{"schema":"scienceos.atlas.feed-shard/1","feed":"today","date":"2026-10-22","generatedAt":"2026-10-22T08:36:18Z","sources":["arxiv","openalex","huggingface"],"items":[{"paper":{"id":"arxiv:2610.30108","arxivId":"2610.30108","doi":"10.48550/arXiv.2610.30108","title":"Multimodal Embodied Agents: A Unified Capability-centric Perspective from Computer-Use to Robot-Use","abstract":"Multimodal agents (MMAs) sustain interaction through reasoning, memory, tools, and feedback, most visibly as computer-use agents, while robotic systems (RSs) couple sensing and actuation under physical dynamics. We define multimodal embodied agents (MMEAs) as goal-directed systems that couple multimodal task reasoning with physical action and revise decisions from the resulting feedback. This raises a central question: what changes when a multimodal agent moves from computer-use to robot-use? We introduce PAPAV, a unified capability-centric framework of five recurring functions: Perceive the current state, Anticipate action effects, Plan a feasible course, Act through an interface or body, and Verify the outcome. Defined by function rather than architecture, these capabilities let PAPAV identify what all three share and then compare MMEAs with MMAs and with RSs. Five physical constraints, from partial observability to unverifiable outcomes, leave fewer choices fixed in advance and less room to reverse errors. Across 64 benchmarks, explicit evaluation centers on Act; only 5 assess Anticipate and 3 assess Verify, leaving both largely hidden behind task success. These constraints also frame the open challenges we identify. PAPAV therefore provides a common basis for designing reliable physical agents and measuring progress across all five capabilities rather than task success alone.","authors":[{"name":"Yanzhe Chen","position":0},{"name":"Ziyi Yang","position":1},{"name":"Jifeng Zhu","position":2},{"name":"Qiming Huang","position":3},{"name":"Ruihe An","position":4},{"name":"Peiyao Xu","position":5},{"name":"Hesen Yang","position":6},{"name":"Runda Liu","position":7},{"name":"Chang Gong","position":8},{"name":"Zhijun Cao","position":9},{"name":"Zechen Bai","position":10},{"name":"Wenzheng Zeng","position":11},{"name":"Yiqi Lin","position":12},{"name":"Guoqiang Liang","position":13},{"name":"Kevin Yuchen Ma","position":14},{"name":"Kevin Qinghong Lin","position":15},{"name":"Mike Zheng Shou","position":16}],"categories":["cs.RO","cs.AI","cs.CV","cs.MA","cs.MM"],"primaryCategory":"cs.RO","publishedAt":"2026-10-21T17:52:48Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30108"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30108"},{"kind":"code","url":"https://github.com/showlab/Awesome-Multimodal-Embodied-Agent","label":"showlab/Awesome-Multimodal-Embodied-Agent"},{"kind":"project","url":"https://showlab.github.io/Awesome-Multimodal-Embodied-Agent/"}],"signals":{"hfUpvotes":{"value":14,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubStars":{"value":137,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubRepo":"showlab/Awesome-Multimodal-Embodied-Agent"},"brief":{"text":"PAPAV organizes embodied-agent capabilities into perceiving, anticipating, planning, acting, and verifying, clarifying how robot-use differs from computer-use. Across 64 benchmarks, evaluation focuses on action, while anticipation and verification remain largely unmeasured.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":1,"reason":"trending","score":0.6992},{"paper":{"id":"arxiv:2610.29989","arxivId":"2610.29989","doi":"10.48550/arXiv.2610.29989","title":"Looking Ahead: Modeling Future Evolution in Autoregressive Video Generation","abstract":"Autoregressive video diffusion models generate videos chunk by chunk, yet producing plausible continuations does not guarantee that they capture how scenes evolve. Under teacher forcing, strong temporal redundancy may allow local visual continuity to serve as a shortcut for current-chunk denoising, reducing the pressure to model visual evolution. We introduce Future Chunk Prediction (FCP), an auxiliary objective jointly optimized with teacher-forced denoising that predicts temporally separated chunks without access to the intervening visual context. Drawing supervision directly from training videos, FCP requires no additional pretrained visual encoder and leaves autoregressive inference unchanged. Our method achieves superior physical plausibility over standard teacher-forced autoregressive training with a lower overall training compute budget, highlighting the effectiveness and efficiency of future-chunk supervision for modeling visual evolution. Intriguingly, our visualizations suggest that FCP fosters a more structured representation space, which may be valuable for building autoregressive world models.","authors":[{"name":"Ruopeng Gao","position":0},{"name":"Yukun Huang","position":1},{"name":"Jiarui Hu","position":2},{"name":"Shuai Wang","position":3},{"name":"Wenbo Hu","position":4},{"name":"Ying Shan","position":5},{"name":"Limin Wang","position":6}],"categories":["cs.CV"],"primaryCategory":"cs.CV","publishedAt":"2026-10-21T16:28:46Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29989"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29989"},{"kind":"code","url":"https://github.com/TencentARC/Looking-Ahead","label":"TencentARC/Looking-Ahead"},{"kind":"project","url":"https://hellorpg.github.io/looking-ahead/"}],"signals":{"hfUpvotes":{"value":62,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubStars":{"value":3,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubRepo":"TencentARC/Looking-Ahead"},"brief":{"text":"Future Chunk Prediction trains autoregressive video models to predict later frames without intervening visual context, improving physical plausibility over standard training while using less compute.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":2,"reason":"trending","score":0.6923},{"paper":{"id":"arxiv:2610.25043","arxivId":"2610.25043","doi":"10.48550/arXiv.2610.25043","title":"VLA-Rolling: Fewer Denoising Steps, Better Control via Adaptive Action Reuse","abstract":"Flow-matching vision-language-action (VLA) models generate an action chunk by iteratively denoising Gaussian noise, but typically execute only a short prefix before replanning with updated observations. Under this receding-horizon paradigm, the unexecuted remainder---the action tail---is discarded, and the next chunk is regenerated from fresh noise. We show that this discarded tail retains transferable denoising progress: action-tail initialization can match or outperform a native 10-step Gaussian baseline using fewer denoising steps. However, the optimal denoising budget varies substantially across samples, indicating that tail reuse should be adapted to the updated environment. Based on this observation, we introduce VLA-Rolling, a lightweight inference framework that reuses action tails through three stages: action-tail inheritance, per-action confidence estimation, and confidence-guided refinement through renoising and denoising budget allocation, while the underlying VLA remains frozen. Across $π_{0.5}$, SmolVLA, X-VLA, and MolmoAct2 on LIBERO, MetaWorld, LIBERO-Plus, and RoboTwin 2.0, VLA-Rolling reduces action-chunk inference cost in all 16 model--benchmark pairs, achieving up to a $1.48\\times$ speedup, while improving success rate in 14 pairs by up to $15.08$ percentage points. These results show that action tails are not merely discarded predictions, but reusable intermediate states that enable more efficient and accurate VLA inference. These results demonstrate that the action tail can support the generation of better actions with fewer denoising steps. Our project is available at https://chenbao123ww.github.io/VLA-Rolling/.","authors":[{"name":"Sicheng He","position":0},{"name":"Yi Pan","position":1},{"name":"Qiuting Yu","position":2},{"name":"Jiaming Huang","authorId":"oa:A5139988067","position":3},{"name":"Qi Lu","position":4},{"name":"Shuyan Wu","position":5},{"name":"Hongyan Feng","authorId":"oa:A5147481297","position":6},{"name":"Xuanyu Liu","position":7},{"name":"Jun Zhang","position":8},{"name":"Jie Zhang","position":9},{"name":"Wenqi Zhang","position":10},{"name":"Xuhong Zhang","position":11},{"name":"Jianwei Yin","position":12}],"categories":["cs.AI","cs.RO"],"primaryCategory":"cs.AI","publishedAt":"2026-10-18T06:01:43Z","updatedAt":"2026-10-21T03:43:29Z","version":3,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.25043"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.25043"},{"kind":"code","url":"https://github.com/ZJU-OmniAI/VLA-Rolling","label":"ZJU-OmniAI/VLA-Rolling"},{"kind":"project","url":"https://chenbao123ww.github.io/VLA-Rolling/"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.25043.jpg","label":"thumbnail"}],"citations":{"value":0,"source":"openalex","fetchedAt":"2026-10-22T08:09:23Z"},"signals":{"hfUpvotes":{"value":60,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubStars":{"value":10,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubRepo":"ZJU-OmniAI/VLA-Rolling"},"brief":{"text":"VLA-Rolling reuses discarded action tails and adapts denoising to current observations, reducing inference cost across all 16 tested model–benchmark pairs and improving success rates in 14, with up to 1.48× speedup.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":3,"reason":"trending","score":0.6805},{"paper":{"id":"arxiv:2610.25404","arxivId":"2610.25404","doi":"10.48550/arXiv.2610.25404","title":"MT Forcing: Towards Stable and Consistent Multi-Turn Image Editing with Autoregressive Diffusion","abstract":"Multi-turn image editing aims to modify an image according to a sequence of instructions, enabling more effective interactive visual content creation for users. However, existing methods often exhibit noticeable drift after only a few editing turns, manifesting in two aspects: 1) unstable visual quality, with accumulating artifacts and degraded details; and 2) inconsistent image content, with unintended changes to preserved regions and prior edits. To address these challenges, we propose MT Forcing, a novel training paradigm for multi-turn image editing with unified multimodal models (UMMs) based on autoregressive diffusion. Specifically, we propose Single-Turn Distribution Matching Distillation (Single-Turn DMD), which supervises student self-rollouts with a teacher that applies cumulative edits to the source image in a single turn, providing visual-quality guidance without recursive error accumulation. Additionally, we introduce Teacher-Forcing Guidance, which supervises history-dependent transitions on reference trajectories during warmup and as a regularizer during distillation, encouraging cross-turn content consistency beyond what per-turn distribution matching alone can ensure. Extensive experiments demonstrate that our approach significantly improves the multi-turn image editing capabilities of UMMs, yielding more stable visual quality and stronger cross-turn consistency.","authors":[{"name":"Houyuan Chen","authorId":"oa:A5159041480","position":0},{"name":"Fanzhou Wang","authorId":"oa:A5021955233","position":1},{"name":"Xiangyu Fan","position":2},{"name":"Zesong Qiu","authorId":"oa:A5046875669","position":3},{"name":"Yukun Wei","position":4},{"name":"Songlin Yang","position":5},{"name":"Zhitao Yang","authorId":"oa:A5119736214","position":6},{"name":"Lvmin Zhang","authorId":"oa:A5079385579","position":7},{"name":"Pei Yang","position":8},{"name":"Hao Zhao","position":9},{"name":"Dahua Lin","position":10},{"name":"Lei Yang","position":11},{"name":"Anyi Rao","authorId":"oa:A5067715162","position":12}],"categories":["cs.CV"],"primaryCategory":"cs.CV","publishedAt":"2026-10-18T17:01:40Z","updatedAt":"2026-10-21T08:45:58Z","version":2,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.25404"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.25404"},{"kind":"code","url":"https://github.com/houyuanchen111/MT-Forcing","label":"houyuanchen111/MT-Forcing"},{"kind":"project","url":"https://houyuanchen111.github.io/mt-forcing.github.io/"}],"citations":{"value":0,"source":"openalex","fetchedAt":"2026-10-22T08:09:23Z"},"signals":{"hfUpvotes":{"value":41,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubStars":{"value":22,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubRepo":"houyuanchen111/MT-Forcing"},"brief":{"text":"MT Forcing trains autoregressive diffusion models to edit images across multiple instruction turns, using single-turn quality guidance and reference trajectories to preserve prior edits. Experiments show more stable visual quality and stronger consistency across turns.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":4,"reason":"trending","score":0.6197},{"paper":{"id":"arxiv:2610.28698","arxivId":"2610.28698","doi":"10.48550/arXiv.2610.28698","title":"Scaling in Games: Continued Pre-Training For Embodied Agents In Diverse Virtual Worlds","abstract":"Scaling data, diversity and model capacity has improved embodied learning in physical environments, where diverse experiences remain grounded in shared physical laws. Whether these gains extend to virtual worlds with different visual styles, dynamics, controls, and objectives remains unclear. We investigate this question through a controlled study of continued pre-training (CPT) across three axes: data volume, diversity, and model size. Our 120 training configurations span four model sizes from 0.8B to 27B, six data budgets from 50 to 1,600 hours, and five mixtures ranging from Minecraft-only data to diverse multi-world corpora. We assess performance and generalization using three-level validation losses and gameplay benchmarks from in-corpus worlds to unseen worlds. Our results show: (1) more data improves in-corpus performance, diversity improves transfer to related worlds with limited gains in distant worlds, and model size improves data efficiency but generalizes worse to distant worlds under single-world training; (2) we observe an early interface phase with broad cross-world benefits and a later world phase with gains narrowed to in-corpus and related worlds, and fit an empirical scaling law relating validation loss to data volume, diversity and model size; and (3) our models are competitive with Minecraft specialists and frontier models in 3D worlds, and perform best in real-time settings. However, the limits of imitation and persistent gaps in distant worlds suggest that building generalist embodied agents requires richer training stages beyond continued pre-training.","authors":[{"name":"Kuan Zhang","position":0},{"name":"Yukun Chen","position":1},{"name":"Zhihao Yang","position":2},{"name":"Yue Su","position":3},{"name":"Xiangnan Wu","position":4},{"name":"Zirong Chen","position":5},{"name":"Run Luo","position":6},{"name":"Jinkun Hou","position":7},{"name":"Tao Tan","position":8},{"name":"Yinhe Zheng","position":9},{"name":"Yiming Li","position":10}],"categories":["cs.AI","cs.CV","cs.LG"],"primaryCategory":"cs.AI","publishedAt":"2026-10-20T11:09:09Z","version":1,"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28698"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28698"},{"kind":"code","url":"https://github.com/THUSI-Lab/GameScaling","label":"THUSI-Lab/GameScaling"},{"kind":"project","url":"https://gamescaling.github.io"}],"signals":{"hfUpvotes":{"value":17,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubStars":{"value":9,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubRepo":"THUSI-Lab/GameScaling"},"brief":{"text":"Continued pre-training across virtual worlds shows that more data boosts in-world performance, while diverse data helps transfer mainly to related worlds. Larger models use data more efficiently, but training on a single world can hurt transfer to distant worlds.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":5,"reason":"trending","score":0.6149},{"paper":{"id":"arxiv:2610.29997","arxivId":"2610.29997","doi":"10.48550/arXiv.2610.29997","title":"DONIC: Generalist Dexterous Bimanual Tracking Control from Human Demonstrations","abstract":"Despite recent progress in dexterous reference tracking, existing models focus primarily on single-hand manipulation and a limited range of interactions, leaving the bimanual dexterous behaviors common in real life unaddressed. The scarcity of contact-rich bimanual manipulation data and the increased complexity of coordinating contacts make developing a generalist bimanual tracker particularly challenging. Therefore, we present 1) DONIC, the first generalist bimanual tracker jointly trained across operation categories and human-motion sources, together with 2) BiHOI, a multi-view dataset of 1,575 sequences spanning 229 objects and emphasizing sustained bimanual interaction, where reconstructed interactions are retargeted into robot references. DONIC combines proximal policy optimization with behavioral cloning in a feedback loop that recycles verified successful executions into training demonstrations. An object-centered per-hand reward maintains object-motion feedback while supervising each hand according to its reference contact role. Clip-Centered Advantages remove reference-dependent advantage offsets while preserving within-trajectory differences. Together, these designs enable one single policy to learn coordinated manipulation across heterogeneous references. Without finetuning, a single checkpoint achieves 86.79% success on unseen trajectories of training objects and 57.32% on held-out objects, surpassing the state-of-the-art bimanual tracking baselines by 77.36 and 45.73 percentage points.","authors":[{"name":"Mingze Sun","position":0},{"name":"Ziang Cao","position":1},{"name":"Yuan Fei","position":2},{"name":"Qiankuan Tao","position":3},{"name":"Zihan Wang","position":4},{"name":"Tianqi Liu","position":5},{"name":"Junhao Chen","position":6},{"name":"Haiwen Diao","position":7},{"name":"Fangzhou Hong","position":8},{"name":"Zhaoxi Chen","position":9},{"name":"Ziwei Liu","position":10}],"categories":["cs.RO","cs.CV"],"primaryCategory":"cs.RO","publishedAt":"2026-10-21T16:36:02Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29997"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29997"},{"kind":"code","url":"https://github.com/Ropedia/DONIC","label":"Ropedia/DONIC"},{"kind":"project","url":"https://ropedia.github.io/DONIC/"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.29997.jpg","label":"thumbnail"}],"signals":{"hfUpvotes":{"value":18,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubStars":{"value":4,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubRepo":"Ropedia/DONIC"},"brief":{"text":"DONIC is a single policy for tracking varied bimanual manipulation demonstrations, trained with a new dataset and a loop that reuses successful executions. Without finetuning, it outperforms prior bimanual tracking methods on unseen trajectories and held-out objects.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":6,"reason":"trending","score":0.5997},{"paper":{"id":"arxiv:2610.25736","arxivId":"2610.25736","doi":"10.48550/arXiv.2610.25736","title":"Learning from Fleet Experience: Collective Experience Learning Across 96 Robots","abstract":"Robot deployment generates experience that can support policy improvement, but continued reliance on expert corrective actions makes learning difficult to scale across growing fleets. We propose learning from fleet experience: using deployed robots' own executions as a source of supervision for further policy improvement. We call this paradigm Collective Experience Learning (CEL), in which robots pool execution experience to update a shared policy that returns to the fleet. Following expert-supervised initialization, we realize CEL through failure-guided experience retrieval fine-tuning (FERT). FERT uses failure contexts to retrieve relevant segments from successful trajectories and prioritizes their recorded actions for supervised updates, without collecting additional expert demonstrations or corrective actions. Operators provide binary episode-outcome labels and environment resets. A fault-tolerant asynchronous training system (FATS) coordinates robot execution, cloud inference and policy optimization despite local interruptions. Across five real-world manipulation tasks, the resulting policies achieve success rates of 93.3-99.2%, evaluated over 120 trials per task on robots excluded from training data collection. With collection time per robot held fixed, larger fleets provide more experience and generally improve performance across the evaluated sizes. On multi-stage phone packaging, learning with 96 robots raises success from 79.2% to 98.3%. These results support CEL as a scalable approach to improving robot policy reliability through experience generated during deployment.","authors":[{"name":"Chao Yu","position":0},{"name":"Shu'ang Yu","position":1},{"name":"Hao Lin","position":2},{"name":"Hongzhi Zang","position":3},{"name":"Yuxiang Yan","position":4},{"name":"Chenxi Jiang","authorId":"oa:A5130650466","position":5},{"name":"Zhen Guo","position":6},{"name":"Chuankang Li","position":7},{"name":"Yihao Chen","position":8},{"name":"Xin Xu","position":9},{"name":"Xu Fu","authorId":"oa:A5114222301","position":10},{"name":"Jiakai Zhou","authorId":"oa:A5102020434","position":11},{"name":"Liangzhi Shi","authorId":"oa:A5132582789","position":12},{"name":"Hongyi Ren","authorId":"oa:A5062327519","position":13},{"name":"Yiwei Liu","position":14},{"name":"Jing Wu","position":15},{"name":"Yuanqing Wang","authorId":"oa:A5148805754","position":16},{"name":"Chen Wu","position":17},{"name":"Bo Dai","authorId":"oa:A5079863650","position":18},{"name":"Chunyu Yang","position":19},{"name":"Hongliang Huang","authorId":"oa:A5084860180","position":20},{"name":"Shuoyu Yue","authorId":"oa:A5113430958","position":21},{"name":"Weiqi Yin","position":22},{"name":"Yinuo Chen","position":23},{"name":"Ruihua Chen","authorId":"oa:A5130615511","position":24},{"name":"Chunyang Zhu","position":25},{"name":"Han Xiao","position":26},{"name":"Yanghao Li","position":27},{"name":"Ruipu Wu","position":28},{"name":"Zhikai Cai","position":29},{"name":"Xiu Chen","position":30},{"name":"Xu Tan","position":31},{"name":"Si Liu","position":32},{"name":"Yuchen Yan","position":33},{"name":"Jialin Zhang","position":34},{"name":"Wen Zeng","position":35},{"name":"Yiming Zhou","position":36},{"name":"Zixiao Huang","position":37},{"name":"Xinlei Chen","position":38},{"name":"Guanghui Ren","position":39},{"name":"Quanlu Zhang","position":40},{"name":"Boxun Li","position":41},{"name":"Maoqing Yao","position":42},{"name":"Yu Wang","position":43}],"categories":["cs.RO"],"primaryCategory":"cs.RO","publishedAt":"2026-10-19T03:41:27Z","updatedAt":"2026-10-21T05:52:57Z","version":2,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.25736"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.25736"},{"kind":"project","url":"https://cel-robot-learning.github.io"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.25736.jpg","label":"thumbnail"}],"citations":{"value":0,"source":"openalex","fetchedAt":"2026-10-22T08:09:23Z"},"signals":{"hfUpvotes":{"value":116,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"}},"brief":{"text":"Collective Experience Learning pools robots’ deployment experience to improve a shared policy without further expert corrections. Across five manipulation tasks, policies reached 93.3–99.2% success, and a 96-robot fleet raised phone-packaging success from 79.2% to 98.3%.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":7,"reason":"trending","score":0.575},{"paper":{"id":"arxiv:2610.28444","arxivId":"2610.28444","doi":"10.48550/arXiv.2610.28444","title":"EvaOmni: An Evidence-Grounded Benchmark for Audio-Visual Reasoning","abstract":"Omni LLMs can naturally locate key audio and visual evidence and integrate information across modalities for audiovisual reasoning. Existing audio-visual reasoning benchmarks that mainly rely on multiple-choice questions usually score models solely based on answer correctness, making it difficult to assess whether a model truly forms a reasoning chain from evidence to the correct answer. In addition, textual or unimodal shortcuts may allow models to answer correctly without analyzing the necessary audio-visual evidence, further weakening the reliability of evaluation benchmarks. To address these issues, we introduce EvaOmni, an evidence-grounded benchmark containing 392 videos and 1,027 purely human-annotated question-answer pairs, covering 15 capabilities and three levels: retrieval and localization, basic understanding, and complex reasoning. We design a series of rigorous annotation procedures in which experts annotate audio and visual evidence and construct video-based distractors; humans review whether questions meet strict inspection criteria, and finally models filter textual and unimodal shortcuts. Beyond high-quality question-answer pairs, we require models to simultaneously return the answer and supporting evidence intervals in order to evaluate the complete reasoning chain. We propose the Coverage Grounded Score (C.G.) to measure how much annotated evidence a model identifies, and the Bottleneck Grounded Score (B.G.) to penalize poor temporal localization in either modality; both are incorporated into answer correctness. Across 7 open-weight and 17 proprietary Omni LLMs, Gemini3.7-Flash achieved the highest overall accuracy (79.4%), C.G. (70.8%), and B.G. (35.1%). At Level 3, it reached 79.1% accuracy, but its B.G. was only 12.3%. These results show that even the most advanced models still have room for development in grounded audio-visual reasoning.","authors":[{"name":"Xiaoxing Hu","position":0},{"name":"Wenxi Yao","position":1},{"name":"Chaoyuan Shi","position":2},{"name":"Ziqi Ye","position":3},{"name":"Xiaorong Guo","position":4},{"name":"Peixuan Li","position":5},{"name":"Zhuoran Sun","position":6},{"name":"Ziyang Gong","position":7},{"name":"Peiyuan Zhang","position":8},{"name":"Mingxin Liu","position":9},{"name":"Shuran Ma","position":10},{"name":"Hongbo Liu","position":11},{"name":"Xiaomin Yu","position":12},{"name":"Yu Tian","position":13},{"name":"Wentao Wang","position":14},{"name":"Zhenjie Yang","position":15},{"name":"Jian Hu","position":16},{"name":"Yunhang Shen","position":17},{"name":"Haoyu Cao","position":18},{"name":"Xue Yang","position":19}],"categories":["cs.CV"],"primaryCategory":"cs.CV","publishedAt":"2026-10-20T16:03:30Z","updatedAt":"2026-10-21T15:25:03Z","version":2,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28444"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28444"},{"kind":"code","url":"https://github.com/VisionXLab/EvaOmni","label":"VisionXLab/EvaOmni"},{"kind":"project","url":"https://xiaoxing2001.github.io/EvaOmni.github.io/"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.28444.jpg","label":"thumbnail"}],"signals":{"hfUpvotes":{"value":18,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubStars":{"value":19,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubRepo":"VisionXLab/EvaOmni"},"brief":{"text":"EvaOmni evaluates audiovisual reasoning by requiring models to identify supporting audio and video evidence as well as answer questions. Across 24 models, Gemini3.7-Flash led overall, but its weak evidence localization on complex questions highlights a gap in grounded reasoning.