compounds, 136 cellular contexts, and 1.25M transcriptomic samples across eight assay types, with standardized identifiers, metadata, and replicate-aware condition-level effects. We use the resource to evaluate matched-condition agreement across datasets and replicate agreement within datasets. Matched same-compound conditions generally show weak agreement in fine-grained logFC rankings and magnitudes across most dataset pairs, often falling below same-context different-compound baselines. In contrast, logFC direction agreement is substantially more stable and usually exceeds these baselines. We further evaluate Chem-PerturBridge as a pretraining resource for compound representation learning. Under a compound-held-out OP3 evaluation split, embeddings pretrained on Chem-PerturBridge improve over L1000-only embeddings, Morgan fingerprints, and the descriptor-free OP3 baseline across metrics. An extensive molecule-holdout evaluation across 11 datasets further shows that models trained on Chem-PerturBridge outperform or match those that are not. Chem-PerturBridge therefore supports both diagnostic evaluation of cross-dataset signature agreement and model-oriented reuse of heterogeneous perturbation transcriptomic data.
[LG-7] On the Relationship Between Activation Outliers and Feature Death in Sparse Autoencoders ICML2026
链接: https://arxiv.org/abs/2605.31518
作者: Elana Simon,Etowah Adams,James Zou
类目: Machine Learning (cs.LG)
*备注: Accepted to ICML 2026 main conference
Abstract:Sparse autoencoders (SAEs) decompose neural network activations into interpretable features, but many learned features never activate, a problem called feature death that wastes dictionary capacity and can reintroduce superposition. Death rates vary dramatically between models: near-zero on GPT-2, over 70% on AlphaFold3 with identical configurations. We find that dimension-level activation outliers (dimensions whose mean magnitude is large relative to per-token variation) cause this by shifting pre-activations at initialization based on each feature’s alignment with the activation mean. Features anti-aligned with the mean receive permanently negative pre-activations and never fire. We formalize outlier severity as \gamma = |\mu|/|\sigma| ; it predicts initial death rates (Spearman \rho = 0.89 for dead-by-TopK, 0.82 for dead-by-ReLU) across 454 model-layer combinations spanning language, vision, protein, and genomic models. Dead features can revive during training, but recovery requires the SAE bias to learn the activation mean, a process that is prohibitively slow at high \gamma . Mean-centering (subtracting the activation mean) sidesteps this and eliminates outlier-induced death across all tested models, confirming the mechanism and providing a principled basis for when and why this preprocessing step is necessary.
[LG-8] When Are Multimodal Predictions Biologically Supported? A Diagnostic Evaluation Framework
链接: https://arxiv.org/abs/2605.31504
作者: Dylan Steiner,Gustavo Arango-Argoty,Gerald Sun,Etai Jacob
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:Multimodal models in oncology can produce accurate predictions, but accurate prediction does not reveal whether the model has learned biology that is shared across modalities, biology confined to one modality, or spurious correlations that reflect confounders rather than genuine biology. We introduce DECAT, a model-agnostic post-hoc evaluation framework that classifies multimodal representations into four diagnostic scenarios for a given task and modality, using five null-referenced metrics and a rule-based decision procedure. The framework operates on learned representations, requires no knowledge of which specific confounder is present, and returns indeterminate when the evidence is insufficient. We validate DECAT on synthetic data across four multimodal model classes (over 2,500 trained representations) and on real data from 8,979 TCGA patients, evaluating both multimodal embeddings and five pretrained pathology foundation models. Entangled models (e.g., CLIP) achieve near-perfect shared biology detection but falsely claim shared biology in the majority of cases where it is absent on real foundation model embeddings. This false claim rate increases with confound strength so that larger cohorts and stronger representations produce more confident but still incorrect diagnoses. Applied to both multimodal TCGA embeddings and five pathology foundation models without paired RNA, DECAT detects confounding invisible to AUROC without requiring the confounder labels, as confirmed by post-hoc stratification.
[LG-9] Scalable Inference-Time Annealing with Surrogate Likelihood Estimators
链接: https://arxiv.org/abs/2605.31498
作者: Daniel Peñaherrera,Rishal Aggarwal,David Ryan Koes
类目: Machine Learning (cs.LG); Biomolecules (q-bio.BM)
*备注: 26 pages, 5 figures, submitted to JMLR 2026
Abstract:A long standing challenge in computational chemistry and biophysics is efficiently sampling the Boltzmann distribution of molecules. Advances in generative modeling have been proposed to address the limitations of conventional sampling techniques by eliminating the computational cost of simulation. A promising direction is iteratively finetuning diffusion models along a temperature ladder whereby training data is generated via importance sampling during inference-time annealing. Unfortunately, these methods require computing a divergence over the score field to estimate importance weights, rendering them intractable for larger systems. Here we present scalable inference-time annealing (SITA), which retrains flow-based models to generate samples at progressively lower temperatures using an energy-based model to facilitate fast surrogate likelihoods. We demonstrate state-of-the-art performance on both Alanine Dipeptide and Alanine Tripeptide while avoiding costly divergence terms. Our code is available at: this https URL
[LG-10] Assign and Add: A Mechanistic Study of Compositional Arithmetic
链接: https://arxiv.org/abs/2605.31497
作者: Brady Exoo,Alberto Bietti,John Sous
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:Large language models are able to compose skills in order to perform complex tasks, many of which might not have been seen during training. The details of how exactly this composition occurs remain elusive. In this paper, we study a mechanism for compositional generalization in transformers by considering a simple controlled setting involving variable assignment and modular addition. By partitioning our training data into disjoint sets, we observe that small transformers are able to generalize to previously unseen combinations of variables and numbers. Our mechanistic analysis shows that the same ``modular addition’’ MLP module is used whether the inputs are given directly or indirectly through a separate variable assignment mechanism. We also analyze the training dynamics from an empirical lens, which reveals three phases of learning: first, modular addition is learned, then the structure required for variable assignment, and finally a refinement phase where the model generalizes to some hard sequences not seen in training. Finally, we provide a theoretical framework to explain how compositionality emerges from training dynamics. These results suggest that compositional generalization can be a natural consequence of the compositionality of internal mechanisms in~transformers.
[LG-11] Graphical einops: bridging tensor networks and computation graphs
链接: https://arxiv.org/abs/2605.31485
作者: Vincent Wang-Maścianica,Nikhil Khatri
类目: Machine Learning (cs.LG); Category Theory (math.CT)
*备注:
Abstract:Architecture diagrams are ubiquitous in deep learning, but they are usually only representational: the tensor-program identities they suggest are still proved by prose and tensor-axis manipulation. We introduce a formal graphical calculus for the structural fragment of tensor programming underlying einops, making such diagrams proof-enabling. Our calculus represents tensor axes as nested graded tubes around a base type. The tube boundary recovers the undirected tensor-network view of axes, while the directed interior retains the operational reading of computation graphs. The key rewrite is grade-naturality: sliding spectacles over tubes. Standard equivariance proofs become short diagrammatic derivations. We additionally demonstrate how our rewrite system may be applied to convert attention masks into pre-processing operations, recovering efficient implementations of sparse attention blocks.
[LG-12] Balanced LoRA: Removing Parameter Invariance to Accelerate Convergence ICML2026
链接: https://arxiv.org/abs/2605.31484
作者: Valérie Castin,Kimia Nadjahi,Pierre Ablin,Gabriel Peyré
类目: Machine Learning (cs.LG)
*备注: Accepted at ICML 2026
Abstract:Low-Rank Adaptation (LoRA) is the most widely adopted method for fine-tuning large language models. Notably, LoRA is inherently overparameterized: multiple pairs of low-rank factors can yield the same adapted weight matrix. We show–both theoretically and empirically–that these pairs exhibit significantly different condition numbers. As a result, converging to different loss minimizers directly impacts the convergence rate of LoRA. Building on this observation, we introduce Balanced Low-Rank Adaptation (BaLoRA), a variant of LoRA that projects iterates onto a balanced manifold. This manifold improves the conditioning of the loss landscape while preserving the adapted matrix. The projection step is computationally lightweight and integrates seamlessly into existing fine-tuning pipelines. Empirically, BaLoRA converges faster than standard LoRA and achieves superior performance across a range of fine-tuning tasks.
[LG-13] Flow map learning in nonlinear vector autoregressive models: influence of the feature-library structure on the training error
链接: https://arxiv.org/abs/2605.31438
作者: Markus Gross
类目: Machine Learning (cs.LG)
*备注: 35 pages, 12 figures
Abstract:Time series forecasting often requires learning nonlinear and time-delayed dependencies. A paradigmatic class of forecasting models are nonlinear vector autoregressive processes (NVAR), also known as next-generation reservoir computers (NG-RCs). These models approximate the Koopman operator on the space spanned by their explicit feature library. We consider the identifiability problem for learning Markovian nonlinear dynamical systems and show that the training error as a function of time resolution follows characteristic (pre-)asymptotic scaling laws. These laws depend on whether the feature library can represent the early Lie-series coefficients of the flow map (propagator) exactly or merely approximately. For dynamical systems governed by polynomial vector fields, we demonstrate the mechanism for NVAR/NG-RC models with monomial and Fourier feature libraries. We determine the dependence of the training error on the temporal resolution, the involved nonlinear degree, and the number of delay terms. While delay terms reduce the optimal one-step training error, they improve long-horizon forecasts only when the library provides sufficient nonlinearity. Thus, small training error coexists with weak generalization as the model class is mismatched to the true data-generating process. Numerical experiments on various chaotic dynamical systems confirm the theoretical predictions.
[LG-14] DG-CoLearn: An Efficient Collaborative Learning Framework for Dynamic Graphs
链接: https://arxiv.org/abs/2605.31427
作者: Ashley Hoi-Ting Au,Zikun Zhang,Ligang He,Qiang Ni
类目: Machine Learning (cs.LG); Distributed, Parallel, and Cluster Computing (cs.DC)
*备注:
Abstract:Dynamic graph learning (DGL) is essential for modelling evolving graph data, but existing methods suffer from significant computational overhead due to repeated full-snapshot retraining and are not well-suited for collaborative settings with partitioned data. In realistic graph systems, cross-partition edges are unavoidable, but direct sharing of graph structure between clients may violate privacy constraints. We propose DG-CoLearn, a client-oblivious collaborative dynamic graph learning framework built on incremental graph snapshot processing, which focuses computation on graph regions affected by temporal updates while preserving historical information through temporal modelling. This incremental design is consistently applied across the entire graph processing pipeline, including a server-mediated embedding exchange mechanism to enable accurate multi-hop message passing without exposing raw cross-client structural information. Extensive experiments demonstrate that DG-CoLearn achieves up to 33.8 \times speedup in training time and 27.4 \times reduction in communication overhead, while consistently improving predictive performance on both node classification (up to 13.36% F1 improvement) and link prediction (up to 8.27% MAP improvement) tasks. These results highlight the effectiveness of DG-CoLearn in bridging efficiency, scalability, and client-to-client structural privacy in collaborative dynamic graph learning.
[LG-15] Fixed Universal Transformers
链接: https://arxiv.org/abs/2605.31423
作者: Jingwen Liu,Alexandr Andoni,Daniel Hsu
类目: Machine Learning (cs.LG)
*备注:
Abstract:We introduce \emphuniversal transformers: fixed transformers that can simulate any transformer in a given class via a suitable input embedding. Analogous to a universal Turing machine, the input embedding encodes a description of the target model while all internal parameters remain fixed. We provide explicit sparse constructions achieving universality when the embedding dimension is sufficiently large, and further show that universality is generic: randomly initialized transformers are universal almost surely, which aligns with recent empirical results of Zhong and Andreas (2024). We empirically validate our theory on the algorithmic tasks of parenthesis balancing and multi-hop reasoning. Our results suggest that much of a transformer’s expressive power may reside in its input representation rather than its learned weights.
[LG-16] Constrained Multi-Objective Reinforcement Learning with Max-Min Criterion ICML2026
链接: https://arxiv.org/abs/2605.31388
作者: Giseung Park,Hyunyoung Nam,Woohyeon Byeon,Amir Leshem,Youngchul Sung
类目: Machine Learning (cs.LG)
*备注: Accepted to ICML 2026
Abstract:Multi-Objective Reinforcement Learning (MORL) extends standard RL by optimizing policies with respect to multiple, often conflicting, objectives. While max-min MORL has emerged as an effective approach for promoting fairness, its applicability remains limited, particularly when constraints must be incorporated. In this paper, we propose a MORL framework that integrates the max-min criterion with explicit constraint satisfaction. We establish a theoretical foundation for the proposed framework and validate the resulting algorithm through convergence analysis and experiments in tabular settings. We further demonstrate the practical relevance of our approach in simulated building thermal control, multi-objective locomotion control, and greenhouse-gas-emission-aware traffic management. Across these domains, our method effectively balances fairness and constraint satisfaction in multi-objective decision-making.
[LG-17] Softsign: Smooth Sign in Your Optimizer For Better Parameter Heterogeneity Handling
链接: https://arxiv.org/abs/2605.31371
作者: Dmitrii Feoktistov,Timofey Belinsky,Andrey Veprikov,Amir Zainullin,Aleksandr Beznosikov
类目: Machine Learning (cs.LG)
*备注: 9 pages, 3 tables, 4 Figures
Abstract:Sign-based and LMO-inspired optimizers have recently attracted substantial attention in deep learning due to their strong performance and low memory footprint. However, their fixed-magnitude updates can hurt terminal convergence: they decouple update mechanisms from gradient magnitudes and fail to account for parameter heterogeneity, often leading to oscillation rather than convergence. We propose SoftSignum, a smooth relaxation of sign-based optimization that replaces the hard sign map with a temperature-controlled soft-sign transformation, enabling a parameter-wise transition from sign-like updates to magnitude-sensitive SGD-like steps. We complement it with an adaptive quantile-based temperature schedule and extend the same principle to matrix-valued optimizers, obtaining SoftMuon. We also develop a generalized geometry-relaxation framework based on strongly convex regularizers and Fenchel conjugates, proving convergence in stochastic non-convex setting. Experiments on diverse deep learning tasks, including LLM pretraining, show that SoftSignum and SoftMuon consistently improve over their hard sign-based counterparts and standard AdamW.
[LG-18] Forgetting Has Neighbors: Localized Collateral Forgetting in Machine Unlearning
链接: https://arxiv.org/abs/2605.31317
作者: Polina Dolgova,Sebastian U. Stich
类目: Machine Learning (cs.LG)
*备注:
Abstract:Machine unlearning aims to remove the influence of selected training examples without full retraining. Standard evaluations often summarize unlearning quality with aggregate metrics, such as accuracy- and forgetting-based scores, which can hide localized failures. We study this failure mode at the example level by comparing the predictions of an unlearned model to those of the model retrained after deletion. We show that this pointwise discrepancy can be highly non-uniform: for gradient-ascent and random-labeling methods, with and without retain-set fine-tuning, it grows with geometric proximity to the forget set. We call this phenomenon localized collateral forgetting. Our analysis identifies a mechanism behind the effect: surrogate targets used during unlearning can be inconsistent with the local prediction structure induced by retraining, and this inconsistency propagates through shared representations to nearby examples. Motivated by this mechanism, we propose Local Teacher Distillation, a simple mitigation strategy that replaces random targets with soft labels from a small teacher trained only on retained neighbors of the forget set. On CIFAR-100 partial-class deletion, this local teacher brings the unlearned model substantially closer to retraining, especially near the forget set, while maintaining competitive aggregate unlearning metrics.
[LG-19] Graph Neural Networks Are Not Continuous Across Graph Resolutions
链接: https://arxiv.org/abs/2605.31315
作者: Christian Koke,Yuesong Shen,Abhishek Saroha,Marvin Eisenberger,Bastian Rieck,Michael Bronstein,Daniel Cremers
类目: Machine Learning (cs.LG)
*备注: arXiv admin note: text overlap with arXiv:2310.00431
Abstract:We show that contrary to conventional wisdom in the community, graph neural networks (GNNs) are not continuous with respect to all natural modes of graph convergence. As a result, GNNs may generate substantially different latent representations for graphs that are very similar. In particular they assign vastly different latent embeddings to graphs that represent the same underlying object at different resolution scales. We trace this failure of continuity back to a structural obstruction arising from commonly used information-propagation schemes. Building on this insight we then derive a principled modification to standard GNN architectures which equips models with continuity across scales. The proposed modification enables consistent integration of distinct resolutions and reliable generalization between them. We systematically validate our theoretical findings in a wide range of numerical experiments.
[LG-20] Non-Asymptotic Convergence of Stochastic Iterative Algorithms: A Lyapunov Framework
链接: https://arxiv.org/abs/2605.31309
作者: Zaiwei Chen,Siva Theja Maguluri
类目: Machine Learning (cs.LG); Probability (math.PR); Machine Learning (stat.ML)
*备注: 44 pages
Abstract:We survey Lyapunov-based techniques for the finite-time analysis of stochastic iterative algorithms, also known as stochastic approximation (SA) algorithms, for solving fixed-point equations \barF(x)=x , where the operator \barF(\cdot) can only be accessed through a noisy oracle. We first focus on the standard setting in which \barF(\cdot) is contractive with respect to some norm and the noise is i.i.d., and explain how generalized Moreau envelopes serve as universal Lyapunov functions, regardless of the underlying norm. We then show how this framework yields mean-square convergence guarantees and applies to stochastic gradient descent, linear SA, and value-based reinforcement learning algorithms such as Q-learning and temporal-difference learning. Finally, we discuss extensions to Markovian noise, seminorm-contractive operators, dissipative operators, and high-probability bounds, and conclude with open problems. The goal is to present a unified and self-contained roadmap for the finite-time analysis of SA and its applications, especially in reinforcement learning.
[LG-21] GETA: Generalized Encrypted Traffic Analysis
链接: https://arxiv.org/abs/2605.31277
作者: Ransika Gunasekara,Rahat Masood,Salil Kanhere
类目: Cryptography and Security (cs.CR); Machine Learning (cs.LG)
*备注:
Abstract:Traditional traffic analysis is being fundamentally challenged by the rapid adoption of encryption, tunnelling, and privacy-preserving protocols, which increasingly obscure packet payloads and limit the usefulness of Deep Packet Inspection (DPI). Although machine learning has advanced encrypted traffic analysis, existing approaches often remain tied to protocol-specific header features, depend on large labelled datasets, and degrade when deployed across heterogeneous network environments. We present GETA, a protocol-agnostic framework for encrypted traffic analysis that models network flows as multivariate time series using only traffic metadata, thereby avoiding reliance on packet payloads or header semantics. GETA combines meta-learning, embedding refinement, and self-attention to support few-shot adaptation to previously unseen domains with minimal labelled data. Across nine public datasets spanning application identification, VPN traffic classification, IoT device fingerprinting, and attack detection, GETA consistently outperforms state-of-the-art baselines. These results show that GETA offers a practical and generalisable foundation for robust traffic analysis in modern encrypted networks.
[LG-22] Learning Parametric Nitrogen Fertilizer Response Curves Using Neuro Symbolic Regression CEC
链接: https://arxiv.org/abs/2605.31276
作者: Giorgio Morales,John Sheppard
类目: Machine Learning (cs.LG)
*备注: Accepted at the Workshop on Symbolic Regression and Equation Discovery, part of the 2026 IEEE World Congress on Computational Intelligence (WCCI) and the IEEE Congress on Evolutionary Computation (CEC)
Abstract:Accurately modeling crop response to Nitrogen (N) fertilization is a fundamental challenge in precision agriculture, as it impacts both economic returns and environmental sustainability. Existing approaches either rely on predefined parametric forms or opaque machine learning models, limiting their ability to interpret or discover site-specific functional relationships from data. In this work, we propose a neuro symbolic regression (SR) approach to learn parametric N-response curves without assuming a predefined functional form. Our approach integrates a transformer-based Multi-Set Symbolic Skeleton Prediction strategy, enabling the discovery of shared functional structures across multiple subdomains or management zones (MZs). By constructing diverse input subsets and enforcing consistency across them, the method recovers robust symbolic skeletons that are subsequently fitted to observed data using a genetic algorithm. This framework was first evaluated on synthetic one-dimensional problems to assess its robustness under varying levels of epistemic uncertainty. The results demonstrate the ability of the proposed SR approach to recover correct expressions even in data-scarce regimes. In this work, we present the results of applying our method to real-world winter wheat data, learning distinct parametric N-response curves for different MZs within a field. The results show that the discovered expressions not only achieve lower fitting errors than traditional models such as quadratic-plateau and exponential functions, but also capture diverse functional behaviors across spatial regions. This demonstrates the potential that neuro SR has to enable the discovery of site-specific agronomic relationships and support informed decision-making in precision agriculture.
[LG-23] Survival Reinforcement Learning: Toward Scalable Self-Supervised RL
链接: https://arxiv.org/abs/2605.31273
作者: Franki Nguimatsia-Tiofack,Fabian Schramm,Théotime Le Hellard,Justin Carpentier
类目: Machine Learning (cs.LG)
*备注:
Abstract:While self-supervised Contrastive Reinforcement Learning (CRL) has shown remarkable depth-scaling capabilities, successfully using networks over 64 layers, scaled CRL still struggles with long-horizon goal-conditioned planning due to the uniformity-tolerance dilemma inherent in contrastive losses. We introduce Survival Reinforcement Learning (SRL), an online classification-based alternative that extends the survival value learning framework by maximizing the agent’s dwell time at target goals. SRL bypasses the structural constraints of CRL and mitigates the “bang-bang” control solutions inherent to survival frameworks, which often induce undesirable behavior in complex dynamical systems. Evaluated across diverse robotic benchmarks, scaled SRL matches state-of-the-art CRL on manipulation tasks and outperforms it by 2x to 8x on stable, long-horizon locomotion tasks. Our results provide strong additional evidence that classification-based methods may serve as a key primitive in the broader effort to scale reinforcement learning.
[LG-24] Algorithmic Recourse of In-Context Learning for Tabular Data ICML2026
链接: https://arxiv.org/abs/2605.31272
作者: Wenshuo Dong,Jiaming Zhang,Shaopneg Fu,Hongbin Lin,Di Wang,Lijie Hu
类目: Machine Learning (cs.LG)
*备注: Accepted by ICML 2026
Abstract:As predictive models are increasingly deployed in high-stakes settings such as credit approval, there is a growing need for post-hoc methods that provide recourse to affected individuals. Many such models operate on tabular data, where features correspond to real-world attributes. Recently, in-context learning (ICL) has enabled large language models to perform tabular prediction by conditioning on labeled examples at inference time, without explicit training. However, algorithmic recourse for tabular decision-making under ICL remains largely unexplored. In this work, we present the first study of algorithmic recourse for tabular data under ICL. We carry out a theoretical analysis, showing that recourse remains well-defined and bounded, and we characterize how recourse converges toward classical solutions as the context size increases. In practice, we propose a novel zeroth-order recourse framework, Adaptive Subspace Recourse for In-Context Learning (ASR-ICL), that efficiently generates actionable and sparse recourse for black-box ICL models. The proposed framework naturally extends to multi-class tabular tasks. Experiments across multiple real-world datasets and models demonstrate that ASR-ICL achieves recourse quality comparable to existing methods with fewer queries and empirically confirm the predicted convergence behavior, supporting our theoretical analysis.
[LG-25] Lightweight CNN-Based Anomaly Detection for High Voltage Converter Modulators in the Spallation Neutron Source
链接: https://arxiv.org/abs/2605.31259
作者: Alberto D. Cencillo,Leonardo Concepción,Julián Luengo,Isaac Triguero
类目: Machine Learning (cs.LG)
*备注: 21 pages, 8 figures
Abstract:Unscheduled trips of high-power pulsed converters are a leading source of downtime at large accelerator facilities. At the Spallation Neutron Source (SNS), the High Voltage Converter Modulators (HVCMs) are consistently the second-largest contributor to lost beam time. Each HVCM pulse is recorded across sensor channels spanning currents, voltages, and magnetic fluxes, whose mutual interactions encode the operating state of the system. Fault precursors do not manifest uniformly across these channels: depending on fault type, they may alter the temporal structure of individual signals, change the statistical dependencies among channels, or both. Existing deep-learning approaches typically process multi-channel signals with standard convolutional pipelines that entangle temporal and cross-channel operations from the first layer, giving the model no explicit mechanism to represent channel independence or structured inter-channel interaction. We hypothesise that architectural inductive bias, specifically the ordering of temporal filtering and cross-channel mixing, plays a central role in detection performance on this class of data. To test this, we vary the order in which these two operations are applied, and examine whether per-pulse adaptive channel reweighting further improves sensitivity. Evaluated on the public HVCM dataset across all four SNS subsystems (RFQ, DTL, CCL, SCL), our best variant achieves a pooled AUC-PR of 0.816 and AUC-ROC of 0.934, outperforming the state of the art on most subsystems and five of the six fault families. Ablations identify three dominant input channels and link per-fault-family performance to whether precursors manifest as amplitude shifts in individual channels or as subtler patterns requiring joint channel representations to surface.