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":8,"reason":"trending","score":0.5663},{"paper":{"id":"arxiv:2610.28763","arxivId":"2610.28763","doi":"10.48550/arXiv.2610.28763","title":"Automating ML for Science: Can Frontier Agents Climb Scientific Hills in the Wild?","abstract":"Frontier coding agents are increasingly framed as end-to-end automated researchers, and ML for science offers a demanding test of that promise: improving a scientific model often requires domain insight beyond routine ML engineering. We introduce Agent4Sci, a long-horizon testbed for measuring how far today's agents can improve the published best models on scientific ML benchmarks, spanning 20 benchmarks across 11 domains. Rather than ranking agents on a leaderboard, Agent4Sci asks what kinds of progress agents can make on these tasks under in-the-wild conditions, and where they fall short. Empirically, today's frontier agents advance the published best on every benchmark, with Claude Opus reaching 10.4% mean relative improvement and no run flagged for academic misconduct. These gains, however, come almost entirely from a bag of ML-engineering tricks rather than research-level insight. Trajectory analysis surfaces three recurring limitations in this setting: novelty miscalibration, resource under-utilization, and rapid collapse into local engineering sweeps. Lightweight plug-in skills for research ideation, experiment management, and scheduled checklists raise Claude Opus from 10.4% to 20.1% and Codex from 4.4% to 8.1%, while yielding more diverse, domain-relevant interventions. Agent4Sci gives the community a way to track and push frontier agents from ML engineering toward domain insight on real-world ML-for-science problems. Code and data are available on GitHub (https://github.com/maszhongming/agent4sci) and Hugging Face (https://huggingface.co/datasets/MingZhong/agent4sci).","authors":[{"name":"Ming Zhong","position":0},{"name":"Stacy Li","position":1},{"name":"Nicholas Carlini","position":2},{"name":"Matthew Jagielski","position":3}],"categories":["cs.AI","cs.CL","cs.LG"],"primaryCategory":"cs.AI","publishedAt":"2026-10-20T18:03:48Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28763"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28763"},{"kind":"code","url":"https://github.com/maszhongming/agent4sci","label":"maszhongming/agent4sci"}],"signals":{"hfUpvotes":{"value":10,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubStars":{"value":1,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubRepo":"maszhongming/agent4sci"},"brief":{"text":"Agent4Sci tests frontier coding agents on 20 scientific ML benchmarks: agents beat published best models across all benchmarks, mostly through routine engineering rather than research insight. Lightweight research skills increased mean relative gains for Claude Opus from 10.4% to 20.1%.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":9,"reason":"trending","score":0.5332},{"paper":{"id":"arxiv:2610.25369","arxivId":"2610.25369","doi":"10.48550/arXiv.2610.25369","title":"Painting Change Map as guidance for generating painting processes","abstract":"Digital art generation has undergone a massive paradigm shift, moving from static image synthesis to dynamic video generation that can mimic the human painting process. While existing models can synthesize visually stunning time-lapses of paintings being created or \"unpainted,\" they largely treat the creative process as a black box. Current conditioning mechanisms are either too rigid, relying on static geometric cues like Canny edges or depth maps, or too holistic, utilizing simple scalar progress variables that offer no localized control. Consequently, users lack the agency to steer how a painting evolves over time. In this work, we argue that true control over the video generation of artwork requires a joint understanding of two fundamental dimensions: where an artist acts, and when those actions take place. To achieve this, we introduce the Painting Change Map, a novel spatio-temporal control mechanism that explicitly tracks normalized structural changes across a 1,000-timestep continuum, a resolution sufficient to capture fine-grained stroke-level changes while remaining computationally tractable. By mapping localized visual differences into an encoded RGB space, our representation enables fine-grained guidance for both the forward painting and reverse unpainting processes. Extensive benchmarking demonstrates the efficacy of our approach. We outperform existing control baselines and achieve competitive or superior results compared to state-of-the-art painting methods. Furthermore, under oracle conditions, our method surpasses all baselines by a significant margin. Code is available at https://github.com/Markus-Pobitzer/painting-change-map.","authors":[{"name":"Markus Pobitzer","authorId":"oa:A5094018538","position":0},{"name":"Chang Liu","position":1},{"name":"Niccolò Biondi","authorId":"oa:A5036222429","position":2},{"name":"Nicu Sebe","authorId":"oa:A5129650669","position":3}],"categories":["cs.CV"],"primaryCategory":"cs.CV","publishedAt":"2026-10-18T16:15:14Z","updatedAt":"2026-10-21T14:35:17Z","version":2,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.25369"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.25369"},{"kind":"code","url":"https://github.com/Markus-Pobitzer/painting-change-map","label":"Markus-Pobitzer/painting-change-map"}],"citations":{"value":0,"source":"openalex","fetchedAt":"2026-10-22T08:09:23Z"},"signals":{"hfUpvotes":{"value":17,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubStars":{"value":12,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubRepo":"Markus-Pobitzer/painting-change-map"},"brief":{"text":"Painting Change Maps guide painting and unpainting video generation by encoding where and when visual changes occur. The method outperforms existing control baselines and, with oracle guidance, surpasses all baselines by a significant margin.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":10,"reason":"trending","score":0.5178},{"paper":{"id":"arxiv:2610.30128","arxivId":"2610.30128","doi":"10.48550/arXiv.2610.30128","title":"Think Harder Where It Matters: Item-Level Adaptive Recursion for Sequential Recommendation","abstract":"Existing sequential recommenders usually allocate a fixed computational budget to every item in a user's interaction history, regardless of the item's contribution to the final prediction. We challenge this de facto design with two diagnostic findings. (1) Non-uniform computational need: items within the same sequence vary drastically in how much computation they require. Most converge after shallow processing while a critical minority only stabilizes at full depth. (2) Suboptimality of fixed allocation: given oracle access to the target item, a non-uniform depth assignment consistently outranks the fixed one, indicating that existing uniform computation both wastes capacity on already-converged items and sacrifices overall recommendation quality. These observations motivate a shift from fixed-depth computation to item-level adaptive reasoning. To this end, we propose LoopRec. LoopRec recursively applies a single weight-shared block and introduces a lightweight gate at each recursion boundary that makes a binary exit-or-continue decision for every item independently. Items whose representations have sufficiently matured exit early, while those requiring deeper reasoning continue to subsequent recursions. This design converts the fixed global depth into a learned, per-item property without auxiliary depth labels, and is readily adaptable to diverse backbone architectures. Extensive experiments on public benchmarks with four backbone families demonstrate that LoopRec consistently outperforms matched fixed-depth counterparts. Further analysis reveals that LoopRec's learned allocation exhibits interpretable patterns aligned with our post-hoc oracle analysis: lower-frequency items and more recent interactions naturally receive deeper computation. Our code is available at https://github.com/Elpmis117/LoopRec.","authors":[{"name":"Zhanshuo Cao","position":0},{"name":"Jiakai Tang","position":1},{"name":"Yuqi Zhou","position":2},{"name":"Sunhao Dai","position":3},{"name":"Jun Xu","position":4}],"categories":["cs.IR","cs.AI","cs.CL","cs.LG"],"primaryCategory":"cs.IR","publishedAt":"2026-10-21T17:58:06Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30128"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30128"},{"kind":"code","url":"https://github.com/Elpmis117/LoopRec","label":"Elpmis117/LoopRec"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.30128.jpg","label":"thumbnail"}],"signals":{"hfUpvotes":{"value":5,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubStars":{"value":0,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubRepo":"Elpmis117/LoopRec"},"brief":{"text":"LoopRec gives each interaction item an independent choice to exit or continue through a shared reasoning block. Across four backbone families, it outperforms matched fixed-depth recommenders, with deeper computation tending to go to rarer items and recent interactions.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":11,"reason":"trending","score":0.5132},{"paper":{"id":"arxiv:2610.30120","arxivId":"2610.30120","doi":"10.48550/arXiv.2610.30120","title":"ALPS: Visual State Encoders for World Models","abstract":"World models turn visual representations into predictions and actions. We argue that their visual foundation should combine physical-state information, native language alignment, and pixel reconstruction, so that a single representation can support prediction, instruction, and inspection. However, existing encoders typically specialize in only one or two of these capabilities; we therefore introduce ALPS, a pretrained visual state encoder that unifies all three in a single set of patch tokens. Using only still images and text, we pretrain a ViT-B/16 from scratch with language, reconstruction, and spatial correspondence objectives, and then refine it via cross-view self-distillation. In a unified evaluation across fifteen public encoders, ALPS achieves the lowest aggregate LIBERO state-probe error among ViT-B models (matching the 3.5x larger V-JEPA 2-L) while maintaining strong language alignment and accurate pixel reconstruction. Downstream, frozen ALPS tokens reach the highest mean policy success on LIBERO (85.5%) when paired with our native text tower, match or exceed leading encoders on world-model planning across PushT, Wall, and Granular, and allow predicted future tokens to be decoded directly into pixels for visual inspection. We hope this work helps establish visual pretraining as a shared foundation for understanding, prediction, and action in world models.","authors":[{"name":"Yanqing Liu","position":0},{"name":"Letian Zhang","position":1},{"name":"Chengzhi Liu","position":2},{"name":"Zijun Wang","position":3},{"name":"Jieru Mei","position":4},{"name":"Hongliang Fei","position":5},{"name":"Yuyin Zhou","position":6},{"name":"Cihang Xie","position":7}],"categories":["cs.CV"],"primaryCategory":"cs.CV","publishedAt":"2026-10-21T17:57:11Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30120"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30120"},{"kind":"code","url":"https://github.com/UCSC-VLAA/ALPS","label":"UCSC-VLAA/ALPS"},{"kind":"project","url":"https://ucsc-vlaa.github.io/ALPS/"}],"signals":{"hfUpvotes":{"value":11,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubStars":{"value":1,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubRepo":"UCSC-VLAA/ALPS"},"brief":{"text":"ALPS combines physical-state encoding, language alignment, and pixel reconstruction in one visual representation. It achieves the lowest state-probe error among ViT-B encoders and strong results in robot policy success and world-model planning.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":12,"reason":"trending","score":0.5005},{"paper":{"id":"arxiv:2610.29458","arxivId":"2610.29458","doi":"10.48550/arXiv.2610.29458","title":"Harness Scaling for Generalizable Agentic Reinforcement Learning","abstract":"Agent harnesses have become a central component of modern agentic systems, shaping how models use tools, manage context, and coordinate long-horizon interaction. Realizing the benefits of these harnesses depends on how effectively the underlying model can make use of them, motivating reinforcement learning directly through the harness. However, existing approaches typically train through a single fixed harness, which can lead to specialization and limited transfer across harness configurations. To address this, we introduce \\textbf{Harness Scaling}, a new scaling dimension for agentic RL that extends training from a single harness to diverse execution configurations. Concretely, we construct a factorized harness space, select a compact training subset with balanced factorial coverage, and organize cross-harness training to combine global coverage with controlled variation for the same task. Across multiple benchmarks, model scales, and RL algorithms, Harness Scaling consistently improves generalization to unseen white-box configurations. Moreover, models trained through structured white-box variation transfer effectively to unseen external black-box harnesses, outperforming models trained through a single complex black-box harness. These results establish harness diversity as a meaningful scaling dimension for agentic RL and highlight the importance of how this diversity is organized for cross-harness generalization.","authors":[{"name":"Fei Bai","position":0},{"name":"Huatong Song","position":1},{"name":"Chuan Hao","position":2},{"name":"Ran Tao","position":3},{"name":"Jian Yang","position":4},{"name":"Bryan Dai","position":5},{"name":"Wayne Xin Zhao","position":6},{"name":"Ji-Rong Wen","position":7}],"categories":["cs.LG","cs.AI","cs.CL"],"primaryCategory":"cs.LG","publishedAt":"2026-10-21T09:15:50Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29458"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29458"},{"kind":"code","url":"https://github.com/RUCAIBox/Harness-Scaling","label":"RUCAIBox/Harness-Scaling"}],"signals":{"hfUpvotes":{"value":6,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubStars":{"value":0,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubRepo":"RUCAIBox/Harness-Scaling"},"brief":{"text":"Harness Scaling trains agents across a structured variety of tool-use and execution setups, improving generalization to unseen configurations and external harnesses compared with training through a single complex harness.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":13,"reason":"trending","score":0.4965},{"paper":{"id":"arxiv:2610.26453","arxivId":"2610.26453","doi":"10.48550/arXiv.2610.26453","title":"Learning What to Remember: Resource-Efficient Self-Improvement for Lifelong Agents","abstract":"Self-improving agents aim to continuously improve their behavior by learning from experience after deployment. However, accumulating and reusing more experience does not necessarily lead to better decisions, particularly when generation, interaction, and memory resources are jointly constrained. Under such stringent constraints, we find that several existing lifelong-memory methods provide limited benefits over simple retrieval or even no-memory agents. This reveals a fundamental challenge for resource-efficient self-improvement: agents must learn not only which past experiences are relevant, but which are actually worth reusing to improve future decisions. When memory capacity and opportunities for reasoning and recovery are limited, retrieving an ineffective experience can consume scarce resources and hinder downstream execution. To address this challenge, we propose TokenWorth, a lightweight feedback-driven memory adaptation framework that learns query-dependent memory worth from sparse execution feedback. Without updating the backbone model, TokenWorth uses the learned worth to guide both experience retrieval and bounded memory management, enabling the agent to continually adapt how it reuses accumulated experience. Across three LifelongAgentBench environments and both local and API-based deployments, TokenWorth achieves the highest task success rate while using the fewest output tokens among the evaluated methods under the specified resource budgets. These results suggest that effective self-improvement under resource constraints depends not on accumulating more experience, but on learning what is worth remembering. Our code is publicly available at https://github.com/AIGNLAI/TokenWorth.","authors":[{"name":"Guannan Lai","authorId":"oa:A5125074031","position":0},{"name":"Zi-Yang Bai","authorId":"oa:A5092799708","position":1},{"name":"Jun-Peng Jiang","authorId":"oa:A5034185483","position":2},{"name":"Han-Jia Ye","authorId":"oa:A5065180062","position":3}],"categories":["cs.AI"],"primaryCategory":"cs.AI","publishedAt":"2026-10-19T14:09:45Z","updatedAt":"2026-10-21T14:25:06Z","version":2,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.26453"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.26453"},{"kind":"code","url":"https://github.com/AIGNLAI/TokenWorth","label":"AIGNLAI/TokenWorth"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.26453.jpg","label":"thumbnail"}],"citations":{"value":0,"source":"openalex","fetchedAt":"2026-10-22T08:09:23Z"},"signals":{"hfUpvotes":{"value":11,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubStars":{"value":3,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubRepo":"AIGNLAI/TokenWorth"},"brief":{"text":"TokenWorth learns from sparse execution feedback which past experiences are worth retrieving and retaining, without changing the backbone model. Across three benchmark environments and local and API deployments, it achieved the highest task success rate while using the fewest output tokens under the tested budgets.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":14,"reason":"trending","score":0.4686},{"paper":{"id":"arxiv:2610.26542","arxivId":"2610.26542","doi":"10.48550/arXiv.2610.26542","title":"Breaking Diffusion: Parallel Reasoning with Draft Refinement","abstract":"Drafting an answer allows a reasoning model to reconsider each decision in the context of the others. Yet repeated diffusion denoising does not ensure that individually plausible token updates form a better answer together. We introduce Draft Refinement Language Models (DRLMs), which use a single denoiser to propose parallel revisions and evaluate a candidate draft. We train the denoiser to reconstruct every token from uniformly corrupted drafts, teaching both preservation and repair. Its predictions specify revisions directly, without a diffusion reverse transition. Evaluation requires a different use of these predictions: confidence in an existing token reflects both contextual support and evidence from observing the token itself. Using the known corruption process, we correct for this self-token evidence to estimate how well each token is supported by the prompt and the remaining draft. Aggregating these estimates yields a cavity energy computed in one forward pass, without an additional scoring model. Each round proposes a revision, reads the completed candidate in its revised context, and retains it only if its energy is lower. Across Countdown, Sudoku-Extreme, and Maze-Hard, our method outperforms autoregressive and discrete diffusion baselines. With 6M parameters, DRLM reaches 44.4% on Countdown-5 and 93.9% on Sudoku-Extreme. The results establish complete-draft parallel refinement as an effective alternative to reverse diffusion for complex reasoning.","authors":[{"name":"Jaesang Yu","authorId":"oa:A5124990830","position":0},{"name":"Seoyeon Byun","authorId":"oa:A5112666713","position":1},{"name":"Yun Wang","position":2},{"name":"Yan Lu","position":3},{"name":"Juho Kim","authorId":"oa:A5147434150","position":4}],"categories":["cs.AI","cs.CL","cs.LG"],"primaryCategory":"cs.AI","publishedAt":"2026-10-19T15:02:45Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.26542"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.26542"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.26542.jpg","label":"thumbnail"}],"citations":{"value":0,"source":"openalex","fetchedAt":"2026-10-22T08:09:23Z"},"signals":{"hfUpvotes":{"value":19,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"}},"brief":{"text":"Draft Refinement Language Models propose parallel edits and keep them only when a one-pass score indicates a better-supported draft. They outperform autoregressive and discrete diffusion baselines on Countdown, Sudoku-Extreme, and Maze-Hard.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":15,"reason":"trending","score":0.4662},{"paper":{"id":"arxiv:2610.29357","arxivId":"2610.29357","doi":"10.48550/arXiv.2610.29357","title":"Memory-Centric Continual Learning for AI Agents","abstract":"Autonomous agents powered by foundation models can reason, plan, and use tools, but their fixed weights and context limits hinder sustained learning from experience over long-term deployment. While external memory systems attempt to bridge this gap, they mostly act as passive retrieval caches, evaluated primarily by whether they can recall stored facts or past interactions. Whether memory can drive genuine continual learning remains an open question. In this paper, we introduce Memory-Centric Continual Learning (MCCL), arguing that memory should be judged by whether it improves future actions, not merely what it retrieves. MCCL has three main parts. First, a three-axis taxonomy organizes memory by where state is stored (substrates), what form it takes (assets), and how it is managed (policies). This defines a two-loop lifecycle that updates memory state in an inner loop while refining management policies in an outer loop based on task feedback. Second, an evaluation protocol measures retention under interference, adaptation to change, and task transfer to isolate genuine behavioral gains from simple context expansion. Third, a memory scaling framework models how agent performance scales with experience, storage capacity, and compute budgets across non-parametric and parametric memory. Together, these components reframe agent memory from passive lookup to continuous learning, offering a principled route toward self-improving autonomous agents.","authors":[{"name":"Dongfang Li","position":0},{"name":"Zhuoen Chen","position":1},{"name":"Bonian Jia","position":2},{"name":"Zixuan Liu","position":3},{"name":"Zhengxuan Lu","position":4},{"name":"Can Deng","position":5},{"name":"Yukun Shi","position":6},{"name":"Yuhang Meng","position":7},{"name":"Xin Zhang","position":8},{"name":"Jiahe Huang","position":9},{"name":"Fuhao Li","position":10},{"name":"Junmai Wang","position":11},{"name":"Shunyao Wen","position":12},{"name":"Jingjun Sun","position":13},{"name":"Jiaxu Tian","position":14},{"name":"Jiangling Zhang","position":15},{"name":"Zihan Wang","position":16},{"name":"Liangyou Li","position":17},{"name":"Yitong Yao","position":18},{"name":"Shuangyong Song","position":19},{"name":"Beilun Wang","position":20},{"name":"Pengxu Wei","position":21},{"name":"Hanlin Xu","position":22},{"name":"Chang Liu","position":23},{"name":"Yongxiang Li","position":24},{"name":"Baotian Hu","position":25},{"name":"Xiangyang Ji","position":26},{"name":"Min Zhang","position":27}],"categories":["cs.CL","cs.AI"],"primaryCategory":"cs.CL","publishedAt":"2026-10-21T07:49:42Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29357"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29357"}],"signals":{"hfUpvotes":{"value":17,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"}},"brief":{"text":"Memory-Centric Continual Learning frames agent memory as a driver of better future actions, not just a store for retrieval. It proposes a memory taxonomy, evaluation protocol, and scaling framework to assess learning from experience.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":16,"reason":"trending","score":0.4655},{"paper":{"id":"arxiv:2610.29680","arxivId":"2610.29680","doi":"10.48550/arXiv.2610.29680","title":"TraceForge: Distilling with Your Own Thinking","abstract":"Using an LLM to generate training data for another is among the most general ways to train one, and it remains common practice. Some frontier teams have closed that route for their own models: what comes back is an answer, or a short summary, without the detailed reasoning trace behind it. That is a little ironic: they were themselves built on distilling human knowledge, and distilling an open-source model is still entirely routine; the restriction applies to their own outputs alone. Recent work recovers the hidden reasoning by replaying encrypted blocks across models. It somehow works, but it breaks the providers' terms of service, so it is not something a product can be built on long-term. We built the alternative. We take the context and the teacher's answer as input, and generate the reasoning trace ourselves, in the style of the student's own model family. Training is two-stage, supervised and then adversarial: a generator learns to regenerate the family's own reasoning traces from their context and answer, and a critic learns how real a reasoning trace looks. Applied to a teacher's answer, the generator writes one in the family's style. An independent judge rates how likely traces are to be genuine on a scale from 0 to 1: ours average 0.68, real traces 0.66, and traces selected by the critic from four candidates 0.82. These are authenticity scores, not correctness rates. We test the idea on five benchmarks with everything else held fixed. On general tasks the five training corpora are about the same; on coding, the generated reasoning beats Fable 5's own thinking content. With the same 5,000 questions and the same teacher answers, HumanEval+ improves by 3.7 points and MBPP+ by 7.1 at the final checkpoint, and the MBPP+ lead holds at every checkpoint. We suspect that reasoning written in the student's family style is easier for it to learn from.","authors":[{"name":"Yan Wang","position":0}],"categories":["cs.AI","cs.LG"],"primaryCategory":"cs.AI","publishedAt":"2026-10-21T12:57:29Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29680"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29680"},{"kind":"project","url":"https://alleai.cloud/cot/"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.29680.jpg","label":"thumbnail"}],"signals":{"hfUpvotes":{"value":15,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"}},"brief":{"text":"TraceForge generates reasoning traces in a student model’s style from a teacher’s answers, avoiding restricted access to hidden traces. On coding benchmarks, it outperforms training on the teacher’s answers alone, improving HumanEval+ by 3.7 points and MBPP+ by 7.1.