[LG-26] Fraud Type Decomposition and the Observation-Mechanism Taxonomy:Class-Specific Detection Limits in Payment Networks
链接: https://arxiv.org/abs/2605.31257
作者: Gaurav Dhama
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注: 59 pages
Abstract:Fraud detection in payment networks relies on labels generated through heterogeneous and imperfect observation processes, yet existing approaches treat fraud as a homogeneous binary variable. We show that this assumption is structurally incorrect and leads to provable inefficiency. We introduce an observation-mechanism taxonomy that partitions fraud into five classes, each defined by a distinct censorship and labeling pipeline. We prove that estimating fraud rates separately by class and aggregating strictly dominates pooled estimation, with the efficiency gap characterized as a Jensen penalty arising from heterogeneous observation rates. For each class, we derive the binding theoretical constraint on detection, including endogenous label corruption, structural non-observability, and feature non-informativeness. These results establish that fraud detection is fundamentally a collection of distinct estimation problems, each governed by its own observation structure and detection limit.
[LG-27] oward Identifiable Sparse Autoencoders ICML
链接: https://arxiv.org/abs/2605.31245
作者: Walter Nelson,Theofanis Karaletsos,Francesco Locatello
类目: Machine Learning (cs.LG)
*备注: International Conference on Machine Learning (ICML) 2026
Abstract:Recently, sparse autoencoders (SAEs) have emerged as an attractive tool for interpreting and interacting with representations in practical neural networks. While it is common empirical folklore, we also show theoretically that SAEs are highly unstable: different training runs are likely to produce different concept dictionaries and sparse codes. We characterize the model properties that hinder the stability of real-world SAEs, and address each of these problems through minimal changes to the architecture and training procedure. Together, these changes yield two versions of an \textbfidentifiable SAE (iSAE), a variant of the standard TopK SAE with lower reconstruction error and improved stability. We explain this improvement theoretically by connecting SAEs with traditional dictionary learning approaches, and show that the dictionaries learned in practice satisfy an approximate restricted isometry condition, rendering the corresponding sparse codes in those models near-identifiable.
[LG-28] Spectral Reach: Understanding Neural Scaling as Progress into the Spectral Tail
链接: https://arxiv.org/abs/2605.31244
作者: Konstantin Nikolaou,Jonas Scheunemann,Sven Krippendorf,Samuel Tovey,Christian Holm
类目: Machine Learning (cs.LG); Computational Physics (physics.comp-ph)
*备注:
Abstract:Neural scaling laws describe predictable power-law relationships between model size, dataset size, compute, and performance. While these laws guide the development of modern foundation models, the mechanisms underpinning them remain poorly understood, in part due to the absence of scalable analysis tools. To close this gap, we introduce “spectral position”: a scalable measure of which eigenvalues of the empirical neural tangent kernel (eNTK) currently drive loss reduction. Applying this measure to scaling experiments, we find that spectral position decreases throughout training: learning shifts from dominant eigenmodes into the spectral tail. Larger models reach further into the tail than smaller models, revealing a size-dependent capacity we call “spectral reach”. This suggests why larger models achieve lower losses: they sustain learning on weak spectral signals inaccessible to smaller models. We further identify feature learning as a key enabler of spectral reach. It adaptively amplifies gradient magnitudes as learning advances, sustaining progress where frozen representations stall. This points to concrete interventions through architecture and optimizer design.
[LG-29] Bifurcated Remaining Useful Life Prediction: A Hybrid Approach for Realistic Uncertainty Characterization ALT
链接: https://arxiv.org/abs/2605.31241
作者: Xabier Belaunzaran,Antonio Nappa,Arkaitz Artetxe,Basilio Sierra
类目: Machine Learning (cs.LG)
*备注: Submitted to 9th European Conference of the Prognostics and Health Management Society 2026
Abstract:This study presents a novel hybrid prognostic framework for uncertainty-aware Remaining Useful Life (RUL) estimation in turbofan engines using the NASA C-MAPSS dataset. The framework employs a state-aware strategy that bifurcates the engines operational lifespan into “healthy” and “degraded” regimes. An LSTM-based autoencoder, trained strictly on nominal data (RUL 150 cycles), monitors reconstruction error to act as a robust state classifier. For the healthy regime, a Conditional Weibull Survival Analysis is used for Mean Residual Life estimation. For the degraded regime, a Probabilistic Neural Network with Monte Carlo Dropout captures both aleatoric and epistemic uncertainties. Rather than using rigid binary labels, a calibrated sigmoid function converts the autoencoders output into continuous state probabilities, dynamically weighting the final ensemble prediction. The primary strength of this framework is its generation of physically consistent uncertainty bands, yielding high-confidence predictions near end-of-life while accurately reflecting the inherent variance of early operation, providing a robust tool for risk-informed maintenance.
[LG-30] A holomorphic neural network framework for 3D boundary value problems governed by harmonic potentials
链接: https://arxiv.org/abs/2605.31231
作者: Enrico Ballini,Allan Peter Engsig-Karup,Tito Andriollo
类目: Numerical Analysis (math.NA); Machine Learning (cs.LG)
*备注:
Abstract:We present a neural-network-based framework for the solution of three-dimensional boundary value problems where the solution is expressible in terms of harmonic potentials. The approach leverages the Whittaker integral formula, which allows representing the solution through functions that are holomorphic with respect to a suitable complex variable. These functions are subsequently approximated using holomorphic neural networks, which guaranty fulfillment of the holomorphicity requirement. A key feature of the proposed formulation is that the governing partial differential equations (PDEs) are satisfied exactly by construction. Therefore, in contrast to standard physics-informed neural networks, no residual minimization of PDEs is required in the interior of the domain, and training is based exclusively on boundary collocation points. The method is validated against three-dimensional Laplace and linear elasticity problems, where, in the latter case, displacement and stress fields are expressed via the Papkovich-Neuber potentials. The numerical results show an accurate approximation of both scalar and vector fields, with errors remaining controlled throughout the domain. Overall, the work demonstrates that the incorporation of analytical structures into neural network architectures provides a natural and effective framework for the meshless approximation of three-dimensional boundary value problems while preserving the underlying properties of the governing equations.
[LG-31] Multivariate Distributional Reinforcement Learning Using Sliced Divergences
链接: https://arxiv.org/abs/2605.31222
作者: Baptiste Debes,Tinne Tuytelaars
类目: Machine Learning (cs.LG)
*备注:
Abstract:Distributional reinforcement learning (DRL) models the full return distribution rather than expectations, but extending it to multivariate settings remains challenging. Many common metrics do not naturally generalize beyond one dimension or lose computational tractability, and the multivariate case introduces additional difficulties such as general matrix discounting, for which no contraction results are available. We introduce Sliced Distributional Reinforcement Learning (SDRL), which lifts tractable one-dimensional divergences to multivariate return distributions via projections. We prove Bellman contraction for uniform slicing under shared scalar discounting, and introduce a maximum-slicing variant with contraction under general dense discount matrices. SDRL supports a broad class of base divergences; we analyze Wasserstein, Cramér, and Maximum Mean Discrepancy (MMD), and characterize which SDRL variants suit the standard single-sample Bellman update used in distributional RL. We evaluate SDRL on a toy chain problem and a gridworld image-based environment as well as a subset of Atari games.
[LG-32] Beyond Additive Decompositions: Interpretability Through Separability ICML2026
链接: https://arxiv.org/abs/2605.31200
作者: Jinyang Liu,Munir Eberhardt Hiabu
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注: To appear in Proceedings of the 43rd International Conference on Machine Learning (ICML 2026)
Abstract:Interpretable machine learning requires models that are accurate and structurally faithful to the this http URL explainability methods rely heavily on additive representations (e.g., Generalized Additive Models (GAMs), SHapley Additive exPlanations (SHAP), functional ANOVA), which can suffer from signal cancellation and off-support extrapolation in the presence of strong interactions. We propose Tensor Separation Learning (TSL), a regression model that learns a sum of rank-1 products of univariate per-feature functions via a stagewise greedy procedure with orthogonal refitting. By enforcing separability, TSL avoids the information loss inherent in additive projections caused by marginalizing higher-order interactions. The learned TSL model can be fully reconstructed from first-order partial dependence functions, up to constant factors. This stage-wise correspondence ensures that the resulting visualizations are faithful to the fitted components. We establish approximation-rate guarantees for functions with bounded mixed p -th order partial derivatives and demonstrate that TSL competes with black-box models on regression benchmarks.
[LG-33] Geometry-based Schrödinger Bridges for Trustworthy Multimodal Fusion ICML2026
链接: https://arxiv.org/abs/2605.31193
作者: Jiayu Xiong,Jing Wang,Qi Zhang,Wanlong Wang,Jun Xue
类目: Machine Learning (cs.LG)
*备注: ICML 2026 accepted paper
Abstract:Real-world multimodal systems must be robust against low-quality data, such as sensor noise, incomplete multimodal data and conflicting inputs. However, existing trustworthy fusion methods rely on the model’s own prediction confidence to judge data quality. This creates a circular dependency: when a model is confident but wrong, these methods fail to detect the error. To break this loop, we propose Geometry-based Multimodal Fusion (GMF). Instead of relying on predictions, we evaluate reliability by measuring how much transport correction the input needs in latent space. We implement Diffusion Schrödinger Bridge transport with Rectified Flow, where the squared initial velocity gives an efficient learned correction score. Valid data has low squared velocity magnitude, while noisy, incomplete data or conflicting data requires stronger transport correction. This geometry-based reliability signal acts as an independent judge, effectively flagging unreliable inputs even when the classifier is fooled. Extensive experiments demonstrate that GMF significantly improves robustness against severe sensor noise and semantic conflicts compared to confidence-based baselines.
[LG-34] FlagGAM: Rule-Based Generalized Additive Modeling for Explainable Tabular Prediction
链接: https://arxiv.org/abs/2605.31189
作者: Zijie Zhao,Roy E. Welsch
类目: Machine Learning (cs.LG)
*备注:
Abstract:Tabular prediction in high-stakes domains requires models that are accurate, transparent, and robust to imperfect inputs. We propose FlagGAM, a rule-defined basis framework that separates feature-level rule construction from prediction. A Flag Core Module converts numerical and categorical variables into sparse, human-readable univariate bases, including threshold flags, category-level flags, tail-deviation bases, and categorical step functions; a default additive head then combines these bases as a restricted GAM-style predictor. Rather than reducing triggered rules to compact count summaries, FlagGAM retains a sparse rule-basis matrix that supports mixed-type classification and regression, feature-specific weighting, and optional flexible prediction heads. Across tabular benchmarks, default FlagGAM remains close to EBM in transparent additive mode, improves substantially over ridge regression on mixed-type regression, and shows smaller AUROC degradation than common baselines under missing and noisy perturbations. Flexible heads further improve accuracy and approach strong tree-based baselines, with the caveat that the resulting model should be interpreted as a rule-basis representation followed by a nonlinear predictor rather than as a fully additive GAM. Overall, FlagGAM provides a practical middle ground for tabular settings that require competitive accuracy, communicable rules, and robustness to imperfect inputs.
[LG-35] How well does Classification Accuracy capture Concept Drift Detection Quality? An overview of Concept Drift Detection evaluation
链接: https://arxiv.org/abs/2605.31186
作者: Joanna Komorniczak
类目: Machine Learning (cs.LG)
*备注:
Abstract:Data streams are nowadays among the most frequently analyzed data structures, with the concept drift posing a major challenge encountered by processing systems. Despite the proposition of numerous solutions to counteract the accuracy degeneration due to concept drift, the scientific community has not yet established a unified framework for evaluating the concept drift detection task. Existing research often relies on classification quality metrics, but these can be affected by multiple factors and may not reliably reflect drift detection quality. In this work, we present an in-depth overview of the relationship between metrics for quantifying drift detection quality and classification performance in synthetic nonstationary data streams. The proposed research studies eight drift detection quality metrics in relation to the classifier’s performance across seven synthetic data stream generation tools, additionally considering drift dynamics as a factor. The studies aim to identify the most informative set of drift detection quality metrics and provide a deep understanding of the method’s evaluation.
[LG-36] Retriever Portfolios: A Principled Approach to Adaptive RAG ICML2026
链接: https://arxiv.org/abs/2605.31176
作者: Miltiadis Stouras,Vincent Cohen-Addad,Silvio Lattanzi,Ola Svensson
类目: Machine Learning (cs.LG); Data Structures and Algorithms (cs.DS)
*备注: Accepted at ICML 2026. Code available at: this https URL
Abstract:Retrieval-augmented generation (RAG) systems typically rely on a single retriever and a single set of hyperparameters, despite facing highly heterogeneous queries that range from simple factoid questions to complex multi-hop reasoning. We propose a method that automatically selects a small, diverse subset of retrievers (a portfolio) from a large pool of candidates, to cover different regions of the target query distribution. We formalize this setting via an expected best-of- k objective over the query distribution and show that it admits an efficient portfolio construction algorithm with near-optimal guarantees. Across multiple QA benchmarks, our learned portfolios and router pipeline consistently outperform single-retriever and naive multi-retriever baselines on both retrieval metrics and answer quality. In addition, compared to inference-time hyperparameter tuning approaches, fixed portfolios enable parallel retrieval and LLM calls, achieving comparable (and sometimes better) accuracy with substantially lower latency and token cost.
[LG-37] Convergence of Two-Timescale Markovian Stochastic Approximations with Applications in Reinforcement Learning ICML2026
链接: https://arxiv.org/abs/2605.31172
作者: Vagul Mahadevan,Claire Chen,Shuze Daniel Liu,Shangtong Zhang
类目: Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注: ICML 2026
Abstract:This work studies the convergence of two-timescale stochastic approximations (SA), a class of iterative algorithms that update two sets of parameters in fast and slow timescales respectively. Notable examples of two-timescale SA in reinforcement learning (RL) include temporal difference learning with gradient correction (TDC) and actor-critic methods. Previously, the stability (i.e., boundedness) and convergence of two-timescale SA were only established under i.i.d. noise. This work instead establishes the stability and convergence of two-timescale SA under Markovian noise, a setup that is more realistic in RL. Notably, we do not need to use any projection operator and the noise does not need to live in a compact space. Our key technical novelty is to control the fast timescale parameter with the running max of the slow timescale parameter, instead of with the current slow timescale parameter, as most prior works do. As a key application, we establish the first almost sure convergence of TDC with eligibility traces under off-policy learning with linear function approximation.
[LG-38] abCausal: Pretraining Across Causal Environments for Tabular Causal Discovery
链接: https://arxiv.org/abs/2605.31156
作者: Zi-Rong Li,Si-Yang Liu,Tian-Zuo Wang,Han-Jia Ye
类目: Machine Learning (cs.LG)
*备注:
Abstract:Causal discovery aims to recover directed causal relations from observational and interventional data, providing a basis for mechanistic understanding and reliable decision-making. Causal discovery foundation models (CDFMs) seek to amortize this problem by mapping a dataset directly to a causal graph in a single forward pass, avoiding per-dataset testing, search, or optimization. However, existing CDFMs remain limited, often failing to consistently match strong classical methods, and we find that a key bottleneck is how causal pretraining tasks are constructed. Based on this observation, we propose TabCausal, a data-driven CDFM trained with broad causal pretraining over diverse graph priors, structural mechanisms, noise models, dimensions, sample sizes, and intervention regimes. A dynamic task construction strategy composes these causal environments into varied discovery tasks, enabling more transferable structural learning from observational and mixed-interventional data. On large-scale synthetic benchmarks, TabCausal achieves better macro-averaged performance than a diverse set of causal discovery baselines. To further bridge abstract synthetic generators and realistic causal reasoning scenarios, we introduce a protocol-guided and LLM-audited semantic causal environment benchmark, where domain-grounded SCMs generate interpretable observational and interventional datasets for out-of-distribution analysis. Across both synthetic and semantic environments, TabCausal demonstrates robust structure recovery, especially under interventional evidence, highlighting broad causal pretraining as a key ingredient for transferable amortized causal discovery.
[LG-39] Learning Hyperspherical Time-Frequency Representations for Time-Series Out-of-Distribution Detection IJCAI ECAI2026
链接: https://arxiv.org/abs/2605.31155
作者: Willian T. Lunardi,Samridha Shrestha,Martin Andreoni
类目: Machine Learning (cs.LG)
*备注: 14 pages, 2 figures, 4 tables, accepted at IJCAI-ECAI 2026
Abstract:Out-of-distribution (OOD) detection for time-series data remains comparatively underexplored compared to vision and language, with a limited principled understanding of how supervised time-series representations can be leveraged for reliable detection under distributional shifts. This work formulates time-series OOD detection as representation learning with hyperspherical embeddings, where class-conditional structure is induced by a von Mises-Fisher (vMF) likelihood-based objective on the unit sphere. The learned representation combines time- and frequency-domain views of the input signal via domain-specific encoders, integrating them into a joint embedding space for OOD detection. Detection uses distance-based scores over the learned embeddings, including k-nearest neighbors (k-NN) and Mahalanobis scores. We evaluate the approach at scale on the complete UCR and UEA time-series archives under a cross-dataset protocol. Empirical results show consistent improvements under both k-NN and Mahalanobis scoring over strong contrastive learning and post-hoc baselines in the same setting. Code is available at this https URL.
[LG-40] Generalizing Multi-Scale Time-Series Modeling with a Single Operator ICML2026
链接: https://arxiv.org/abs/2605.31129
作者: Cheonwoo Lee,Dooho Lee,Doyun Choi,Jaemin Yoo
类目: Machine Learning (cs.LG)
*备注: Accepted at ICML 2026
Abstract:Multi-scale modeling has emerged as an effective design principle for time-series forecasting by capturing temporal dynamics at multiple resolutions. As no principled foundation has been established in the literature, we unify existing scaling methods into a scaling operator family, revealing a fundamental limitation of existing approaches: reliance on fixed and discrete scaling. To address this limitation, we propose SiGMA (Single Generalized Multi-scale Architecture), which enables distance-aware scaling via the learnable discrete Gaussian (LDG) kernel grounded in scale-space theory. We evaluate SiGMA comprehensively on long- and short-term forecasting benchmarks against state-of-the-art multi-scale baselines. SiGMA outperforms all competitors on both tasks, especially achieving the best performance in 13 out of 16 long-term evaluation settings. Beyond accuracy, SiGMA significantly improves training speed by up to 5.3 times and reduces memory consumption by up to 3.8 times over the strongest competitors. Code is available at this https URL.
[LG-41] Scalable Bayesian Inference for Nonlinear Conservation Laws
链接: https://arxiv.org/abs/2605.31127
作者: Tim Weiland,Philipp Hennig
类目: Machine Learning (cs.LG); Numerical Analysis (math.NA)
*备注: 27 pages, 13 figures, 3 tables
Abstract:Nonlinear conservation laws are at the heart of many of the most important dynamical systems in science and engineering. In practical applications, such systems are often subject to various sources of uncertainty, e.g. due to sparse or noisy measurements. Inferring physical quantities and fields of interest then becomes an ill-posed problem which both classical numerical methods and modern deep learning-based methods struggle to treat appropriately. Recent work has framed classical numerical methods as Bayesian inference under Gaussian process priors, resulting in a physics-aware treatment of uncertainties. Following this line of work, we develop a novel numerically conservative method for uncertainty-aware simulations of nonlinear conservation laws. We use recent sparse approximation techniques to scale up to large-scale forward and inverse problems. For forward simulation, we inherit the accuracy of classical solvers while providing structured uncertainty quantification. On inverse problems, we recover posteriors over nonparametric source fields in seconds – outperforming neural baselines that take minutes to produce a less accurate point estimate.
[LG-42] Dont Fool Me Twice: Adapting to Adversity in the Wild with Experience-Driven Reasoning
链接: https://arxiv.org/abs/2605.31119
作者: Navin Sriram Ravie,Andrew Jong,Krrish Jain,John Liu,Omar Alama,Bijo Sebastian,Sebastian Scherer
类目: Robotics (cs.RO); Machine Learning (cs.LG)
*备注:
Abstract:In robotics, dangers and adversity modes are often embodiment-specific and relative to each agent. A frontier of autonomous mobile robotics is to enable agents to operate effectively in the wild in unseen unstructured environments. A significant challenge in unseen unstructured environments is that it may not be possible to predict all the dangers to the specific robot. Although recent work has used large foundation vision-language models (VLMs) to preemptively predict an exhaustive list of common-sense dangers, it remains difficult to capture possible interaction and embodiment-dependent adversities. We propose a continual learning framework for a mobile embodied agent to learn online from disturbances and attribute anomalous behaviours to causes through semantics, enabling better prediction and planning of the world in the future. Our framework, “Don’t Fool Me Twice”, first observes disturbances and describes their effects on the robot; this description is augmented with visual context to query a VLM to predict possible causes; the local disturbance is characterized using kernel regression, which allows for efficient, few-shot modeling of transient anomalies. We leverage semantic voxel-centric modeling to estimate epistemic uncertainty, enabling richer downstream recovery by treating interaction-driven disturbances as learnable spatial behaviors. We present four hypotheses and validate them in simulation and on hardware across embodiments and adversity modes.
[LG-43] Subspace-Decomposed JEPAs: Disentangling Progression and Content in Latent World Models
链接: https://arxiv.org/abs/2605.31111
作者: Lucas Thil,Jesse Read,Rim Kaddah,Guillaume Doquet
类目: Machine Learning (cs.LG)
*备注:
Abstract:Joint-Embedding Predictive Architectures (JEPAs) learn compact latent world models by predicting future embeddings, but no single coordinate of the latent is designated to encode task progression. We carve the JEPA latent into two orthogonal subspaces with disjoint roles: a low-dimensional progression subspace shaped by a cosine-margin triplet loss, and a high-dimensional content subspace regularised by the existing SIGReg objective of LeWM. We prove that the two anti-collapse forces act on disjoint coordinates, so they compose additively rather than competing on the same dimensions. Our method, SD-JEPA improves over the LeWM baseline on the majority of its control benchmarks at matched compute, and outperforms the strongest non-LeWM JEPA baseline on Push-T; a subspace-ablation falsifier confirms the split is the load-bearing ingredient. Beyond planning, the resulting 1-D angular progression coordinate functions as a scene-aware compass on the latent. It advances with task progress, regresses when the agent backtracks, and under controlled perturbations both spikes and relocalises to a semantically appropriate new task-phase sector, separating the moment of surprise from its meaning in a way that prediction-error scalars cannot. Three quantitative tests back this up: |\Delta\theta_t| outperforms the standard latent-prediction-error surprise at localising semantic events on 40 held-out cube episodes by up to +0.18 pooled AUROC (97.5% per-episode win rate at \pm 1 -step tolerance); a within-episode linear probe across all four environments (40 episodes per env) shows the 8-dimensional progression subspace (4.2% of the latent) explains 72-95% of task-progress variance…
[LG-44] Riemannian Diffusion Models on General Manifolds via Physics-Informed Neural Networks
链接: https://arxiv.org/abs/2605.31106
作者: Gyeonghoon Ko,Juho Lee
类目: Machine Learning (cs.LG)
*备注:
Abstract:Riemannian diffusion models generalize score-based generative modeling to manifold-supported data via stochastic diffusion equations on the manifold. However, training requires sampling from and differentiating the manifold heat kernel, which is rarely available in closed form beyond a few highly symmetric manifolds. We propose a general approach that approximates the heat kernel by directly solving the manifold heat equation with a physics-informed neural network (PINN). Given an explicit manifold specification, we choose a coordinate system, derive the corresponding heat (Fokker–Planck) equation and a short-time asymptotic approximation, and then train a PINN to learn the log heat kernel. The resulting surrogate enables both forward noising (heat-kernel sampling) and conditional-score evaluation for denoising score matching. We demonstrate the method on diverse manifolds including S^2 , SO(3) , \mathrmSPD(n) , and permutation-quotiented point clouds.