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":17,"reason":"trending","score":0.4528},{"paper":{"id":"arxiv:2610.30090","arxivId":"2610.30090","doi":"10.48550/arXiv.2610.30090","title":"VX-One: Typed, Calibrated Decision Heads for Real-Time Perimeter Triage - An empirical study of unknown-attack honesty, cascade gray zones, and calibration under drift","abstract":"Perimeter triage faces a latency/semantics trade-off: signature engines are fast and format-safe but blind to unseen attacks, while generative models are semantic but slow and produce unparseable output. In late 2026 a class of commercial System One decision models (TypeSafe Jev; Microsoft-Decision-1) began selling typed decision primitives - binary gates, closed-set choices, ordinal scores - as product APIs, with no academic treatment. We present VX-One: a small multi-head network (100K-10M parameters) whose outputs are these primitives - a calibrated gate, a choice over a closed candidate set that always includes an explicit unknown option, and an ordinal score, trained multi-task, temperature-calibrated per head (~0.7-1.7 ms p50 on CPU). We evaluate it on UNSW-NB15, CIC-IDS2017, EMBER2024 against calibrated XGBoost, Random-Forest, MLP, Kitsune, and measure a two-phase cascade (Suricata with 52,674 Emerging-Threats rules -> VX-One) over five days of CIC-IDS2017. Findings: (i) VX-One wins detection (AUROC/FPR@95%TPR) on UNSW-NB15 and under temporal drift on CIC-IDS2017 (0.711 vs 0.637), but XGBoost wins multi-class attribution on all three benchmarks; (ii) leave-one-family-out gate detects never-seen families on 9/9 UNSW folds yet drops below chance on CIC recon family, and abstention threshold transfers poorly across benchmarks; (iii) on payload-less reconstructions signature engines resolve ~0.1% of flows - gray zone ~99.9% of traffic, recovered by Phase 2 (combined recall >=0.978); (iv) calibration collapses under temporal drift (ECE 0.04->0.40) for every method while ranking partially survives. We provide a measured account of where typed calibrated decisions help perimeter triage, and where they fail.","authors":[{"name":"Juan S. Santillana","position":0}],"categories":["cs.CR","cs.CL","cs.LG","cs.NI"],"primaryCategory":"cs.CR","publishedAt":"2026-10-21T17:39:12Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30090"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30090"},{"kind":"code","url":"https://github.com/vectrayx/vx-one","label":"vectrayx/vx-one"},{"kind":"project","url":"https://vectrayx.com"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.30090.jpg","label":"thumbnail"}],"signals":{"hfUpvotes":{"value":2,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubStars":{"value":0,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubRepo":"vectrayx/vx-one"},"brief":{"text":"VX-One is a fast, calibrated network triage model that detects attacks and flags unknowns, but XGBoost gives better attack-family attribution. Under temporal drift, detection outperforms XGBoost, while calibration degrades for all tested methods.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":18,"reason":"trending","score":0.4381},{"paper":{"id":"arxiv:2610.30023","arxivId":"2610.30023","doi":"10.1145/3829340.3842349","title":"RealMe: Immersive Volumetric Video from Minimal Captures","abstract":"This paper addresses the problem of creating immersive volumetric videos from sparse and low-resolution capture resources. Conventional dense multi-view reconstruction methods can produce high-quality volumetric videos, but their reliance on costly capture setups makes them impractical in real-world scenarios. Recent approaches use generative models to relax capture requirements, but still suffer from limited resolution and 4D inconsistency. To overcome these constraints, we propose RealMe, a unified framework built around a high-resolution video diffusion model that enhances limited captures into high-resolution (2K), 4D-consistent videos for volumetric video reconstruction. Specifically, RealMe constructs auxiliary data from capture videos and encodes them into latents along with layout-aware positional embeddings. The video diffusion model adapts factorized spatio-temporal attention to the multi-view capture layout, denoises these latents across viewpoints and time, and decodes them into enhanced videos to refine the 4D representation. This adaptation helps the model capture temporal dynamics and multi-view relationships, leading to more 4D-consistent generation. To reduce the data gap between generated videos and real captures, we apply a latent frequency loss that provides frequency-domain supervision in latent space. Extensive experiments on the public DNA-Rendering benchmark and our self-captured RealVV dataset demonstrate that RealMe is able to create immersive volumetric videos and outperforms state-of-the-art methods in visual fidelity and 4D consistency.","authors":[{"name":"Chenxu Zhou","position":0},{"name":"Zhiyuan Yu","position":1},{"name":"Yudong Jin","position":2},{"name":"Yinji Shentu","position":3},{"name":"Yu Zhang","position":4},{"name":"Jiaming Sun","position":5},{"name":"Xiaowei Zhou","position":6}],"categories":["cs.CV"],"primaryCategory":"cs.CV","publishedAt":"2026-10-21T16:49:14Z","version":1,"license":"http://creativecommons.org/licenses/by-sa/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30023"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30023"},{"kind":"code","url":"https://github.com/zju3dv/RealMe","label":"zju3dv/RealMe"},{"kind":"project","url":"https://zju3dv.github.io/realme/"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.30023.jpg","label":"thumbnail"}],"signals":{"hfUpvotes":{"value":5,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubStars":{"value":2,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubRepo":"zju3dv/RealMe"},"brief":{"text":"RealMe uses video diffusion to turn sparse, low-resolution captures into high-resolution, temporally consistent volumetric videos. It outperforms existing methods in visual fidelity and 4D consistency on two datasets.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":19,"reason":"trending","score":0.4325},{"paper":{"id":"arxiv:2610.23751","arxivId":"2610.23751","doi":"10.48550/arXiv.2610.23751","title":"MTEdit: Packed Trajectories Beat Rollouts for Multi-Turn Image Editing","abstract":"Multi-turn image editing aims to modify an image according to a sequence of instructions, making it better aligned with human-AI collaborative creation than single-turn editing. However, constructing high-quality multi-turn training data remains challenging, as conventional sequential rollout accumulates errors and introduces visual-quality drift across editing turns. To address this challenge, we introduce MTEdit, a multi-turn image editing dataset built through a pipeline centered on packed trajectory canvas (PTC) generation. By jointly generating multiple editing states within a unified visual context, this pipeline reduces visual-quality drift across turns. Covering both step-wise and context-aware editing scenarios, MTEdit comprises 500K multi-turn image editing trajectories, with up to 8 turns and an average of 5.38 turns per trajectory, making it one of the most extensive datasets for multi-turn image editing. We further introduce MTEdit-Bench, a benchmark designed to evaluate both scenarios by assessing models' ability to follow editing instructions, maintain content consistency, resist visual-quality drift, and effectively use historical context across turns. Using MTEdit, we fine-tune Qwen-Image-Edit-Causal to obtain Qwen-Image-MTEdit. On MTEdit-Bench, we evaluate a broad range of closed-source and open-source image editing models, including Qwen-Image-MTEdit. Experimental results show that Qwen-Image-MTEdit achieves the best performance among the evaluated open-source models, demonstrating the effectiveness of MTEdit in improving models' multi-turn image editing capabilities.","authors":[{"name":"Houyuan Chen","authorId":"oa:A5159041480","position":0},{"name":"Fanzhou Wang","authorId":"oa:A5021955233","position":1},{"name":"Xiangyu Fan","authorId":"oa:A5159077022","position":2},{"name":"Songlin Yang","authorId":"oa:A5159081009","position":3},{"name":"Zhitao Yang","authorId":"oa:A5119736214","position":4},{"name":"Lvmin Zhang","authorId":"oa:A5079385579","position":5},{"name":"Pei Yang","authorId":"oa:A5159060303","position":6},{"name":"Hao Zhao","authorId":"oa:A5159051291","position":7},{"name":"Dahua Lin","position":8},{"name":"Lei Yang","authorId":"oa:A5159080957","position":9},{"name":"Anyi Rao","authorId":"oa:A5067715162","position":10}],"categories":["cs.CV"],"primaryCategory":"cs.CV","publishedAt":"2026-10-16T16:56:48Z","updatedAt":"2026-10-21T08:15:50Z","version":2,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.23751"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.23751"}],"citations":{"value":0,"source":"openalex","fetchedAt":"2026-10-22T08:09:23Z"},"signals":{"hfUpvotes":{"value":34,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"}},"brief":{"text":"MTEdit introduces a dataset of 500,000 multi-turn image-editing trajectories and a benchmark for testing instruction-following, consistency, and use of prior turns. Fine-tuning on the dataset gives Qwen-Image-MTEdit the strongest performance among evaluated open-source models.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":20,"reason":"trending","score":0.4227},{"paper":{"id":"arxiv:2610.29560","arxivId":"2610.29560","doi":"10.48550/arXiv.2610.29560","title":"From VLM Planners to Robotic Agents: Self-Evolving Harnesses for History-Dependent Manipulation","abstract":"Vision-language-action (VLA) models achieve strong performance across embodied manipulation tasks by unifying visual understanding, language conditioning, and action generation. However, long-horizon tasks can involve non-Markovian dependencies: successful decisions may require retaining task progress and past interaction outcomes, while partial observability and execution errors compound over time. Existing work has begun to address these challenges by augmenting foundation-model policies with memory and by introducing hierarchical architectures that separate high-level planning from low-level execution. We organize the high-level VLM as a robotic agent through an editable harness to maintain task context and improve planning through execution experience. We instantiate this approach as SERA (Self-Evolving Robotic Agent), with a fixed VLA executor providing low-level control. The harness maintains and updates task context using observations and execution feedback to guide subsequent planning. Self-evolution uses environment rollouts to propose and evaluate harness revisions, with the selected configuration fixed before held-out testing. On RoboMME, SERA achieves 51.38% overall success with autonomous selection, compared with 44.38% without self-evolution. Our exploration suggests that agentic planning is a promising direction for history-dependent robotic manipulation.","authors":[{"name":"Tingwen Du","position":0},{"name":"Haoyu Guo","position":1},{"name":"Hanxue Zhang","position":2},{"name":"Xuanxu Wang","position":3},{"name":"Junting Dong","position":4},{"name":"Chao Gao","position":5},{"name":"Mingfei Han","position":6},{"name":"Tao Lu","position":7},{"name":"Mulin Yu","position":8},{"name":"Chunhua Shen","position":9},{"name":"Xiaojun Chang","position":10},{"name":"Zhihui Li","position":11}],"categories":["cs.RO"],"primaryCategory":"cs.RO","publishedAt":"2026-10-21T10:49:49Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29560"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29560"},{"kind":"code","url":"https://github.com/StevenKKXS/SERA","label":"StevenKKXS/SERA"},{"kind":"project","url":"https://stevenkkxs.github.io/sera-webpage/"}],"signals":{"hfUpvotes":{"value":4,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubStars":{"value":2,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubRepo":"StevenKKXS/SERA"},"brief":{"text":"SERA adds an editable, self-improving harness around a vision-language planner, using task history and execution feedback to guide a fixed low-level controller. On RoboMME, autonomous harness selection raises overall success to 51.38%, versus 44.38% without self-evolution.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":21,"reason":"trending","score":0.4128},{"paper":{"id":"arxiv:2610.29441","arxivId":"2610.29441","doi":"10.48550/arXiv.2610.29441","title":"Sparsity Should Be ALLOTted by Output: Training-Free Allocation of Block-Sparse Attention for Video Diffusion Transformers","abstract":"Full 3D attention is costly in video Diffusion Transformers. Block-sparse attention reduces this cost by computing only selected attention blocks. However, existing methods either overlook the potential to optimize sparsity allocation or rely on heuristic rules or local metrics to guide it. In this work, we observe that sparsifying attention in different layers and at different denoising steps affects final-video fidelity to substantially different degrees. Yet the identities of the most sensitive layer-step pairs remain nearly unchanged across inputs. Building on these findings, we propose Allocating Layer-step Latency by Output Tests (ALLOT), a training-free sparsity allocation method. Treating each layer-step pair as a cell, ALLOT estimates each cell's contribution to final-video error through (i) cell probing and (ii) trajectory calibration, then jointly optimizes cell densities under a measured-latency budget. It preserves existing selection strategies and kernels, replacing only the allocation component. At equal or shorter wall-clock time, ALLOT outperforms every compared allocation strategy with XAttention, with an 11.8 dB PSNR gain on HunyuanVideo. Extensive experiments across four additional published pipelines on Wan2.1, Wan2.2, and HunyuanVideo show that ALLOT outperforms or ties the default in all nine settings.","authors":[{"name":"Tianze Xia","position":0},{"name":"Xiaoling Zhou","position":1},{"name":"Ruya Lan","position":2},{"name":"Yiqiao Liao","position":3},{"name":"Chengru Song","position":4},{"name":"Shiming Xiang","position":5}],"categories":["cs.CV"],"primaryCategory":"cs.CV","publishedAt":"2026-10-21T09:01:48Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29441"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29441"}],"signals":{"hfUpvotes":{"value":12,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"}},"brief":{"text":"ALLOT allocates block-sparse attention across layers and denoising steps using output-error tests and a measured-latency budget, without training. It beats compared allocation strategies at equal or shorter runtime, with an 11.8 dB PSNR gain on HunyuanVideo.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":22,"reason":"trending","score":0.3969},{"paper":{"id":"arxiv:2610.29968","arxivId":"2610.29968","doi":"10.48550/arXiv.2610.29968","title":"CulturalRiddles: A Multicultural Benchmark of Riddles","abstract":"As LLMs expand their language coverage, evaluating how well they understand and represent diverse cultures becomes increasingly important. However, many benchmarks remain English-centric, with limited cultural coverage and a focus on factual knowledge over its application in real-world cultural contexts. We propose CulturalRiddles, a benchmark of original, culturally grounded riddles created entirely by humans through a participatory community effort, spanning 62 cultures (45 countries, 51 languages). These riddles are deeply rooted in their respective cultures and require abductive and compositional reasoning, alongside detailed factual and social knowledge, language skills, and an understanding of culturally specific associations. We benchmark 28 LLM configurations using both automatic and large-scale human evaluation. Even frontier models achieve at most 22% accuracy in some communities. We publicly release all artifacts alongside an open contribution process to expand the benchmark's linguistic and cultural coverage.","authors":[{"name":"My Chiffon Nguyen","position":0},{"name":"Tianyi Hu","position":1},{"name":"Vinod Anbalagan","position":2},{"name":"Henry Gagnier","position":3},{"name":"Pouya Sadeghi","position":4},{"name":"Farah Abdou","position":5},{"name":"Tidor-Vlad Pricope","position":6},{"name":"Diego MMS Oliveira","position":7},{"name":"Jocelyn K D'Arcy","position":8},{"name":"Madeline Smith","position":9},{"name":"Oluwaseun A Ajayi","position":10},{"name":"Yusser Al Ghussin","position":11},{"name":"Zena Al-Khalili","position":12},{"name":"Micol Altomare","position":13},{"name":"Shayan Amir","position":14},{"name":"Ahanaf Aziz","position":15},{"name":"Bryan Chen","position":16},{"name":"Ruchira Dhar","position":17},{"name":"Abdellah El Mekki","position":18},{"name":"Rawan Mohammed Elframawy","position":19},{"name":"Akhil Eppa","position":20},{"name":"Silvia Fernandez","position":21},{"name":"Alex Gimei","position":22},{"name":"Arya Hariharan","position":23},{"name":"Christian Hoang","position":24},{"name":"RamMohanRao Kadiyala","position":25},{"name":"Diana Kozachek","position":26},{"name":"Peerat Limkonchotiwat","position":27},{"name":"Ankita Maity","position":28},{"name":"Di Mo","position":29},{"name":"Ayush Munot","position":30},{"name":"Aylin Naebzadeh","position":31},{"name":"Anastasiia Nikitina","position":32},{"name":"Dina Pisarevskaya","position":33},{"name":"Divas Poudel","position":34},{"name":"Jebish Purbey","position":35},{"name":"Md Rabius Sany Apu","position":36},{"name":"Ahmad Mufli Ramadhan","position":37},{"name":"Ralitza Nikolova Rizova","position":38},{"name":"Diogo M. Serrano-Miranda","position":39},{"name":"Vivek Gopalkrishna Silimkhan","position":40},{"name":"Luis Fernando Solis Navarro","position":41},{"name":"Kaiser Sun","position":42},{"name":"Arya Suneesh","position":43},{"name":"Manaswitha Sunkara","position":44},{"name":"Tejaswi Sunkara","position":45},{"name":"Jonathan E Swindell","position":46},{"name":"Hanna Yukhymenko","position":47},{"name":"Jingyi Zheng","position":48},{"name":"Esther Adenuga","position":49},{"name":"Ilia Afanasev","position":50},{"name":"Wafa Aissa","position":51},{"name":"Lisa Alazraki","position":52},{"name":"Lauren C Altomare","position":53},{"name":"Priscilla Amondi Amuok","position":54},{"name":"Emile Anand","position":55},{"name":"Gabriel Assis","position":56},{"name":"José Matías Avalos","position":57},{"name":"Silvia Lorena Avila","position":58},{"name":"Zakaria Baannou","position":59},{"name":"Niyati Bafna","position":60},{"name":"Yash Bagla","position":61},{"name":"Sarah Bahrami","position":62},{"name":"Bontu F Balcha","position":63},{"name":"Vanya Bannihatti Kumar","position":64},{"name":"Irene Baucells","position":65},{"name":"Abdumutalib Begmuratov","position":66},{"name":"Naman Bhatia","position":67},{"name":"Souvik Bhattacharyya","position":68},{"name":"Isabel Blanco Ferreiro","position":69},{"name":"Felipe Bochini","position":70},{"name":"Sayuru Bopitiya","position":71},{"name":"Angana Borah","position":72},{"name":"Tulika Bose","position":73},{"name":"Olena Bugaiova","position":74},{"name":"Juliana Carvalho","position":75},{"name":"Nancy F Chen","position":76},{"name":"Stephy Chen","position":77},{"name":"Rochelle Choenni","position":78},{"name":"Jayanth Krishna Chundru","position":79},{"name":"Leandro A Cossio","position":80},{"name":"Nebojša Damnjanović","position":81},{"name":"Ona de Gibert","position":82},{"name":"Marco De Santis","position":83},{"name":"Pedro Delgado","position":84},{"name":"Kaustubh Dholé","position":85},{"name":"Adnan El Assadi","position":86},{"name":"Wesam El-Sayed","position":87},{"name":"Nadine M Elframawy","position":88},{"name":"Oumayma Essarhi","position":89},{"name":"Naome A Etori","position":90},{"name":"Reuben Chagas Fernandes","position":91},{"name":"Saanidhi Pravin Gade","position":92},{"name":"Javier García Gilabert","position":93},{"name":"Álvaro González-Escapa","position":94},{"name":"Bin Hu","position":95},{"name":"Kaiyu Huang","position":96},{"name":"Abderrahmane Issam","position":97},{"name":"Danka Jokić","position":98},{"name":"Ulugbek Khalilov","position":99},{"name":"Dominik Krzemiński","position":100},{"name":"Roy Ka-Wei Lee","position":101},{"name":"Haichang Li","position":102},{"name":"Chen Cecilia Liu","position":103},{"name":"Zhengyuan Liu","position":104},{"name":"Cedric Lothritz","position":105},{"name":"Houman Mehrafarin","position":106},{"name":"Antonio Miranda-Escalada","position":107},{"name":"Fengran Mo","position":108},{"name":"Mehrnaz Mofakhami","position":109},{"name":"Wafaa Mohammed","position":110},{"name":"Felipe Moreno-Vera","position":111},{"name":"Prahitha Movva","position":112},{"name":"Jannatul Nayem","position":113},{"name":"Van Ngo","position":114},{"name":"Joy Olusanya","position":115},{"name":"Marc Pàmies","position":116},{"name":"Avinash Kuney Pandey","position":117},{"name":"Yurii Paniv","position":118},{"name":"Fred Philippy","position":119},{"name":"Vladislav Poritski","position":120},{"name":"Dikshya Poudel","position":121},{"name":"Joanna Radoła","position":122},{"name":"Sara Rajaee","position":123},{"name":"Swati Rajwal","position":124},{"name":"Prerana Rane","position":125},{"name":"Muhammad Ravi Shulthan Habibi","position":126},{"name":"Lucas Resck","position":127},{"name":"Shubhashis Roy Dipta","position":128},{"name":"Ananya Sahu","position":129},{"name":"Eduardo Sánchez","position":130},{"name":"Borja Sánchez Clemente","position":131},{"name":"Aleix Sant","position":132},{"name":"Nobin Sarwar","position":133},{"name":"Istiak Shihab","position":134},{"name":"Santhoshini Pranathi Singaraju","position":135},{"name":"Jane Smith","position":136},{"name":"David Stap","position":137},{"name":"Emily Z Su","position":138},{"name":"Ananth K Suresh","position":139},{"name":"Sajag Swami","position":140},{"name":"Ruiyang Tan","position":141},{"name":"Nazia Tasnim","position":142},{"name":"Khalid Tounoussi","position":143},{"name":"Patteera Triamamornwooth","position":144},{"name":"Muhammad Saad Umar","position":145},{"name":"Shankar Venkitachalam","position":146},{"name":"Oksana Volchek","position":147},{"name":"Lingbo Wang","position":148},{"name":"Debela Desalegn Yadeta","position":149},{"name":"Rushikesh Zawar","position":150},{"name":"Jinghan Zhang","position":151},{"name":"Mike Zhang","position":152},{"name":"Weihua Zheng","position":153},{"name":"Marzieh Fadaee","position":154},{"name":"Akhil Arora","position":155},{"name":"Károly Boczka","position":156},{"name":"Julia Kreutzer","position":157}],"categories":["cs.CL","cs.AI"],"primaryCategory":"cs.CL","publishedAt":"2026-10-21T16:20:30Z","version":1,"license":"http://creativecommons.org/licenses/by-sa/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29968"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29968"},{"kind":"code","url":"https://github.com/Cohere-Labs-Community/CulturalRiddles","label":"Cohere-Labs-Community/CulturalRiddles"},{"kind":"project","url":"https://culturalriddles.ai/"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.29968.jpg","label":"thumbnail"}],"signals":{"hfUpvotes":{"value":2,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubStars":{"value":1,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubRepo":"Cohere-Labs-Community/CulturalRiddles"},"brief":{"text":"CulturalRiddles introduces a human-created benchmark of culturally grounded riddles spanning 62 cultures. Across 28 LLM configurations, even frontier models reach no more than 22% accuracy in some communities.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":23,"reason":"trending","score":0.3836},{"paper":{"id":"arxiv:2610.29784","arxivId":"2610.29784","doi":"10.48550/arXiv.2610.29784","title":"Evolving with Open Worlds: Towards Recursive Self-Improvement for Future Forecasting","abstract":"Most contemporary artificial intelligence (AI) systems are released with frozen parameters and improve through external updates, whereas an ideal system would keep improving from its own experience, a capacity we call recursive self-improvement (RSI). Yet studying RSI requires an environment that is continuously changing, objectively verifiable, and naturally grounded, posing a fundamental challenge for agent evolution. Here we introduce DeepEvolve, a sustainable evolution agent system grounded in future event prediction, in which agents continuously interact with a dynamically evolving environment and improve by internalizing and externalizing the experience gained through these interactions. A living environment synthesizes verifiable tasks, and a recursive-evolution engine turns the reasoning trajectory behind each prediction back into capability: decoupling search from prediction adapts parameter-level evolution to the contamination-prone live environment, while distilling reusable skills lets the agent evolve from the limited live trajectories such an environment affords. On the external FutureX leaderboard, which evaluates predictions against real-world outcomes and is updated weekly, our self-improved forecaster ranked first across all systems over three consecutive weeks, achieving up to a 195% improvement in rating. On our synthesized evaluation, it ranked among the top three across four base LLMs, outperforming the strongest baseline by up to 10.7 Bradley--Terry points. More broadly, these results suggest that a system built for self-improvement is best understood as an agent that co-evolves with a genuinely living, open-ended environment, pointing to a shift in AI from training static models on fixed data toward cultivating agents that grow stronger from their own experience, with such an open world as the indispensable basis of that growth.","authors":[{"name":"Zhipeng Chen","position":0},{"name":"Yunsen Liang","position":1},{"name":"Yanxi Chen","position":2},{"name":"Hongkang Zhao","position":3},{"name":"Lingyun Feng","position":4},{"name":"Ruiyang Ren","position":5},{"name":"Guoxin Chen","position":6},{"name":"Jia Deng","position":7},{"name":"Feng Chang","position":8},{"name":"Chuan Hao","position":9},{"name":"Ran Tao","position":10},{"name":"Jian Yang","position":11},{"name":"Ji-Rong Wen","position":12},{"name":"Bryan Dai","position":13},{"name":"Wayne Xin Zhao","position":14}],"categories":["cs.CL"],"primaryCategory":"cs.CL","publishedAt":"2026-10-21T14:09:57Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29784"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29784"},{"kind":"code","url":"https://github.com/RUCAIBox/DeepEvolve","label":"RUCAIBox/DeepEvolve"}],"signals":{"hfUpvotes":{"value":3,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubStars":{"value":1,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubRepo":"RUCAIBox/DeepEvolve"},"brief":{"text":"DeepEvolve helps forecasting agents improve by learning from predictions in a changing, verifiable environment. Its forecaster ranked first on the FutureX leaderboard for three consecutive weeks, with ratings improving by up to 195%.