[LG-45] Learning to Bid in FCR Markets: A Best-of-Both-Worlds Approach
链接: https://arxiv.org/abs/2605.31070
作者: Marius Potfer,Cheng Wan,Pierre Gruet
类目: Machine Learning (cs.LG); Computer Science and Game Theory (cs.GT)
*备注: Algorithms and data available at this https URL
Abstract:Bidding in the European Frequency Containment Reserve (FCR) market is challenging for flexibility providers because competing offers are hidden and bidders observe only partial feedback form the market, such as, clearing price and awarded quantity. For a participant active in a single country, we show that the multi-country FCR clearing problem can be recast as a repeated multi-unit uniform-price auction against an endogenous vector of opposing bids. This reformulation yields an online learning problem and allows us to adapt a Best-of-Both-Worlds combinatorial semi-bandit algorithm implementable from this standard market feedback. The resulting bidder achieves logarithmic pseudo-regret in stochastic environments and \mathcalO(\sqrtT) regret in adversarial ones. Synthetic experiments confirm the expected scaling, and backtests on historical European FCR data show competitive performance in practice: the method performs especially well on stable products, while EXP3-type baselines can be safer under stronger non-stationarity. Overall, the results show that learning-based bidding in FCR markets is theoretically grounded and practically useful when the learning rule matches product-level market stability.
[LG-46] Best-Arm Identification-Based Trust Region Selection for Bayesian Optimization on Multimodal Functions
链接: https://arxiv.org/abs/2605.31050
作者: Nobuo Namura,Sho Takemori
类目: Machine Learning (cs.LG)
*备注: 19 pages, 13 figures
Abstract:Gaussian process-based Bayesian optimization (BO) is a popular approach for expensive black-box optimization, but its performance often degrades on complex multimodal or high-dimensional problems. Trust region-based BO mitigates this issue by focusing on local regions, and recent studies suggest that selecting an effective region can be formulated as a multi-armed bandit problem. We propose a trajectory-aware framework that integrates best-arm identification (BAI) with trust region-based BO to efficiently solve multimodal optimization problems. Our method extrapolates the optimization trajectories of multiple locally initialized optimizers to predict their final performance and progressively eliminates suboptimal candidates via BAI. We theoretically show that the proposed BAI-guided BO converges faster to the global optimum than conventional BO under mild assumptions, and demonstrate its effectiveness through extensive experiments on synthetic and real-world benchmarks.
[LG-47] he Challenges of Using Reinforcement Learning for Controlling Industrial Energy Systems
链接: https://arxiv.org/abs/2605.31044
作者: Tobias Lademann,Théo Vincent,Jan Peters,Matthias Weigold
类目: Machine Learning (cs.LG)
*备注: Submitted to Finding the Frame Workshop at RLC 2026
Abstract:Reinforcement learning has shown promising results for optimizing the control of industrial energy systems, yet most existing studies remain limited to the application in simulation environments. We investigate the challenges of deploying reinforcement learning in a real-world industrial energy system, considering a thermal heating network as a use case. We formulate the task as a Markov Decision Process and systematically analyze the associated challenges along the structure of the formal description, including partial observability, action space design, reward design, and the simulation-to-reality gap. The challenges are grounded in an existing real-world deployment, where reinforcement learning achieves operational stability but shows a significant performance gap compared to simulation.
[LG-48] UniRTL: Unifying Code and Graph for Robust RTL Representation Learning ICML2026
链接: https://arxiv.org/abs/2605.31040
作者: Yi Liu,Hongji Zhang,Lei Chen,Mingxuan Yuan,Qiang Xu
类目: Machine Learning (cs.LG)
*备注: Forty-Third International Conference on Machine Learning (ICML 2026)
Abstract:Developing effective representations for register transfer level (RTL) designs is crucial for accelerating the hardware design workflow. Existing approaches, however, typically rely on a single data modality, either the RTL code or its associated graph-based representation, limiting the expressiveness and generalization ability of the learned representations. For RTL, the control data flow graph (CDFG) offers a comprehensive structural representation that preserves complete information, while the code modality explicitly encodes semantic and functional information. We argue that integrating these complementary modalities is essential for a thorough understanding of RTL designs. To this end, we propose UniRTL, a multimodal pretraining framework that learns unified RTL representations by jointly leveraging code and CDFG. UniRTL achieves fine-grained alignment between code and graph through mutual masked modeling and employs a hierarchical training strategy that incorporates a pretrained graph-aware tokenizer and staged alignment of text (i.e., functional summary) and code prior to graph integration. We evaluate UniRTL on two downstream tasks, performance prediction and code retrieval, under multiple settings. Experimental results show that UniRTL consistently outperforms prior methods, establishing it as a more robust and powerful foundation for advancing hardware design automation.
[LG-49] Model Monotonicity in Autobidding Auctions: When Do Better Predictions Lead to Better Outcomes?
链接: https://arxiv.org/abs/2605.31036
作者: Ashwinkumar Badanidiyuru
类目: Computer Science and Game Theory (cs.GT); Machine Learning (cs.LG)
*备注:
Abstract:Online advertising platforms rely on machine learning models to predict click-through rates (pCTR) and conversion rates (pCVR) for auction mechanisms. We introduce a novel framework to study the interaction between recommender system model quality, auction format, and autobidder behavior. We formalize when model improvements – defined via a refinement relation inspired by filtrations in probability theory – lead to improvements in platform-level Evaluation Criteria Metrics (ECM) such as revenue, welfare, or liquid welfare. Our main contributions are: (1) a formal definition of model improvement based on cluster refinement, and (2) a systematic characterization of ECM monotonicity across different combinations of bidder types (tCPA, max-CPA), auction formats (first-price, second-price, VCG), and budget constraints. We show that first-price auctions with uniform bidding guarantee revenue monotonicity for tCPA bidders without budgets (via Jensen’s inequality), while second-price auctions and budget constraints can break this property. We provide full numerical constructions for the non-monotonicity results. Our findings have practical implications for advertising platforms seeking to align model improvements with business outcomes.
[LG-50] Multi-Scale Separable Fourier Neural Networks for Solving High-Frequency PDEs
链接: https://arxiv.org/abs/2605.31027
作者: Qihong Yang,Qiaolin He
类目: Machine Learning (cs.LG)
*备注: 51 pages, 27 figures
Abstract:We propose a novel neural network architecture, termed Multi-Scale Separable Fourier Neural Networks (MS-SFNN), for the accurate and efficient solution of linear and nonlinear high-frequency partial differential equations (PDEs). MS-SFNN exploits a separable representation: given a d -dimensional input, it employs d independent subnetworks – each acting on a single coordinate – and constructs basis functions via element-wise multiplication of their outputs. The PDE solution is approximated as a linear combination of these basis functions, with coefficients determined by least squares. Critically, all network weights and biases are randomly initialized once, from a uniform distribution with unit variance, and remain fixed thereafter. To enhance expressivity, a tunable scaling factor is introduced in each subnetwork to modulate the frequency content of the resulting basis functions. Fourier features are explicitly embedded through cosine activations, endowing the method with strong spectral approximation capabilities. To mitigate the memory bottleneck associated with dense collocation in high-frequency or three-dimensional problems, we replace automatic differentiation with analytically derived basis function derivatives and develop a memory-efficient batched QR decomposition algorithm for solving large-scale least-squares systems. Numerical experiments demonstrate that MS-SFNN achieves unprecedented accuracy across a range of challenging PDEs, significantly outperforming state-of-the-art methods such as Physics-Informed Neural Networks (PINN) and Separated-Variable Spectral Neural Networks (SV-SNN).
[LG-51] Augmented Lagrangian Predictive Coding
链接: https://arxiv.org/abs/2605.31022
作者: Jeffrey Seely,Julian Gould
类目: Machine Learning (cs.LG)
*备注: 22 pages, 10 figures
Abstract:Predictive coding (PC) is a local-learning alternative to backpropagation (BP), training deep networks via local energy-minimization dynamics rather than a global backward pass. We introduce Augmented Lagrangian Predictive Coding (PC-ALM), which maintains PC’s inference budget but aligns each weight update toward BP by accumulating per-layer constraint errors into a layer-local Lagrange multiplier. In linear PC networks, PC-ALM converges to an equilibrium with exact BP gradients distributed across the network via only layer-local updates. We analyze PC-ALM in nonlinear PC networks up to depth 128 and show that it matches BP performance across all width-depth regimes, notably in deep narrow networks where PC underperforms. PC-ALM introduces recurrent dynamics in each layer’s activations. Compared to PC’s heat flow on a scalar energy, PC-ALM dynamics are driven by dual ascent on the augmented Lagrangian. We observe “ballistic” credit propagation across very deep networks, with credit signals evenly distributed across layers, compared to PC’s slow, diffusive credit propagation. Beyond the algorithm itself, the augmented Lagrangian framework offers a generalization of PC, and may yield insights into how distributed systems could compute and propagate BP-like credit signals through purely local dynamics.
[LG-52] An Efficient and Scalable Graph Condensation with Structure-Preserving
链接: https://arxiv.org/abs/2605.31016
作者: Yulin Hu,Fuyan Ou,Ye Yuan
类目: Machine Learning (cs.LG)
*备注:
Abstract:Graph condensation (GC) is pivotal for enabling Graph Neural Networks (GNNs) deployment in resource-constrained scenarios by compressing large-scale graphs into compact synthetic counterparts. Existing GC methods commonly suffer from computational inefficiency due to coupled optimization as well as encountering poor generalization across GNN architectures. To address these challenges, this study proposes an Efficient and Scalable Graph Condensation with Structure-Preserving (SP-ESGC), which possesses a decoupled design that separates node condensation from graph structure generation. Specifically, it first employs heat kernel feature propagation to generate node representation via spectral graph theory-inspired diffusion. Further, a novel hybrid clustering strategy is designed to extracts discriminative intra-class centroids from the node representation. Finally, a pre-trained edge predictor infers transferable structural patterns from the original graph, ensuring accurate synthetic graph generation. Extensive experiments on real-world graph datasets demonstrate that the proposed SP-ESGC implementes a precise GC with significantly high computational efficiency. Moreover, SP-ESGC also generalizes well across diverse GNN architectures.
[LG-53] SDM-Q: Cost-Aware Staged Decision-Making for Multi-Omics Classification with Deep Q-Learning
链接: https://arxiv.org/abs/2605.31014
作者: Nan Mu,Xiaoyang Fan,Chen Zhao
类目: Machine Learning (cs.LG)
*备注:
Abstract:Multi-omics data provide complementary molecular characterizations of disease phenotypes and play an important role in disease diagnosis and subtype classification in precision medicine. However, acquiring complete multi-omics profiles is expensive and time-consuming, while most existing deep learning methods assume full modality availability during inference, resulting in substantial redundancy and limited practicality in clinical settings. To address this issue, we propose SDM-Q, a reinforcement learning framework for adaptive and cost-aware multi-omics classification. Specifically, multi-omics diagnosis is reformulated as a finite-horizon sequential decision problem, where the currently acquired omics modalities define the diagnostic state at each stage. An action–value function determines whether to acquire an additional modality or terminate the decision process and output the final prediction. To balance diagnostic utility and acquisition cost, the reward is defined only at the terminal stage and jointly determined by classification correctness and cumulative modality acquisition cost. A backward stage-wise optimization strategy is introduced to improve policy consistency and training stability. Experiments on four public multi-omics datasets, including ROSMAP, LGG, BRCA, and KIPAN, demonstrate that SDM-Q effectively reduces redundant modality acquisition while maintaining competitive classification performance compared with methods using complete multi-omics inputs. In the BRCA and KIPAN datasets, more than 99% and 95% of subjects, respectively, achieve accurate classification using only a single omics modality, while the average number of acquired modalities remains below two for ROSMAP and LGG. These results suggest that cost-aware sequential decision-making provides an effective paradigm for improving the efficiency of precision medicine workflows.
[LG-54] Physics-Informed Coarsening for Multigrid Graph Neural Surrogates ICML2026
链接: https://arxiv.org/abs/2605.31013
作者: Amir Bazzi,David Cardinaux,Ramy Nemer,Jose Alaves,Arjun Kalkur Matpadi Raghavendra,Elie Hachem
类目: Machine Learning (cs.LG)
*备注: Accepted at ICML 2026. 16 pages, 5 figures
Abstract:Learning-based surrogates for partial differential equations have recently matched the accuracy of classical solvers while achieving orders-of-magnitude speedups, predominantly in fluid settings and structured geometries. In contrast, robust surrogates for deformable solids remain underexplored, despite the presence of nonlinear elasticity, plasticity, and transient behavior that challenge standard architectures. We introduce a multigrid graph neural network for solid mechanics that couples an encoder-processor-decoder backbone with a physics-informed coarsening strategy. Instead of downsampling via geometric heuristics, our method scores nodes using a residual-based measure of local physical activity and preferentially retains regions of high strain or stress concentration, allocating multiscale capacity where it is most needed. This preserves long-range interactions through hierarchical message passing while improving stability over long rollouts. We evaluate on multiple datasets covering linear, nonlinear, and transient regimes, and observe consistent gains in accuracy and rollout stability compared to standard sampling baselines. Our results highlight the importance of physics-informed coarsening for scalable surrogate modeling in solid mechanics.
[LG-55] Learning Multi-Agent Coordination via Sheaf-ADMM ICML2026
链接: https://arxiv.org/abs/2605.31005
作者: Jeffrey Seely,Bartłomiej Cupiał,Llion Jones
类目: Machine Learning (cs.LG)
*备注: 17 pages, 8 figures, 6 tables. Accepted at ICML 2026
Abstract:We present a differentiable optimization framework for multi-agent coordination. An input is decomposed into overlapping local views, each processed by an agent that solves a convex subproblem parameterized by a neural encoder. Agents coordinate through the Alternating Direction Method of Multipliers (ADMM) with inter-agent constraints specified by a cellular sheaf. The sheaf specifies which aspects of neighboring solutions must agree, allowing for heterogeneous notions of global consensus. Backpropagating through the unrolled optimization jointly trains all components of the multi-agent system. We evaluate on maze pathfinding, image classification, and Sudoku, where agents with individually insufficient local views learn to coordinate to produce correct global outputs. On MNIST, the local-view decomposition yields improved robustness to distribution shifts relative to a standard CNN. On Sudoku, the optimization-derived structure yields markedly higher solve rates than parameter-matched MPNN baselines. Finally, the ADMM structure exposes distinct primal, consensus, and dual state variables, opening the coordination dynamics to direct analysis and intervention – a property unavailable in standard message-passing architectures.
[LG-56] HetCCL: Enabling Collective Communication For Mixed-Vendor Heterogeneous Clusters
链接: https://arxiv.org/abs/2605.31000
作者: Yuejie Wang,Tao Chang,Yuanyuan Zhao,Yulong Ao,Zeyu Gu,Zhiyu Li,Yanmin Jia,Yan Zhang,Mingjun Zhang,He Liu,Yongzhe He,Yonghua Lin,Guyue Liu
类目: Networking and Internet Architecture (cs.NI); Machine Learning (cs.LG)
*备注:
Abstract:Training Large Language Models (LLMs) on heterogeneous clusters presents significant challenges for collective communication, as hardware from multiple vendors introduces diverse network and computational characteristics. Existing collective communication frameworks (e.g., NCCL, RCCL) designed for homogeneous environments fail to address mixed-hardware setups, while communication libraries with heterogeneous support (e.g., Gloo, OpenMPI) incur heavy overhead in the data path. This paper presents HetCCL, a framework that enables heterogeneous collective communication by efficient P2P transport across heterogeneous devices (e.g., GPUs), eliminating the host-device memory copy overhead while offloading the control to the CPUs. For combining collectives (e.g., AllReduce, ReduceScatter), HetCCL introduces a border-communicator mechanism that achieves vendor independence by using the intrinsic reduction in the combining collectives in vendor collective communication libraries. With efficient heterogeneous P2P transport and portable reduction mechanism, HetCCL proposes a hierarchical topology abstraction for heterogeneous clusters, dissecting collective communication into cluster-level primitives that guarantee optimal cross-cluster data transfer volume and optimal bandwidth utilization. We implement HetCCL with 4 different vendor support and evaluate it in 4 heterogeneous settings with benchmarks and end-to-end LLM tasks. Our evaluation shows that HetCCL achieves 17-19x higher bandwidth than Gloo in heterogeneous communications, and speeds up end-to-end training by up to 16.9% in the per-step-time. Subjects: Networking and Internet Architecture (cs.NI); Machine Learning (cs.LG) Cite as: arXiv:2605.31000 [cs.NI] (or arXiv:2605.31000v1 [cs.NI] for this version) https://doi.org/10.48550/arXiv.2605.31000 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-57] Eigenvectors of Experts are Training-free Non-collapsing Routers
链接: https://arxiv.org/abs/2605.30992
作者: Giang Do,Hung Le,Truyen Tran
类目: Machine Learning (cs.LG)
*备注: 24 pages
Abstract:Sparse Mixture of Experts (SMoE) architectures improve the training efficiency of Large Language Models (LLMs) by routing input tokens to a selected subset of specialized experts. Despite their remarkable success, both training and inference in SMoE models suffer from the expert collapse issue (Chi et al., 2022), which degrades model performance. Prior studies primarily focus on improving the router; however, such methods rely on training from scratch or fine-tuning, which requires high computational and data-processing costs. Furthermore, we demonstrate that, despite these efforts, the issue persists when advancing well-pretrained SMoE models, as evidenced by both theoretical and empirical results. To fill that gap, we analyze the advanced SMoE models and observe that the eigenvectors of expert weight matrices encode rich semantic information, pointing to an effective alternative to conventional routing strategies. Building on this insight, we propose Singular Value Decomposition SMoE (SSMoE), a novel and training-free framework that leverages spectral properties of the expert weights to address the collapse issue and enhance model performance. Extensive experiments across diverse language and vision tasks, under both clean and corrupt data settings, demonstrate the strong generalization and robustness of SSMoE. Our findings highlight how a deeper understanding of model internals can guide the development of more effective SMoE architectures. Our implementation is publicly available at this https URL.
[LG-58] Revisiting Zeroth-Order Hessian Approximation: A Single-Step Policy Optimization Lens
链接: https://arxiv.org/abs/2605.30960
作者: Junbin Qiu,Zhaowei Hong,Renzhe Xu,Yao Shu
类目: Machine Learning (cs.LG)
*备注:
Abstract:Accurate Zeroth-Order (ZO) Hessian estimation is a cornerstone of derivative-free methods, essential for tasks such as bilevel optimization, Bayesian inference, and uncertainty quantification. However, obtaining a complete suite of low-variance estimators for the Hessian and its inverse in high-dimensional settings remains a significant challenge. To address this, we propose a unified framework that reinterprets ZO Hessian approximation through the lens of single-step Policy Optimization (PO). This perspective establishes a theoretical equivalence between general ZO Hessian estimators and the Hessian of a smoothed PO objective, unifying distinct classical randomized estimators as specific instances of baseline selection. Building on this foundation, we introduce ZoVH, a comprehensive suite of variance-reduced estimators for the full Hessian matrix, its regularized inverse, and the bias-corrected inverse Hessian-gradient product. ZoVH leverages two key techniques: (1) a unique optimal baseline derived to provably minimize variance, and (2) a query reuse strategy that incorporates historical function queries to enhance sample efficiency without inflating costs. Our rigorous theoretical analysis confirms the unbiasedness of the Hessian estimator, validates the variance optimality of our baseline, provides error bounds for the entire ZoVH suite, and establishes convergence guarantees for the resulting curvature-aware ZO algorithm. Extensive empirical results validate our theoretical findings, demonstrating that ZoVH achieves superior estimation accuracy and convergence performance in real-world applications. Code is available at this https URL
[LG-59] Spectral Anatomy of Quantum Gaussian Process Kernels
链接: https://arxiv.org/abs/2605.30952
作者: Jian Xu,Chao Li,Guang Lin,Yuning Qiu,Delu Zeng,John Paisley,Qibin Zhao
类目: Machine Learning (cs.LG)
*备注:
Abstract:Two recent results have reshaped quantum Gaussian processes (QGPs). On the one hand, \citetlowe2025assessing rule out the exponential speedups claimed by HHL-based QGP regression in the typical, well-conditioned regime; on the other, an independent line of work shows that highly expressive quantum kernels suffer posterior pathologies that break Bayesian optimization. We show that these seemingly unrelated phenomena are governed by the same quantity: the normalized spectral entropy S(K)/\log n of the kernel Gram matrix. We prove a Cauchy–Schwarz tail bound on Nyström approximation error, a finite-sample variance-contraction identity in terms of Bach’s degrees of freedom d_\sigma(K) , and a characterization of the \emphtarget-dependent optimal entropy via the intrinsic dimension of the target in the kernel eigenbasis. Empirically, the diagnostic is kernel-agnostic: hardware-efficient, matchgate, IQP \emphand RBF/Matérn/RFF/deep-kernel families all collapse onto identical S/\log n curves on dequantization, ECE, and variance-contraction panels. The NLL sweet spot lives at high entropy for smooth targets and at low entropy for band-limited quantum-data targets. The diagnostic transfers from simulator to IBM Heron hardware with median absolute error 3.2% and mean 5.2% in S/\log n across 24 configurations at n_q = 4 , with matchgate and IQP within 5% mean and a single HE configuration returning a 30% outlier that drops to 0.5% on rerun (attributed to calibration drift); the same diagnostic transfers to a second Heron backend (mean error 2.7% ) and to a n_q = 6 scale-up on the original backend (mean error 1.7% ). No error mitigation is applied throughout.
[LG-60] Local linear convergence of gradient methods for overparameterized Gaussian mixtures
链接: https://arxiv.org/abs/2605.30936
作者: Jingxing Wang,Vasileios Charisopoulos,Maryam Fazel
类目: Machine Learning (cs.LG); Optimization and Control (math.OC); Machine Learning (stat.ML)
*备注: 45 pages, 7 figures
Abstract:We study the problem of learning Gaussian mixture models under overparameterization. Prior work has shown that while overparameterization is essential for avoiding spurious local optima and enables global recovery of the ground-truth model using the gradient-EM (expectation-maximization) algorithm, it can dramatically slow down the local rate of convergence. Under certain assumptions on the mixture weights, we show that a standard divergence measure minimized by statistical learning procedures possesses a manifold of slow growth on which the well-known Polyak stepsize reduces the loss geometrically, and design a gradient-based method that converges to minimizers at a locally linear rate. Additionally, we show that our method converges to nearly optimal solutions – up to a natural misspecification threshold – for mixtures with arbitrary weights. At a high level, the method alternates between several “short” gradient descent steps that approach the manifold and “long” Polyak steps that contract the distance to minimizers. Our results suggest that slow convergence is not an intrinsic challenge of overparameterization, but can be overcome by exploiting the favorable structure of the loss landscape.
[LG-61] Unsupervised Diffusion Solver for Combinatorial Optimization via Combinatorial Adjoint Matching ICML26
链接: https://arxiv.org/abs/2605.30920
作者: Shengyu Feng,Tarun Suresh,Yiming Yang
类目: Machine Learning (cs.LG)
*备注: ICML26
Abstract:Diffusion-based neural solvers have shown strong promise for combinatorial optimization (CO), but existing methods typically rely on supervised training with large collections of near-optimal solutions. In this work, we extend adjoint-based trajectory optimization methods to discrete combinatorial domains. We formulate diffusion-based CO as a stochastic control problem over Continuous-Time Markov Chains and introduce discrete adjoint dynamics for propagating optimization signals through discrete generative trajectories. Building on this formulation, we propose Combinatorial Adjoint Matching (CAM), an unsupervised training framework for discrete diffusion solvers with structured and low-variance trajectory-level optimization signals. Empirically, CAM consistently outperforms existing unsupervised diffusion baselines and achieves performance competitive with strong supervised diffusion solvers and even traditional solvers across diverse combinatorial optimization problems. Our code is available at this https URL.
[LG-62] Welfare Improvability and Variance: A Principal-Agent Approach to Optimal Benchmark Item Aggregation
链接: https://arxiv.org/abs/2605.30916
作者: Andreas Haupt,Justin Hartenstein,Anka Reuel,Mykel Kochenderfer,Sanmi Koyejo
类目: Machine Learning (cs.LG); Computer Science and Game Theory (cs.GT); Theoretical Economics (econ.TH)
*备注:
Abstract:AI benchmarks have well-documented limitations, with prior work examining contamination, saturation, and construct underspecification. Aggregation has received far less attention: benchmarks are typically summarized by uniformly averaging item-level scores, implicitly treating every test item as equally valuable. We model benchmarking as a multitask principal-agent game and show that the welfare loss from a benchmark is determined jointly by three item-level primitives: alignment with normative welfare priorities, marginal improvability, and performance variance. We translate the theory into an audit framework that ranks items along each of these three axes, and apply it to OLMES items using WORKBank for welfare, the EvoLM 4B suite for improvability, and the PolyPythias 410M panel for variance. The framework surfaces items that are Pareto-inferior within OLMES subject to a pro-worker welfare operationalization. All code is available at this https URL.