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":24,"reason":"trending","score":0.3815},{"paper":{"id":"arxiv:2610.28887","arxivId":"2610.28887","doi":"10.48550/arXiv.2610.28887","title":"Building AI Research Teams for Autonomous Discovery of Linear Attention Architectures","abstract":"Autonomous agents can increasingly propose ideas, modify code, and run experiments, creating an opportunity to accelerate scientific discovery using teams of collaborative AI researchers. To escape local minima in the discovery, we propose to encourage orthogonal exploration by constructing the team using agents with varying expertise. However, realizing this design requires more than diverse backgrounds. When experiments are costly, a team can test only a limited number of ideas, thereby creating the need to balance exploration and exploitation when selecting a subset of researchers to propose and execute new experiments. Furthermore, each result should shape what the team tries next. We therefore ask how AI researchers with different scientific backgrounds can build on one another's findings and sustain progress under a limited experimental budget. To address these challenges, we build a team of persistent AI researchers, each grounded in a different discipline, who keep private research histories, share experimental evidence, and receive research opportunities through adaptive allocation. We demonstrate the effectiveness of this design by deploying the team to search for linear attention architectures, where it outperforms GDN2 at 1.3B scale on language modeling, common-sense reasoning, and synthetic and real-world retrieval tasks. In controlled comparisons, removing any single organizational component from our design yields markedly weaker discoveries: without disciplinary backgrounds, the search plateaus early; without private histories, researchers make less progress despite identical shared evidence; and uniform selection falls short of adaptive allocation within the same budget. These results suggest that sustained autonomous discovery depends not only on stronger agents, but also on who explores, what they remember, and how research effort is allocated.","authors":[{"name":"Yujian Liu","position":0},{"name":"Li An","position":1},{"name":"Zelin He","position":2},{"name":"Jacob Helwig","position":3},{"name":"Jiabao Ji","position":4},{"name":"Rohit Jain","position":5},{"name":"Siyu Zhu","position":6},{"name":"Shiyu Chang","position":7}],"categories":["cs.CL"],"primaryCategory":"cs.CL","publishedAt":"2026-10-20T20:09:31Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28887"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28887"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.28887.jpg","label":"thumbnail"}],"signals":{"hfUpvotes":{"value":12,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"}},"brief":{"text":"A team of persistent AI researchers with varied expertise and adaptive experiment allocation discovered linear attention architectures that outperform GDN2 at 1.3B scale across language modeling, reasoning, and retrieval tasks. Removing disciplinary backgrounds, private research histories, or adaptive selection weakened results.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":25,"reason":"trending","score":0.3724},{"paper":{"id":"arxiv:2610.29434","arxivId":"2610.29434","doi":"10.48550/arXiv.2610.29434","title":"RubatoCodec: Budgeted Adaptive Speech Tokenization with Streaming Decoding","abstract":"Neural speech codecs provide discrete tokens for speech generation. Lower frame rates shorten token sequences, but each token must then carry more linguistic and acoustic information. Variable-frame-rate codecs address this trade-off by adapting token allocation to speech content, but threshold-based allocation lets content determine the token count, so the sequence length a speech generator must produce varies across utterances and datasets at a fixed setting. We present RubatoCodec, a variable-frame-rate speech codec built on X-Codec-2 with an exact utterance-level event budget and streaming decoding. Its Adaptive Firing mechanism uses changes in codec features and waveform spectra to divide speech into variable-duration events without a learned boundary predictor. Discrete codes and an explicit duration represent each event, allowing a speech generator to predict and decode one event at a time. Controlled experiments show that adaptive allocation outperforms fixed-rate coding in reconstruction fidelity and intelligibility at low event rates. A model trained on large-scale multilingual speech achieves competitive reconstruction quality across English, Japanese, and Mandarin, with the best intelligibility preservation among the compared codecs on the evaluated English and Japanese benchmarks at comparable bitrates. In a matched text-to-speech comparison, the adaptive representation yields lower WER and higher UTMOS and UTMOSv2 with comparable speaker similarity. The speech generator also learns the codec's utterance-level event budget without an explicit generation-time rate constraint. These results demonstrate that budgeted adaptive speech representations support both high-quality reconstruction and streaming text-to-speech generation. Code and weights are available at https://github.com/third-intelligence/RubatoCodec and https://huggingface.co/third-intelligence/RubatoCodec.","authors":[{"name":"Chihiro Arata","position":0},{"name":"Shohei Taniguchi","position":1},{"name":"Gouki Minegishi","position":2},{"name":"Kan Hatakeyama-Sato","position":3}],"categories":["cs.SD","eess.AS"],"primaryCategory":"cs.SD","publishedAt":"2026-10-21T08:57:55Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29434"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29434"},{"kind":"code","url":"https://github.com/third-intelligence/RubatoCodec","label":"third-intelligence/RubatoCodec"}],"signals":{"hfUpvotes":{"value":2,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubStars":{"value":0,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"},"githubRepo":"third-intelligence/RubatoCodec"},"brief":{"text":"RubatoCodec allocates an exact number of speech tokens per utterance and decodes them as a stream. Adaptive allocation improves low-rate reconstruction and intelligibility, while text-to-speech achieves lower word error and higher quality scores with comparable speaker similarity.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":26,"reason":"trending","score":0.3714},{"paper":{"id":"arxiv:2610.29545","arxivId":"2610.29545","doi":"10.48550/arXiv.2610.29545","title":"PACE: Periodic and Channel-Aligned Encoding of Four-Bit Recurrent States","abstract":"Linear-attention language models keep their context in a recurrent state of fixed size, where an inference system stores one copy of this state for every sequence it decodes at the same time. Storing the state at lower precision therefore lets more sequences share an accelerator. Existing state quantizers round the state after every generated token and report that four bits lose accuracy unless the bit width is raised or the model is fine-tuned.\\par We introduce PACE, a post-hoc four-bit state quantizer. PACE rounds the state only once every few generated tokens, as chunked kernels already do once per chunk during prefill, and keeps the tokens generated in between in a small buffer. Rounding less often reduces the error that accumulates in the state. Additionally, we find that the large entries of the gated-delta state gather in columns, because some channels of the update written at each token remain large across many tokens. PACE therefore shares each quantization scale within one column. It also keeps the heads that retain rounding error longest at eight bits. On four hybrid models from 4B to 35B parameters, PACE with mixed four- and eight-bit heads is competitive with eight-bit DAMP on the three larger models at about two thirds of its memory, PACE with four-bit heads throughout scores higher than Q-Mamba at equal memory, and rounding every few tokens preserves the decoding speedup of the buffer.","authors":[{"name":"Patrick Haller","position":0},{"name":"Filipe Laitenberger","position":1},{"name":"Alan Akbik","position":2}],"categories":["cs.CL"],"primaryCategory":"cs.CL","publishedAt":"2026-10-21T10:31:52Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29545"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29545"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.29545.jpg","label":"thumbnail"}],"signals":{"hfUpvotes":{"value":8,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"}},"brief":{"text":"PACE compresses recurrent states by rounding every few tokens, sharing scales by column and retaining eight-bit precision for error-prone heads. Across four hybrid models, it outperforms Q-Mamba at equal memory, while mixed-precision PACE matches eight-bit DAMP on three larger models using about two-thirds the memory.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":27,"reason":"trending","score":0.3571},{"paper":{"id":"arxiv:2610.24899","arxivId":"2610.24899","doi":"10.48550/arXiv.2610.24899","title":"HybridResearcher: Bridging the Structured-Unstructured Boundary for Deep Research Agents","abstract":"While LLM agents are increasingly capable of gathering and synthesizing evidence from a single homogeneous source, such as structured tables or unstructured data, real-world analytical questions frequently span heterogeneous sources. For example, answering one enterprise question may require both a result computed over an internal database and evidence found on the web. Recent evaluations on benchmarks like HybridDeepResearch show that even frontier agents struggle when evidence must cross this boundary. While these benchmarks have exposed the limitation, methodologies to explicitly teach cross-source reasoning remain absent. To address this, we introduce HybridResearcher, a 4B agent trained to bridge evidence across the structured and unstructured boundary. To build it, we synthesize cross-source bridge trajectories by starting with validated single-source tasks-such as existing text-to-SQL or web search queries-and fabricating the missing source around a bridge entity drawn from the seed's own output. We retain a sample only when strict construction gates confirm that neither side alone suffices, ensuring the data requires true cross-source interdependence. Finally, we train on this curated dataset using supervised fine-tuning followed by reinforcement learning. Evaluated on the existing HybridDeepResearch benchmark, HybridResearcher improves over our base model, Qwen3.5-4B, by 30.0 points pass@8 and surpasses the far larger Qwen3.5-397B-A17B by 8.3 points, while also improving over its base model on established, held-out single-source benchmarks (BrowseComp and BrowseComp-Plus), demonstrating that this complex cross-source reasoning can be taught directly to small models without triggering catastrophic forgetting.","authors":[{"name":"Soyoung Yoon","position":0},{"name":"Peiran Xu","position":1},{"name":"Han Wang","position":2},{"name":"Ruofan Wu","position":3},{"name":"Xiaodong Yu","position":4},{"name":"Feng Yan","position":5},{"name":"Mohit Bansal","authorId":"oa:A5001987532","position":6},{"name":"Debiao Li","position":7},{"name":"Seung-Won Hwang","position":8},{"name":"Yuxiong He","position":9},{"name":"Zhewei Yao","authorId":"oa:A5125681225","position":10}],"categories":["cs.AI","cs.CL"],"primaryCategory":"cs.AI","publishedAt":"2026-10-17T22:46:56Z","updatedAt":"2026-10-21T00:41:07Z","version":2,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.24899"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.24899"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.24899.jpg","label":"thumbnail"}],"citations":{"value":0,"source":"openalex","fetchedAt":"2026-10-22T08:09:23Z"},"signals":{"hfUpvotes":{"value":9,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"}},"brief":{"text":"HybridResearcher trains a 4B agent to connect evidence from databases and web sources using carefully constructed tasks. It improves on its base model by 30 points on HybridDeepResearch and also boosts performance on held-out web-search benchmarks.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":28,"reason":"trending","score":0.33},{"paper":{"id":"arxiv:2610.29317","arxivId":"2610.29317","doi":"10.48550/arXiv.2610.29317","title":"SpatialBrain: Spatial Intelligence Growing from Experience","abstract":"Spatial reasoning requires more than access to geometric tools: a reasoner must decide when direct observation is sufficient and when to reconstruct, measure, or change reference frames. We present SpatialBrain, a framework that retains experience about these decisions while keeping its vision-language backbone fixed. The key distinction is between evidence about a scene and experience about how to reason with that evidence. 3DFusionMemory fuses current-scene observations into queryable spatial relations. A Spatial Skill Library provides reusable tool procedures, and Spatial Cognition associates task conditions with experience about when and how to use memory and skills. Completed trajectories are abstracted into process records and consolidated between task blocks, allowing later decisions to draw on earlier execution without transferring scene coordinates or reference answers. Across five spatial-reasoning benchmarks, SpatialBrain improves on direct answers from the same backbone on four and ties on the fifth. Component comparisons, experience-coverage curves, and traced resource decisions characterize how the framework accumulates and uses these resources.","authors":[{"name":"Yifan Zhou","position":0},{"name":"Ziyang Gong","position":1},{"name":"Xiaojun Ye","position":2},{"name":"Yifei Liu","position":3},{"name":"Donggang Li","position":4},{"name":"Xiaolong Zhou","position":5},{"name":"Zhihang Zhong","position":6},{"name":"Xue Yang","position":7}],"categories":["cs.AI"],"primaryCategory":"cs.AI","publishedAt":"2026-10-21T07:15:54Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29317"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29317"}],"signals":{"hfUpvotes":{"value":6,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"}},"brief":{"text":"SpatialBrain combines scene memory with reusable reasoning procedures and learns when to apply them from past tasks, without changing its vision-language backbone. It outperforms direct answers on four of five spatial-reasoning benchmarks and ties on the fifth.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":29,"reason":"trending","score":0.3299},{"paper":{"id":"arxiv:2610.28711","arxivId":"2610.28711","doi":"10.48550/arXiv.2610.28711","title":"Qwen-Audio-3.1-ASR Technical Report","abstract":"We present Qwen-Audio-3.1-ASR, a family of audio-language models that builds upon its predecesor Qwen-Audio-3.0-ASR with improved speech recognition and expanded capabilities in speaker-diarized transcription and general audio understanding. The family retains support for 30 languages and 16 Chinese dialects, together with instruction-controlled capabilities including historical-context conditioning, hierarchical hotword customization, and native transcription polishing. Speaker-diarized ASR produces structured outputs containing speaker labels, timestamps, and recognized text, maintaining consistent speaker labels across turns and the representation of overlapping speech. The general audio understanding capabilities cover audio-, speech-, and music-captioning, temporal grounding and sound event detection, as well as audio question answering and multiple-choice reasoning. These tasks broaden the application scope from recognizing spoken words to interpreting speakers, acoustic events, and their temporal relations. To accommodate different deployment requirements, the family includes variants for asynchronous file transcription, low-latency streaming, interactive voice input, and non-streaming recognition, with speaker-diarized transcription and general audio understanding available in certain variants. Evaluations span conventional ASR, multilingual and dialectal speech, speaker-diarized transcription, and general audio understanding, demonstrating the capabilities of the Qwen-Audio-3.1-ASR family across transcription and broader audio analysis tasks.","authors":[{"name":"Chuanmeng Bian","position":0},{"name":"Daren Chen","position":1},{"name":"Peixin Chen","position":2},{"name":"Zhigao Chen","position":3},{"name":"Chong Deng","position":4},{"name":"Zhiyun Fan","position":5},{"name":"Zhifu Gao","position":6},{"name":"Bo Gong","position":7},{"name":"Qing Gu","position":8},{"name":"Jiajun He","position":9},{"name":"Yawei Hu","position":10},{"name":"Yunjie Ji","position":11},{"name":"Wei Ju","position":12},{"name":"Jingbei Li","position":13},{"name":"Xiangang Li","position":14},{"name":"Xu Li","position":15},{"name":"Zengxi Li","position":16},{"name":"Zheng Li","position":17},{"name":"Chengdong Liang","position":18},{"name":"Baiji Liu","position":19},{"name":"Yaming Liu","position":20},{"name":"Ying Liu","position":21},{"name":"Bin Ma","position":22},{"name":"Yiping Peng","position":23},{"name":"Yuezhang Peng","position":24},{"name":"Zhendong Peng","position":25},{"name":"Yu Pu","position":26},{"name":"Yang Shi","position":27},{"name":"Xin Shu","position":28},{"name":"Jian Tang","position":29},{"name":"Biao Tian","position":30},{"name":"Peiyao Wang","position":31},{"name":"Tianzi Wang","position":32},{"name":"Wen Wang","position":33},{"name":"Wupeng Wang","position":34},{"name":"Cheng Wen","position":35},{"name":"Yuzhong Wu","position":36},{"name":"Zijian Xia","position":37},{"name":"Yunchong Xiao","position":38},{"name":"Nan Yang","position":39},{"name":"Jianwei Yu","position":40},{"name":"Jixing Yu","position":41},{"name":"Binbin Zhang","position":42},{"name":"Lei Zhang","position":43},{"name":"Sitong Zhao","position":44},{"name":"Guangdong Zhou","position":45},{"name":"Yuan Zhou","position":46},{"name":"Jianheng Zhuo","position":47}],"categories":["cs.SD","cs.CL"],"primaryCategory":"cs.SD","publishedAt":"2026-10-20T17:08:12Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28711"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28711"}],"signals":{"hfUpvotes":{"value":4,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"}},"brief":{"text":"Qwen-Audio-3.1-ASR adds speaker-labeled transcription and broader audio understanding to speech recognition across 30 languages and 16 Chinese dialects. Evaluations cover transcription, diarization and audio analysis tasks.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":30,"reason":"trending","score":0.3022},{"paper":{"id":"arxiv:2610.30136","arxivId":"2610.30136","doi":"10.48550/arXiv.2610.30136","title":"Agentic World Action Models","abstract":"Recent world action models provide visual foresight and future prediction for robot manipulation, but lack explicit mechanisms for compositional skill reuse. Agentic robotics combines code-based planning and reasoning with reusable skills, but often lacks predictive grounding in physical dynamics. We introduce AW-0, an agentic world action model framework that integrates code-based planning with joint video and action prediction. We jointly model code, future video, and robot actions as three coupled generative streams within a unified mixture-of-transformers architecture. Within this architecture a code expert generates parameterized skill routines and control logic while video and action experts generate future visual states and low-level motor commands. Code serves two complementary roles: it structures multi-step prediction and control within the world model and represents skills as reusable programs for subsequent agentic execution. We further introduce an automated curation pipeline that labels robot and human video trajectories with aligned programs and consolidates recurring routines into reusable skill libraries. Evaluations on RoboTwin 2.0 and physical bimanual manipulation tasks show improved task performance over VLA and WAM baselines, alongside gains in language following, semantic reasoning and robustness, demonstrating that agentic code enables connection between planning, prediction and execution within WAMs.","authors":[{"name":"Zhuoyang Liu","position":0},{"name":"Dantong Niu","position":1},{"name":"Botong Zhang","position":2},{"name":"Boning Shao","position":3},{"name":"Jing Wang","position":4},{"name":"Ruijie Zheng","position":5},{"name":"Yuvan Sharma","position":6},{"name":"Runyu Lu","position":7},{"name":"Jiaming Liu","position":8},{"name":"Renrui Zhang","position":9},{"name":"Roei Herzig","position":10},{"name":"Yutong Bai","position":11},{"name":"Ken Goldberg","position":12},{"name":"Danfei Xu","position":13},{"name":"Yuke Zhu","position":14},{"name":"Guanzhi Wang","position":15},{"name":"Linxi \"Jim\" Fan","position":16},{"name":"Trevor Darrell","position":17}],"categories":["cs.RO"],"primaryCategory":"cs.RO","publishedAt":"2026-10-21T17:59:56Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30136"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30136"},{"kind":"project","url":"https://agentic-wam.github.io/"}],"signals":{"hfUpvotes":{"value":4,"source":"hf","fetchedAt":"2026-10-22T08:36:18Z"}},"brief":{"text":"AW-0 combines code-based planning with predictions of future video and robot actions, while organizing reusable skills as programs. It outperforms VLA and world action model baselines on RoboTwin 2.0 and physical bimanual tasks.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":31,"reason":"trending","score":0.2934},{"paper":{"id":"arxiv:2610.30131","arxivId":"2610.30131","doi":"10.48550/arXiv.2610.30131","title":"Scaling H2 Factorization to Billion-Unknown Three-Dimensional Problems","abstract":"Strong recursive skeletonization provides approximate direct solvers for dense matrices from integral equations and kernel methods, but distributed implementations face substantial communication costs and limited parallelism, especially in three dimensions. We present a distributed-memory $\\mathcal{H}^2$ factorization solver for symmetric matrices that combines spatial box coloring, delayed Schur complement updates, streamed sketching-based interpolative decomposition, and BLAS-3 and shared-memory kernels. Spatial coloring enables every process to participate in each boundary stage, while delayed updates and streamed sketching reduce dense storage and communication. To our knowledge, this is the first distributed-memory strong recursive skeletonization factorization demonstrated for three-dimensional volume integral equations at billion-unknown scale. Experiments on up to 512 compute nodes demonstrate factorization of a three-dimensional Laplace system with 3.62 billion unknowns, alongside three-dimensional Helmholtz tests and a two-dimensional Matérn system with 268 million unknowns. Combining streamed sketching and fully delayed updates reduces factorization time by $6.9\\times$ for Laplace and $8.5\\times$ for Helmholtz relative to the unsketched, eager configuration, while reducing maximum per-process resident memory by 57\\% and 54\\%, respectively.","authors":[{"name":"Tianyu Liang","position":0},{"name":"Chao Chen","position":1},{"name":"James Demmel","position":2},{"name":"Yang Liu","position":3}],"categories":["math.NA","cs.NA","math-ph","math.MP"],"primaryCategory":"math.NA","publishedAt":"2026-10-21T17:59:12Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30131"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30131"}],"brief":{"text":"A distributed-memory H² factorization solver scales strong recursive skeletonization to three-dimensional problems, demonstrating a Laplace system with 3.62 billion unknowns. Streamed sketching and delayed updates cut factorization time by up to 8.5× and per-process memory by up to 57%.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":32,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.30123","arxivId":"2610.30123","doi":"10.48550/arXiv.2610.30123","title":"Temporal Exploration and Rendezvous via Weisfeiler-Leman Colorings","abstract":"Temporal graphs are graphs whose edge relation changes in discrete time steps. The Temporal Exploration Problem (TEXP) asks for a temporal walk that visits all vertices. We show that every connected temporal graph with $n$ vertices can be explored in $O(qn\\log^3(n))$ time steps, where $q$ is the number of vertex color classes induced by the 2-dimensional Weisfeiler-Leman (2-WL) coloring. For small $q$, our bound substantially improves on the known worst-case bound of $Θ(n^2)$ in general connected temporal graphs. A corresponding walk can be constructed in deterministic polynomial time. The upper bound is complemented by lower bounds of $Ω(n\\log(n))$ when $q$ is constant and $Ω(qn)$ when $1 \\leq q \\leq n$. We also apply our techniques to the Temporal Rendezvous Problem (TRP), in which two agents that do not communicate have to meet on a vertex of the temporal graph. Our exploration algorithm yields that, with only polynomial-time computational power, the agents can meet in $O(n\\log^3(n))$ time steps, leaving a gap of only $\\log^2(n)$ to the known lower bound. Dogeas, Erlebach, Kammer, Meintrup, and Moses Jr. [ICALP'24] have shown an $O(rn^{1+ε})$ bound for TEXP, for every fixed $ε> 0$, where $r$ is the number of orbits, and an $O(n^{1+ε})$ bound for TRP. Both results require superpolynomial computation time. Our contribution improves upon these results in several ways. First, the dependence on $n$ in the upper bounds on the number of time steps is decreased from $n^{1+ε}$ to $n\\log^3(n)$ for both TEXP and TRP. Second, for TEXP, we bound the number of time steps linearly by the parameter q, which is efficiently computable and upper bounded by $r$. Third, we provide novel deterministic polynomial-time algorithms for both TEXP and TRP.","authors":[{"name":"Frank Kammer","position":0},{"name":"Sandra Kiefer","position":1},{"name":"Johannes Meintrup","position":2}],"categories":["cs.DS","cs.CC","cs.DM","math.CO"],"primaryCategory":"cs.DS","publishedAt":"2026-10-21T17:57:42Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30123"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30123"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.30123.jpg","label":"thumbnail"}],"brief":{"text":"Using 2-WL color classes, the study gives a polynomial-time algorithm to explore connected temporal graphs in O(qn log³ n) steps, improving bounds when q is small. It also enables two noncommunicating agents to meet in O(n log³ n) steps.