[LG-63] Automating Formal Verification with Reinforcement Learning and Recursive Inference
链接: https://arxiv.org/abs/2605.30914
作者: Max Tan
类目: Machine Learning (cs.LG); Software Engineering (cs.SE)
*备注: Master’s thesis, 140 pages, 16 figures, 17 tables
Abstract:Automated formal verification remains challenging for large language models because data for proof assistants and verification-aware languages is scarce, and correctness depends on satisfying precise machine-checkable specifications rather than producing plausible code. This thesis studies how verifier environments can improve LLM generation of verified programs and proofs through reinforcement learning from verifiable rewards (RLVR) and verifier-guided inference-time search. First, we train open-source models in Dafny with RLVR using Group Relative Policy Optimization (GRPO) and related variants, assembling generated candidates into complete programs and scoring them with compiler and verifier outcomes. Initial experiments on an APPS-derived Dafny dataset increased verified reward from 2.2% to 58.1%, but revealed specification hacking, where models exploit weak formal specifications instead of implementing the intended solutions. After filtering underspecified and vulnerable tasks, multi-turn RLVR on the refined benchmark improves the verified pass rate from 9.7% to 31.1%. Second, we develop a verifier-guided inference scaffold in Lean that treats proof generation as structured search over decomposed subgoals, verifier feedback, diagnostics, and repair. With a fixed base model, the full scaffold with proof reviser improves pass rate on an initial VeriCoding pilot set from 46.2% under direct repair to 69.2%. On the larger VERINA dataset, whole-task decomposition plus proof reviser solves 7 of 42 previously unsolved tasks. We also introduce Dalek-Bench, a repository-scale Lean benchmark derived from the Rust \textttcurve25519-dalek verification project; preliminary results remain weak, indicating that stronger progress evaluation and task-specific tool-use policies are still needed.
[LG-64] PINNs Failure Modes are Overfitting
链接: https://arxiv.org/abs/2605.30910
作者: Nigel T. Andersen,Takashi Matsubara
类目: Machine Learning (cs.LG)
*备注:
Abstract:Physics-Informed Neural Networks (PINNs) are a common class of machine learning-based partial differential equation (PDE) solvers which train a network to represent a solution by minimizing a residual loss that encodes the PDE. Despite their successes, they are known to fail on certain simple equations, converging to an incorrect solution despite low loss. These failure modes have garnered significant attention in the literature over the past several years, motivating both architectural and optimization based solutions. By directly visualizing the residual, we show that failure modes are the result of overfitting: the loss is minimized on the collocation points, but not elsewhere. Applying regularization causes the failure modes to vanish. Finally, we extend double backpropagation over the full set of residuals, and use it to achieve state-of-the-art performance on four standard failure mode equations with up to 23\times fewer collocation points and a vanilla architecture.
[LG-65] Density-Guided Robust Counterfactual Explanations on Tabular Data under Model Multiplicity ICML2026
链接: https://arxiv.org/abs/2605.30901
作者: Jun Tan,Qing Guo,Zicheng Xu,Jinglin Li,Qi Fang,Ning Gui
类目: Machine Learning (cs.LG)
*备注: 26 pages, 11 figures, accepted by ICML 2026
Abstract:Counterfactual explanations (CEs) are essential for actionable recourse, yet their reliability is often compromised in low-density regions, where classifiers exhibit high variance. Unlike existing methods that rely on expensive ensemble intersections to define stability, we propose \textitDensityFlow, a generative framework that constructs robust CEs by adhering to the high-confidence data manifold. Specifically, we model the counterfactual generation as continuous-time dynamics parameterized by Neural ODE, guided by a differentiable density score to actively avoid uncertain, low-density areas. This density score is learned via Noise Contrastive Estimation, effectively leveraging a (K+1) -way discriminator to estimate density ratios. For black-box settings, we introduce a local proxy distillation mechanism that aligns a lightweight surrogate with the target model strictly within the trajectory of CE generation, enabling efficient gradient-based optimization with minimal queries. Experiments demonstrate that \textitDensityFlow achieves superior validity under model multiplicity while significantly reducing query costs compared to ensemble-based baselines. Our implementation is available at this https URL.
[LG-66] Zero Collapse: A Failure Mode of Policy Gradient Methods in Discontinuous Reward Environments
链接: https://arxiv.org/abs/2605.30896
作者: Nishant Kumar,Enrique Areyan Viqueira,Amy Greenwald
类目: Machine Learning (cs.LG)
*备注: 20 pages, 7 figures; includes Appendix
Abstract:Bidding in repeated auctions is a central challenge for reinforcement learning (RL), combining continuous control with the strategic complexities of digital advertising. While policy gradient and value-based methods seem well-suited for these settings, they often struggle with the discontinuous, “cliff-like” nature of auction reward landscapes. In a first-price auction, for example, a bidder receives zero reward until they cross a specific threshold, after which the reward decreases as the bid increases. This creates a landscape of flat, zero-reward regions separated by sharp boundaries. We identify a fundamental failure mode in this setting termed “zero collapse.” We show that stochastic exploration and gradient-based updates can cause policies to overshoot optimal high-reward regions and enter flat, zero-reward regimes. Once there, the lack of an informative gradient signal makes recovery extremely sample-inefficient, effectively trapping the agent. We find that actor-critic methods are particularly susceptible, as biased value estimates can accelerate this movement toward unstable regions. Our contributions include: (1) a mechanistic explanation of how discontinuous rewards lead to vanishing signals and zero collapse; (2) an analysis of the interaction between policy stochasticity and step size; and (3) an empirical demonstration of this phenomenon across REINFORCE and actor-critic variants. We propose practical mitigation strategies involving initialization and architectural choices to improve stability. Finally, we introduce a formal RL framework for auction environments highlighting their unique structural properties. Comments: 20 pages, 7 figures; includes Appendix Subjects: Machine Learning (cs.LG) Cite as: arXiv:2605.30896 [cs.LG] (or arXiv:2605.30896v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2605.30896 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-67] Bandwidth Allocation with Device Partitioning for Federated Learning over Industrial IoT networks
链接: https://arxiv.org/abs/2605.30892
作者: Kangmin Kim,Jaeyoung Song
类目: Machine Learning (cs.LG)
*备注:
Abstract:We consider a federated learning (FL) system in which Industrial Internet-of-Things (IIoT) devices collaboratively train a global model over wireless channels without sharing local data. In such systems, communication time is a primary bottleneck that constrains overall training efficiency. Unlike conventional networks that prioritize individual quality-of-service requirements, FL systems collectively aim to converge to an optimal global model as efficiently as possible, which calls for a fundamentally different approach to bandwidth allocation. In this paper, we propose a novel bandwidth allocation policy that exploits the heterogeneity of device computing capabilities to minimize total training time. Rather than distributing bandwidth among all selected devices simultaneously, the proposed policy partitions the participating devices into ordered subsets and sequentially grants each subset exclusive access to the full bandwidth. We formally prove that this partitioning-based policy achieves a strictly lower training time than any bandwidth allocation scheme without partitioning, irrespective of the underlying scheduling algorithm. Furthermore, by reducing per-device transmission duration, the proposed policy also minimizes uplink energy consumption, which is particularly beneficial for battery-constrained IIoT devices. Extensive experiments on real-world datasets - including GC10-Det, an industrial surface defect benchmark, and CIFAR-10, a standard image classification benchmark - demonstrate that the proposed policy consistently reduces training time and energy consumption compared to existing bandwidth allocation schemes, approaching the theoretical lower bound on round time.
[LG-68] GlucoFM: A Dual-Stream Foundation Model for Continuous Glucose Monitoring
链接: https://arxiv.org/abs/2605.30865
作者: Zechen Li,Keerthana Natarajan,Weizhi Zhang,Menglian Zhou,Simon A. Lee,Yuwei Zhang,Maxwell A. Xu,Zeinab Esmaeilpour,Flora D. Salim,Mark Malhotra,Lindsey Sunden,Shwetak Patel,Yuzhe Yang,Ahmed A. Metwally
类目: Machine Learning (cs.LG)
*备注:
Abstract:Continuous glucose monitoring (CGM) provides a dense view of daily metabolic physiology, yet existing generic time-series and CGM-specific foundation models often encode glucose traces as entangled single-stream sequences, leaving the distinct temporal structure of glycemic dynamics only implicitly modeled. We present GlucoFM, a lightweight CGM foundation model that aligns irregular recordings to a 24-hour chronological grid, preserves observation masks, and decomposes glucose dynamics into slow physiological state and transient event streams, capturing low-frequency glycemic baselines and short-term deviations that may reflect acute physiological responses or sensor artifacts. GlucoFM is pretrained on 109,066 hours of unlabeled CGM recordings from 477 subjects with two complementary objectives: masked contextual latent prediction over fused daily representations and temporal dynamics prediction over state and event streams. Across four diverse cohorts and seven clinical prediction tasks, GlucoFM achieves the strongest subject-disjoint linear-probing performance among evaluated baselines, improving average PR-AUC by 4.1 points over the best CGM-specific foundation model. Its gains are most pronounced on core metabolic outcomes, leading PR-AUC on all diabetes-risk and \beta -cell dysfunction tasks and on 3 of 4 insulin-resistance tasks. GlucoFM also achieves the best overall cross-dataset transfer performance and strong few-shot adaptation among evaluated methods, and consistent gains when aggregating multiple days for subject-level prediction, highlighting physiology-aware decomposition as an effective inductive bias for transferable CGM representation learning.
[LG-69] ForecastCompass: Guiding Agent ic Forecasting with Adaptive Factor Memory
链接: https://arxiv.org/abs/2605.30858
作者: Yurui Chang,Yongkang Du,Yuanpu Cao,Jinghui Chen,Lu Lin
类目: Machine Learning (cs.LG)
*备注:
Abstract:Agentic forecasting is important for decision-making in dynamic environments, but it remains challenging because agents must reason from incomplete, time-limited evidence and produce calibrated probabilities before outcomes are resolved. Memory provides a natural mechanism for transferring experience from resolved forecasts to future prediction tasks. However, existing agent-memory methods are not tailored to forecasting, as they typically store past interactions, reflections, or factual associations without explicitly representing reusable predictive factors or calibration knowledge. We propose ForecastCompass (FoCo), an adaptive factor-based memory framework for agentic forecasting. FoCo organizes forecasting experience with a hierarchical forecasting-task taxonomy, enabling retrieval task-relevant forecasting knowledge. It maintains two complementary memory components: factor memory, which captures reusable predictive dimensions, and reasoning memory, which encodes probability updating, uncertainty handling, and calibration principles. Using retrospective analyses as learning signals, FoCo iteratively revises memory through a verbalized memory-revision procedure, enabling the agent to accumulate transferable forecasting knowledge over time. Experiments on Prophet Arena and FutureX with GPT-5-mini and Gemini-2.5-Flash show that FoCo improves both probabilistic accuracy and calibration.
[LG-70] A Lecture Note on Offline RL and IRL Part II: Foundations of Inverse Reinforcement Learning and Dynamic Discrete Choice Models
链接: https://arxiv.org/abs/2605.30843
作者: Enoch Hyunwook Kang
类目: Machine Learning (cs.LG); Econometrics (econ.EM)
*备注:
Abstract:In the forward reinforcement-learning problem, the reward is fixed and known; the learner is asked to find a good policy or value function. Here we turn the question around. Given offline data generated by an expert, can we recover the reward the expert was optimizing? This is the inverse reinforcement learning problem, and remarkably, two communities, structural econometricians studying dynamic discrete choice (DDC) and machine learners studying entropy-regularized IRL, have been working on exactly the same probabilistic model under different names. We begin by proving their equivalence. We then develop the classical identification result of Magnac and Thesmar and the classical computational paradigms that grew out of it: Rust’s nested fixed-point algorithm, the conditional-choice-probability approach of Hotz and Miller, and the two temporal-difference approaches of Adusumilli and Eckardt: linear semi-gradient TD and approximate value iteration. Each route has its limits: dimensionality, transition-kernel estimation, the deadly triad, or projected fixed-point bias. We then walk through the modern ML/IRL strand: adversarial IRL, occupancy matching, IQ-Learn, and offline ML-IRL, deriving each method’s actual objective and stating precisely what it does and does not identify. We close with the empirical-risk-minimization framework of Kang et al., which yields a gradient-based estimator for offline IRL/DDC.
[LG-71] CoMem: Context Management with A Decoupled Long-Context Model
链接: https://arxiv.org/abs/2605.30842
作者: Yuwei Zhang,Chengyu Dong,Shuowei Jin,Changlong Yu,Hejie Cui,Hongye Jin,Xinyang Zhang,Hamed Bonab,Colin Lockard,Jianshu Chen,Zhenyu Shi,Jingbo Shang,Xian Li,Bing Yin
类目: Machine Learning (cs.LG)
*备注: Work in progress
Abstract:Context management enables agentic models to solve long-horizon tasks through iterative summarization of previous interaction histories. However, this process typically incurs substantial decoding overhead for the extra summarization tokens, which significantly affect the end-to-end response latency at deployment. In this paper, we introduce CoMem, a novel framework that decouples memory management from the primary agent workflow, enabling these processes to execute in parallel. We propose a k -step-off asynchronous pipeline that overlaps the memory model’s summarization with the agent’s inference, effectively masking the latency of context processing. To ensure robustness under this asynchronous setting, we introduce a reward-driven training strategy that aligns the memory model to capture sufficient statistics for the agent’s decision-making. Theoretical analysis confirms that CoMem offers a superior efficiency-effectiveness trade-off compared to coupled architectures. Our extensive experimental results on SWE-Bench-Verified show that CoMem provides 1.4x latency improvements upon vanilla long-context solutions while preserving most of the performance. Furthermore, we demonstrate that these latency gains scale favorably with increased system throughput, offering a modular path forward for the independent optimization of agent reasoning and memory compression.
[LG-72] Send a SCOUT First: Pre-hoc Reasoning for Adaptive Detector Allocation in Prompt-Injection Defense
链接: https://arxiv.org/abs/2605.30837
作者: Shuhao Zhang,Jiarui Li,Qi Cao,Ruiyi Zhang,Pengtao Xie
类目: Cryptography and Security (cs.CR); Machine Learning (cs.LG)
*备注: We propose SCOUT, a detector allocation framework that predicts each detector’s accuracy and latency on a given input before running it, letting operators control the safety-utility trade-off with a single threshold and route to an LLM judge only when needed
Abstract:Prompt-injection detectors are heterogeneous: each is strong on a different slice of attacks, and none is always reliable. Yet existing systems still treat detection as a fixed single-detector pipeline, committing every request to one detector’s blind spots. We reframe defense as detector allocation: given a heterogeneous pool, decide per request which detectors to run and whether to escalate to an LLM judge. Our framework SCOUT (Scalable and Controllable Outcome-prediction for Uncertainty-aware Triage) makes this decision dynamic by predicting each detector’s per-sample reliability and latency from how it behaved on similar past inputs, and exposes a single safety-utility threshold to the operator (where utility bundles benign-pass rate and wall-clock). To evaluate this setting, we build SCOUT-450, a benchmark that captures the structurally complex, agent-facing injections that older prompt-injection sets under-represent. On SCOUT-450, a safety-oriented operating point reduces attack-success rate by 46% and total wall-clock by 40% relative to an always-on GPT-4o judge, at a 5.1-point benign-utility drop. SCOUT also transfers to three external benchmarks (BIPIA, IPI, and IHEval), improving the safety-utility frontier.
[LG-73] Cross-Layer Subspace Coupling for LLM Compression: A Unifying Framework and Its Empirical Limits
链接: https://arxiv.org/abs/2605.30836
作者: Snigdha Chandan Khilar
类目: Machine Learning (cs.LG); Differential Geometry (math.DG)
*备注:
Abstract:Recent SVD based compression methods for large language models like SVD LLM and Basis Sharing can be unified under one optimization problem. While mathematical proofs and tests on Pythia models show this unified approach improves weight reconstruction error by up to 46% percent it fails in practical tasks. Downstream metrics like perplexity and accuracy severely degrade compared to standard per layer SVD LLM. The authors explain this failure mechanistically. Although the bundle method mathematically couples adjacent layers the transformer residual stream actually decouples them during forward passes. Thus per layer optimality matters more than joint cross layer optimization. The paper concludes that weight space reconstruction is a flawed objective for cross layer compression and future methods must focus on per layer activation reconstruction instead.
[LG-74] Learning Permutation-invariant Macroscopic Dynamics ICML2026
链接: https://arxiv.org/abs/2605.30812
作者: Zhichao Han,Mengyi Chen,Qianxiao Li
类目: Machine Learning (cs.LG); Computational Physics (physics.comp-ph)
*备注: ICML 2026 submission
Abstract:Accurately modeling the macroscopic dynamics of high-dimensional microscopic systems is of broad interest across the sciences. Many data-driven approaches learn a low-dimensional latent state through an autoencoder trained for pointwise input reconstruction. These methods typically assume a fixed ordering of microscopic degrees of freedom in the input. However, in many settings, such as particle systems, the microscopic state is inherently unordered. This motivates an autoencoder framework that learns permutation-invariant latent representations. To this end, we adopt a permutation-invariant encoder and design the decoder to reconstruct the mass distribution centered at the observed points rather than per-sample reconstruction. We then jointly learn the macroscopic dynamics of the observables together with the latent states. We demonstrate the effectiveness and robustness of the proposed method across a range of microscopic settings, including learning the energy dynamics in interacting particle systems, predicting mixing dynamics in Lennard-Jones fluids, and modeling the stretching dynamics from video data of polymers moving in an elongational force field.
[LG-75] Non-destructive Identification of Oyster Species is possible from Hyperspectral Images with Machine Learning
链接: https://arxiv.org/abs/2605.30811
作者: Ethan Kane Waters,Max Wingfield,Aiden Mellor,Paul Stewart,Iman Tahmasbian
类目: Machine Learning (cs.LG)
*备注: 13 pages, 9 figures
Abstract:Differentiating between oyster species is important for developing new commercial oyster species suited to production systems and is critical for traceability in seafood supply chains. Common methods, such as DNA profiling, are destructive and time consuming. The possibility of using hyperspectral imaging (HSI) for discriminating between Black-Lip rock (BL) and Sydney rock (SR) oysters was investigated. Live BL and SR samples (N = 156) were scanned with a HSI camera (950-2515nm). Partial Least Square Discriminant Analysis and Convolutional Neural Networks were trained with Monte Carlo Cross Validation to distinguish BL and SR oysters from the spectral reflectance of their left and rights valves. The PLS-DA model successfully distinguished between the species from both the left and right valves with a median test set classification accuracy of 100%, out performing the CNN with 83% and 96% respectively. Elemental and mineralogical composition in the surface and cross-section of oyster valves were measured with electron microscopy. Analysis of the right valve revealed a greater number of layers in BL compared to SR (4 vs 2). The concentrations of carbon and oxygen varied in the outer layer of the right valves, with BL being rich in carbon and SR being rich in oxygen. The variation in carbon and oxygen concentrations observed between BL and SR right valves may reflect differences in the relative abundance or composition of chitin and glycoproteins. This is supported by model-derived wavelength importance corresponding to vibrational modes of functional groups characteristic of these compounds. Transmittance analysis revealed that light was transmitted through the valves, around the valve edges, indicating that the spectral signatures may have been influenced by the other valve or the meat. Ultimately, the findings highlight an effective rapid, non-destructive methodology for oyster species.
[LG-76] IRIS: time-structured manifold projections
链接: https://arxiv.org/abs/2605.30810
作者: Brian Ondov,Chia-Hsuan Chang,Weipeng Zhou,Xingjian Zhang,Xueqing Peng,Yutong Xie,Huan He,Qiaozhu Mei,Hua Xu
类目: Machine Learning (cs.LG)
*备注:
Abstract:High-dimensional biomedical data, such as cell-by-gene matrices, are increasingly generated temporally. However, Manifold Learning algorithms, like t-SNE and UMAP, cannot incorporate time-ordering in their layouts, obfuscating the dynamics of cell types or other classes. As a solution, we present IRIS, a new Manifold Learning algorithm that structures layouts both chronologically and by manifold topology. IRIS can visualize a wide range of dynamic biomedical data, including scRNA-seq, comparative metagenomics, and literature.
[LG-77] Conformal Reliability: A New Evaluation Metric for Conditional Generation ICML2026
链接: https://arxiv.org/abs/2605.30807
作者: Yachen Gao,Xinwei Sun,Yikai Wang,Ye Shi,Jingya Wang,Jianfeng Feng,Yanwei Fu
类目: Machine Learning (cs.LG)
*备注: Accepted at ICML 2026
Abstract:Conditional generative models have recently achieved remarkable success in various applications. However, a suitable metric for evaluating the reliability of these models, which takes into account their inherent uncertainty, is still lacking. Existing metrics, which typically assess a single output, may fail to capture the variability or potential risks in generation. In this paper, we propose a novel evaluation metric called reliability score based on conformal prediction, which measures the worst-case performance within the prediction set at a pre-specified confidence level. However, computing this score is challenging due to the high-dimensional nature of the output space and the nonconvexity of both the metric function and the prediction set. To efficiently compute this score, we introduce Conformal ReLiability (CReL), a framework that can (i) construct the prediction set with desired coverage; and (ii) accurately optimize the reliability score within the constructed prediction set. We provide theoretical results on coverage and demonstrate empirically that our method produces more informative prediction sets than existing approaches. Experiments on synthetic data and the image-to-text and text-to-image tasks further demonstrate the interpretability of our new metric, and the validity and effectiveness of our computational framework. Source code can be found at this https URL.
[LG-78] AbstainGNN: Teaching Graph Neural Networks to Abstain for Graph Classification KDD2026
链接: https://arxiv.org/abs/2605.30786
作者: Xixun Lin,Zhiheng Zhou,Zhengyin Zhang,Yancheng Chen,Shuai Zhang,Ge Zhang,Shichao Zhu,Lixin Zou,Chuan Zhou,Peng Zhang,Shirui Pan,Yanan Cao
类目: Machine Learning (cs.LG)
*备注: Accepted at KDD 2026
Abstract:Graph classification is a core task in graph data mining with widespread real-world applications. Recent advances in graph neural networks (GNNs) have led to substantial performance improvements for graph classification. However, existing GNNs are typically forced to make predictions even under high uncertainty or unknown conditions, resulting in unreliable decisions that can severely impact downstream tasks, particularly in safety-critical scenarios. To address this critical limitation, we propose AbstainGNN, a novel and theory-driven framework for graph classification with abstention, which enables GNNs to reject uncertain predictions instead of producing incorrect decisions. Specifically, AbstainGNN explicitly models both the predictive function and the abstention function, allowing for effective utilization of graph structural information. Moreover, unlike existing heuristic abstention methods, we theoretically characterize the trade-off between classification errors and rejection costs from a PAC-Bayesian generalization perspective, and derive a unified learning objective for model optimization. Guided by this theoretical insight, we further develop an efficient two-stage training strategy consisting of predictive function warm-start and abstention function calibration. Extensive experiments on five benchmark datasets show that AbstainGNN outperforms existing abstention methods, achieving superior classification performance under the same rejection rates.
[LG-79] Efficient and Uncertainty-Aware Diffusion Framework for Offline-to-Online Reinforcement Learning
链接: https://arxiv.org/abs/2605.30776
作者: Ha Manh Bui,Metod Jazbec,Eric Nalisnick,Anqi Liu
类目: Machine Learning (cs.LG)
*备注: International Conference on Machine Learning, 2026
Abstract:Offline-to-Online Reinforcement Learning (O2O-RL) leverages an offline, pre-trained policy to minimize costly online interactions. Although data-efficient, O2O-RL is susceptible to shifts between offline and online distributions. Existing work aims to mitigate the harm of this shift by finetuning the policy on trajectory data sampled from a diffusion model. Inspired by this line of work, we propose DUAL: an efficient \textbfDiffusion \textbfUncertainty-\textbfAware framework for offline-to-online reinforcement \textbfLearning. DUAL utilizes the prior knowledge of the diffusion model to distill a fast-sampling diffusion actor policy and transition model in the offline phase. DUAL also employs a Laplace approximation and distance transition-state-shift detection, thereby using uncertainty quantification to improve exploration versus exploitation in the online phase. We formally show that our actor loss with the Laplace approximation provides a proxy for a principled estimate of epistemic uncertainty. Empirically, DUAL improves the online expected return over O2O-RL baselines across multiple settings and environments.
[LG-80] Chain-of-Thought and Compressed Looped Transformers: A Memory-Budget Separation
链接: https://arxiv.org/abs/2605.30757
作者: Haozhou Zhang
类目: Machine Learning (cs.LG)
*备注:
Abstract:Chain-of-thought prompting and looped Transformers both give a fixed model more test-time computation, but they differ in what they remember. Chain-of-thought stores intermediate state in generated tokens that remain in the context, whereas a looped Transformer carries state through recurrent hidden activations. We argue that this persistent mutable memory is a central resource for test-time reasoning. We compare three memory regimes, the compressed latent loop, the full sequence-state loop, and the chain-of-thought scratchpad. Our main result shows that a compressed loop is limited by the size of its recurrent state. Running the loop longer adds computation but does not by itself create a growing scratchpad, so a loop with a small recurrent state remains a small-space reasoner even when run for many steps. Under a standard complexity assumption, such loops cannot decide problems that are P-complete under logspace reductions, whereas polynomial-length chain-of-thought can. The separation is specific to compressed loops, as full sequence-state loops carry state at every input position and live in a memory-rich regime closer to explicit scratchpads. Controlled pointer-chasing and associative-recall sweeps illustrate this memory-budget view, with performance sensitive to whether the persistent-state budget matches the task’s working-memory demand. Subjects: Machine Learning (cs.LG) Cite as: arXiv:2605.30757 [cs.LG] (or arXiv:2605.30757v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2605.30757 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-81] FLAG: Flow Policy MaxEnt-RL by Latent Augmented Guidance
链接: https://arxiv.org/abs/2605.30749
作者: Sungha Kim,Gawon Lee,Jusuk Lee,Jonghae Park,H. Jin Kim,Daesol Cho
类目: Machine Learning (cs.LG); Robotics (cs.RO)
*备注:
Abstract:Maximum entropy reinforcement learning (MaxEnt-RL) enables robust exploration, yet practical implementations often restrict policies to simple Gaussians. While recent approaches incorporate expressive generative policies via importance-weighted supervised learning, they are prone to importance weight collapse, which limits their scalability in high-dimensional action spaces. Our key insight is to mitigate this limitation by localizing the sampling region, avoiding the weight degeneracy induced by importance sampling over the entire action space. To instantiate this insight, we introduce \textbfFLAG (\textbfFlow policy with \textbfLatent-\textbfAugmented \textbfGuidance). FLAG augments the state space with a flow latent variable and optimizes a provably consistent proxy MaxEnt-RL objective. We empirically demonstrate that FLAG enables expressive policy optimization with limited importance samples and scales to high-dimensional control tasks. Furthermore, FLAG achieves state-of-the-art performance across challenging benchmarks. Our project webpage: this https URL Subjects: Machine Learning (cs.LG); Robotics (cs.RO) Cite as: arXiv:2605.30749 [cs.LG] (or arXiv:2605.30749v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2605.30749 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Sungha Kim [view email] [v1] Fri, 29 May 2026 02:25:03 UTC (7,941 KB) Full-text links: Access Paper: View a PDF of the paper titled FLAG: Flow Policy MaxEnt-RL by Latent Augmented Guidance, by Sungha Kim and 5 other authorsView PDFHTML (experimental)TeX Source view license Current browse context: cs.LG prev | next new | recent | 2026-05 Change to browse by: cs cs.RO References Citations NASA ADSGoogle Scholar Semantic Scholar export BibTeX citation Loading… BibTeX formatted citation loading… Data provided by: Bookmark checked="checked"class=“labs-tab-input”> Bibliographic Tools Bibliographic and Citation Tools Bibliographic Explorer Toggle Bibliographic Explorer (What is the Explorer?) Connected Papers Toggle Connected Papers (What is Connected Papers?) Litmaps Toggle Litmaps (What is Litmaps?) scite.ai Toggle scite Smart Citations (What are Smart Citations?) Code, Data, Media Code, Data and Media Associated with this Article alphaXiv Toggle alphaXiv (What is alphaXiv?) Links to Code Toggle CatalyzeX Code Finder for Papers (What is CatalyzeX?) DagsHub Toggle DagsHub (What is DagsHub?) GotitPub Toggle Gotit.pub (What is GotitPub?) Huggingface Toggle Hugging Face (What is Huggingface?) ScienceCast Toggle ScienceCast (What is ScienceCast?) Demos Demos Replicate Toggle Replicate (What is Replicate?) Spaces Toggle Hugging Face Spaces (What is Spaces?) Spaces Toggle TXYZ.AI (What is TXYZ.AI?) Related Papers Recommenders and Search Tools Link to Influence Flower Influence Flower (What are Influence Flowers?) Core recommender toggle CORE Recommender (What is CORE?) IArxiv recommender toggle IArxiv Recommender (What is IArxiv?) 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(); About Help contact arXivClick here to contact arXiv Contact subscribe to arXiv mailingsClick here to subscribe Subscribe Copyright Privacy Policy Web Accessibility Assistance arXiv Operational Status
[LG-82] Reducing the GPU Memory Bottleneck with Lossless Compression for ML – Extended EUROSYS’26
链接: https://arxiv.org/abs/2605.30728
作者: Aditya K Kamath,Arvind Krishnamurthy,Marco Canini,Simon Peter
类目: Machine Learning (cs.LG); Distributed, Parallel, and Cluster Computing (cs.DC)
*备注: Extended version of paper published at 21st European Conference on Computer Systems (EUROSYS '26), April 27-30, 2026, Edinburgh, Scotland Uk
Abstract:Machine learning (ML) training and inference often process data sets far exceeding GPU memory capacity, forcing them to rely on PCIe for on-demand tensor transfers, causing critical transfer bottlenecks. Lossy compression has been proposed to relieve bottlenecks but introduces workload-dependent accuracy loss, making it complex or even prohibitive to use in existing ML deployments. We explore lossless compression as an alternative that avoids this deployment complexity. We identify where lossless compression can be integrated into ML pipelines while minimizing interference with GPU execution. Based on our findings, we introduce Invariant Bit Packing (IBP), a novel lossless compression algorithm designed to minimize data transfer time for ML. IBP identifies and eliminates invariant bits across groups of tensors, improving throughput through GPU-optimized decompression that leverages warp parallelism, low-overhead bit operations, and asynchronous PCIe transfers. We provide easy-to-use APIs, showcasing them by adding IBP support to GNN training, as well as DLRM and LLM inference frameworks. IBP achieves, on average, 74% faster GNN training, 180% faster DLRM embedding lookup, and 24% faster LLM inference.
[LG-83] Self-Certifying Transport MCMC via Dual Spectral-Gap Certificates
链接: https://arxiv.org/abs/2605.30722
作者: Jun Hu
类目: Machine Learning (cs.LG); Computation (stat.CO); Methodology (stat.ME)
*备注: 35 pages, 3 figures, 9 tables. Submitted to JASA
Abstract:We propose CerT-MCMC, a framework that equips learned-transport Markov chain Monte Carlo with automatic, rigorous convergence certificates. A normalising flow maps a Gaussian reference to an approximation of the target posterior; the same flow then serves as both the independence Metropolis-Hastings proposal and the basis for a computable spectral-gap bound. We develop two complementary certificates. The covering certificate bounds the weight-ratio oscillation over the full proposal support via finite-sample covering arguments, yielding full-support spectral-gap bounds when a conservative gradient bound is available; its correction term scales as O(n^-1/D), making it rapidly weak and eventually vacuous as dimension increases. We prove a matching Omega(n^-1/D) lower bound, establishing that this barrier is intrinsic to pointwise Lipschitz certification. The quantile-core certificate restricts attention to a high-probability residual core on which the oscillation is controlled by one-dimensional empirical quantiles, with a finite-sample probability slack of O(n^-1/2), independent of the ambient dimension. On synthetic targets (D=2-20), structural-engineering posteriors (D=6,8), real-data logistic regression on the Heart Disease data set (D=13), and synthetic Bayesian logistic regression (D=20), the quantile-core certificate delivers non-vacuous spectral-gap bounds where the covering certificate is vacuous, and its spectral-gap proxy tracks empirical effective sample sizes within 7%. A negative control experiment confirms that the certificate discriminates flow quality by a factor exceeding 10x, whereas acceptance rates differ by only 1.15x. To our knowledge, the dual-certificate framework is the first to provide automatic, dimension-aware convergence certificates for learned-transport MCMC, distinguishing genuine transport failure from proof-technique limitations.
[LG-84] Universal Decision Learners
链接: https://arxiv.org/abs/2605.30694
作者: Sridhar Mahadevan
类目: Machine Learning (cs.LG)
*备注: 15 pages
Abstract:Many theories of decision making – planning, reinforcement learning, causal intervention, online learning, and game-theoretic equilibrium – turn local information into globally coherent behavior. This paper proposes a common categorical formulation: a Universal Decision Learner (UDL) extends a partially specified decision functor from observed contexts to new contexts by a pair of universal constructions. Left Kan extensions express rollout, aggregation, and candidate generation; right Kan extensions express consistency, constraint satisfaction, and fixed-point semantics. The central claim is not that every decision problem has the same algorithm, but that many decision formalisms instantiate the same universal problem: extend local behavioral data canonically, then characterize the globally coherent extensions. We give the abstract UDL construction, prove its universal comparison property, define Kan-invariant behavioral equivalence and minimal abstractions, and show how Bellman equations, planning recursions, causal interventions, online regret, and equilibria arise as special cases. The supplementary material develops the reinforcement-learning specialization in more detail.
[LG-85] Spatio-temporal stochastic graph-based learning for infectious disease forecasting
链接: https://arxiv.org/abs/2605.30662
作者: Luz Stefani Sotomayor Valenzuela,Susanna Cramb,Darren Wraith
类目: Machine Learning (cs.LG); Populations and Evolution (q-bio.PE)
*备注: Preprint under review
Abstract:Spatio-temporal graph-based models have typically been used to forecast new cases of infectious diseases such as COVID-19 and chickenpox outbreaks. However, the use of stochastic modelling into their learning process has been surprisingly under-investigated and rarely considered entire data sets of large countries. As a result, it is unknown whether these models would provide accurate forecasts in real-world disease spread scenarios. In this work, we propose a spatio-temporal stochastic graph-based architecture that integrates a stochastic formulation and uncertainty approximation process to forecast new infectious disease cases. We find that our approach can adapt to encode large and small population geographical networks within a single model architecture. Using two real-world data sets, COVID-19 in the US and chickenpox in Hungary, we report an enhanced effect of the proposed architecture across predictions of the 2022 first wave for COVID-19 in the US and comparative results of chickenpox waves during 2012-2014 in Hungary. By benchmarking with four spatio-temporal graph-based models, quantitative results show competitive overall weekly performance of the proposed approach on forecasting new cases for all 3,218 US counties and all 20 Hungary counties. The proposed approach can represent overall epidemic progression relative to baselines, though with a one-step delay; while exhibiting a reduced sensitivity to high-frequency and low-amplitude variability.
[LG-86] BOKBO (Best of K Bad Options): Calibrated Abstention for VLA Policies
链接: https://arxiv.org/abs/2605.30660
作者: Anya Singh,Cabrel Happi,Jai Relan,Varun Nair,Vidyut Baradwaj
类目: Machine Learning (cs.LG); Robotics (cs.RO)
*备注:
Abstract:Test-time scaling for vision-language-action (VLA) policies, methods such as RoboMonkey, SEAL, MG-Select, and V-GPS, samples K candidate action chunks at inference and executes the verifier-best. When all K candidates are unsafe, the system executes a violating action with no warning. We propose BOKBO, the first conformal abstention layer for K-sample VLA inference, providing finite-sample distribution-free guarantees on executed-violation rate. We provide both global and per-task (Mondrian) variants, with the per-task variant closing the conditional gap on the hardest tasks. Our analysis exposes a structural failure of policy-internal nonconformity scores under perturbation-based K-sampling: the base-policy confidence proxy and K-sample disagreement correlate at 0.98 with the action-noise hyperparameter \sigma , while correlating at the noise floor with actual safety violations. We test the failure’s scope by replicating the analysis under token-level temperature sampling and find the failure is mechanism-specific and partially mitigated under policy-stochasticity-based sampling. A learned violation predictor conditioned on semantic visual features and task identity supports tight calibration: at \epsilon = 0.05 on libero_object_temp_x0.1 with OpenVLA-OFT, the conditional CRC bound holds on 86% of bootstrap splits with 78% coverage and 70% net task success. Mondrian-BOKBO raises the minimum per-task conditional hold fraction from 0.71 to 0.93. Results are stable across 5 training seeds, replicate within bootstrap noise on \pi_0 -FAST, hold on libero_spatial_temp_x0.1 as a co-equal benchmark, and survive four within-suite distribution shifts. We additionally identify and correct a methodological pitfall: globally-set force thresholds well below expert-typical manipulation forces conflate unsafe behavior with normal manipulation, inflating violation rates by 5\times . Subjects: Machine Learning (cs.LG); Robotics (cs.RO) Cite as: arXiv:2605.30660 [cs.LG] (or arXiv:2605.30660v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2605.30660 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-87] Learning to Perceive the World Through Control: Empowerment-Based Representation Learning
链接: https://arxiv.org/abs/2605.30656
作者: Mahsa Bastankhah,Sophie Broderick,Benjamin Eysenbach
类目: Machine Learning (cs.LG)
*备注:
Abstract:In many practical reinforcement learning environments, observations are far higher-dimensional than the variables that matter for control. In this work, we ask: can we learn representations that capture only control-relevant features of the environment? We study this question through the empowerment objective, which maximizes an agent’s influence over the environment and is widely used for unsupervised skill learning. We show that empowerment agents induce two distinct representations – forward and backward – that capture complementary aspects of the state, and both of which are invariant to control-irrelevant features. Thus, empowerment maximization leads agents to learn an implicit, control-centric model of the world. Our analysis highlights the importance of learning representations through interaction rather than from passive datasets: interaction aimed at maximizing control is essential for learning useful invariance properties, a perspective that aligns closely with the causal learning literature.
[LG-88] Bridging the Gap Between Natural Language and Market Dynamics via High-Dimensional Representation Learning
链接: https://arxiv.org/abs/2605.30652
作者: Yujin Jeong,Noelle Jung,Brian Y. C. Leung(Mike)
类目: Machine Learning (cs.LG)
*备注:
Abstract:Traditional multi-modal financial forecasting often relies on scalar sentiment scores, which fail to capture the nuances of financial news. To address this information loss, this paper explores high-dimensional representation learning by replacing discrete polarity ratings with dense FinBERT embeddings within a Transformer-based forecasting architecture. We benchmarked various embedding strategies on the FNSPID dataset, including raw embeddings, attention-weighted aggregation, and a custom Siamese network. While the attention-based mechanism struggled with the low signal-to-noise ratio typical of financial data, the integration of Siamese-optimized embeddings outperformed both the scalar baseline and raw embedding approaches, demonstrating that preserving high-dimensional narrative context yields improved predictive accuracy for short-term stock price movements.
[LG-89] Convergence of Steepest Descent and Adam under Non-Uniform Smoothness ICML2026
链接: https://arxiv.org/abs/2605.30648
作者: Sharan Vaswani,Yifan Sun,Reza Babanezhad
类目: Machine Learning (cs.LG); Optimization and Control (math.OC)
*备注: ICML 2026
Abstract:Recent work has analyzed the convergence of first-order methods under non-uniform smoothness assumptions that better model the loss landscape in machine learning tasks. We generalize this assumption to objectives whose curvature is an affine function of the objective value. This property is satisfied by a broad class of problems, including logistic regression, generalized linear models with a logistic link function, softmax policy gradient in reinforcement learning, and a class of neural networks. Under this assumption and gradient domination conditions, we establish a general convergence rate for the steepest descent method, and deterministic, diagonal variants of RMSProp and Adam. Our results imply that for logistic regression on separable data and the softmax policy gradient objective, sign GD converges linearly and is provably faster than GD. Furthermore, we show that for a class of two-layer neural networks on separable data, RMSProp and Adam can converge at a linear rate with a constant step-size and momentum parameter. Finally, we present a lower bound demonstrating that, under our assumption, RMSProp and Adam are provably faster than AdaGrad, AMSGrad, gradient descent, and heavy-ball momentum.
[LG-90] Diffusion Models Preferentially Memorize Prototypical Examples or: Why Does My Diffusion Model Love Slop?
链接: https://arxiv.org/abs/2605.30642
作者: Marta Aparicio Rodriguez,Anastasia Borovykh,Grigorios A. Pavliotis,Daniel J. Korchinski
类目: Machine Learning (cs.LG)
*备注:
Abstract:Generative models have a persistent limitation: their tendency to memorize training data can create legal liabilities and erode creative diversity. Understanding which samples are memorized in whole or in part, and under what conditions, therefore remains an important open problem. Here we answer the question “Are atypical or rare samples memorized first?” in the negative. We train diffusion models on strings generated according to the production rules of the Random Hierarchy Model (RHM), and find that samples composed of common substrings are preferentially memorized. This holds true even if the training data consists of entirely unique samples, indicating that deduplication at the data point level does not provide a meaningful privacy guarantee. Correspondingly we predict, then observe, delayed memorization for fat-tailed datasets (i.e., those with more atypical samples). This effect is amplified when fat-tails are introduced into high-level production rules. These together suggest that dataset diversity, particularly at higher levels of abstraction, plays an important role in staving off memorization. Finally, we identify an intermediate regime of partial memorization in which common substrings are learned first and subsequently overproduced during generation. If training is stopped in this regime, models will exhibit the reversion-to-the-mean blandness often derided as “slop”.
[LG-91] CellBRIDGE: Learning Cellular Trajectories via Interaction-Aware Alignment
链接: https://arxiv.org/abs/2605.30635
作者: Silas Ruhrberg Estévez,Nicolas Huynh,Tennison Liu,Roderik M. Kortlever,Gerard I. Evan,David L. Bentley,Mihaela van der Schaar
类目: Machine Learning (cs.LG); Genomics (q-bio.GN)
*备注:
Abstract:Inferring dynamics from population snapshots is a fundamental challenge in machine learning and biology. In scRNA-sequencing (scRNA-seq), destructive measurements preclude direct tracking of individual cells across time, making trajectory inference underdetermined. Optimal Transport (OT) provides a principled framework for snapshot alignment, but a long-standing modeling question is which cost functions yield biologically meaningful couplings. Standard OT approaches rely on gene-expression distances, implicitly treating cells as independent points and neglecting structured cell-cell communication mediated by ligand-receptor signaling. We introduce CellBRIDGE (Cell-Based Regularized Interaction-Driven Gene Expression), which augments feature-based OT with a directed, typed interaction cost derived from ligand-receptor activity. By explicitly modeling cell-cell communication, CellBRIDGE improves cross-snapshot couplings and downstream trajectory estimates across synthetic and real scRNA-seq datasets relative to feature-only baselines. Notably, CellBRIDGE enables mechanistically interpretable in silico perturbations: on lung cancer data, silencing specific ligand-receptor pairs induces trajectory shifts that recapitulate expected effects of targeted pathway inhibition.
[LG-92] Jamming-Resilient PRB Reservation for Latency-Critical O-RAN Network Slicing
链接: https://arxiv.org/abs/2605.30622
作者: Elahe Delavari,Junaid Farooq
类目: Networking and Internet Architecture (cs.NI); Machine Learning (cs.LG)
*备注: Accepted at ML-Spec Workshop in IEEE DySPAN 2026
Abstract:Open radio access network (O-RAN) architectures enable near real-time, software-driven control of network slicing through programmable xApps deployed on the near-real-time RAN Intelligent Controller (near-RT RIC). In industrial 5G downlink systems, adversarial jamming can abruptly reduce the effective physical resource block (PRB) capacity, triggering queue buildup and persistent latency violations, particularly in the presence of low spectral efficiency cell edge user equipments. This paper proposes a reserve-based resilience framework for PRB allocation in sliced O-RAN deployments. A finite pool of reserved PRBs is controlled by a near-RT RIC xApp that provides hybrid mitigation by proactively clearing backlog to build latency margin and reactively allocating reserve capacity during jammer active intervals. We formulate reserve activation as a constrained sequential decision problem and design a masked Deep Q-Network to learn effective control policies under non-stationary jamming. Simulation results show substantial reductions in URLLC latency violations and improved reserve efficiency compared to reactive baselines.
[LG-93] Improving Selective Classification with Pairwise Queries for Binary Classification
链接: https://arxiv.org/abs/2605.30615
作者: Harsh Vardhan,Sunav Choudhary,Natwar Modani,Arya Mazumdar
类目: Machine Learning (cs.LG)
*备注:
Abstract:In selective classification, a model predicts the labels of data samples where it is confident, and abstains from predicting labels for samples on which it is not confident. The rejected samples are often labeled by an expert, which is expensive. The budget for the expert is best utilized when the model has low error on non-rejected samples. However, the estimate of a model’s confidence might be inconsistent with the model’s predictions, which can lead to high error on non-rejected points. Such situations can readily occur in in-context binary classification by LLMs. To remedy this, we propose making additional pairwise queries to the same model. These pairwise queries can detect high-error samples and be incorporated into selective classification techniques to reduce the error on non-rejected samples. Theoretically, we establish the conditions under which a simple algorithm using pairwise queries outperforms an inconsistent confidence estimate. We support this insight through extensive experiments for 1 synthetic and 4 in-context learning-based real binary classification datasets. In all these cases, we show that our algorithms, using pairwise queries, obtain a better accuracy-cost tradeoff than using only the raw confidence estimates, for instance, the LLM’s next-token logits.
[LG-94] CacheProbe: Auditing Prompt Cache Isolation in Gateway APIs
链接: https://arxiv.org/abs/2605.30613
作者: Ryan Fahey
类目: Cryptography and Security (cs.CR); Machine Learning (cs.LG)
*备注: 11 pages, 8 figures, 2 tables Accepted at SAGAI '26 (Workshop on Secure Agents for Generative AI), co-located with IEEE Symposium on Security and Privacy 2026
Abstract:Over the past year, prompt caching in Large Language Models (LLMs) has become increasingly more popular across inference APIs. Prompt caching helps save precious compute resources and speeds up response times by reusing parts of the KV cache of a specific prompt for another request. However, many implementations of prompt caching are not secure against timing attacks or even basic metadata disclosure. Gu et al. (ICML 2025) develop a method to audit prompt caching in LLMs. This paper investigates whether OpenRouter’s API gateway architecture introduces prompt caching vulnerabilities that bypass provider-level prompt cache isolation guarantees. Most LLM inference providers implement per-account or per-organization prompt caching to prevent data leaks, but does routing through OpenRouter with shared organizational credentials inadvertently create global cache sharing across all OpenRouter users?
[LG-95] ZAPS-DA: Zero-Phase Action Policy Smoothing with Decoupled Actor for Continuous Control in Reinforcement Learning
链接: https://arxiv.org/abs/2605.30612
作者: Faiq Shamass
类目: Robotics (cs.RO); Machine Learning (cs.LG); Systems and Control (eess.SY)
*备注: 7 pages, 5 figures, 5 tables. Submitted to IEEE RA-L
Abstract:Continuous control policies trained with off-policy reinforcement learning frequently exhibit high-frequency action jitter, rendering direct deployment on physical actuators impractical. Post-hoc filtering attenuates jitter but introduces phase lag; embedding smoothness penalties in the actor’s loss couples them with the RL gradient and conflates reward regression with over-aggressive smoothing. We present ZAPS-DA, a framework that reduces action jitter at deployment with negligible phase lag and no post-processing. ZAPS-DA pairs an unmodified main actor (trained by the base RL loss) with a separate decoupled actor trained via supervised imitation of zero-phase filtered targets stored in the replay buffer. The deployed policy is the decoupled actor: a feed-forward map from the current observation to a smooth action, with no inference-time filter and no action-history input – a mechanism we term causal distillation of a non-causal filter. A magnitude-matched MSE loss provides zero-hyperparameter portability across optimizer classes. Validated with Soft Actor-Critic and a Savitzky–Golay filter in two driving simulators using paired n=150 evaluation protocols: on MetaDrive, ZAPS-DA reduces steering jitter by 14–21x and throttle jitter by 3–5x (all p 10^-4 , Bonferroni-corrected) while matching task-completion (p=0.28 success, p=0.31 crash) at a 6.3% reward cost; on a custom Webots adaptive cruise control environment, the same SG configuration produces a Pareto improvement – reward parity (p=0.121), 8–45x steering jitter reduction, and total task-failure rate reduced from 2.0% to 0.7%.