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":33,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.30122","arxivId":"2610.30122","doi":"10.48550/arXiv.2610.30122","title":"A Quasipolynomial Algorithm for $\\ell_\\infty$ Contractions","abstract":"We give an algorithm for computing an $\\eps$-approximate fixed point of any $d$-dimensional nonexpansion with respect to the $\\ell_\\infty$ norm in time $\\log(1/\\eps)^{O(\\log^2 d)}$. At its core, our method uses a boosting argument to geometrically shrink a search region. Among other consequences, our algorithm implies that simple stochastic games can be solved in quasipolynomial time.","authors":[{"name":"Ashkan Soleymani","position":0},{"name":"Gabriele Farina","position":1}],"categories":["cs.GT","cs.CC","cs.LG","math.OC"],"primaryCategory":"cs.GT","publishedAt":"2026-10-21T17:57:19Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30122"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30122"}],"brief":{"text":"An algorithm finds an approximate fixed point of any ℓ∞ nonexpansion in time quasipolynomial in dimension and accuracy, and also yields a quasipolynomial-time method for solving simple stochastic games.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":34,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.30119","arxivId":"2610.30119","doi":"10.48550/arXiv.2610.30119","title":"Self-normalized Bernstein Inequalities for Regenerative Processes and Applications to Learning Without Independence","abstract":"We prove a self-normalized Bernstein inequality for bounded functions of a regenerative process with independent, identically distributed marked cycles of finite mean length. The bound uses observed cycle lengths and retains explicit terms for truncation and the cycle ending after the observation horizon. Selecting a tuning parameter from a finite deterministic collection gives a version of Bernstein inequality where all terms are empirical but one, related to the tails of the regeneration times. We then study this term in many regimes: it turns out that it is of second-order in many cases of interest. This is important for applications in machine learning: we prove generalization error bounds for classification with non-i.i.d., regenerative data. These bounds can be used to define learning objectives: by minimizing the first-order term only in the bound, we define agnostic classifiers that still satisfy oracle inequalities. We also show that the bound on the average deviation provided by our results is minimax-optimal in the sample size and in the expected regeneration time. Finally, we combine the Bernstein inequality with a regenerative strong approximation to obtain a Gaussian approximation using an observable block variance estimator.","authors":[{"name":"Imon Banerjee","position":0},{"name":"Pierre Alquier","position":1}],"categories":["cs.LG","math.PR","math.ST","stat.TH"],"primaryCategory":"cs.LG","publishedAt":"2026-10-21T17:56:37Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30119"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30119"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.30119.jpg","label":"thumbnail"}],"brief":{"text":"A self-normalized Bernstein inequality for regenerative data uses observed cycle lengths and accounts for truncation and unfinished cycles. It yields minimax-optimal deviation bounds and supports classification generalization guarantees and Gaussian approximation with an observable variance estimate.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":35,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.30118","arxivId":"2610.30118","doi":"10.48550/arXiv.2610.30118","title":"The role of locality and weight sharing in convolutional neural networks","abstract":"The success of convolutional neural networks is often attributed to locality and weight sharing. However, how these two properties separately shape learning, and how their effects depend on the spatial correlations of natural images, remain poorly understood. We address these questions in a teacher-student model with a convolutional teacher acting on $P$ input positions with correlation $c$ between positions. We compare convolutional and locally connected students trained by online stochastic gradient descent, and derive ordinary differential equations that describe their dynamics exactly in the limit of large feature dimension $D$. For uncorrelated positions ($c = 0$), the two students follow the same learning trajectory, with the convolutional student learning exactly $P$ times faster. We also find that locality affects the sample complexity: a student that matches the kernel size of the teacher recovers it from $\\mathcal{O}_{D}(D)$ samples, whereas $\\mathcal{O}_{D}(D\\log(D))$ samples are needed with a larger kernel. For $c > 0$, spatial correlations can reduce the advantage of weight sharing. The asymptotic error of the convolutional student grows with $c$, whereas that of the locally connected student is independent of $c$ and overall lower when evaluated at the same learning rate. When the learning rate is instead tuned to the training horizon, however, the locally connected student wins at early times and the convolutional student at late times, with a transition that we predict analytically.","authors":[{"name":"Francesco Mori","position":0},{"name":"Melanie Weber","position":1}],"categories":["cs.LG","cond-mat.dis-nn","cond-mat.stat-mech","stat.ML"],"primaryCategory":"cs.LG","publishedAt":"2026-10-21T17:56:22Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30118"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30118"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.30118.jpg","label":"thumbnail"}],"brief":{"text":"A teacher–student model disentangles locality and weight sharing in convolutional learning. Matching the teacher’s kernel reduces sample needs, while spatial correlations can weaken weight sharing’s advantage: locally connected students do better early, but convolutional students win later with tuned learning rates.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":36,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.30109","arxivId":"2610.30109","doi":"10.48550/arXiv.2610.30109","title":"Minimax Optimality of the NPMLE for Gaussian Mixtures in Hellinger Distance","abstract":"We analyze the problem of estimating univariate Gaussian mixture densities using the nonparametric maximum likelihood estimator (NPMLE) based on $n$ independent and identically distributed samples. We show that the unconstrained NPMLE achieves the minimax optimal rate $\\log(n)/n$ in squared Hellinger distance, when the true mixing measure is sub-Gaussian. In addition, we show that when the true mixing measure has compact support, the NMPLE constrained to a compact interval also achieves the minimax optimal rate, which is slightly faster and of order $\\log(n)/(n \\log \\log n)$. Our proof consists of sharply characterizing the local Hellinger bracketing entropy of the class of finite Gaussian mixtures. One of our key technical innovations is to sharply bound the $L^2$ bracketing entropy of balls of bounded-degree polynomials, which may be of independent interest. As an application, we establish the regret-optimality of $g$-modeling procedures based on the NPMLE in the Gaussian Empirical Bayes problem with bounded priors.","authors":[{"name":"Shayan Hundrieser","position":0},{"name":"Tudor Manole","position":1}],"categories":["math.ST","cs.IT","cs.LG","math.IT","stat.TH"],"primaryCategory":"math.ST","publishedAt":"2026-10-21T17:53:10Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30109"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30109"}],"brief":{"text":"Gaussian mixture NPMLEs achieve minimax-optimal squared Hellinger error rates: log(n)/n for sub-Gaussian mixing measures, and log(n)/(n log log n) with compact support. The results also establish regret optimality for Gaussian Empirical Bayes procedures with bounded priors.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":37,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.30102","arxivId":"2610.30102","doi":"10.48550/arXiv.2610.30102","title":"Implementation of Dynamic Microsegmentation for a Multi-Robot Network","abstract":"This report describes a laboratory instantiation of a dynamic microsegmentation function for an Internet-of-Things network, specifically a multi-robot network, using a commercial software-defined microsegmentation tool. Fine-grained dynamic segmentation policies are defined, where segments are adapted based on asset functional and health tags. Automated implementation of the dynamic policies in software is also undertaken. The microsegmentation solution is shown to automatically isolate an indicated threat in real time, while maintaining unaffected operations. The demonstration indicates that microsegmentation can be implemented in IoT networks, in a way that allows for robust operation in the face of threats and disruptions.","authors":[{"name":"Vishwam Raval","position":0},{"name":"Jaewon Kim","position":1},{"name":"Ravi Pattabhi","position":2},{"name":"Sandip Roy","position":3}],"categories":["eess.SY","cs.MA","cs.NI","cs.RO","cs.SY"],"primaryCategory":"eess.SY","publishedAt":"2026-10-21T17:50:18Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30102"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30102"}],"brief":{"text":"A laboratory multi-robot network uses dynamic microsegmentation policies that adapt to robots’ functions and health status. The system automatically isolates a detected threat in real time while unaffected operations continue.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":38,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.29909","arxivId":"2610.29909","doi":"10.48550/arXiv.2610.29909","title":"Tractable Inference with Kolmogorov-Arnold Networks","abstract":"Machine-learning surrogates in computational science often require not only accurate predictions but also expectations, variances, conserved integrals, and normalization constants. Conventional neural networks, including Kolmogorov--Arnold networks (KANs), generally evaluate these quantities through Monte Carlo sampling or multidimensional quadrature, making repeated inference costly, particularly in high dimensions. We introduce KANs with Tractable Operators for Uncertainty and Constraints (TOUCKANs), whose finite-degree polynomial outer functions yield a decomposable sum--product representation. Under factorizing input measures, expectations and higher moments then reduce algebraically to deterministic recursions over one-dimensional leaf integrals. Independent and shared inner-block variants form a parameter-efficiency hierarchy, and a universal-approximation proof for its most restricted member extends to the complete class. The same deterministic operators allow conserved integrals and normalization to be imposed throughout training. Across uniform and Gaussian benchmarks through 50 dimensions, deterministic mean and variance errors remained below the median fixed-budget Monte Carlo errors in every case. TOUCKANs also conserved Cahn--Hilliard mass to machine precision and imposed normalization and orthogonality through deterministic moments, enabling joint computation of the lowest four Schrödinger eigenpairs in two, three, and five dimensions. TOUCKANs therefore connect KAN approximation with tractable probabilistic circuits, making statistical operators and integral constraints directly available within an expressive neural surrogate.","authors":[{"name":"Juan Diego Toscano","position":0},{"name":"Sascha Ranft","position":1},{"name":"Zhiwei Gao","position":2},{"name":"Li-Lian Wang","position":3},{"name":"George Em Karniadakis","position":4}],"categories":["physics.comp-ph","cs.LG","cs.NA","math.NA"],"primaryCategory":"physics.comp-ph","publishedAt":"2026-10-21T15:49:10Z","version":1,"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29909"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29909"}],"brief":{"text":"TOUCKANs use polynomial KANs to compute expectations, variances, and integral constraints through deterministic recursions. Across benchmarks up to 50 dimensions, their mean and variance errors beat median fixed-budget Monte Carlo errors, while conserved mass reached machine precision.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":39,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.29835","arxivId":"2610.29835","doi":"10.48550/arXiv.2610.29835","title":"Linear Complementary Pairs of MDS Codes from Elliptic Curves","abstract":"In this paper, we investigate the Linear Complementary Pairs (LCPs) of Maximum Distance Separable (MDS) codes that arise from elliptic curves. We establish geometric conditions for LCPs of elliptic codes and prove that they cannot consist entirely of MDS codes when evaluated over all rational points of the curve. To address this, we construct LCPs of MDS elliptic codes by restricting evaluation divisors to subgroup cosets. We apply this method to obtain explicit infinite families of LCPs of MDS codes over Mordell and supersingular elliptic curves, including large-length constructions over quadratic extensions.","authors":[{"name":"Adler Marques","position":0}],"categories":["cs.IT","math.AG","math.IT","math.NT"],"primaryCategory":"cs.IT","publishedAt":"2026-10-21T14:57:26Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29835"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29835"}],"brief":{"text":"Elliptic-curve codes form linear complementary pairs of MDS codes when evaluation divisors are restricted to subgroup cosets; the work gives explicit infinite families, including long codes over quadratic extensions.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":40,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.29812","arxivId":"2610.29812","doi":"10.48550/arXiv.2610.29812","title":"Real PMIDs, Wrong Papers: Harness Design for Biomedical Claim-Verification Agents","abstract":"Literature-searching language agents are becoming a standard component of scientific assistants for the life sciences, yet there is little controlled evidence on which parts of their harness matter. We run a controlled ablation on biomedical claim verification (SciFact) with the live PubMed API as the only tool. Holding the backbone and output format fixed, we compare four harnesses: closed-book answering, single-shot retrieval using the claim as the query, a ReAct-style agent that writes its own PubMed queries, and an oracle given the gold abstracts, each with two efficient frontier backbones. Naive single-shot retrieval finds the gold paper for only 6% of claims and turns overconfident errors into abstentions without improving accuracy. Letting the model formulate its own queries triples gold recall and yields the best verdicts (62% accuracy vs. 49-59% for the other realistic harnesses), although these differences are not significant at our sample size and agents use fewer than half of their search budget. With gold abstracts, the same models reach 79-86% accuracy, a significant 17-24 point gap showing that retrieval, not reasoning, is the main bottleneck. Finally, a citation audit shows that closed-book models cite PubMed IDs that almost always exist (327/335) but are almost never relevant: a cross-family LLM judge rates only 3 of 327 existing citations as relevant to the claim, versus 67% for citations produced by the search agent. Existence checks are therefore insufficient safeguards for scientific agents. We release code, prompts, and all cached tool calls at https://github.com/saqlain2204/real-pmids-wrong-papers.","authors":[{"name":"Mohammed Saqlain","position":0}],"categories":["cs.AI","cs.CL","cs.IR","cs.LG"],"primaryCategory":"cs.AI","publishedAt":"2026-10-21T14:35:56Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29812"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29812"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.29812.jpg","label":"thumbnail"}],"brief":{"text":"A controlled PubMed study finds that agents writing their own search queries retrieve more relevant papers and achieve the best claim-verification accuracy, while access to gold abstracts boosts accuracy further, identifying retrieval as the main bottleneck. Most PubMed IDs cited by closed-book models exist but point to irrelevant papers.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":41,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.29798","arxivId":"2610.29798","doi":"10.48550/arXiv.2610.29798","title":"Asymmetric Likelihood-Free Hypothesis Testing","abstract":"In likelihood-free hypothesis testing (LFHT), given samples from two unknown distributions $P,Q$ on $[n]$ with $\\mathrm{d_{TV}}(P,Q)\\geq\\varepsilon$, and from a third distribution $D\\in\\{P,Q\\}$, the goal is to identify whether $D=P$ or $D=Q$. We study asymmetric likelihood-free hypothesis testing, where the sample budgets $s_P$ and $s_Q$ for $P$ and $Q$ may differ, and characterize the feasible triples $(s_P,s_Q,s_D)$ up to logarithmic factors. We establish that unequal reference budgets can give a polynomial saving. For instance, for constant separation, $n$ samples from $P$ and $n^{3/4}$ from $Q$ allow testing with $\\widetilde O(n^{1/4})$ samples from $D$, whereas both existing baselines---symmetric LFHT and unequal-sample closeness testing---require $Ω(\\sqrt n)$ samples from $D$. We further apply our results to non-adaptive distribution 2-clustering, where the sample allocation among $k$ distributions must be fixed before observing any data. We obtain matching upper and lower bounds for this problem up to logarithmic factors in terms of the domain size, number of distributions, smaller cluster size, and separation, both with one cluster distribution known and with both unknown. Comparing with known bounds for adaptive sample allocation gives a worst-case adaptivity gap of $\\widetildeΘ(k^{1/4})$.","authors":[{"name":"Gunjan Kumar","position":0},{"name":"Yash Pote","position":1},{"name":"Jonathan Scarlett","position":2}],"categories":["cs.LG","cs.DS","cs.IT","math.IT","math.ST","stat.ML","stat.TH"],"primaryCategory":"cs.LG","publishedAt":"2026-10-21T14:26:53Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29798"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29798"}],"brief":{"text":"Asymmetric likelihood-free testing characterizes how unequal sample budgets affect the cost of identifying an unknown distribution, and shows that unequal reference samples can reduce the required test samples from order √n to about n¹ᐟ⁴. The results also give matching bounds for non-adaptive distribution clustering and show a worst-case adaptivity gap of about k¹ᐟ⁴.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":42,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.29785","arxivId":"2610.29785","doi":"10.48550/arXiv.2610.29785","title":"Where Do LLMs Learn to Recommend AI?","abstract":"Asked what to study, which career to pursue, or what product to build, large language models (LLMs) often recommend AI-related options. We ask where this tendency comes from by comparing models before and after fine-tuning and by varying how questions are asked. First, pretrained models already include explicit AI/ML terms in the recommendations of 27% of their responses, and their released instruction-tuned versions do so in 59%. Second, fine-tuning models ourselves shows that this rise is not inevitable: general instruction tuning leaves the rate unchanged, but removing the 7% of instruction examples that concern AI cuts it by 14 percentage points, below the level of the pretrained models it starts from. Data rewarding helpfulness raise it by 9 points, while data ranked by the preferences of diverse simulated users lower it by 9. Finally, with models held fixed, asking for the \"best\" option or for what to do \"right now\" raises the rate by 8 and 18 points, whereas telling the model it is an AI advisor adds 4. Overall, the tendency starts in pretraining, and what models learn afterward, together with how a question is phrased, decides how strongly it appears.","authors":[{"name":"Benaya Trabelsi","position":0},{"name":"Jonathan Shaki","position":1},{"name":"Sarit Kraus","position":2}],"categories":["cs.CL","cs.AI","cs.CY","cs.LG"],"primaryCategory":"cs.CL","publishedAt":"2026-10-21T14:13:52Z","version":1,"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29785"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29785"}],"brief":{"text":"LLMs’ tendency to recommend AI begins in pretraining, but fine-tuning data and question wording can strengthen or weaken it; removing AI-related examples from instruction data reduces the tendency below its pretrained level.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":43,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.29720","arxivId":"2610.29720","doi":"10.48550/arXiv.2610.29720","title":"Bio-Inspired Auditory Processing: Coupling Parallel Resonate-and-Fire Dynamics with Ternary Recurrent Networks for Speech Recognition","abstract":"Spiking Neural Networks (SNNs) promise energy-efficient inference through sparse, event-driven computation, yet remain constrained on continuous, context-dependent tasks such as automatic speech recognition (ASR). A key bottleneck is the Leaky Integrate-and-Fire (LIF) neuron, whose membrane decay erases information across pauses and phonetic boundaries while it must both extract features and retain long-range context. Inspired by cortical separation of nonlinear filtering and somatic integration, we introduce the Spiking Ternary GRU (stGRU): a ternary-weight gated recurrent unit performs temporal filtering, while downstream Resonate-and-Fire (RF) neurons generate the binary spike stream driving the next block's event pathway. This makes binary event transmission a native training constraint rather than a separate deployment-time activation quantization step; together with ternary weights, learned projections avoid dense floating-point matrix MACs. Trained end-to-end on LibriSpeech-960, Small, Medium, and Large stGRUs reach 8.09%, 7.04%, and $6.44 \\pm 0.17\\%$ WER on test-clean without an external language model. To our knowledge, these are the first matrix-MAC-free spiking networks without a learned ANN sequence decoder to achieve single-digit WER on LibriSpeech. The architecture also achieves 10.88% test CER on Aishell-1 (9.28% on dev) when trained from scratch. Internally, trained networks develop a depth-graded hierarchy: shallow layers act as fast acoustic filters, while deep layers form slower semantic integrators whose latent representations segregate into part-of-speech manifolds and word-level spike events. Deep-layer firing rates follow lognormal distributions with a sparsity gradient, mirroring hierarchical processing observed in human cortex in vivo. Our biologically grounded architecture combines competitive ASR with favorable modeled compute and memory cost.","authors":[{"name":"Pavlos Poulos","position":0},{"name":"Katelyn Sadorf","position":1},{"name":"Themos Stafylakis","position":2},{"name":"Costas A. Anastassiou","position":3}],"categories":["cs.SD","cs.AI","cs.NE","eess.AS","q-bio.NC"],"primaryCategory":"cs.SD","publishedAt":"2026-10-21T13:29:46Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29720"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29720"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.29720.jpg","label":"thumbnail"}],"brief":{"text":"A spiking network combines ternary recurrent units with resonate-and-fire neurons to recognize speech using binary event signals. It achieves single-digit word error rates on LibriSpeech without an external language model, while developing layered representations from acoustic features to word-level events.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":44,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.29696","arxivId":"2610.29696","doi":"10.48550/arXiv.2610.29696","title":"Quantum f-divergences from retrodiction: a unifying classical variational principle","abstract":"The same classical f-divergence admits inequivalent extensions to quantum states. The question is whether these extensions can be obtained from a common operational variational principle. Here we generalize Matsumoto's reverse-test optimization by introducing a reference prior state. More explicitly, we minimize the statistical distinguishability between a forward and a reverse prepare-and-measure process, related by retrodiction through the chosen reference and subject to prescribed quantum averages. Distinguishability is quantified by the classical f-divergence between their complete records. For faithful states in finite dimension, we prove that the same spectral implementation attains the minimum simultaneously for every operator-convex f. Varying the reference allows us to interpolate between standard and maximal quantum f-divergences of the prescribed states, passing through the geodesic family. More precisely, a flat reference yields the standard quantum f-divergences, thereby giving the Nussbaum-Szkoła distributions a variational characterization as optimal records. Choosing either prescribed state as reference recovers maximal quantum f-divergences and Matsumoto's reverse tests, while power references realize the geodesic interpolation. In the relative-entropy case, this gives the passage from Umegaki to Belavkin-Staszewski relative entropy by changing the prior for retrodiction.","authors":[{"name":"Francesco Buscemi","position":0}],"categories":["quant-ph","cs.IT","math-ph","math.IT","math.MP"],"primaryCategory":"quant-ph","publishedAt":"2026-10-21T13:09:11Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29696"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29696"}],"brief":{"text":"A reference-prior variational principle unifies quantum f-divergences through retrodiction: changing the prior recovers standard and maximal divergences and interpolates between them, including Umegaki and Belavkin–Staszewski relative entropy.