[LG-96] Constrained Flow Optimization via Sequential Fine Tuning for Molecular Design ICML2026
链接: https://arxiv.org/abs/2605.30610
作者: Sven Gutjahr,Riccardo De Santi,Luca Schaufelberger,Kjell Jorner,Andreas Krause
类目: Machine Learning (cs.LG)
*备注: ICML 2026
Abstract:Adapting generative foundation models, in particular diffusion and flow models, to optimize given reward functions (e.g., binding affinity) while satisfying constraints (e.g., molecular synthesizability) is fundamental for their adoption in real-world scientific discovery applications such as molecular design or protein engineering. While recent works have introduced scalable methods for reward-guided fine-tuning of such models via reinforcement learning and control schemes, it remains an open problem how to algorithmically trade-off reward maximization and constraint satisfaction in a reliable and predictable manner. Motivated by this challenge, we first present a rigorous framework for Constrained Generative Optimization, which brings an optimization viewpoint to the introduced adaptation problem and retrieves the relevant task of constrained generation as a sub-case. Then, we introduce Constrained Flow Optimization (CFO), an algorithm that automatically and provably balances reward maximization and constraint satisfaction by reducing the original problem to sequential fine-tuning via established, scalable methods. We provide convergence guarantees for constrained generative optimization and constrained generation via CFO. Ultimately, we present an experimental evaluation of CFO on both synthetic, yet illustrative, settings, and a molecular design task. Across these evaluations, CFO achieves consistent increases in reward while ensuring high constraint satisfaction, showcasing its practical utility for constrained generative optimization.
[LG-97] ASER: Task-Aware Stein Regularisation for Geometry-Driven Robustness
链接: https://arxiv.org/abs/2605.30601
作者: Michał Kozyra,Gesine Reinert
类目: Machine Learning (cs.LG)
*备注:
Abstract:Modern deep networks remain fragile under distribution shift and adversarial perturbations, often due to excessive or poorly structured input sensitivity. We introduce TASER (Task-Aware Stein Regularisation), a training-time regularisation framework derived from Langevin Stein operators. By penalising pointwise Stein residuals under the training distribution, TASER encourages geometric compatibility between predictors and data density, inducing anisotropic, data-aware smoothness. We provide theoretical links between Stein regularisation and reduced first-order shift sensitivity, develop scalable implementation variants compatible with modern architectures, and demonstrate improved robustness and stability across regression and vision benchmarks. Across CIFAR-10 experiments, TASER consistently improves the adversarial robustness of established training methods without incurring statistically significant clean-accuracy degradation.
[LG-98] he Fast Mixing Mechanism for Differential Privacy
链接: https://arxiv.org/abs/2605.30600
作者: Omri Lev,Moshe Shenfeld,Vishwak Srinivasan,Katrina Ligett,Ashia C. Wilson
类目: Machine Learning (cs.LG); Information Theory (cs.IT)
*备注:
Abstract:Randomized sketching is a central tool for compressing large-scale optimization problems while preserving accuracy. In particular, sketches that are based on structured matrices, such as the Hadamard matrix, can be applied efficiently and often yield solutions that approximate those of the original problem at much lower computational cost. In differential privacy (DP), Gaussian sketching has been used to solve DP linear regression, beginning with \citetsheffet2017differentially, sheffet2019old and later refined by \citetlev2025gaussianmix, lev2026near. However, although these methods achieve strong utility guarantees, they usually do not improve runtime over classical DP approaches. In this work, we introduce a new DP sketching mechanism based on fast transforms, which, in certain cases, matches the runtime of classical fast sketching methods. We prove state-of-the-art privacy guarantees for this mechanism and show that, in favorable regimes, they match those of the Gaussian sketch up to a constant factor. As an application, we combine this mechanism with recent sketch-based methods for DP linear regression to obtain a new algorithm with strong utility and improved runtime. We establish privacy and accuracy guarantees for this algorithm, yielding, to the best of our knowledge, the first fast method for DP ordinary least squares.
[LG-99] ScaleMAP: Preserving Local Density and Neighborhood Structure in Low-Dimensional Embeddings
链接: https://arxiv.org/abs/2605.30597
作者: Rajas Poorna,Marcus T. Cicerone(Georgia Institute of Technology)
类目: Machine Learning (cs.LG)
*备注: 23 pages, 16 figures
Abstract:Nonlinear dimensionality-reduction methods such as UMAP and PaCMAP adaptively normalize local distances during graph construction, erasing neighborhood scale from the data. This distorts more than relative cluster sizes: sparse structures like bridges between transitioning cell types and narrow spectral spikes in hyperspectral images can be suppressed or lost entirely. DensMAP adds a density penalty to correct this, but this penalty competes with UMAP’s attraction-repulsion forces, scattering points far from their neighborhoods. ScaleMAP takes a different approach: each pairwise embedding displacement is divided by the geometric mean of the two endpoints’ original-space local radii, re-injecting scale information as a change of variables rather than as a competing objective. Across standard benchmarks and scientific datasets from transcriptomics, hyperspectral imaging, and flow cytometry, ScaleMAP matches DensMAP on density preservation while maintaining UMAP-level neighborhood preservation. In transcriptomic data, it recovers sparse bridges between cell populations that UMAP collapses; in flow cytometry, it faithfully represents density structure across 17 orders of magnitude. The same principle applied to PaCMAP yields consistently improved density preservation, suggesting the approach generalizes beyond UMAP.
[LG-100] Improving Relative Representations with Learned Anchors and Whitened Inner Products
链接: https://arxiv.org/abs/2605.30596
作者: Oscar Thorsted Svendsen,Nikolaj Holst Jakobsen,Fabian Mager,Hiba Nassar
类目: Machine Learning (cs.LG)
*备注: 14 pages, 5 figures
Abstract:Independently trained neural models typically converge to incompatible latent representations, creating a fundamental barrier to highly modular AI systems. While Relative Representations (RR) address this by mapping absolute coordinates to a shared space defined by similarities to common anchor points, traditional implementations rely on randomly sampled anchors and cosine similarity, which frequently fail to capture the anisotropic geometries of modern architectures like Transformers. In this work, we propose a robust framework for cross-model communication based on two improvements. We learn anchors as robust semantic prototypes and utilize a geometry-aware similarity metric which preserves discriminative magnitude information and is invariant to affine shifts. Our approach demonstrates significant gains in performance and consistency across vision and language tasks. Notably, it enables nearly lossless information transfer and stable zero-shot communication even between highly heterogeneous architectures, such as small language models of varying scales.
[LG-101] Learning Transferable Predictability Representations
链接: https://arxiv.org/abs/2605.30592
作者: Diyali Goswami,Auroop R. Ganguly
类目: Machine Learning (cs.LG)
*备注: 27 pages, 3 figures
Abstract:We study the problem of assigning a scalar score to a short trajectory window that reflects its position on an ordered continuum of predictability regimes, spanning structured deterministic dynamics to unstructured stochastic noise. Existing methods address deterministic-versus-stochastic discrimination within a single system and do not produce scores with a consistent numerical interpretation across systems. We formalize this as ordinal estimation over a five-level predictability ladder and identify a structural source of cross-system ambiguity: ranking supervision alone leaves the score coordinate unfixed up to a monotone reparameterization, which we term the gauge freedom of ordinal scoring. We propose the Gauge-Fixed Ordinal Network (GON), a temporal convolutional model trained with an anchor-and-variance objective that pins level-wise score means to shared target coordinates. GON operates on 2-jet features that expose local trajectory geometry, preserved by smooth flows and disrupted by stochastic surrogate procedures. On five held-out dynamical systems, initializing from a pretrained GON checkpoint consistently outperforms training from scratch across all window budgets, with adaptation depth reflecting geometric proximity to the training family. Zero-shot scores retain ordinal structure at the stochastic boundary, where surrogate procedures most strongly disrupt nonlinear geometry, and pretrained initialization consistently beats scratch across all window budgets. Pairwise discrimination and globally coherent ordinal scoring are distinct properties requiring a stable score coordinate for cross-system transfer, with direct implications for predictability assessment, model selection, and early-warning diagnostics across natural and engineered dynamical systems.
[LG-102] Zeroth-Order Non-Log-Concave Sampling with Variance Reduction and Applications to Inverse Problems ICML2026
链接: https://arxiv.org/abs/2605.30573
作者: M. Berk Sahin,Behzad Sharif,Abolfazl Hashemi
类目: Machine Learning (cs.LG)
*备注: Accepted to ICML 2026
Abstract:Sampling from high-dimensional, non-log-concave distributions with unnormalized densities remains a fundamental challenge in machine learning, particularly in black-box settings where gradient information is inaccessible or computationally prohibitive. While Langevin dynamics provides a principled framework for sampling when gradients are accessible, its extension to the black-box settings suffers from high variance and lacks non-asymptotic convergence guarantees for non-log-concave sampling. To address these limitations, we propose a variance-reduced zeroth-order Langevin sampling method. Our method employs a gradient estimator that substantially reduces the variance of the classical batched zeroth-order estimator and eliminates the unfavorable dimensional dependence of the batch size required for accurate estimation, enabling practical and stable sampling. We establish the first non-asymptotic convergence guarantees for zeroth-order non-log-concave sampling in terms of \varepsilon -relative Fisher information, and, under a Poincaré inequality assumption, squared total variation distance. We further propose ZO-APMC, a posterior sampling algorithm for black-box inverse problems with pre-trained score-based generative priors, establishing the first non-asymptotic convergence guarantees for such methods. We validate our theory through synthetic experiments and demonstrate strong empirical performance on practical linear and nonlinear inverse problems.
[LG-103] Supervised Training Rapidly Degrades Early Visual Cortex Alignment Across Biologically Plausible Learning Rules
链接: https://arxiv.org/abs/2605.30556
作者: Nils Leutenegger
类目: Machine Learning (cs.LG); Neurons and Cognition (q-bio.NC)
*备注: 7 pages, 4 figures
Abstract:Random, untrained neural networks consistently match or exceed trained networks in representational similarity to early visual cortex. This puzzling finding challenges the assumption that learning improves brain alignment. We investigate it by tracking representational similarity analysis (RSA) alignment to human fMRI data across training for four learning rules: backpropagation (BP), feedback alignment (FA), predictive coding (PC), and spike-timing-dependent plasticity (STDP). Using 720 object images from the THINGS database and fMRI data from three subjects across six visual ROIs, we measure Spearman correlations between model and brain representational dissimilarity matrices at eight training checkpoints (epochs 0-40). We find that (1) a single epoch of training reduces V1 alignment by 25-90%, depending on the learning rule; (2) backpropagation reduces V1 alignment most severely (delta r = -0.080), while predictive coding and STDP preserve substantially more (delta r ~ -0.04); and (3) a weaker, opposite tendency appears in object-selective cortex (LOC), where BP shows the largest increase in alignment during training, although the absolute change is small. These results suggest that untrained architectures capture low-level visual statistics through inductive biases alone, and that global error signals (BP) reshape early representations more aggressively than local learning rules (PC, STDP), which better preserve brain-like structure.
[LG-104] Destruction is a General Strategy to Learn Generation; Diffusions Strength is to Take it Seriously; Exploration is the Future ICLR
链接: https://arxiv.org/abs/2605.30553
作者: Pierre-André Noël
类目: Machine Learning (cs.LG); Information Theory (cs.IT)
*备注: Published April 27th, 2026 as an ICLR blogpost this https URL
Abstract:I present diffusion models as part of a family of machine learning techniques that withhold information from a model’s input and train it to guess the withheld information. I argue that diffusion’s destroying approach to withholding is more flexible than typical hand-crafted information withholding techniques, providing a rich training playground that could be advantageous in some settings, notably data-scarce ones. I then address subtle issues that may arise when porting reinforcement learning techniques to the diffusion context, and wonder how such exploration problems could be addressed in more diffusion-native ways. I do not have definitive answers, but I do point my fingers in directions I deem interesting. A tutorial follows this thesis, expanding on the destroy-then-generate perspective. A novel kind of probabilistic graphical models is introduced to facilitate the tutorial’s exposition.
[LG-105] Early Prediction of Future Behavioral Strategy from Process Traces
链接: https://arxiv.org/abs/2605.30550
作者: Robert Kasumba,Dennis Barbour,Chien-Ju Ho
类目: Machine Learning (cs.LG)
*备注:
Abstract:Adaptive systems often need to make task-specific decisions about people from limited evidence: a tutor may need to anticipate how a learner will approach a new problem, a game may need to adapt when a player enters a new level, and a human-AI system may need to infer whether a partner will persist with a plan or switch goals. These decisions depend on person-level tendencies that shape how people solve related tasks, but such tendencies are difficult to infer from standard behavioral evidence. One approach is to use aggregate outcome summaries, such as scores, completion rates, or productivity; these summaries are compact and available across tasks, but can collapse distinct behavioral processes into similar outcomes. Another approach is to use process-level traces, which record how behavior unfolds; however, process modeling within one task can entangle stable person-level tendencies with task-specific layout and affordances. In this work, we study early cross-task behavioral inference: whether partial source-task process traces can reveal transferable person-level structure that predicts strategy in a held-out target task. We introduce a Process-Level Latent Variable Model (PLVM), which encodes task-specific traces and fuses them into a shared person-level latent representation for cross-task prediction. In PowerWash Simulator, a naturalistic telemetry dataset of human gameplay, PLVM uses partial traces from two cleaning tasks to predict locally persistent Zone Planner behavior versus frequent Zone Hopper behavior in the held-out Fire Station level. Controlled simulations with known latent types show that cross-task fusion helps when source tasks reveal complementary dimensions of a shared latent process. These results suggest that process-level cross-task modeling can support early prediction of target-task strategy when observing sufficient target-task behavior is impractical.
[LG-106] SubsurfaceGen: Procedural Generation of Field-Scale Earth Models and Seismic Data
链接: https://arxiv.org/abs/2605.30541
作者: Joseph Stitt,Pratik Rathore,Madeleine Udell,Ching-Yao Lai
类目: Machine Learning (cs.LG); Geophysics (physics.geo-ph)
*备注: 38 pages
Abstract:Full waveform inversion (FWI) is the gold standard for subsurface imaging, with applications from carbon sequestration to energy and mineral exploration to earthquake hazard assessment. Machine learning approaches to FWI need field-scale, geologically diverse, and physically realistic training data, but existing resources such as Marmousi, SEAM, and OpenFWI fall short on spatial extent, temporal extent, geological diversity, and physical realism. We address these limitations with SubsurfaceGen, a GPU-accelerated generator for 3D velocity models and seismic data. Along with SubsurfaceGen, we release a paired dataset of 4,276 2D velocity slices, 5 s wavefields, and 8 s shot gathers drawn from 42 realistic, field-scale 3D velocity models, each spanning 10 km x 10 km laterally and 6.19 km deep at 10 m resolution. The dataset spans six geological settings – four built with SubsurfaceGen and two drawn from prior sources – relevant for carbon sequestration and hydrocarbon exploration. We use this dataset to evaluate neural operators on wavefield prediction and encoder-decoders on end-to-end velocity inversion, holding out one geological setting for out-of-distribution testing. These experiments surface failure modes at field-scale and demonstrate how SubsurfaceGen and the associated dataset can impact ML-based FWI.
[LG-107] DisasterLex: An Expert Concept-to-Schema Knowledge Graph for Geospatial Reasoning in Disaster Analytics
链接: https://arxiv.org/abs/2605.30538
作者: Yiming Xiao,Ankit Basu,Kai Yin,Sahil Vartak,Christian Swords,Ali Mostafavi
类目: Machine Learning (cs.LG)
*备注:
Abstract:Disasters are inevitable and increasingly costly, and effective response depends on querying structured tabular data: precise, information-dense records of hazard, exposure, vulnerability, and lifeline infrastructure that underpin disaster management. Current text-to-SQL methods enable natural-language access to such tables but transfer poorly to the disaster domain, where queries span heterogeneous geospatial schemas and require reasoning over causal relations. We introduce DisasterLex, a knowledge-graph-mediated framework that inserts an Expert Knowledge Graph (EKG) of curated concepts and typed causal edges between the user query and the database, bridged to schema by concept-to-table links. The orchestration runs four stages (identifying query entities, routing to the operational domain, planning over causal edges, and grounding the SQL), restricting the schema passed to the model at each step. We instantiate it on a disaster-analytics database (36 geospatial tables, 150 columns) with an EKG of 107 concepts, 117 causal edges, and 52 concept-to-schema links, evaluated on a 75-query test set. On all seven base models spanning proprietary and open-weight families, DisasterLex beats four state-of-the-art baselines (LightRAG, HippoRAG 2, ReFoRCE, CHESS) by 1.4x to 2.75x, with absolute scores of 1.65 to 3.56 (of 5.0). Error analysis shows baseline failures cluster in routing and multi-table SQL composition, the operations our orchestration explicitly addresses. Code, data, and the EKG artifact are available at this https URL and on Zenodo at this https URL.
[LG-108] he Long-Term Effects of Data Selection in LLM Fine-Tuning
链接: https://arxiv.org/abs/2605.30537
作者: Yuxin Yang,Aoxiong Zeng,Xiangquan Yang
类目: Machine Learning (cs.LG)
*备注: work in process
Abstract:Data selection is increasingly used to reduce the cost of large language model (LLM) fine-tuning, with recent methods prioritizing samples by current utility, diversity, quality, or influence. This paper studies a different question: when fine-tuning occurs over multiple stages, can selection strategies that look optimal now make the model less adaptable later? We introduce a long-horizon view of LLM data selection in which a selector is evaluated not only by immediate task performance, but also by future adaptation speed, forgetting, capability imbalance, and out-of-distribution robustness. We compare representative random, loss-based, gradient-based, diversity-based, quality-based, and utility-diversity selection families under a unified multi-stage protocol. Through controlled experiments designed to instantiate this protocol, we show how short-term selectors can exhibit rank reversal: they improve the current stage while slowing subsequent learning and increasing forgetting. We formalize this behavior as \emphmyopic selection, provide a simple local analysis of why it can occur, and propose a diagnostic Long-Horizon Aware Selection (LHAS) objective that augments immediate utility with coverage, future-proxy transfer, and anti-concentration terms. The study argues that data selection should be evaluated as a training intervention that shapes the model’s learning trajectory, rather than only as a local data-efficiency mechanism.
[LG-109] Representation Collapse in Sequential Post-Training of Large Language Models
链接: https://arxiv.org/abs/2605.30524
作者: Yichen Liu,Mingyu Chen,Hao Wang,Xiaoran Xu,Chenxi Lin,Rui Zhang,Yutong Zhou,Yuxin Yang,Jiarui Wu,Wei Sun
类目: Machine Learning (cs.LG)
*备注: work in progress
Abstract:Large language models are now adapted through chains of post-training stages rather than through a single instruction-tuning pass. This paper studies whether such sequential post-training gradually compresses internal representations into low-rank, anisotropic, and homogeneous feature spaces. We define a measurement suite for hidden states, logits, token trajectories, and LoRA updates, and we use it to analyze supervised fine-tuning, preference optimization, safety/refusal tuning, math and code specialization, and long chain-of-thought tuning under controlled stage orderings. The central hypothesis is that excessive representation concentration is not merely a geometric curiosity: it predicts reduced plasticity during later adaptation, weaker out-of-domain generalization, and poorer calibration. We further evaluate lightweight interventions, including mixed-domain replay, feature refresh, representation diversity regularization, and LoRA update decorrelation, as ways to preserve future learnability without giving up the behavioral gains of post-training.
[LG-110] Discovering a Zeta Map Algorithm on Dyck Paths via Mechanistic Interpretability
链接: https://arxiv.org/abs/2605.30482
作者: Xiaoyu Huang,Blake Jackson,Kyu-Hwan Lee
类目: Machine Learning (cs.LG)
*备注:
Abstract:Machine learning is increasingly used in mathematical discovery, but in mathematics the desired output is often not a prediction itself, but an explicit construction that can be checked independently. We study this setting through the zeta map on Dyck paths, a classical bijection in the combinatorics of the q,t-Catalan numbers. We train a deliberately small one-layer, one-head encoder-decoder transformer on this map and analyze its learned computation using mechanistic interpretability tools, including decoder cross-attention analysis, linear probing, and causal intervention. The analysis reveals a level-based mechanism: encoder representations make path levels linearly accessible, while the decoder selects and traverses input positions in a structured way. Translating these signals into combinatorics leads to the scaffolding map, an explicit peak-centered traversal algorithm for Dyck paths. We prove that this algorithm agrees with the zeta map, modulo a reversal convention in the labeling. This gives a controlled example of AI-assisted mathematical discovery in which mechanistic interpretability turns model behavior into a precise, human-verifiable combinatorial algorithm.
[LG-111] Universal Multiclass Transductive Online Learning
链接: https://arxiv.org/abs/2605.30479
作者: Steve Hanneke,Hongao Wang
类目: Machine Learning (cs.LG)
*备注:
Abstract:We consider the problem of universal transductive online classification with a possibly unbounded label space. This setting considers online learning, with the sequence of instances (without labels) known to the learner in advance. We say a concept class \mathcalH is learnable if there is a learning algorithm \mathcalA , such that for every realizable sequence, the number of mistakes made by \mathcalA grows at most sublinearly with the number of predictions. We characterize the learnability of this setting and show that there are only two possible optimal rates for the learnable classes: either bounded or increasing logarithmically. We introduce a new combinatorial structure, called ``Level-Constrained-Littlestone-Littlestone (LCLL) tree’', which, along with the indifference property, characterizes the learnability. We also extend the learnability result to the agnostic case and the case where only the stochastic process that generates the instance sequence is known.
[LG-112] Local Differential Privacy with Correlated Noise Achieves Central-DP Optimal Cost
链接: https://arxiv.org/abs/2605.30476
作者: Madhura Pathegama,Srikanth Avasarala,Viveck R. Cadambe,Juba Ziani
类目: Information Theory (cs.IT); Cryptography and Security (cs.CR); Machine Learning (cs.LG)
*备注:
Abstract:We study privately estimating the sum of n user-held values in the presence of an honest-but-curious server. This motivates requiring privacy not only at data release but also throughout server-side computation. We therefore adopt the local (pure) differential privacy model, in which each user transmits a noise-perturbed value. It is well known that independent local noise typically incurs a substantial utility loss compared to the centralized model, where noise is added only after aggregation. We show that this gap is not fundamental. By carefully designing correlations among the locally added noise variables, we construct \varepsilon -DP mechanisms whose estimation cost matches the optimal cost achievable in the centralized setting, up to an arbitrarily small error. Subjects: Information Theory (cs.IT); Cryptography and Security (cs.CR); Machine Learning (cs.LG) Cite as: arXiv:2605.30476 [cs.IT] (or arXiv:2605.30476v1 [cs.IT] for this version) https://doi.org/10.48550/arXiv.2605.30476 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-113] Can Subgraph Explanations Be Weaponized to Steal Graph Neural Networks? NEURIPS2026
链接: https://arxiv.org/abs/2605.30470
作者: Ojas Nimase,Jiate Li,Yue Zhao,Yushun Dong
类目: Machine Learning (cs.LG)
*备注: 28 pages, 8 figures, 10 tables. Under review at NeurIPS 2026
Abstract:Graph Machine Learning as a Service (GMLaaS) platforms increasingly implement explainability interfaces to meet regulatory transparency requirements. However, this transparency creates exploitable vulnerabilities for model extraction attacks. We present the first model extraction attack specifically designed for graph classification under strict black-box constraints where the attacker observes only discrete class labels and binary explanation masks (no probability scores, gradients, or confidence values). Our method (1) uses model explanation outputs to guide Monte Carlo edge sensitivity estimation toward decision boundaries, with Hoeffding concentration guarantees on estimation accuracy and (2) exploits explanation subgraphs to efficiently narrow the boundary search space. Extensive experiments on benchmark graph datasets across multiple domains demonstrate our method’s superiority over comparable baselines. These findings demonstrate that such explainability interfaces create exploitable attack surfaces, informing both defensive mechanisms and policy frameworks for explainable AI mandates. The implementation code is provided in this https URL.
[LG-114] DisjunctiveNet: Neural Symbolic Learning via Differentiable Convexified Optimization Layers ICML2026
链接: https://arxiv.org/abs/2605.30456
作者: Shraman Pal,Can Li
类目: Machine Learning (cs.LG); Optimization and Control (math.OC)
*备注: ICML 2026
Abstract:Many learning tasks in science and engineering are characterized by sparse datasets, which limits the effectiveness of purely data-driven approaches. At the same time, these problems are often accompanied by rich domain knowledge derived from physical laws, operational requirements, and expert heuristics. Such knowledge is frequently expressed as rules involving logical propositions and linear inequalities. Existing neuro-symbolic methods typically enforce these rules approximately through soft penalties, assume input-independent rules when designing specialized architectures, or rely on non-differentiable post-processing at inference time to achieve hard constraint satisfaction. While recent advances in differentiable optimization layers enable end-to-end feasibility enforcement within neural networks, extending these approaches to logical or mixed-integer rules remains challenging due to inherent nonconvexity. In this work, we propose a unified end-to-end framework for enforcing hard, input-dependent mixed integer linear constraints within neural networks. Our approach represents rules as disjunctive constraints and applies hierarchical convex relaxations to obtain convex hull formulations. These relaxations yield tractable linear constraints that can be embedded as differentiable optimization layers while enabling exact rule satisfaction. We demonstrate the effectiveness of the proposed framework on real-world datasets, achieving perfect rule satisfaction and strong predictive performance.