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":45,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.29669","arxivId":"2610.29669","doi":"10.48550/arXiv.2610.29669","title":"Conflicted Records: Typed Adversarial Contracts for Auditable LLM Output Quality Enforcement","abstract":"LLM-generated records increasingly populate production information systems where they are treated as data assets subject to the same quality expectations as records from operational databases. A single such record can simultaneously satisfy and violate a quality rule: a grounded fact adjacent to a hallucination, a useful answer adjacent to a stereotype. Classical data cleaning cannot resolve this because the two judgments are incompatible over the same span; a single LLM judge cannot either, since it shares the training biases that produced the conflict. Existing interventions such as prompt engineering, chain-of-thought, and RLHF reduce surface error rates but provide no audit trail and cannot distinguish genuine remediation from blanket suppression. The question is whether a deterministic check over structured inter-agent records can catch the defects that referential checks alone miss. We present BiasGuard-D: two adversarial roles drawn from distinct model families exchange typed claim documents, a three-model quorum adjudicates by majority vote, and a combined referential-plus-semantic invariant validates the verdict. The semantic component fires on verdicts referential checks miss; removing the iterative loop drops invariant satisfaction sharply. The deliberation layer transfers from stereotype to hallucination remediation on HaluEval QA with only role-framing changes, raising right-answer recovery by 49 percentage points; the triage gate required a cross-family second auditor to maintain recall on objective hallucination. Every claim, vote, and instruction is stored as a typed record giving operators a provenance trail auditable under emerging AI regulations.","authors":[{"name":"Kardelen Subasi","position":0},{"name":"Jarosław A. Chudziak","position":1}],"categories":["cs.DB","cs.AI","cs.CL","cs.MA"],"primaryCategory":"cs.DB","publishedAt":"2026-10-21T12:43:35Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29669"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29669"}],"brief":{"text":"BiasGuard-D uses typed records and cross-model adversarial review to catch quality defects that referential checks miss, with a stored audit trail. On HaluEval QA, it raised right-answer recovery by 49 percentage points with role-framing changes alone.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":46,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.29641","arxivId":"2610.29641","doi":"10.48550/arXiv.2610.29641","title":"Recursive Code Evolution for Task-Oriented Semantic Communications","abstract":"Semantic communication jointly optimizes inference accuracy, channel use, and computation, yet the underlying communication program is typically specified before training. This paper presents ComRSI (Communications-Oriented Recursive Self-Improvement), a framework for recursively evolving that program through large-language-model-assisted code edits. A fixed agent modifies the encoder, channel adapters, decoder, loss, and training policy; a protected evaluator screens candidates under prescribed budgets and resource constraints. Accepted code and textual feedback inform subsequent proposals, while candidate weights are trained from scratch. In the original ten-generation CIFAR-100 lineage, eight accepted revisions produce an attention-pooled code achieving 74.3% test accuracy over an Additive White Gaussian Noise (AWGN) channel at 32 complex symbols per image, compared with 67.2% at 256 symbols for the initial design. The program also improves fading-channel accuracy and transfers to Tiny-ImageNet after retraining. A repeat study with ten searches per method compares ComRSI against non-recursive agents with and without textual history. Mean final search scores balancing accuracy, transmission rate, and computational cost are 0.7587, 0.7376, and 0.7089, respectively; the mean numbers of accepted revisions per search are 6.0, 5.0, and 2.1. The advantage over the memoryless agent is supported by a two-sided rank test, whereas the difference from the history-equipped agent remains statistically inconclusive at this sample size. Archived proposals show that pooling can yield a large improvement, while retained history supports subsequent refinement. These results support recursive program optimization within the evaluated setting and expose its remaining rate-robustness and compute-placement trade-offs.","authors":[{"name":"Yan Gong","position":0},{"name":"Zheng Chu","position":1},{"name":"Wei Koong Chai","position":2},{"name":"Hari Mohan Pandey","position":3},{"name":"Jingrui Hou","position":4},{"name":"Miao Zhang","position":5},{"name":"Yuanjian Li","position":6},{"name":"Pei Xiao","position":7}],"categories":["cs.IT","cs.LG","eess.SP","math.IT"],"primaryCategory":"cs.IT","publishedAt":"2026-10-21T12:23:33Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29641"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29641"}],"brief":{"text":"ComRSI recursively evolves semantic-communication programs through language-model code edits, improving accuracy while reducing channel use in the tested setting. Its search outperforms memoryless agents, though its advantage over agents with textual history remains inconclusive.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":47,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.29512","arxivId":"2610.29512","doi":"10.48550/arXiv.2610.29512","title":"Dissipation Bounds and Hardness in Globally Asymptotically Stable Quadratic Systems","abstract":"We study worst-case finite-time energy dissipation in quadratic systems with proved global asymptotic stability (GAS). Estimating the maximum retained energy fraction to any fixed positive additive error below $1/2$ is NP-hard, even when the input includes a strict quartic Lyapunov function and proof of strictness, both constructed in polynomial time. NP denotes nondeterministic polynomial time. For a partially damped family, the unit-sphere minimum norm of the coupling map equals the structured coefficient distance to loss of GAS. Positive distances at most $1/2$ give two-sided inverse-square halving-time bounds with explicit coefficient factors. Polynomial-factor approximation of the least dissipated fraction, halving time or coefficient distance is NP-hard. Additive hardness persists throughout an explicit coefficient neighborhood and under direct damping with exponential stability, a strict quadratic Lyapunov function and a Hurwitz Jacobian. Unless $\\mathrm{P}=\\mathrm{NP}$, where P denotes deterministic polynomial time, no polynomial-time algorithm attains these guarantees on all inputs.","authors":[{"name":"Subhransu S. Bhattacharjee","position":0}],"categories":["eess.SY","cs.CC","cs.SY","math.OC"],"primaryCategory":"eess.SY","publishedAt":"2026-10-21T10:01:45Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29512"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29512"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.29512.jpg","label":"thumbnail"}],"brief":{"text":"Worst-case energy dissipation in globally stable quadratic systems is NP-hard to estimate, even with a strict quartic Lyapunov proof supplied. For partially damped systems, coupling strength determines stability margins and halving-time bounds, but key quantities remain hard to approximate.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":48,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.29510","arxivId":"2610.29510","doi":"10.48550/arXiv.2610.29510","title":"A unified approach for estimation of unnormalized models via Stein operators","abstract":"In this paper, we develop a unified framework connecting proper scoring rules and Stein's method for the estimation of unnormalized models. We construct a class of proper scoring rules that encompasses score matching, ratio matching, pseudolikelihood, and noise contrastive estimation. We show that these proper scoring rules naturally induce Stein operators. This connection yields a new construction principle for Stein operators, and enables us to study estimators based on these scoring rules through the geometry associated with the induced Stein operators. Using this geometric perspective, we provide a unified characterization of the asymptotic efficiency of the resulting estimators.","authors":[{"name":"Mitsuki Nagai","position":0},{"name":"Keisuke Yano","position":1}],"categories":["stat.ML","cs.LG","math.ST","stat.TH"],"primaryCategory":"stat.ML","publishedAt":"2026-10-21T10:01:25Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29510"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29510"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.29510.jpg","label":"thumbnail"}],"brief":{"text":"A unified framework links proper scoring rules for unnormalized models to Stein operators, encompassing several existing estimation methods and enabling a geometric characterization of their asymptotic efficiency.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":49,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.29475","arxivId":"2610.29475","doi":"10.48550/arXiv.2610.29475","title":"Density-Level Backward Error for Low-NFE Generative Flows","abstract":"Low-NFE generative ODE sampling is constrained not only by solver order, but also by how a fixed budget of network evaluations is distributed along the flow. We develop a density-level backward-error framework that identifies the component of a solver's modified-equation defect that changes the evolving marginal, while excluding solenoidal trajectory distortion that is invisible in Wasserstein tangent geometry. The resulting residual induces a solver-order-aware compute-allocation rule and, through gauge freedom, an objective for selecting numerically favorable velocity representatives of the same density path. This separates our approach from schedule search driven by trajectory or endpoint surrogates and from higher-order solver design. Exact Gaussian and mixture models establish the mechanism, including cases where curvature-based allocation fails and continuum-optimal transport is not discretization-optimal. On a pretrained CIFAR-10 EDM probability-flow ODE, density-aware allocation consistently improves fixed-update Euler sampling in the low-NFE regime. Matched-seed convergence to a dense Euler reference confirms that the method accelerates approximation of the same high-NFE flow rather than changing the asymptotic sampler.","authors":[{"name":"Ewsey Obzherin","position":0},{"name":"Valentin Khrulkov","position":1}],"categories":["cs.LG","cs.NA","math.NA","stat.ML"],"primaryCategory":"cs.LG","publishedAt":"2026-10-21T09:30:21Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29475"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29475"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.29475.jpg","label":"thumbnail"}],"brief":{"text":"A density-level error framework allocates network evaluations to reduce changes in the evolving distribution and guides selection of velocity fields. On pretrained CIFAR-10 flows, it improves low-budget Euler sampling while converging toward the same high-evaluation flow.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":50,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.29439","arxivId":"2610.29439","doi":"10.48550/arXiv.2610.29439","title":"How Far Can Foundation Models Be Distilled for Efficient Jet Tagging?","abstract":"Jet tagging is one of the most important applications of machine learning at the Large Hadron Collider. Recent models build in more physics knowledge or rely on large-scale pretraining, but the experiments place strict limits on their inference cost. Knowledge distillation can transfer their performance to small students that are cheap to run. We ask which properties of a teacher help the student, how small a student can get and how close it can come to the state of the art. On a jet tagging benchmark, we distil six teachers that differ in architecture, symmetry and pretraining, including two foundation models, into five small students. Every teacher improves every student, and the gain depends mainly on the strength of the teacher. We then shrink our two strongest distilled students, Transformer and L-GATr-slim, in width and in depth. With 20 thousand parameters, a distilled L-GATr-slim outperforms every published jet tagger trained from scratch at a fraction of the inference cost of every teacher. Even with two thousand parameters, it matches ParT, a model with a thousand times as many parameters. We also study what the students learn from their teachers. Our benchmark of inference cost shows that students of the same size can differ by almost an order of magnitude in speed. We release the code and the weights of all trained models.","authors":[{"name":"Jay Agarwal","position":0},{"name":"Siddharth Khare","position":1},{"name":"Dhruv Kumar","position":2}],"categories":["cs.LG","hep-ex","hep-ph","physics.data-an"],"primaryCategory":"cs.LG","publishedAt":"2026-10-21T09:00:56Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29439"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29439"}],"brief":{"text":"Distilling six jet-tagging teachers into small models shows that stronger teachers generally give larger gains. A 20,000-parameter student beats published taggers trained from scratch, while a 2,000-parameter student matches ParT with a thousand times more parameters.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":51,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.29351","arxivId":"2610.29351","doi":"10.48550/arXiv.2610.29351","title":"Closed-Loop Dual-Reference Guidance for Accent Control in Cross-Lingual Voice Cloning","abstract":"Zero-shot cross-lingual voice cloning often carries the reference speaker's accent into the target language. We propose closed-loop dual-reference guidance, which pairs the original reference with a reusable native anchor constructed by voice conversion (VC). The references initialize separate language-model (LM) states that guide a shared acoustic trajectory; each completed patch then updates both states, keeping their generated histories aligned with the emitted speech. Anchor construction runs once per speaker, and online synthesis uses a frozen TTS model without per-utterance VC. On 20 held-out French speakers, the fixed setting $λ^{\\!*}=0.33$ reduces residual French LID mass by 73\\% versus direct cloning, at an ECAPA similarity cost of only $0.042$; relative to anchor-only cloning, it recovers $0.155$. The method achieves the lowest mean WER ($0.009$) and highest mean GOP ($4.750$) among the held-out test arms. In blinded preference tests, native-English listeners favor it over direct cloning on pronunciation naturalness, and over both anchor-only cloning and a TTS$\\to$VC cascade on speaker identity. Online RTF is approximately 26\\% lower than the cascade's ($0.59$ vs.\\ $0.80$), and the same setting reduces Spanish LID mass by 56\\% without re-tuning.","authors":[{"name":"Yuanzhong Zheng","position":0},{"name":"Wei Huang","position":1},{"name":"Ming Ma","position":2},{"name":"Kailun Jin","position":3},{"name":"Youhai Jiang","position":4}],"categories":["cs.SD","cs.CL","cs.LG","eess.AS"],"primaryCategory":"cs.SD","publishedAt":"2026-10-21T07:46:22Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29351"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29351"}],"brief":{"text":"Closed-loop dual-reference guidance combines a speaker’s voice with a reusable native-language anchor to reduce accent transfer in cross-lingual voice cloning. On held-out French speakers, it cut residual French accent evidence by 73% versus direct cloning and improved pronunciation and speaker identity in listener tests.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":52,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.29345","arxivId":"2610.29345","doi":"10.48550/arXiv.2610.29345","title":"Decodability Is Not Controllability: Local and Anisotropic Speaker Control in Continuous-Latent TTS","abstract":"High linear decodability does not entail causal controllability in continuous-latent text-to-speech (TTS). In VoxCPM2, position-matched probes and equal-norm interventions across 30 speakers and 60 directed dyads expose this separation: reference-audio states at the final residual acoustic language model layer (RALM L7) retain 0.970 probe accuracy, yet aligned edits yield 0/60 target-dominant dyads. Edits at early reference-audio positions produce the opposite pattern. At L0, pair-aligned updates increase target similarity by 0.326 and achieve target dominance in all 60 dyads, against at most 1/60 under any equal-norm direction or position control. Independent speaker encoders, retrieval against held-out speaker prototypes, and double-blind listening confirm that the shifted voice is heard as the intended target. Cross-architecture interventions in dots.tts distinguish backbone leverage from direct acoustic control: backbone edits displace the source without target attraction, while x-vector ablation reduces source similarity from 0.849 to 0.006. These findings identify two requirements for training-free speaker control: geometric specificity and a computational route that carries the intended speaker condition into acoustic synthesis.","authors":[{"name":"Yuanzhong Zheng","position":0},{"name":"Wei Huang","position":1},{"name":"Ming Ma","position":2},{"name":"Kailun Jin","position":3},{"name":"Youhai Jiang","position":4}],"categories":["cs.SD","cs.CL","cs.LG","eess.AS"],"primaryCategory":"cs.SD","publishedAt":"2026-10-21T07:38:25Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29345"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29345"}],"brief":{"text":"Speaker control in continuous-latent TTS depends on where and how states are edited: early, pair-aligned updates produced the intended voice in all 60 speaker pairs, while highly decodable late-layer states did not. Cross-architecture tests show that edits must also reach acoustic synthesis to control speaker identity.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":53,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.29327","arxivId":"2610.29327","doi":"10.48550/arXiv.2610.29327","title":"OACap: Caption-Guided Active Observation for Detailed Video Captioning","abstract":"Unlike question answering, where a query specifies a perceptual target, open-ended detailed captioning must determine what evidence to acquire and how to integrate it into a coherent global description. We propose Observation-Adaptive Video Captioner (OACap), a caption-centered agent that turns unsupported or missing information into targeted observation requests, uses spatiotemporal tools to acquire evidence, and iteratively verifies and revises the caption. OACap captions consistently improve performance across general video-understanding benchmarks using a fixed text-only reader. To better diagnose the information they preserve, we introduce Factual-Semantic Captioning Bench (FSCap-Bench). Its Core-Support Factuality Score (CSFS) combines claim precision with importance-aware fact coverage and responds as expected to four controlled perturbation types on 93.0-98.3% of valid caption pairs. Its Cross-Modal Semantic Round-Trip Score (CSRS) measures semantic preservation through a video-caption-video-caption round trip by matching atomic facts from source and recovered captions. Together, they connect downstream textual-memory utility with factual and semantic preservation.","authors":[{"name":"Hao Shi","position":0},{"name":"Hong Yan","position":1},{"name":"Xiaoyang Zhang","position":2},{"name":"Qing Cui","position":3},{"name":"Chao Huang","position":4}],"categories":["cs.CV","cs.CL","cs.MA","cs.MM"],"primaryCategory":"cs.CV","publishedAt":"2026-10-21T07:26:26Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29327"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29327"}],"brief":{"text":"OACap actively seeks and verifies missing video evidence, producing captions that consistently improve performance on general video-understanding benchmarks with a fixed text-only reader. FSCap-Bench adds measures of factual coverage and semantic preservation to evaluate what captions retain.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":54,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.29230","arxivId":"2610.29230","doi":"10.48550/arXiv.2610.29230","title":"PQ-Vault: Provably Quantum-Safe Asset Custody for Unmodified Ethereum","abstract":"A quantum computer could recover the key of any Ethereum account whose public key appears on chain. We prove that no sidechain or bridge that leaves Ethereum's rules unchanged can protect such accounts: their funds must move into a contract whose release condition relies only on quantum-safe primitives. We design \\emph{PQ-Vault}, a custody system for unmodified Ethereum combining this vault, a quantum-safe proof-of-stake sidechain whose checkpoints carry compact certificates, and an optimistic dispute protocol. We give the first security proof of compact certificates against quantum attackers and a matching attack: with classically chosen parameters, the guarantee falls from 128 to about 60 bits, and a simple rule restores it. We show that a common seed-update rule in hash-chain proof of stake lets an attacker steer chosen slots and profit from splitting its stake, and give an election whose steering gain and inclusion delay are provably bounded against quantum attackers. Since compressing certificates with a pairing-based SNARK breaks quantum security, we use fraud proofs that need a single signature check. For 65,536 validators, a certificate is 0.8 MB, 260 times smaller than all signatures, and verifies in about 20 ms; the vault accepts a checkpoint for about 420,000 gas.","authors":[{"name":"Hassan Khodaiemehr","position":0},{"name":"Khadijeh Bagheri","position":1},{"name":"Chen Feng","position":2}],"categories":["quant-ph","cs.CR","cs.DC","cs.IT","math.IT"],"primaryCategory":"quant-ph","publishedAt":"2026-10-21T05:39:46Z","version":1,"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29230"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29230"}],"brief":{"text":"PQ-Vault protects Ethereum assets with a quantum-safe contract vault and sidechain, while proving limits on protecting exposed accounts. Its certificate design restores security against quantum attackers, and a 65,536-validator checkpoint is 260 times smaller than all signatures and verifies in about 20 ms.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":55,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.29216","arxivId":"2610.29216","doi":"10.48550/arXiv.2610.29216","title":"Preference Is Not Permission: Personalizing the Auto-Purchase Line in Agentic Commerce","abstract":"Personalized shopping benchmarks ask which item fits the user; we ask how far an agent may commit on that item, and pose learning the individual's auto-purchase line as a task. The boundary is evidence-conditioned and three-valued: whether committing is appropriate depends on the sufficiency of product information (warrant), reversibility, price against the user's norm, and urgency, not only on fit; and the natural output is don't buy, confirm first, or buy automatically. Across sixteen instruction-tuned models, autonomous commitment is misaligned with a human authorization reference in which fifteen annotators, role-playing the delegating principal from a given profile, choose the furthest action they would allow without confirmation: models commit even when fit is low and when their own evidence flag reports the warrant missing. This is overreach. The line is individual: these raters disagree, yet the disagreement is structured and partly predictable from a person's own other labels. A shallow learner on a few dozen of a person's own labels predicts their held-out auto-purchase line at a median of 0.79 by 30-40 labels, outperforming every evaluated prompt$\\times$model configuration, including a per-annotator oracle over 80 configurations; the learned boundary does not transfer across domains. We release the counterfactual benchmark, human authorization reference, and annotation protocol after acceptance.","authors":[{"name":"Ryotaro Shimizu","position":0},{"name":"Sai HtaungKham","position":1},{"name":"Mo Fan","position":2}],"categories":["cs.IR","cs.CL","cs.HC","cs.LG"],"primaryCategory":"cs.IR","publishedAt":"2026-10-21T05:15:31Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29216"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29216"}],"brief":{"text":"Auto-purchase authorization depends on evidence, reversibility, price, and urgency—not just product fit. Across 16 models, agents often overcommit, while learning from 30–40 personal labels predicts an individual’s purchase boundary better than tested prompting strategies.