[LG-115] VeriGate: Verifier-Gated Step-Level Supervision for GRPO
链接: https://arxiv.org/abs/2605.30451
作者: Aakriti Agrawal,Minghui Liu,Furong Huang
类目: Machine Learning (cs.LG)
*备注:
Abstract:Group Relative Policy Optimization (GRPO) is an effective recipe for training reasoning models with verifier-based outcome rewards, but its supervision is sparse: when all sampled trajectories for a prompt receive the same verifier reward, the group-relative advantage collapses to zero and learning stalls. Outcome-only rewards also provide no step-level credit assignment, limiting exploration and making it harder to learn robust reasoning. We present VeriGate (Verifier-Gated Step-Level GRPO), a verifier-gated extension of GRPO that addresses these limitations with three design choices. First, VeriGate keeps the verifier in charge whenever verifier rewards induce a meaningful preference among sampled trajectories, and uses process supervision only when verifier rewards are degenerate. Second, instead of collapsing Process Reward Model (PRM) step scores into a single trajectory reward, VeriGate converts them into future-cumulated rewards to assign continuation-aware credit. Third, VeriGate transforms these rewards into group-normalized token-level advantages, restoring informative gradients and fine-grained credit assignment while remaining less susceptible to reward hacking than methods that optimize aggregated PRM scores. Empirically, training on MATH with 1.5B and 7B Qwen2.5-Instruct models and evaluating on six reasoning benchmarks, VeriGate improves average accuracy by about 20% and 12% for 1.5B and 7B models respectively, substantially reduces zero-gradient failures, decreases reward-hacking behavior, and improves reasoning quality relative to outcome-only GRPO and PRM-as-outcome baselines.
[LG-116] Smaller and Faster 3DGS via Post-Training Dictionary Learning
链接: https://arxiv.org/abs/2605.30396
作者: Jiarong Gong,Jonas Unger,Ehsan Miandji
类目: Graphics (cs.GR); Machine Learning (cs.LG)
*备注:
Abstract:3D Gaussian Splatting (3DGS) is a promising neural scene representation for real-time rendering, but trained models often suffer from large memory footprints, limiting deployment on less powerful devices. Existing compression techniques often lead to architectures with several additional trainable parameters. While achieving outstanding compression ratios, they introduce noticeable drops in image quality. In this work, we introduce the first dictionary-learning-based compression framework for 3DGS. The proposed post-training compression pipeline can be deployed in virtually any 3DGS model without the need for re-training or modifications to existing 3DGS models. Our compression framework is straightforward to implement, yet provides significant compression capabilities, preserves image quality, and improves real-time rendering performance. Across 13 benchmark scenes, our approach achieves an average compression ratio of 3.95x, 3.10x, and 4.55x when applied to 3DGS, 3DGS-MCMC, and PixelGS, respectively. This yields consistent rendering speedups of 23.3%, 24.3%, and 25.3%, while maintaining image quality.
[LG-117] he Inclusion Depth of Pattern Languages: An Open Problem in Algorithmic Learning Theory COLT2005
链接: https://arxiv.org/abs/2605.30389
作者: Wei Luo
类目: Formal Languages and Automata Theory (cs.FL); Machine Learning (cs.LG)
*备注: 2 pages. Open problem from COLT 2005. Generic author-prepared version for arXiv. Originally appeared in Learning Theory, 18th Annual Conference on Learning Theory, COLT 2005, Bertinoro, Italy, June 2005
Abstract:Pattern languages are a classical model in formal language theory and algorithmic learning theory. This note formulates the problem of computing the inclusion depth of a pattern language: the length of the longest strict inclusion chain from the universal pattern language to the language generated by a given pattern. Inclusion depth captures the mind-change complexity of pattern identification from positive data. The central open question is whether the inclusion depth ID_Sigma§ is computable for every pattern p over every finite alphabet Sigma with at least two symbols, and whether it is computable in polynomial time. A simple conjectured formula, ID_Sigma§ = 2|p| - #var§ - 1, would imply a linear-time algorithm. The problem connects pattern language inclusion, combinatorics on words, language identification in the limit, and mind-change-bounded learning.
[LG-118] A Novel Evaluation Metric for Unsupervised Learning in AIS-Based Maritime Anomaly Detection: MADQI
链接: https://arxiv.org/abs/2605.30388
作者: Ismet Gocer,Zakirul Bhuiyan,Raza Hasan,Shakeel Ahmad
类目: Machine Learning (cs.LG)
*备注: 26 pages, A new Eval Metric for Unsupervised Machine Learning
Abstract:This paper introduces a new systematic framework for detecting anomalies in maritime Automatic Identification System (AIS) datasets. These anomalies include abnormal vessel behaviours related to speed, position jumps, time gaps, and turn angles. Although unsupervised learning algorithms such as Isolation Forest are widely used for detecting anomalous vessel movements, they often lack systematic and meaningful evaluation measures. To address this limitation, we propose a novel quality metric called Maritime Anomaly Detection Quality Index (MADQI). The prosed MADQI is a composite index designed to evaluate the anomaly detection performance of machine learning models without requiring labelled data. The proposed framework uses Haversine distance calculations to analyse AIS datasets and identify anomalies based on their spatial and behavioural characteristics. The proposed MADQI evaluation framework integrates four interconnected metrics: Anomaly Rate Consistency (ARC), Physical Plausibility Score (PPS), Score Distribution Separation (SDS), and Extreme Case Evidence (ECE). These metrics are combined through automatic normalisation using multi-chunk evaluation and adaptive scaling techniques. Experimental results on the AIS dataset show that the proposed framework achieved a MADQI score of 80.37%, demonstrating its effectiveness for unsupervised anomaly detection. In particular, the algorithm performed strongly in identifying abnormal vessel behaviour. Among the individual MADQI components, ECE and ARC achieved scores of 0.907 and 1.000, respectively, indicating excellent capability in detecting extreme anomalies and maintaining anomaly rate consistency. Overall, these results are encouraging and demonstrate that the proposed framework provides a reliable and meaningful approach for evaluating unsupervised anomaly detection in maritime AIS data.
[LG-119] Gait2Hip-60: A Unified Deep Learning Benchmark for Predicting Hip Muscle Forces and Joint Moments from Multi-Cadence Gait Kinematics
链接: https://arxiv.org/abs/2605.30374
作者: Jiaqi Zhang,Ji Hou,Qing Sun,Xianzhi Gao,Bo Huo
类目: Machine Learning (cs.LG)
*备注: 16 pages, 9 figures. Code and dataset publicly available
Abstract:Estimating hip muscle forces and joint moments during gait typically relies on musculoskeletal simulation, which is informative but time-consuming and difficult to apply in clinical settings. This study developed a deep learning framework to predict these hip dynamics parameters directly from lower-limb gait kinematics and compared three representative sequence models under a unified protocol. Gait data were collected from 60 healthy adults under three metronome-guided cadence conditions. Ten bilateral lower-limb joint angles were used as inputs, and OpenSim-derived hip muscle forces and hip joint moments were used as reference outputs. Three deep learning models of LSTM, Transformer, and Mamba were trained and evaluated using the same subject-level split, preprocessing pipeline, and metrics. The best model was then directly tested on an external cohort of 9 patients with osteonecrosis of the femoral head (ONFH) without retraining. In the healthy-subject benchmark, Transformer achieved the best subject-level mean performance for both hip muscle force prediction (RMSE = 1.33 N/kg, MAE = 0.57 N/kg, R2 = 0.819) and hip joint moment prediction (RMSE = 0.11 Nm/kg, MAE = 0.07 Nm/kg, R2 = 0.862), with similar advantages across walking cadences. In zero-shot external validation, Transformer retained moderate predictive ability in ONFH for hip muscle force prediction (RMSE = 1.51 N/kg, MAE = 0.70 N/kg, R2 = 0.537) and hip joint moment prediction (RMSE = 0.17 Nm/kg, MAE = 0.12 Nm/kg, R2 = 0.569). These findings support the feasibility of estimating hip dynamics from gait kinematics, identify Transformer as a strong baseline, and highlight the need for broader pathological validation and improved generalization before clinical application.
[LG-120] From Mean-Field Limits to Semiclassical Concentration: Global Convergence of the Canonical Evolutionary Strategy
链接: https://arxiv.org/abs/2605.30371
作者: Matías Neto,Nicolás Garay,Luis Martí,Nayat Sanchez-Pi
类目: Neural and Evolutionary Computing (cs.NE); Machine Learning (cs.LG); Dynamical Systems (math.DS)
*备注:
Abstract:We address the issue of global convergence in stochastic continuous optimization. For that purpose, we formulate the Canonical Evolutionary Strategy (CES) as a controlled mathematical framework to analyze global convergence in evolutionary algorithms via the semiclassical limit of a Schrödinger-type replicator-mutator equation. We provide a rigorous hierarchy from a discrete individual-based dynamics to a deterministic mean-field limit, demonstrating that global convergence is governed by the principal eigenfunction of the underlying operator. This property, defined as Geometric Selection, naturally prioritizes robust, flat optima over narrow local traps, offering a mathematical justification for the ‘‘survival of the flattest’’ phenomenon. Moreover, unlike consensus-driven methods that are prone to premature variance collapse when the global minimizer resides outside the initial support, the replicator-mutator dynamics of CES facilitate intrinsic mass transport. High-dimensional benchmarks (d = 30) confirm this advantage, showing that CES achieves lower residual errors in shifted initialization scenarios where standard consensus-driven and gradient-based methods fail to migrate effectively. By shifting the focus from point-wise consensus to spectral concentration, our framework provides a robust theoretical foundation for global convergence in Evolution Strategies (ES) without the need for additional numerical heuristics.
[LG-121] Kernel Foundry: A Diagnosis-driven Evolutionary Kernel Optimizer with Multi-Experts
链接: https://arxiv.org/abs/2605.30359
作者: Zixuan Huang,Da Chen,Kecheng Huang,Lihao Yin,Xing Li,Huiling Zhen,Mingxuan Yuan,Zili Shao
类目: Neural and Evolutionary Computing (cs.NE); Distributed, Parallel, and Cluster Computing (cs.DC); Machine Learning (cs.LG); Performance (cs.PF); Software Engineering (cs.SE); Systems and Control (eess.SY)
*备注:
Abstract:Generating high-performance GPU kernels remains challenging due to the need for both correctness and hardware-aware optimization. While large language models (LLMs) show promise in code generation, they often fail to produce kernels that are both correct and efficient. We propose Kernel Foundry, a diagnosis-driven evolutionary framework for automatic GPU kernel optimization. Our method combines expert-guided, retrieval-augmented initialization with a multi-island evolutionary search, where candidate kernels are iteratively refined using structured diagnostic feedback. A centralized experience library accumulates reusable optimization knowledge to guide subsequent evolution, while explicit mechanisms prevent cheating behaviors that bypass kernel-level computation. Experiments on KernelBench show that our method consistently improves both correctness and performance over strong baselines, achieving up to 100% correctness on Level~2. Subjects: Neural and Evolutionary Computing (cs.NE); Distributed, Parallel, and Cluster Computing (cs.DC); Machine Learning (cs.LG); Performance (cs.PF); Software Engineering (cs.SE); Systems and Control (eess.SY) Cite as: arXiv:2605.30359 [cs.NE] (or arXiv:2605.30359v1 [cs.NE] for this version) https://doi.org/10.48550/arXiv.2605.30359 Focus to learn more arXiv-issued DOI via DataCite
[LG-122] QASM-Eval: A Dataset to Train and Evaluate LLM s on OpenQASM-3 Beyond Quantum Circuits
链接: https://arxiv.org/abs/2605.30358
作者: Zhenxiao Fu,Lei Jiang,Fan Chen
类目: Machine Learning (cs.LG); Quantum Physics (quant-ph)
*备注:
Abstract:Quantum computing remains in the Noisy Intermediate-Scale Quantum (NISQ) era, where the performance is highly constrained to noise. Addressing the limitation often requires hardware-facing capabilities beyond gate-sequence circuit specification, including mid-circuit measurement and classical feedback for quantum error correction (QEC), precise timing control for dynamical decoupling (DD), and pulse-level waveform access for calibration. OpenQASM-3 was introduced to expose exactly these capabilities, providing a hardware-level programming interface. However, despite the rapid progress of large language models in code generation, there is still no dataset specifically designed to train and evaluate LLMs on OpenQASM-3 programs that involve its advanced hardware-oriented features. To address this gap, we introduce QASM-Eval, the first comprehensive dataset designed to train and evaluate LLMs on OpenQASM-3. Rather than focusing on quantum algorithm design or reasoning, QASM-Eval explicitly targets the language’s hardware-facing features. QASM-Eval comprises an expert-verified test set of 100 tasks and a training set of 4,000 tasks, systematically covering classical logic, timing scheduling, pulse control, and complex real-world workflows. To automatically validate generated programs, we check syntax, quantum states and program timeline using an extended verifier. Our evaluation reveals that while state-of-the-art LLMs struggle heavily in OpenQASM-3 coding tasks, targeted fine-tuning on QASM-Eval yields significant gains. QASM-Eval provides a crucial benchmark and training foundation to accelerate the development of reliable LLM assistants for hardware-facing quantum programming in NISQ era. Data and code: this https URL
[LG-123] Discovering Thermodynamically Admissible Dissipation Potentials via Grammar-Based Symbolic Regression
链接: https://arxiv.org/abs/2605.31532
作者: Federico Califano,Jacopo Ciambella
类目: oft Condensed Matter (cond-mat.soft); Machine Learning (cs.LG)
*备注:
Abstract:Constitutive laws for inelastic materials must satisfy strict thermodynamic admissibility requirements, yet current data-driven approaches sacrifice interpretability, even when formal guarantees are provided by physics-encoded architectures. We propose a symbolic regression framework for the data-driven discovery of dissipation potentials governing the evolution of internal variables within the Generalized Standard Materials (GSM) formalism. Starting from the Clausius–Duhem inequality, we enforce the thermodynamic requirements, convexity and non-negativity, that the dual dissipation potential must satisfy to guarantee non-negative mechanical dissipation. These requirements are formulated in the general subdifferential setting, encompassing rate-dependent (viscoelastic) and viscoplastic dissipative mechanisms, including potentials with genuine elastic domains, within a unified framework. Candidate potentials are generated by a composition-extended convexity-preserving grammar that guarantees thermodynamic admissibility \emphby construction. The framework is validated on synthetic datasets spanning Newtonian, power-law, and Bingham viscoplastic ground truths under process and measurement noise, and on experimental oscillatory shear measurements of a synthetic elastomer across multiple strain amplitudes and frequencies, where the discovered potentials reproduce the amplitude-dependent softening of the dynamic moduli and outperform a calibrated linear Zener baseline.
[LG-124] Modeling Covariate Transition for Efficient Estimation of Longitudinal Treatment Effects in Randomized Experiments ICML’26
链接: https://arxiv.org/abs/2605.31443
作者: Naoki Chihara,Tatsushi Oka,Yasuko Matsubara,Yasushi Sakurai,Shota Yasui
类目: Methodology (stat.ME); Machine Learning (cs.LG); Econometrics (econ.EM); Statistics Theory (math.ST)
*备注: Accepted by ICML’26
Abstract:We present a regression-adjustment framework designed for the estimation of longitudinal treatment effects in randomized experiments under static regimes. While regression-adjustment methods are useful for variance reduction in randomized experiments by using pre-treatment covariates, they usually focus only on average effects, from which we cannot obtain valuable insights into when the effects appear and how long they continue. To address this issue, we consider intermediate outcomes and evolving post-treatment covariates over time, and we represent such dynamic trajectories using transition kernels. Furthermore, we establish the asymptotic normality and the semiparametric efficiency bound for our estimator, enabling more powerful statistical inference. Simulation studies and empirical analysis using A/B test data from a streaming platform in Japan show the practical advantages of our method.
[LG-125] Improved Guarantees for Langevin Monte Carlo with Averag e Smoothness
链接: https://arxiv.org/abs/2605.31413
作者: Arnak S. Dalalyan,Avetik Karagulyan
类目: atistics Theory (math.ST); Machine Learning (cs.LG)
*备注:
Abstract:We establish improved nonasymptotic bounds for Langevin Monte Carlo in the strongly log-concave setting, when the error is measured by the Wasserstein distance. The main result shows that the discretization error is governed by an average coordinate-wise smoothness constant, rather than by the usual global smoothness constant. The proof is short and probabilistic, and relies on a refined use of the synchronous coupling. We further show that the same ideas lead to improved bounds for variable step sizes, for potentials whose Laplacian is Lipschitz-continuous, and for finite-sum problems sampled by stochastic-gradient Langevin dynamics with fixed point control variates. In the Laplacian-smooth case, the usual Hessian-Lipschitz contribution is replaced by a weaker trace-type third-order smoothness quantity. In the finite-sum setting, the resulting SGLD bound improves the dependence on the root mean square smoothness of the component functions. Applications to generalized linear models with Gaussian design show that these refinements can yield substantial, dimension-dependent improvements over previously known bounds, especially for correlated covariates.
[LG-126] Deep-learning-based low-energy trigger algorithms for the Hyper-Kamiokande experiment
链接: https://arxiv.org/abs/2605.31391
作者: Katharina Lachner,Saúl Alonso-Monsalve,Benjamin Richards,Davide Sgalaberna
类目: Instrumentation and Detectors (physics.ins-det); Machine Learning (cs.LG); High Energy Physics - Experiment (hep-ex)
*备注: 16 pages, 6 figures
Abstract:Modern machine learning techniques have become increasingly important in particle physics because of their powerful pattern-recognition capabilities, including in real-time data acquisition where stringent runtime constraints apply. This paper details the performance of deep-learning-based trigger algorithms for a large water Cherenkov detector such as Hyper-Kamiokande aimed at low-energy neutrino events (below 7 MeV). The performance of custom neural-network supervised classifiers is shown alongside two anomaly-detection approaches trained solely on detector noise: a pure autoencoder and an energy-based model based on Manifold Projection–Diffusion Recovery (MPDR). The supervised model shows signal identification efficiencies of 76.7% for single electrons of 3 MeV kinetic energy, significantly exceeding signal efficiencies obtained from a traditional hit-count-based trigger of 26.4%, as does the MPDR approach with 31.8%. Runtime evaluations on GPU yield per-window inference latencies well below the millisecond scale, indicating that real-time operation is feasible.
[LG-127] Wall-Clock Complexity for Zeroth-Order Optimization with Tunable Oracle Fidelity
链接: https://arxiv.org/abs/2605.31346
作者: Alexandra Suvorikova,Igor Pavlov,Artem Vasin,Georgii Bychkov,Anastasia Antsiferova,Darina Dvinskikh,Alexander Gasnikov
类目: Optimization and Control (math.OC); Machine Learning (cs.LG)
*备注:
Abstract:Zeroth-order (black-box) optimization is applied when gradients are unavailable and objective evaluations rely on expensive simulations. In many such applications, the oracle fidelity is tunable: higher-accuracy queries reduce noise but incur higher computational costs. To capture this trade-off, we study an accuracy-aware wall-clock model where each query with fidelity \delta has a cost c(\delta) , and we minimize the total time T_\mathrmtotal = \sum_k=1^N c(\delta_k) , subject to a target accuracy constraint. We show how the choice of oracle type, noise model, and optimization scheme induces explicit wall-clock-optimal choices for the algorithmic parameters. For instance, we demonstrate that accelerated methods can be wall-clock inferior to non-accelerated schemes. Furthermore, we characterize the conditions under which a constant fidelity strategy is optimal in the Big-O sense. Our framework provides a unified methodology to translate convergence guarantees into practical fidelity and batching recommendations.
[LG-128] Log-Ratio Propagation on the Simplex: A Theory of Cellwise Contamination for Compositional Data
链接: https://arxiv.org/abs/2605.31345
作者: Matthias Templ
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Methodology (stat.ME)
*备注: 50 pages, no figures; 11-page supplement included as an ancillary file. A companion methods paper (cellPcaCoDa: cellwise-robust PCA for compositional data) is forthcoming
Abstract:Compositional data must be analysed through log-ratios: scale invariance, the defining axiom of the field, leaves no alternative. The centred log-ratio divides by the geometric mean of every part, so a single contaminated component shifts every centred-log-ratio coordinate at once, displacing the log-ratio vector by a fixed amount that no choice of coordinates can reduce. We develop a theory of cellwise contamination on the simplex around this observation. A scale-invariant contamination model built from multiplicative perturbation combines with a propagation theorem showing that corruption of a single raw part induces a rank-one shift of the log-ratio vector, with direction determined by the contrast matrix. The resulting perturbation pattern is not equivalent to any independent cellwise contamination model in log-ratio coordinates – so standard Euclidean cellwise methods applied to log-ratios are ill-posed under the simplex contamination mechanism. For estimators whose Euclidean cellwise breakdown is witnessed by a column-concentrated configuration – a class including MCD, S -, \tau -, and coordinate-wise M -estimators of location and scatter – the cellwise breakdown value on the simplex is reduced by the factor (D-1)/D relative to its Euclidean counterpart, a reduction that is tight and arises purely from the normalisation mismatch between nD raw cells and n(D-1) ilr cells. The cellwise influence function for the variation matrix carries a diagnostic fingerprint: contamination of a single part inflates exactly one row and column, identifying the responsible component. These results form the theoretical foundation for cellwise-robust methods on the simplex; a companion paper develops a cellwise-robust PCA estimator that exploits the propagation geometry and demonstrates it on simulated and geochemical data.
[LG-129] S3LDBO: A Snapshot Single-Loop Algorithm for Decentralized Bilevel Optimization
链接: https://arxiv.org/abs/2605.31311
作者: Chao Yin,Youran Dong,Shiqian Ma,Bofan Wang,Junfeng Yang
类目: Optimization and Control (math.OC); Distributed, Parallel, and Cluster Computing (cs.DC); Machine Learning (cs.LG)
*备注:
Abstract:Networked AI systems increasingly rely on multiple agents that collaboratively learn and adapt models over communication networks. In such systems, bilevel formulations naturally arise in hyperparameter optimization, data cleaning, and meta-learning, but the repeated evaluation of gradients, Jacobians, and Hessians can impose a substantial computational burden on individual agents. To address this challenge, we propose Snapshot-SLDBO (S ^3 LDBO), an efficient single-loop decentralized bilevel optimization algorithm that enables agents to intermittently skip expensive derivative evaluations through a snapshot mechanism. This mechanism can be interpreted as an autonomous computation-adaptation strategy for networked AI, where agents selectively perform costly local updates while maintaining global collaborative learning. We establish the ergodic iteration complexity and the high probability nonergodic iteration complexity of the proposed algorithm within a deterministic setting. Experimental results on hyperparameter optimization with synthetic and MNIST datasets, data hyper-cleaning on Fashion-MNIST, and decentralized meta-learning on miniImageNet demonstrate that the proposed algorithm improves computational efficiency while maintaining competitive learning performance.
[LG-130] mRNAutilus: Multi-Objective-Guided Discrete Generation of mRNA with Optimized Therapeutic Properties
链接: https://arxiv.org/abs/2605.31296
作者: Sawan Patel,Sophia Tang,Yesol Kim,Yinuo Zhang,Divya Srijay,Ping-Jung Lin,Shambhavi Shubham,Fengmei Pi,Cedric Wu,Sherwood Yao,Pranam Chatterjee
类目: Biomolecules (q-bio.BM); Machine Learning (cs.LG)
*备注:
Abstract:Therapeutic mRNA design requires coordinating multiple interacting sequence features across the full transcript, where codon usage, untranslated regions (UTRs), and their coupling jointly determine stability, translation efficiency, and protein expression. Here, we present mRNA generation via unrolled trajectories and informed latent updates (mRNAutilus), a framework for simultaneous codon optimization and de novo UTR design directly from sequence. mRNAutilus combines a masked discrete diffusion model trained on millions of full-length mRNAs with Monte Carlo Tree Guidance to generate Pareto-efficient sequences under multiple functional objectives, using lightweight regressors over model embeddings to predict half-life, translation efficiency, and protein abundance. Unlike recent methods that design coding sequences and UTRs separately or rely on post hoc assembly and screening, mRNAutilus generates complete transcripts in a single process optimized across properties. Across diverse targets, zero-shot mRNAs encoding P. pyralis luciferase achieve over 400-fold higher expression than wild-type and outperform commercial and machine learning-designed baselines, including zero-shot generative approaches. Zero-shot SARS-CoV-2 Spike mRNAs exceed clinically used and commercial constructs and match or surpass lab-optimized designs with improved durability. We further demonstrate generality in therapeutic settings, including prime editing (PEMax) and programmable proteome modulation, where mRNAutilus-designed constructs enhance expression of peptide-guided E3 ligases (uAbs) for beta-catenin degradation. These results establish a sequence-based, multi-objective framework for generating functional mRNAs tailored to diverse biological applications.
[LG-131] Memory by Design: Probabilistic Sequence Layers
链接: https://arxiv.org/abs/2605.31163
作者: Matthew Dowling,Hyungju Jeon,Cristina Savin,Il Memming Park
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注: Preprint, in submission
Abstract:We introduce the design-model framework: a way to derive efficient recurrent sequence maps from explicit assumptions about memory. A design model writes evidence into memory by exact Bayesian filtering; a query-dependent readout produces a predictive distribution whose mean is the layer output. In our linear-Gaussian instantiation, the \emphBayesian Layer propagates both a mean and a covariance: the covariance tracks uncertainty over stored associations, steering writes toward uncertain directions, attenuating gains as evidence accumulates, and preserving confident memories. The same framework unifies several sub-quadratic recurrences. Linear attention, GLA, and Mamba-2/SSD are exact filters under one design model, whereas DeltaNet and related Delta-rule models arise as covariance-reset reductions under another. Restoring the covariance yields closed-form predictions for retrieval dynamics, verified empirically, and improves robustness beyond the training regime across controlled collision studies, learned associative recall, and the Zoology MQAR benchmark; distilling Bayesian Layers into a pretrained 340M Gated DeltaNet improves RULER long-context retrieval at matched compute.
[LG-132] Approximation and learning of anisotropic and mixed smooth functions by deep ReLU neural networks
链接: https://arxiv.org/abs/2605.31152
作者: Yunfei Yang,Jun Fan
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Numerical Analysis (math.NA)
*备注:
Abstract:This paper studies how efficiently deep ReLU neural networks can approximate and learn smooth functions. When the error is measured in L^p([0,1]^d) norm and the approximator is a network with width W and depth L , recent works have proven the supper approximation rate \mathcalO((WL)^-2s/d) for Besov space \mathcalB^s_q,r([0,1]^d) under the Sobolev embedding condition s/d1/q-1/p . In order to overcome the curse of dimensionality in this rate, we extent this result to anisotropic and mixed smooth function classes. We establish the approximation rate \mathcalO((WL)^-2\tildes) for anisotropic Besov space \mathcalB^\boldsymbols_q,r([0,1]^d) with anisotropic smoothness \boldsymbols=(s_1,\dots,s_d) under the embedding condition \tildes 1/q-1/p , where the mean smoothness \tildes = (\sum_i=1^d s_i^-1)^-1 . For mixed smooth Besov space \mathcalMB^s_q,r([0,1]^d) with mixed smoothness s1/q-1/p , we show that the approximation rate \mathcalO((WL)^-2s) holds up to logarithmic factors. Using these results, we also derive approximation bounds for the composition of anisotropic Besov functions. As an application, it is shown that deep ReLU neural networks can achieve minimax optimal rates up to logarithmic factors for a wide range of smooth function classes.
[LG-133] Free energy Estimation on Any State Space
链接: https://arxiv.org/abs/2605.31063
作者: Jiajun He,Zijing Ou,Francisco Vargas,Yingzhen Li,José Miguel Hernández-Lobato,Carles Domingo-Enrich,Yuanqi Du
类目: Machine Learning (stat.ML); Machine Learning (cs.LG); Chemical Physics (physics.chem-ph); Computational Physics (physics.comp-ph)
*备注:
Abstract:Free energy estimation is a fundamental yet challenging problem, from physics to statistics. Classical approaches rely on thermodynamic transformations, ranging from direct estimation, quasistatic integration, to finite-time averaging. Recent work [He and Du et al., 2025] learns neural transports to significantly accelerate the efficiency in the finite-time regime. In this paper, we generalize this framework to arbitrary state spaces. Building on this view, we develop a generalized neural transport learning approach for efficient estimation. Experiments validate the effectiveness and efficiency of the proposed method beyond continuous settings, extending to discrete and multimodal spaces as well as autoregressive settings. Beyond free energy estimation, we establish algebraic identities and reveal a group-theoretic structure linking infinitesimal time reversal and generalized Doob’s h -transforms, showing that their compositions form a generalized dihedral group.
[LG-134] Hedging on the Frontier: Learning New Tasks with Few Samples
链接: https://arxiv.org/abs/2605.30997
作者: Tobias Wegel,Federico Di Gennaro,Geelon So,Fanny Yang
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注:
Abstract:When a learner faces a new task with few samples, it must leverage any available side information. In practice, this often comes in the form of model evaluations on related tasks in public benchmarks. A key question then is how to model task relatedness such that it is both realistic and the benchmark evaluations lead to provable gains. Empirically, we observe that weak monotonicity is often approximately satisfied: if a model dominates another on many benchmarks, it also tends to outperform on the new task. We explore the statistical complexity of learning under (approximate) weak monotonicity, leveraging it within two learning paradigms: transfer learning and model selection aggregation. We show that not only can we prune the model class based on monotonicity, but we can also further adapt to the geometry of the available trade-offs by hedging on the frontier.
[LG-135] Batched Stochastic Linear Bandits with 1-Bit Communication Constraints
链接: https://arxiv.org/abs/2605.30976
作者: Ivan Lau,Daniel McMorrow,Kevin Jamieson,Jonathan Scarlett
类目: Machine Learning (stat.ML); Information Theory (cs.IT); Machine Learning (cs.LG)
*备注:
Abstract:We study stochastic linear bandits under a natural combination of batching and communication constraints: the time horizon is partitioned into batches of equal size B , and during each batch the learner sends B requested arm pulls to an agent, who then observes the corresponding B rewards and responds with a single bit of feedback to the learner. For each batch, the learner specifies the 1-bit quantization rule the agent uses, which may depend on all previously received bits but not on any past rewards directly. This setting addresses a significant yet unexplored ``middle ground’’ between previous models having per-round quantization only or total bit budgets only. We establish a minimax lower bound showing that \Omega(B\min\d,\log\lvert \mathcalA \rvert) regret is unavoidable due to the 1-bit communication bottleneck, even in the absence of noise. Combined with standard statistical limits, this yields a general lower bound of \widetilde\Omega(B\min\d,\log\lvert \mathcalA \rvert\ + \sqrtdT \min\d,\log\lvert \mathcalA \rvert) . We develop two phased-elimination algorithms based on G -optimal designs and 1-bit mean estimation. The first achieves \widetildeO(dB + d\sqrtT) regret, matching the lower bound up to logarithmic factors when \lvert \mathcalA \rvert = \exp(\Omega(d)) , and the second incorporates a safe-arm identification and warm-start procedure to obtain \widetildeO(B\log\lvert \mathcalA \rvert + d^3/2\sqrtB + \sqrtdT\log\lvert \mathcalA \rvert) regret, which is near-optimal in broad scaling regimes of (\lvert \mathcalA \rvert, B, d, T) . Together, our results demonstrate that a single bit of feedback per batch suffices to nearly match the minimax regret of unconstrained linear bandits in broad scaling regimes, even for batch sizes as large as \Theta(\sqrtT) .
[LG-136] A Unifying View of Anchoring via Operator-Side Tikhonov Regularization
链接: https://arxiv.org/abs/2605.30905
作者: Zihao Chen
类目: Optimization and Control (math.OC); Machine Learning (cs.LG)
*备注:
Abstract:Anchored fixed point and monotone equation methods, including Halpern iteration, extra anchored gradient, and their relatives, add a vanishing pull toward a reference point to obtain last-iterate guarantees. Existing anchored variants often achieve sharp last-iterate guarantees, but from the update-level perspective the placement of the anchor can be algorithm-specific and conceptually opaque. We show that anchoring admits a single operator-side construction: regularize the operator queried by the base method with a vanishing Tikhonov term, then run the unmodified base method. Applied to the Picard iteration, this recipe reproduces the Halpern iteration; applied to the forward step, extragradient (EG), and past extragradient (PEG, also known as Popov’s method), it yields three variants whose anchor placements inherit the base method’s query pattern. The forward-step instantiation gives a new residual convergence guarantee, while the EG and PEG instantiations give new regularized variants. The four analyses share a residual recurrence, recovering the O(1/k) Halpern residual-norm convergence rate, giving O(1/\sqrtk) for the regularized forward step, and giving O(1/k) for the regularized EG and PEG variants in the unconstrained monotone Lipschitz setting.
[LG-137] MLIPilot: LLM -Driven Auto-Research for Machine-Learned Interatomic Potentials
链接: https://arxiv.org/abs/2605.30889
作者: Etinosa Osaro,Santosh Adhikari,Stamatia Zavitsanou,Kelsey Parker,Dario Rocca
类目: Chemical Physics (physics.chem-ph); Machine Learning (cs.LG)
*备注:
Abstract:Constructing production-quality machine-learned interatomic potentials (MLIPs) requires balancing accuracy, dynamical stability, and computational throughput under constraints that are not captured by a single training loss. We introduce MLIPilot, an auto-research framework in which tool-calling large language models propose hypotheses, edit MLIP training code, launch HPC jobs, and accept or revert changes using a fixed, physically constrained scorecard. We evaluate MLIPilot on MACE potential optimization using both commercial and open-weight LLM agents, including GPT-5.5, GPT-4.1, Mistral-24B, and Qwen3-32B. The benchmarks span molecular and periodic settings: a QM7-derived dataset for which we generated B3LYP/6-31G(d) energies and forces, and a Cu EMT dataset with periodic copper supercells labeled by ASE’s Effective Medium Theory calculator. Across these benchmarks, the strongest agents move initially constraint-violating baselines to accepted models by discovering useful training strategies, including output normalization, loss-function changes, progressive training schedules, and model-capacity adjustments. These results suggest that LLM agents can serve as autonomous operators for scientific machine-learning workflows when their search is constrained by domain-specific validation criteria, shifting part of MLIP development from manual trial-and-error toward auditable, automated experimentation.
[LG-138] Generative Quantum Data Embeddings for Supervised Learning
链接: https://arxiv.org/abs/2605.30866
作者: Jaewoong Heo,Daniel K. Park
类目: Quantum Physics (quant-ph); Machine Learning (cs.LG)
*备注: 14 pages, 7 figures
Abstract:Many practically relevant applications of quantum machine learning involve classical data, for which performance depends critically on how inputs are embedded into quantum states. Yet the use of a fixed embedding circuit ansatz remains standard practice. We propose an energy-based generative learning framework that synthesizes gate sequences to optimize embedding structures and refine data-tailored parameters, using a fidelity-based surrogate objective to guide the search toward improved class distinguishability. Empirically, the method improves classification performance across diverse settings, while also revealing datasets where architecture search within the present embedding family yields only limited additional gains. We explain this saturation by deriving bounds on the achievable empirical risk in terms of the Wasserstein distance in the input space, showing that classical data geometry provides an \empha priori diagnostic for regimes in which substantial gains from embedding optimization are unlikely. The results establish a practically useful and theoretically motivated framework for searching effective quantum data embeddings through generative optimization, with the attainable gains diagnosed through the geometry of the underlying classical data.
[LG-139] Bayesian Inference with Shaped Deep Non-linear MLPs
链接: https://arxiv.org/abs/2605.30860
作者: Boris Hanin,Tianze Jiang
类目: atistics Theory (math.ST); Machine Learning (cs.LG); Probability (math.PR)
*备注: 35 Pages
Abstract:A central aim of deep learning theory is to characterize how neural networks make predictions in the regime of simultaneously large model and training set size. Since the limits of diverging number of model parameters and dataset size do not commute it is not clear a priori what limits exist. In this work, we shed new light on these questions by studying Bayesian inference in deep non-linear MLPs in the regime where the number of training samples ( P ), the input dimension ( N_0 ), the hidden layer width ( N ), and the number of hidden layers ( L ) can all be large. We build on the Neural Covariance SDE (Li et al., 2022) to analyze predictive posteriors in the regime where LP/N\in\Theta(1) , playing the role of an effective network depth. Our framework covers both smooth and ReLU activation functions and applies to arbitrary temperature. We find to first order in LP/N a simple criterion for which data generating processes benefit from depth in the sense that larger LP/N increases the Bayesian model evidence. We also give a novel derivation of a prior result from the physics literature that at least to first order in LP/N , the Bayesian predictive posterior is remarkably simple and is simply equivalent to that of a data-dependent kernel method.
[LG-140] he Geometry of Activity Cliffs: Representation Dependence and Multi-Scale Characterization of Activity Landscapes
链接: https://arxiv.org/abs/2605.30831
作者: Pawel Dabrowski-Tumanski,Bartosz Topolski,Dariusz Plewczynski,Tomasz Jetka
类目: Quantitative Methods (q-bio.QM); Machine Learning (cs.LG); Chemical Physics (physics.chem-ph)
*备注:
Abstract:Activity cliffs, structurally similar compounds with large potency differences, are widely treated as intrinsic features of chemical datasets. We argue that apart from target biology, much of our cliff understanding is a consequence of the geometry induced by the chosen molecular representation, not a property of a molecule pair itself. We designed a six-step pipeline to systematically test this hypothesis. The pipeline consists of: assessing pairwise distance geometry, cliff enrichment, activity gradient distribution, persistent homology of the cliff subspace, predictive benchmarking for a chosen pair of an embedding and a metric, and eventually, analysis of the matched molecular pairs and stereoisomers. We applied the pipeline to fifteen configurations of embeddings and metrics to build a benchmark across three distinctive datasets known of activity cliffs challenges. No representation excels on all criteria: Morgan Tanimoto provides the strongest cliff enrichment and cross-scaffold generalization; MolFormer cosine provides the only meaningful stereochemical sensitivity; MACCS and RDKit Dice fingerprints are most sensitive to matched-molecular-pair transformations; ChemBERTa fails uniformly due to embedding collapse. These findings are not a ranking. They reflect the fact that different representations encode different aspects of molecular recognition, and that choosing one implicitly defines what an activity cliff actually is. Subjects: Quantitative Methods (q-bio.QM); Machine Learning (cs.LG); Chemical Physics (physics.chem-ph) Cite as: arXiv:2605.30831 [q-bio.QM] (or arXiv:2605.30831v1 [q-bio.QM] for this version) https://doi.org/10.48550/arXiv.2605.30831 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Pawel Dabrowski-Tumanski [view email] [v1] Fri, 29 May 2026 04:37:00 UTC (3,383 KB)
[LG-141] Is the Last Layer Sufficient for Uncertainty Quantification?
链接: https://arxiv.org/abs/2605.30741
作者: Joseph Wilson,Chris van der Heide,Liam Hodgkinson,Fred Roosta
类目: Machine Learning (stat.ML); Machine Learning (cs.LG)
*备注: 40 pages, 14 figures, 7 tables
Abstract:Epistemic uncertainty quantification (UQ) for deep neural networks (DNNs) is a requirement for safe adoption of AI in mission-critical settings. Several leading methods for UQ linearize DNNs to form Bayesian Generalized Linear Models (GLMs), where epistemic uncertainty is modeled via the predictive posterior distribution. Linearizing around the parameters of the final connected layer of a DNN is a commonly used approximation for reducing the computational burden of such GLMs, though it is often believed to come at the cost of degraded performance. In this work, we compare GLMs arising from full-network and last-layer linearization using both theoretical and empirical approaches. We first employ tools from random matrix theory to conduct a theoretical comparison; this analysis reveals no meaningful improvement in the UQ capabilities of full linearization. Coupled with a large-scale empirical evaluation across a range of modern machine learning tasks, we arrive at the following conclusion: a last-layer approximation yields comparable UQ performance while offering substantially improved computational efficiency.
[LG-142] Learning effective Sargassum transport dynamics from limited drifter observations
链接: https://arxiv.org/abs/2605.30603
作者: F.J. Beron-VEra,M.J. Olascoaga,J. Morell,E. Cruz
类目: Atmospheric and Oceanic Physics (physics.ao-ph); Machine Learning (cs.LG)
*备注:
Abstract:Floating-material transport is influenced by unresolved processes that are often absent from available circulation products. We develop a data-driven transport-learning framework for learning effective transport corrections from limited Lagrangian observations using physically motivated ocean–atmosphere diagnostics and finite-memory representations motivated in part by inertial-particle memory effects. The diagnostic representation is analyzed through predictive and sparse symbolic-discovery approaches under leave-one-trajectory-out validation. Applications to Sargassum-following drifters in the Puerto Rico region and the Gulf Stream show that the diagnostics contain transport-relevant information beyond the baseline circulation products. Multilayer perceptron (MLP) ensembles provide flexible predictive trajectory corrections, while Sparse Identification of Nonlinear Dynamics (SINDy) tests whether instantaneous or delayed sparse symbolic transport structure can be extracted from the diagnostics. The results differ across flow regimes: (i) in Puerto Rico, delayed sparse symbolic corrections provide modest but systematic improvement; (ii) in the Gulf Stream application, dynamically useful sparse symbolic corrections remain primarily instantaneous even though delayed predictive information persists. These results support finite-memory transport effects in coarse-grained floating-material transport while also illustrating the difficulty of obtaining stable delayed sparse symbolic closures.
[LG-143] rue Self-Avoiding Walk for Accelerating Markov-Chain Monte Carlo Integration
链接: https://arxiv.org/abs/2605.30532
作者: Qinghua(Devon)Ding,Venkat Anantharam
类目: Computation (stat.CO); Machine Learning (cs.LG); Machine Learning (stat.ML)
*备注:
Abstract:We study true self-avoiding walk (TSAW) as a mechanism for improving empirical integral estimation via Markov chain Monte Carlo (MCMC). We consider finite-state adaptive sampling dynamics associated with an irreducible Markov kernel P on a finite set, with stationary distribution \pi , in which the transition probabilities are penalized according to empirical overuse. Our main result is that the empirical occupation counts L_t(i) and transition counts N_t(i,j) of the resulting TSAW-based walk satisfy [ L_t(i)-t\pi_i = O(\sqrt\log t) \quad\textand\quad N_t(i,j)-t\pi_iP_ij=O(\sqrt\log t) \qquad\textalmost surely ] for every state i and every edge (i,j) with P_ij0 . Consequently, for every bounded function f:V\to\mathbb R , the error of our integral estimator converges as [ \left|\frac1t\sum_s=0^t-1 f(X_s)-\sum_i\in V\pi_i f(i)\right| = O\left(\frac\sqrt\log tt\right) \qquad\textalmost surely. ] These results show that, in contrast with the usual t^-1/2 error scaling for empirical averages under standard random-walk-based methods, TSAW-based estimator yields empirical integral errors of order O(\sqrt\log t/t) almost surely, thereby achieving a substantially sharper dependence on the sample size t . Subjects: Computation (stat.CO); Machine Learning (cs.LG); Machine Learning (stat.ML) Cite as: arXiv:2605.30532 [stat.CO] (or arXiv:2605.30532v1 [stat.CO] for this version) https://doi.org/10.48550/arXiv.2605.30532 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
[LG-144] Generative Models and Statistical Validation
链接: https://arxiv.org/abs/2605.30453
作者: Sascha Diefenbacher,Sofia Palacios Schweitzer,Gregor Kasieczka
类目: High Energy Physics - Phenomenology (hep-ph); Machine Learning (cs.LG); Data Analysis, Statistics and Probability (physics.data-an)
*备注: 36 pages, 4 figures, Part of the VERaiPHY Initiative
Abstract:Generative machine learning has become an essential tool in theoretical and experimental physics, especially in the context of fast surrogates and density estimators. In this work, we first introduce the underlying framework of modern generative networks and then discuss challenges in quantifying their accuracy, precision, and statistical power.
[LG-145] Learning effective models from network dynamics data with multiple initial conditions using weak form SINDy
链接: https://arxiv.org/abs/2605.30432
作者: Moyi Tian,Daniel A. Messenger,Vanja Dukic,Nancy Rodríguez,David M. Bortz
类目: Dynamical Systems (math.DS); Machine Learning (cs.LG); Social and Information Networks (cs.SI); Adaptation and Self-Organizing Systems (nlin.AO); Physics and Society (physics.soc-ph)
*备注: 24 pages, 14 figures, 1 table
Abstract:Social systems consist of networks of individuals who influence one another through social interactions. Studying how processes evolve on these networks can help us better understand patterns of social behavior. We study a system that couples online and offline social activity and investigate how to learn effective models directly from data using Weak Form Sparse Identification of Nonlinear Dynamics (WSINDy), a method for discovering governing equations. We assess learning performance using data generated by a mean-field approximation model of a stochastic interaction process on networks and test how accurately the system can be recovered under different noise levels. Our results show that using more trajectories improves accuracy when noise is high, but only a small number of additional trajectories is needed to gain most of the benefit, with little improvement beyond that. We also learn effective ODE models from averaged stochastic data on networks. When traditional mean-field approximations fail, identifying continuum ODEs directly from stochastic processes yields efficient models that better match the data and provide deeper insight into the underlying dynamics.
[LG-146] Attention-based optimizer for symmetry finding
链接: https://arxiv.org/abs/2605.30429
作者: Shreya Banerjee,Vinodh Raj Rajagopal Muthu,Charlie Nation,Rick P.A. Simon,Francesco Martini,Alessandro Ricottone,Federico Cerisola,Luca Dellantonio
类目: Quantum Physics (quant-ph); Machine Learning (cs.LG)
*备注: 9+4 pages, 2 Figures, Comments welcome
Abstract:Finding symmetries is crucial for understanding physical models. In this work, we present an optimization framework that searches Pauli symmetries of Hamiltonians, merging the fields of machine learning with automated symmetry finding. Built on a Set-Transformer architecture, our framework uses self-attention to encode the pairwise and higher-order correlations among the Pauli-Strings. The relations are then decoded as a candidate, which is further optimized with a custom commutation-based objective, and mapped to a symmetry of the input Hamiltonian. We apply our method to random Pauli Hamiltonians, periodic one and two dimensional transverse-field Ising model and the Toric code. We show that for physical Hamiltonians (Ising and Toric), our framework succeeds with near-deterministic probability while providing substantial advantage compared to state-of-the-art strategies. For random Pauli Hamiltonians, we estimate the required computational resources, specifically the number of parallel starts and the number of GPUs, to find a symmetry with high success probability under fixed design specifications.
[LG-147] A Novel Computer Vision Approach for Assessing Fish Responses to Intrusive Objects in Aquaculture
链接: https://arxiv.org/abs/2605.30399
作者: Hanne-Grete Alvheim,Stian Mjelde Jakobsen,Martin Føre,Eleni Kelasidi
类目: Quantitative Methods (q-bio.QM); Machine Learning (cs.LG); Image and Video Processing (eess.IV)
*备注:
Abstract:The aquaculture industry needs to address several challenges to secure sustainable seafood production that can serve an increasing global demand. One major challenge is to ensure good fish health and acceptable welfare during production since the improvement of fish welfare is of vital importance in current and future production systems. In this study, this is addressed by developing and implementing methods to identify fish behaviors in response to intrusive objects both on individual and on a group basis. A novel approach for detecting, tracking, and estimating the 3D position of individual fish has thus been developed, and specifically designed to track the caudal fins of farmed fish in industrial sea cages. The tracking data was subjected to a novel stereo-vision method adapted to estimate fish positions, velocities, accelerations, and turning and pitch angles. Datasets obtained from industrial-scale fish farms were then analyzed to identify the impact of structures of varying shapes, sizes, and colors on fish behavior. The method was trained using manually labeled caudal fins, and used YOLOv8 with ByteTrack as an object detector and tracker, SuperGlue for matching detections in the left and right frames, and triangulation to reconstruct the 3D positions of the fish. Different image pre-processing and augmentation methods for enhancing object detection accuracy were tested and their performance compared, while RAFT-Stereo was tested for depth estimation purposes. The obtained results both validate the method’s performance against previous research efforts, and demonstrate the novelty and potential of this method in providing more insight into behavioral dynamics in sea-cages. Subjects: Quantitative Methods (q-bio.QM); Machine Learning (cs.LG); Image and Video Processing (eess.IV) Cite as: arXiv:2605.30399 [q-bio.QM] (or arXiv:2605.30399v1 [q-bio.QM] for this version) https://doi.org/10.48550/arXiv.2605.30399 Focus to learn more arXiv-issued DOI via DataCite
附件下载