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":56,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.29215","arxivId":"2610.29215","doi":"10.48550/arXiv.2610.29215","title":"Third and Fourth Moment Phenomena for Monochromatic Edges and Applications to Graph-Based Two-Sample Tests","abstract":"Given a graph sequence $\\{G_n\\}_{n\\in\\mathbb{N}}$, denote by $T_n$ the number of monochromatic edges when the vertices of $G_n$ are independently colored red or blue with probabilities $p$ and $1-p$, respectively. We characterize the asymptotic normality of $T_n$, after centering and scaling to unit variance, in terms of subgraph (motif) counts and standardized moments. One of the highlights of our results is a {\\it third moment phenomenon}: for every fixed $p\\in(0,1)\\backslash\\{\\frac12\\}$, asymptotic normality holds if and only if the standardized third moment converges to zero. The more familiar {\\it fourth moment condition}, which requires the standardized fourth moment to converge to $3$, gives an equivalent characterization except at the two exceptional probabilities $p=\\frac{1\\pm1/\\sqrt3}{2}$. At these values, the standardized fourth moment converges to $3$ for every graph sequence with a diverging number of edges, even when Gaussian convergence fails. In contrast, in the balanced case $p=\\frac12$, the fourth moment condition characterizes asymptotic normality, while the third moment condition is insufficient. These results identify three distinct regimes for moment characterizations of Gaussian convergence, determined by the coloring probability. We further show that the third moment condition remains sufficient in the sparse regime, where $p=p_n \\to 0$. As an application, we derive the limiting null distribution of two-sample statistics based on undirected $K$-nearest-neighbor graphs in the Poissonized setting, identifying sharp growth conditions on $K$ for asymptotic normality.","authors":[{"name":"Siva Athreya","position":0},{"name":"Bhaswar B. Bhattacharya","position":1},{"name":"Yogesh Dahiya","position":2},{"name":"Karambir Das","position":3},{"name":"Spandan Ghoshal","position":4},{"name":"Sourish Maniyar","position":5},{"name":"Elizabeth Sara Roy","position":6}],"categories":["math.ST","math.CO","math.PR","stat.TH"],"primaryCategory":"math.ST","publishedAt":"2026-10-21T05:15:17Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29215"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29215"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.29215.jpg","label":"thumbnail"}],"brief":{"text":"Monochromatic edge counts have different moment criteria for Gaussian convergence depending on the coloring probabilities: the third moment often suffices, while balanced or exceptional cases require different tests. The results also give sharp conditions for normality of two-sample statistics on undirected K-nearest-neighbor graphs.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":57,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.29211","arxivId":"2610.29211","doi":"10.48550/arXiv.2610.29211","title":"On the gradient conjecture at infinity: Finite spherical length for gradient trajectories of definable potentials","abstract":"We prove that for every open $ U \\subseteq \\mathbb R^n $ the spherical projection of every gradient trajectory $θ(t)$, $t \\geq 0$, of a definable $C^1$-potential $F \\colon U \\to \\mathbb R$ has finite length and, in particular, converges (directional convergence). In other words, every gradient trajectory of a definable $C^1$-potential has finite total angular motion. Here, definable means definable in an arbitrary o-minimal structure expanding the real field. Notably, the gradient trajectory in this result is an arbitrary solution of the gradient flow differential equation (whose solutions may not be unique) with no further assumptions: The gradient trajectory may be unbounded and diverge to infinity or it may be convergent. The established theorem therefore provides affirmative answers to both the gradient conjecture at infinity for definable potentials and the definable Thom gradient conjecture. We apply our theorem to the training of deep neural networks (DNNs) to conclude that, in the case of definable and continuously differentiable activation and loss functions, the spherical projection of the DNN training flow must always converge (directional convergence).","authors":[{"name":"Thang Do","position":0},{"name":"Arnulf Jentzen","position":1},{"name":"Benno Kuckuck","position":2}],"categories":["math.DS","cs.LG","math.AG","math.CA","math.LO"],"primaryCategory":"math.DS","publishedAt":"2026-10-21T05:06:29Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29211"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29211"}],"brief":{"text":"Gradient trajectories of definable \\(C^1\\) potentials have finite angular motion and converge in direction, even when solutions are nonunique or diverge to infinity. This also guarantees directional convergence of deep-network training flows with definable, continuously differentiable activation and loss functions.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":58,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.29199","arxivId":"2610.29199","doi":"10.48550/arXiv.2610.29199","title":"Motivation and Contribution in Open Source Software: A Meta-Analytic Investigation","abstract":"Research on why people contribute to open source software (OSS) and related open collaboration platforms spans more than two decades, but its findings conflict, and studies often treat the work people contribute and their intention to keep contributing as interchangeable. Drawing on self-determination theory and motivation crowding theory, we meta-analyze 41 studies. The main analysis pools 178 correlations of eight antecedents with contribution effort and continuance intention from 29 independent samples (N = 9,632). Seven antecedents relate positively to effort, all eight relate positively to intention, and each relates at least as strongly to intention as to effort. Contrary to a common assumption, intrinsic motivation is not more strongly related to effort than extrinsic motivation, but its gap between intention and effort is larger. Reputation relates to effort mainly on platforms with a reputation system, and we find no sign that monetary rewards crowd out enjoyment. The findings inform how platforms can turn contributors' intentions into work.","authors":[{"name":"Parisa Omidmand","position":0},{"name":"Saeed Nosratabadi","position":1}],"categories":["econ.GN","cs.HC","cs.SE","q-fin.EC"],"primaryCategory":"econ.GN","publishedAt":"2026-10-21T04:44:42Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29199"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29199"}],"brief":{"text":"A meta-analysis of 41 studies finds that eight motivations relate more strongly to intentions to keep contributing to open-source projects than to actual effort; reputation matters most on platforms with reputation systems, and monetary rewards do not reduce enjoyment.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":59,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.29052","arxivId":"2610.29052","doi":"10.48550/arXiv.2610.29052","title":"Our Own Jev: Typed Decisions in One Forward Pass, from an In-House Judge to Open LoRA Models","abstract":"A typed decision maps evidence, a question and a fixed set of options to a probability for every option, computed in one forward pass, by reading a language model's logits or a small head over its hidden states, instead of generating and parsing text. The commercial system Jev popularised this System~1 interface, and open reproductions have followed. We report two uses of the pattern. \\emph{Part I} describes an in-house System~1 judge that triages about 148K agentic coding session records for a training corpus. It replaces a frontier LLM judge that cost on the order of \\$1,000 per pass and whose stored labels matched a fresh two-judge consensus on only 26.9\\% of the sessions where the two judges agreed; a self-hosted frozen backbone with a rank-16 LoRA adapter and a pointer head answers three typed questions in one forward pass, judges 1,000 sessions in about three minutes of one GPU, about \\$0.06 at a reference rate and roughly 1\\% of the frontier judge's cost, and keeps every session inside the organisation. \\emph{Part II} asks whether a readout that adds no parameters can be competitive and whether training one helps. Applying the public Eikos recipe and corpus of about 25K decisions to three further backbones (1B, 2B and 35B), we read the frozen output layer of a LoRA-adapted decoder. On the 111 public hard items of JevBench our single-seed 35B, developed with those items in view, scores 79.3\\%, not significantly different from the commercial reference (73.0\\%, unadjusted $p=0.12$) and ahead of five open systems after Holm--Bonferroni correction; our 2B is ahead of all four kev checkpoints on the leaderboard. A single-seed ablation finds no detectable benefit from training the 26 letter rows of the output layer, and a retrain on a later corpus release moves the 2B's hard-tier accuracy by 11.7 points, which shows how little small evaluations resolve.","authors":[{"name":"Yuchuan Weng","position":0},{"name":"Chi Liu","position":1},{"name":"Xin Chen","position":2},{"name":"Srinivasan Manoharan","position":3}],"categories":["cs.CL","cs.AI","cs.IR","cs.LG"],"primaryCategory":"cs.CL","publishedAt":"2026-10-21T01:28:58Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29052"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29052"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.29052.jpg","label":"thumbnail"}],"brief":{"text":"A one-pass typed judge triages 148K coding sessions at about 1% of a frontier judge’s cost while keeping data in-house; open models using a frozen output layer also perform competitively on JevBench, though small-evaluation results vary.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":60,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.29007","arxivId":"2610.29007","doi":"10.48550/arXiv.2610.29007","title":"A Jump-Modified Wasserstein Gradient Flow for Maximum Mean Discrepancy with Energy Kernels","abstract":"We study a jump-modified Wasserstein gradient flow of the squared maximum mean discrepancy generated by the energy kernels $K(z)=-|z|^q$, $0<q<2$, in every dimension $d\\ge2$ as well as the one-dimensional range $1\\le q<2$. Under subcritical density and finite-moment assumptions, we prove global well-posedness. The added jumps transfer mass toward smaller values of the first variation and dissipate the same MMD energy as the transport term. Under our hypotheses on the target and jump rates, every Borel probability measure with finite MMD energy and zero jump dissipation is equal to the target. Combining this rigidity with compactness and a second-moment identity yields convergence to the target from finite-second-moment data, without additional uniform-in-time density or moment assumptions. We also construct a global collision-free particle process whose energy is nonincreasing along each path. A quantitative mean-field estimate applies to general initial empirical approximations, including deterministic configurations; propagation of chaos for exchangeable data and long-time target approximation follow as consequences.","authors":[{"name":"Ansh Desai","position":0},{"name":"Jason Liu","position":1},{"name":"Srujan Roplekar","position":2},{"name":"Matthew Rosenzweig","position":3}],"categories":["math.AP","cs.LG","math.PR","stat.ML"],"primaryCategory":"math.AP","publishedAt":"2026-10-21T00:16:11Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29007"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29007"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.29007.jpg","label":"thumbnail"}],"brief":{"text":"A jump-modified Wasserstein flow for energy-kernel maximum mean discrepancy is globally well posed and converges to its target under stated assumptions. The work also constructs collision-free particle dynamics with nonincreasing energy and proves mean-field and long-time approximation results.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":61,"reason":"new","score":0.2191},{"paper":{"id":"arxiv:2610.28999","arxivId":"2610.28999","doi":"10.48550/arXiv.2610.28999","title":"Optimal Rényi and Logarithmic Additive Functional Entropy Estimation","abstract":"We characterize the minimax sample complexity of estimating Rényi entropy from independent samples on an alphabet of size at most $k$, when the known order $α>1$ may vary with $k$. For every fixed additive accuracy $\\varepsilon\\in(0,1/2)$ and success probability $2/3$, we prove matching bounds with constants independent of $k$ and $α$. For $1<α<\\log k$, the rate is $Θ_\\varepsilon(αk^{1-1/α}X_α(k,\\varepsilon))$, where a monotone scalar inequality defines $X_α$ at noninteger orders. The bounds remain uniform as the order approaches one or an integer, or grows with the alphabet. Integer orders below $\\log k$ have rate $Θ_\\varepsilon(αk^{1-1/α})$; orders at least $\\log k$ and min-entropy have rate $Θ_\\varepsilon(k\\log k)$. We extend the analysis to logarithms of positive additive functionals. For every fixed continuous $f$ satisfying $f(0)=0$, $f(x)>0$ for $x>0$, and $f(x)\\sim ax^r$ at zero, with $a>0$ and $r>1$, estimating $-β^{-1}\\log\\sum_i f(p_i)$ at fixed accuracy, for fixed $β>0$, has sample complexity $Θ(k^{1-1/r})$ for integer $r$ and $Θ(k/\\log k)$ otherwise. A replication reduction preserves the sampling experiment and transfers the rates for powers. Smooth flat functions show that regularity alone does not determine the complexity. The upper bound combines polynomial approximation, Gamma count statistics, and an independent pilot, with a global affine correction near order one. The lower bound uses interpolation to construct positive mean-one priors and controls their distinguishability and entropy separation after normalization. Exact count-variance identities and weighted approximation duality connect the two arguments.","authors":[{"name":"Arman Adibi","position":0},{"name":"Piotr Krysta","position":1}],"categories":["cs.IT","cs.CC","eess.SP","math.IT","math.ST","stat.TH"],"primaryCategory":"cs.IT","publishedAt":"2026-10-20T23:57:35Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28999"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28999"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.28999.jpg","label":"thumbnail"}],"brief":{"text":"Rényi entropy estimation has matching minimax sample-complexity bounds across orders that may vary with alphabet size, including near one and integer orders. For logarithms of positive additive functionals, the rate depends on the function’s behavior near zero: integer power behavior gives a different rate than noninteger behavior.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":62,"reason":"new","score":0.1945},{"paper":{"id":"arxiv:2610.28978","arxivId":"2610.28978","doi":"10.48550/arXiv.2610.28978","title":"From Common Information to Canonical Correlation Analysis and Beyond: Bridging Information Theory, Signal Processing, and Machine Learning","abstract":"Many problems in signal processing, machine learning, and neuroimaging provide several jointly observed measurements of one process, each mixing a component common to all of them with a component private to itself, and require the common component to be recovered without supervision. This problem has been studied in several communities, largely independently. Statisticians proposed canonical correlation analysis (CCA), introduced by Hotelling in 1936. Information theorists proposed two notions of common information, due to Gács and Körner and to Wyner. Engineers developed subspace methods for sensor arrays and multiuser communications to design multiview receivers. This tutorial presents the main results and tools each community developed for this problem in a single exposition, and explains the connections that emerge where they meet, distinguishing exact equivalences from special cases and analogies. We develop the information-theoretic foundation from first principles. We then introduce CCA and its extension to several views, followed by the engineering thread from blind multiuser detection to CCA as a blind receiver. Next, we turn to nonlinear CCA, covering kernel, deep, and probabilistic formulations, the conditions under which the common factor is identifiable, and connections to self-supervised learning. We conclude with a discussion of further extensions and open directions. The exposition is self-contained, so that readers from each of these communities can follow the parts developed by the others.","authors":[{"name":"Paris A. Karakasis","position":0},{"name":"Walid Saad","position":1}],"categories":["cs.IT","cs.LG","eess.SP","math.IT","stat.ML"],"primaryCategory":"cs.IT","publishedAt":"2026-10-20T22:54:10Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28978"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28978"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.28978.jpg","label":"thumbnail"}],"brief":{"text":"A tutorial unifies information-theoretic, statistical, and engineering approaches to recovering shared signals from multiple observations, tracing connections to CCA, nonlinear methods, and self-supervised learning while distinguishing equivalences from analogies.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":63,"reason":"new","score":0.1945},{"paper":{"id":"arxiv:2610.28971","arxivId":"2610.28971","doi":"10.48550/arXiv.2610.28971","title":"Universality of approximate message passing under a fourth-moment tail condition","abstract":"We prove state evolution for approximate message passing (AMP) with rectangular random matrices whose entries are independent, centered, and variance-normalized. The row and column dimensions grow with a ratio that converges to a positive finite limit. The entry laws may differ across coordinates and dimensions, and we assume uniform integrability of the fourth powers of the rescaled entries. For entries sampled independently from a fixed law, a finite fourth moment suffices, with no requirement for higher moments. The nonlinearities are continuously differentiable and globally Lipschitz, with bounded continuous first state derivatives. The signal and noise have independent and identically distributed coordinates with finite moments of all orders, and are independent of each other and of the matrix. At every fixed number of iterations, empirical averages of pseudo-Lipschitz functions of order two converge in probability to the limits specified by Gaussian state evolution. These averages include empirical second moments and products of trajectory coordinates. The proof combines an aggregate bound on third derivatives with randomized Lindeberg replacement and operator-norm truncation, without matching third moments or invoking a non-Gaussian universality theorem. We also show that independence of complementary projected components for every deterministic rank-one constraint pair characterizes Gaussian entries. This identifies the limitation of extending Gaussian conditioning identities to independent non-Gaussian matrices. Finally, a variance-profile example shows that average variance normalization alone does not ensure the homogeneous state evolution limits.","authors":[{"name":"Ripunjay Dwivedi","position":0},{"name":"Taejoon Kim","position":1}],"categories":["cs.IT","math.IT","math.PR","math.ST","stat.TH"],"primaryCategory":"cs.IT","publishedAt":"2026-10-20T22:42:37Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28971"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28971"}],"brief":{"text":"Approximate message passing with independent matrix entries follows Gaussian state evolution under a uniform fourth-moment tail condition; for entries from a fixed law, a finite fourth moment suffices. The work also shows that Gaussian conditioning identities need not extend to non-Gaussian matrices, and that average variance normalization alone cannot guarantee homogeneous limits.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":64,"reason":"new","score":0.1945},{"paper":{"id":"arxiv:2610.28964","arxivId":"2610.28964","doi":"10.48550/arXiv.2610.28964","title":"Can Quantum Layers offer Performance and Security Advantages in Vision-Language Models?","abstract":"As quantum computing moves from the noisy intermediate-scale quantum (NISQ) era toward early fault-tolerant quantum computing (EFTQC), quantum-enhanced AI is shifting from a conceptual idea to experimental research. Quantum components are expected to extend the capabilities of existing AI systems, yet it remains unclear how such components will help within large multimodal models. In this study, we try to answer that question for vision-language models (VLMs) using classically simulated quantum layers. With LLaVA-OneVision as the backbone, we evaluate 4- and 8-qubit basic-ring and strongly entangling circuits at four insertion and four replacement locations. Insertion adds quantum layers to the classical model, while replacement substitutes them for classical components. Using the same setup, we compare performance and computational cost between classical and quantum-classical hybrid models. On the Visual Spatial Reasoning dataset, inserting a quantum layer after the visual projector reaches 80.7% development accuracy, 2.28 percentage-points above classical full fine-tuning. Replacing the final decoder feed-forward sublayer retains 79.22% accuracy with 13.07 million fewer trainable parameters. Quantum insertion can thus improve accuracy, and selective replacement can reduce parameters. These benefits depend strongly on where the quantum layer is inserted or which classical component it replaces: earlier replacements lower accuracy, and fewer parameters do not reliably shorten training or reduce computation. To assess potential security benefits, we evaluated the robustness of quantum-classical hybrid models under six attack settings. Overall, our study maps where quantum layers help and hurt within VLMs, offering a starting point for hybrid VLM design; whether these gains generalize remains open. The code is available at https://github.com/TITANS-Lab-Artifacts/QuantumVLM.","authors":[{"name":"Md Khalid Syfullah","position":0},{"name":"Alvi Ataur Khalil","position":1}],"categories":["quant-ph","cs.AI","cs.CV","cs.LG"],"primaryCategory":"quant-ph","publishedAt":"2026-10-20T22:28:17Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28964"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28964"}],"brief":{"text":"Classically simulated quantum layers in LLaVA-OneVision improved spatial-reasoning accuracy when added after the visual projector, while replacing the final decoder sublayer reduced trainable parameters with little accuracy loss. Benefits depended on placement, and smaller models did not reliably train faster or use less computation.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":65,"reason":"new","score":0.1945},{"paper":{"id":"arxiv:2610.28950","arxivId":"2610.28950","doi":"10.48550/arXiv.2610.28950","title":"HyperDrone: Simulation-to-Reality Transfer at Millions of Frames per Second","abstract":"We present HyperDrone, an extremely fast quadrotor simulator focused on high-performance batch rendering that is more than 20x faster than existing state-of-the-art batch quadrotor simulators. HyperDrone combines highly optimized multirotor dynamics with HyperLux, a new ray-tracing engine that supports Nvidia OptiX, Vulkan, Apple Metal and WebGPU backends that takes full advantage of modern ray-tracing hardware, capable of running both components entirely on the GPU. HyperLux renders RGB, depth, semantic/instance segmentation, surface normals and optical flow observations at configurable levels of fidelity, including reflections, realistic lighting, motion blur, anti-aliasing, and Physically Based Rendering (PBR). Even in the highest-fidelity settings, HyperLux achieves millions of Frames Per Second (FPS) on consumer-grade GPUs while rendering realistic 3D scenes. We demonstrate the merits of HyperDrone by training a vision-based yaw-control policy purely in simulation and transferring it to a real small-scale quadrotor (Crazyflie) with a camera, running completely onboard.","authors":[{"name":"Jonas Eschmann","position":0},{"name":"Giuseppe Loianno","position":1}],"categories":["cs.RO","cs.AI","cs.LG","cs.SY","eess.SY"],"primaryCategory":"cs.RO","publishedAt":"2026-10-20T22:03:11Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28950"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28950"}],"brief":{"text":"HyperDrone combines GPU-based quadrotor simulation and ray-traced rendering, running over 20 times faster than existing batch simulators. A vision-based yaw-control policy trained only in simulation transferred to a camera-equipped Crazyflie and ran onboard.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":66,"reason":"new","score":0.1945},{"paper":{"id":"arxiv:2610.28907","arxivId":"2610.28907","doi":"10.48550/arXiv.2610.28907","title":"JaxMARL-BC: Multi-Agent Reinforcement Learning for Macroeconomic Modelling, in Seconds","abstract":"Multi-agent reinforcement learning (MARL) offers a principled way of populating simulated economies with many heterogeneous decision makers. Its adoption, pioneered in finance, has nonetheless been slow in macroeconomics because of two main limitations: the lack of a shared frame of reference connecting MARL systems to dynamic stochastic general equilibrium (DSGE) models the profession already trusts, and the very long training times. The first limitation was recently addressed by the Multi-Agent Reinforcement Learning Business Cycle (MARL-BC), which places deep RL households in a real business cycle economy and reduces to classic DSGE models in specific limits. Here we address the second limitation. We present JaxMARL-BC, a JAX-accelerated simulator that trains MARL-BC economies in seconds (between 75x and 724x faster than the reference implementation), and scales to 200,000 independent learning households. We verify its correctness against the known DSGE limits, characterise its scaling through a simple cost law, and demonstrate its value by a proof-of-concept calibration of an extended model with heterogeneous returns on capital, in a two-hour experiment that would previously have taken over a month. We release JaxMARL-BC as open source.","authors":[{"name":"Daniele Montagnani","position":0},{"name":"Andrea Coletta","position":1},{"name":"Antonietta Mira","position":2},{"name":"Aldo Glielmo","position":3}],"categories":["cs.MA","cs.AI","cs.LG","econ.GN","q-fin.EC"],"primaryCategory":"cs.MA","publishedAt":"2026-10-20T20:35:44Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","venue":"Proceedings of the 7th ACM International Conference on AI in Finance (ICAIF '26), Milan, Italy, 2026","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28907"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28907"}],"brief":{"text":"JaxMARL-BC accelerates training in a multi-agent macroeconomic simulator, running 75–724 times faster than its reference implementation and scaling to 200,000 households. It enables a model calibration in two hours that would previously have taken over a month.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":67,"reason":"new","score":0.1945},{"paper":{"id":"arxiv:2610.28875","arxivId":"2610.28875","doi":"10.48550/arXiv.2610.28875","title":"To Manipulate or Not: Opinion Dynamics in LLM Interactions","abstract":"Large language models increasingly take part in extended conversations where they can both shape and be shaped by another party's views, yet most persuasion research measures only one side of that exchange. We introduce a controlled LLM-LLM debate framework in which two language-model roles repeatedly discuss a contested topic (climate-change attribution): one is given an assigned stance and, in some conditions, an explicit instruction to persuade, while the other receives no assigned position. Across four ordered model combinations involving GPT-6 Luna and Gemini 2.5 Flash Lite, we track both each role's explicitly reported position and a separate measure of its turn-by-turn behavior inside the conversation. Opinion dynamics differ substantially across model combinations: same-model interactions narrow the opinion gap on average, clearly for only one of the two models, while cross-model interactions widen it. Explicit reports and in-dialogue behavior capture distinct aspects of opinion dynamics, and a blanket instruction to persuade does not reliably increase influence. These findings suggest that conversational influence depends not only on assigned roles and initial positions but also on the participating models, with implications for how opinion change is measured in extended LLM interactions.","authors":[{"name":"Mohammad Siahkamari","position":0},{"name":"Phuoc Dat Nguyen","position":1},{"name":"Stephen MacNeil","position":2}],"categories":["cs.AI","cs.CL","cs.MA","cs.SI"],"primaryCategory":"cs.AI","publishedAt":"2026-10-20T19:48:20Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28875"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28875"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.28875.jpg","label":"thumbnail"}],"brief":{"text":"A controlled LLM debate framework tracks models’ stated opinions and conversational behavior, finding that opinion gaps narrow in same-model exchanges but widen across models, while instructions to persuade do not reliably increase influence.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":68,"reason":"new","score":0.1945},{"paper":{"id":"arxiv:2610.28847","arxivId":"2610.28847","doi":"10.48550/arXiv.2610.28847","title":"DSINC beyond Heston: the 3/2 and other non-affine stochastic-volatility models","abstract":"This paper extends the damped-sinc (DSINC) method for American options beyond the affine Heston class to general one-factor stochastic-volatility models. We show that the collapse identity, which reduces the early-exercise integral to a single spectral integral, remains valid for these models. Damping between zero and one is admissible whenever the discounted spot is a nonnegative local martingale. An Itô transformation of the volatility factorizes the conditional characteristic function and identifies a natural coordinate for approximating the exercise boundary. The 3/2 model admits closed-form joint transforms in terms of Kummer functions, while uncorrelated SABR admits a Bessel representation. When a practical closed-form transform is unavailable, the required volatility integral is computed directly by solving one backward equation per frequency, without a grid in the log-price. Through a known change of volatility coordinate, this approach also accommodates correlated SABR. Tests on the 3/2, SABR, GARCH and lognormal-volatility models show agreement with fine finite-difference benchmarks to within one basis point of the strike. Computational cost is comparable to or lower than that of our finite-difference implementation, except for the slower Hankel implementation of uncorrelated SABR.","authors":[{"name":"Andrey Itkin","position":0}],"categories":["q-fin.CP","cs.NA","math.NA","q-fin.MF","q-fin.PR"],"primaryCategory":"q-fin.CP","publishedAt":"2026-10-20T19:20:16Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28847"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28847"}],"brief":{"text":"DSINC now handles non-affine stochastic-volatility models, using closed-form transforms where available and backward equations otherwise. Tests across four models match fine finite-difference benchmarks within one basis point of the strike.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":69,"reason":"new","score":0.1945},{"paper":{"id":"arxiv:2610.28818","arxivId":"2610.28818","doi":"10.48550/arXiv.2610.28818","title":"Brittle by Design? Multi-Scenario Reinforcement Learning for Robust Mission Coordination in Systems of Systems","abstract":"As systems engineering (SE) shifts from monolithic systems to Systems of Systems, Mission Engineering (ME) has emerged as a new line of thinking in SE. Mission environments are uncertain and dynamic, and outcomes depend on how mission assets interact with them, rendering static architectures brittle and calling for flexible yet analytically rigorous ME approaches. We propose an intelligent mission coordination methodology integrating high-fidelity digital mission models with Reinforcement Learning (RL) for adaptive task allocation. Using a Digital Engineering (DE) infrastructure comprising a digital mission model and agent-based simulation, we formulate mission tactics management as a Markov Decision Process and train an RL coordinator with Proximal Policy Optimization to map mission states to tactical actions from realized outcomes. An aerial firefighting mission with three scenarios demonstrates the approach. First, all trained coordinators significantly outperform random and heuristic baselines. Second, single-scenario coordinators can be brittle: the coordinator trained on the more forgiving scenario degrades to random tactics performance in more demanding scenarios, whereas the demanding-scenario coordinator retains most of its advantage when transferred. Third, within the same training budget, a coordinator trained on both scenarios concurrently matches the scenario-specific coordinators in their own scenarios and significantly outperforms every other policy. Sensitivity analyses on ignition location, weather, scenario, and training curricula show that the variability represented during training shapes post-deployment performance. This proof of concept shows DE-enabled mission simulations combined with RL offer a reusable methodology for improving ME practice, which can be extended to fleet design and selection from a mission-first perspective.","authors":[{"name":"İbrahim Oğuz Çetinkaya","position":0},{"name":"Sajad Khodadadian","position":1},{"name":"Taylan Topcu","position":2}],"categories":["eess.SY","cs.AI","cs.LG","cs.SY"],"primaryCategory":"eess.SY","publishedAt":"2026-10-20T18:40:12Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28818"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28818"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.28818.jpg","label":"thumbnail"}],"brief":{"text":"Reinforcement learning trained with digital mission simulations improves aerial firefighting coordination over random and heuristic tactics. Training across scenarios matched specialist coordinators in each scenario and outperformed other policies, while single-scenario training could produce brittle tactics.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":70,"reason":"new","score":0.1945},{"paper":{"id":"arxiv:2610.28810","arxivId":"2610.28810","doi":"10.48550/arXiv.2610.28810","title":"Combining Long-Term Memory and LLMs Comes at a High Cost: From Prompt Injection to Persistent Compromise","abstract":"The widespread adoption of generative artificial intelligence in recent years has brought a parallel rise in prompt injection and jailbreak attacks. By embedding malicious instructions, these exploits manipulate language models to bypass intended behavior and operational boundaries. In traditional chatbots, such attacks are confined to a single session because state memory is not preserved across conversations. However, persistent memory features in modern conversational interfaces enable cross-session attack propagation, establishing a threat vector known as memory injection. This work explores this emerging vulnerability in leading LLM-as-a-service platforms, highlighting its real-world severity and the need for countermeasures. While existing literature focuses on violations tailored to specific platform features, we present a general threat model. We demonstrate that all NIST attack goals are collectively achievable across ChatGPT Instant 5.3, Claude Sonnet 4.6, and Gemini 3 Flash under this model. We also show that memory injections can exploit multiple attack vectors, including personal memories, projects, skills, organization instructions, and Model Context Protocol (MCP)-connected resources. Our findings reveal that all evaluated services can be manipulated to persistently spread disinformation, generate harmful content, or simulate offline unresponsiveness, even in fresh sessions. In the particular case of Claude, shared-memory features allow an attacker to compromise multiple users within an organization and also exfiltrate user prompts. To address these vulnerabilities, we derive defense mechanisms from our proposed threat model and foundational cybersecurity principles. Our analysis indicates that nearly half of these mitigations remain unimplemented across the target services. Datasets and attack artifacts are available at https://github.com/PedroBarcha/Memory-injection.","authors":[{"name":"Pedro H. Barcha Correia","position":0},{"name":"Matheus Azevedo de Sá","position":1},{"name":"Paulo H. Gomes de Senna","position":2},{"name":"Giovanni Mandel Martignago","position":3},{"name":"Victor Takashi Hayashi","position":4},{"name":"Charles Christian Miers","position":5},{"name":"Marcos A. Simplicio Jr","position":6}],"categories":["cs.CR","cs.AI","cs.CL","cs.LG"],"primaryCategory":"cs.CR","publishedAt":"2026-10-20T18:33:05Z","version":1,"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28810"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28810"}],"brief":{"text":"A general threat model shows that memory injection can compromise three leading LLM services across sessions, spreading disinformation, producing harmful content, or disrupting responses; shared memory can also expose prompts and affect multiple users. Nearly half of the proposed defenses are not implemented.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":71,"reason":"new","score":0.1945},{"paper":{"id":"arxiv:2610.28797","arxivId":"2610.28797","doi":"10.48550/arXiv.2610.28797","title":"CALLIOPE: Private Insights into AI Use, Without Embeddings","abstract":"We introduce CALLIOPE, a top-down framework for discovering topic hierarchies in large corpora of unstructured data (e.g., LLM conversations), while supporting differential privacy (DP). Unlike traditional and even recent LLM-augmented approaches that ultimately rely on clustering vector embeddings, our framework leverages LLM capabilities directly for the entire pipeline. Experiments demonstrate that this approach provides better utility over existing (embedding-based) methods in the DP setting, especially in the high privacy regime. We evaluate the quality of CALLIOPE outputs using a variety of traditionally studied clustering metrics, as well as some novel utility metrics that we introduce.","authors":[{"name":"Edith Cohen","position":0},{"name":"Vadym Doroshenko","position":1},{"name":"Badih Ghazi","position":2},{"name":"Charlie Harrison","position":3},{"name":"Pritish Kamath","position":4},{"name":"Alexander Knop","position":5},{"name":"Ravi Kumar","position":6},{"name":"Ethan Leeman","position":7},{"name":"Pasin Manurangsi","position":8},{"name":"Mikhail Pravilov","position":9},{"name":"Adam Sealfon","position":10},{"name":"Yurii Sushko","position":11},{"name":"Marika Swanberg","position":12},{"name":"Chiyuan Zhang","position":13}],"categories":["cs.LG","cs.AI","cs.CL","cs.CR"],"primaryCategory":"cs.LG","publishedAt":"2026-10-20T18:23:08Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28797"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28797"}],"brief":{"text":"CALLIOPE uses language models to discover topic hierarchies in unstructured data while preserving differential privacy. It provides better utility than embedding-based methods, especially under strong privacy protections.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":72,"reason":"new","score":0.1945},{"paper":{"id":"arxiv:2610.28747","arxivId":"2610.28747","doi":"10.48550/arXiv.2610.28747","title":"A Reverse-Gold Theorem: Positive-Data Identification by Substrate-Bounded Contrastive Learners","abstract":"Gold's theorem shows that no learner can identify an arbitrary language in a superfinite class from positive presentations alone. We prove a positive result of a different shape. Fix an innate \\emph{recursive contrast operator} $R$ mapping any string to a finite set of structurally bounded perturbations of it, and a discriminative scorer of bounded capacity trained by InfoNCE-style contrastive estimation \\citep{vandenoord2018,macollins2018} on self-generated perturbations. Under a faithfulness condition relating operator, scorer, and target, the trained scorer achieves \\emph{PAC conditional identification} from positive data alone: with high probability it ranks all but an $\\varepsilon$-fraction of positives above every out-of-language perturbation in their own orbit, with sample complexity polynomial in the capacity, the orbit size, the score bound, and $1/\\varepsilon$. The negatives are constructed by the learner, not supplied by an oracle, and no global normalizing constant over strings is ever computed. The misranking bound follows from the geometry of the contrastive loss---per-example loss floors plus a definite loss increment under misranking---yielding an $\\varepsilon^{-2}$ rate, together with an explicit approximation gap that quantifies the cost of in-language (``colliding'') perturbations and vanishes for orbit-pure operators. The ranking guarantee lifts to a global membership classifier under a margin condition, and the same operator composed with the learned scorer is a normalization-free Metropolis--Hastings generator. We instantiate the hypotheses for substitutable context-free languages \\citep{clarkeyraud2007} and the subregular classes \\citep{heinz2010,hayashikamigaito2025}, and state an open conjecture locating the construction in the mildly context-sensitive hierarchy.","authors":[{"name":"Rakesh Sengupta","position":0}],"categories":["stat.ML","cs.CL","cs.LG","q-bio.NC"],"primaryCategory":"stat.ML","publishedAt":"2026-10-20T18:00:47Z","version":1,"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28747"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28747"}],"brief":{"text":"A recursive contrast operator lets a bounded-capacity learner identify languages from positive data by ranking examples above learner-generated perturbations; the result gives polynomial sample complexity and can support membership classification under a margin condition.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":73,"reason":"new","score":0.1945},{"paper":{"id":"arxiv:2610.28141","arxivId":"2610.28141","doi":"10.48550/arXiv.2610.28141","title":"Fixed Sets and the Hardness of Computing Correlated Equilibria","abstract":"We show that computing an approximate correlated equilibrium of a polytope game is PPAD-complete, and thus polynomial-time equivalent to finding a Nash equilibrium. This challenges the conventional view in game theory that correlated equilibria are more tractable objects than Nash equilibria. From a conceptual standpoint, we connect correlated equilibria to a natural generalization of fixed points, which we call fixed sets. A fixed set of a continuous function $f$ is a nonempty finite set $S$ such that $S \\subseteq \\operatorname{conv} f(S)$. Our main technical contribution is to prove that computing a fixed set is PPAD-hard even at inverse polynomial precision.","authors":[{"name":"Ioannis Anagnostides","position":0},{"name":"Constantinos Daskalakis","position":1},{"name":"Gabriele Farina","position":2},{"name":"Noah Golowich","position":3},{"name":"Tuomas Sandholm","position":4},{"name":"Brian Hu Zhang","position":5}],"categories":["cs.GT","cs.CC","cs.DS","cs.LG"],"primaryCategory":"cs.GT","publishedAt":"2026-10-20T12:48:05Z","updatedAt":"2026-10-21T13:06:40Z","version":2,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28141"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28141"}],"brief":{"text":"Computing an approximate correlated equilibrium in polytope games is PPAD-complete, as hard as finding a Nash equilibrium; the work links this result to a new fixed-set concept and proves fixed-set computation PPAD-hard.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":74,"reason":"new","score":0.1945},{"paper":{"id":"arxiv:2610.28699","arxivId":"2610.28699","doi":"10.1016/j.sysarc.2026.103874","title":"Reinforcement learning for hardware acceleration and system-level optimization on FPGA and integrated-circuit platforms: A survey","abstract":"Reinforcement learning (RL) enables adaptive and autonomous decision-making in embedded, edge, and heterogeneous computing systems. Field-programmable gate arrays (FPGAs) and integrated circuits (ICs) offer low latency, energy efficiency, and architectural flexibility for RL deployment, while introducing challenges in resource constraints, implementation complexity, and hardware-algorithm co-design. This review surveys more than 100 studies on RL implementations across FPGA and IC platforms, emphasizing architectures, implementation strategies, performance evaluation, and applications. The literature is organized by RL family, including model-based Markov decision processes, classical value-based methods, deep RL, actor-critic, policy-gradient, and multi-agent approaches. We further classify studies by abstraction level, hardware realization strategy, and application domain, examining how RL algorithms map onto reconfigurable and fixed hardware. Unlike surveys restricted to particular algorithms or platforms, this work provides a unified analysis of FPGA and IC implementations, covering acceleration, reconfigurable architectures, energy-aware control, and system-level optimization. Comparative assessments consider resource utilization, latency, throughput, power consumption, and reported performance gains. Finally, we identify recurring architectural patterns, practical limitations, open challenges, and research directions for efficient RL-driven hardware systems. The survey offers a consolidated reference for researchers developing and evaluating RL solutions on specialized hardware.","authors":[{"name":"Hamza Kheddar","position":0}],"categories":["cs.LG","cs.AI","cs.AR","cs.PF","cs.RO","cs.SY","eess.SY"],"primaryCategory":"cs.LG","publishedAt":"2026-10-20T11:15:34Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","venue":"Journal of Systems Architecture, 2026","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28699"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28699"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.28699.jpg","label":"thumbnail"}],"brief":{"text":"A survey of more than 100 studies maps reinforcement-learning methods, hardware strategies, and applications across FPGA and integrated-circuit platforms, comparing performance and identifying common architectures, limitations, and research directions.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":75,"reason":"new","score":0.1945},{"paper":{"id":"arxiv:2610.30125","arxivId":"2610.30125","doi":"10.48550/arXiv.2610.30125","title":"PhysiCalWorld: In-Context Physics Calibration for World Models","abstract":"Existing world models typically perceive a scene through current visual frame or a narrow history window. However, these observations may not reveal the physics properties that determine transition dynamics, such as an object's mass distribution, joint damping, or surface friction. Without calibrating these physics factors for world models, predictions tend to regress toward an empirical prior and induce severe drift in novel test scenes. To this end, we propose PhysiCalWorld, which calibrates action-conditioned world models to the instance-specific dynamics via in-context prompting, without fine-tuning or analytical physics modeling. Our method encodes an exploratory interaction clip collected within the test scene into a compact physics descriptor token, which grounds subsequent predictions in the scene's actual physical behavior. PhysiCalWorld is trained end-to-end with the latent prediction loss alone, without physics labels. At inference time, the descriptor token is extracted once and can be reused across rollouts with negligible compute overhead. We evaluate PhysiCalWorld across four environments covering diverse physics variations and show that it substantially improves rollout prediction accuracy and visual planning success rate. The learned descriptors also form a physically structured latent space and generalize beyond the physics ranges seen during training.","authors":[{"name":"Yunhao Luo","position":0},{"name":"Wentao He","position":1},{"name":"Yinpei Dai","position":2},{"name":"Hyelin Nam","position":3},{"name":"Hongze Fu","position":4},{"name":"Joyce Chai","position":5},{"name":"Nima Fazeli","position":6},{"name":"Jeong Joon Park","position":7}],"categories":["cs.RO","cs.AI","cs.CV"],"primaryCategory":"cs.RO","publishedAt":"2026-10-21T17:57:59Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30125"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30125"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.30125.jpg","label":"thumbnail"}],"brief":{"text":"PhysiCalWorld calibrates action-conditioned world models using a brief exploratory clip, without fine-tuning or physics labels. Across four environments, it improves rollout accuracy and planning success, while learning descriptors that generalize beyond training physics ranges.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":76,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.30113","arxivId":"2610.30113","doi":"10.48550/arXiv.2610.30113","title":"Quant-Arena: Benchmarking LLM Agents for Quantitative Trading Research","abstract":"Quantitative research is an iterative loop: a researcher proposes a market hypothesis, implements it as an executable signal, tests it on historical data, and decides what to keep. Existing benchmarks score formulas, code, or individual trading decisions, leaving open whether large language models (LLMs) can complete this loop and produce strategies that work beyond the research period. We introduce Quant-Arena, a benchmark and public leaderboard for end-to-end quantitative research on real A-share market data. A model constructs features and stock-scoring code, receives feedback only on 2022--2023, and freezes one strategy for assessment on the sealed year 2024. A cash-account evaluator applies trading constraints and costs, including market impact estimated from a simulated order book. Scores are dominated by realized excess return and complemented by signal quality and the diversity of features used by the strategy. We evaluate nine frontier models, among which GPT-6 Astra scores highest; yet no model beats the equal-weight market in 2024. Quant-Arena is open to submissions from any model or agent harness.","authors":[{"name":"Houde Qian","position":0},{"name":"Zheng Jiang","position":1},{"name":"Wenyi Li","position":2},{"name":"Chenxi Xu","position":3},{"name":"Xinjie Lin","position":4},{"name":"Yuzhao Peng","position":5},{"name":"Ruoyu Han","position":6},{"name":"Youjie Zheng","position":7},{"name":"Kaisen Yang","position":8},{"name":"Qinhuai Na","position":9},{"name":"You Fu","position":10},{"name":"Calvin Xiao","position":11}],"categories":["cs.CL","cs.AI","cs.CE"],"primaryCategory":"cs.CL","publishedAt":"2026-10-21T17:54:01Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30113"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30113"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.30113.jpg","label":"thumbnail"}],"brief":{"text":"Quant-Arena tests whether language models can build and validate trading strategies, then succeed on unseen market data. GPT-6 Astra scores highest among nine models, but none beats the equal-weight market in 2024.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":77,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.30105","arxivId":"2610.30105","doi":"10.48550/arXiv.2610.30105","title":"Asymmetric On-Policy Distillation Learns In-Context Exploration","abstract":"Sequential decision-making under partial observability is difficult because agents must explore to infer hidden, task-relevant information before acting effectively. Learning such information-gathering behaviors tabula rasa via online reinforcement learning is sample-inefficient, whereas offline imitation averages expert actions across hidden states and fails to learn exploration. In this work, we propose ASTEROID, a simple and effective student--teacher supervised learning framework that can learn test-time exploration. Although a privileged expert never needs to explore, we demonstrate that a history-conditioned student can learn to do so when trained with on-policy contexts: the student first collects on-policy context under partial observability, and the privileged expert provides action labels conditioned on that context. By distilling expert actions directly conditioned on the student's on-policy history, the student learns to systematically resolve partial observability, i.e., search while uncertain and exploit once the hidden state is revealed. This enables efficient test-time exploration (similar to Bayesian posterior sampling) without incurring the impractical training-data costs of reinforcement learning. Across a diverse range of eleven different environments, including discrete exploration benchmarks, robotic manipulation, navigation, and Procgen environments, ASTEROID achieves ~2x higher returns under fixed exploration budgets, uses ~100x fewer interactions than RL baselines, and outperforms frontier models at exploration. Finally, our method transfers policies trained in simulation to vision-denied real-world robot manipulation, exhibiting efficient exploration and increasing success by +40% over standard distillation. Overall, it provides a simple yet surprisingly effective framework to learn efficient test-time exploration.","authors":[{"name":"Sriyash Poddar","position":0},{"name":"Yanda Bao","position":1},{"name":"Jacob Krantz","position":2},{"name":"Matthew Chang","position":3},{"name":"Xavier Puig","position":4},{"name":"Roozbeh Mottaghi","position":5},{"name":"Natasha Jaques","position":6},{"name":"Abhishek Gupta","position":7}],"categories":["cs.RO","cs.AI","cs.LG"],"primaryCategory":"cs.RO","publishedAt":"2026-10-21T17:50:31Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30105"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30105"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.30105.jpg","label":"thumbnail"}],"brief":{"text":"ASTEROID trains a history-conditioned student on its own partially observed experiences, using a privileged expert to label actions. Across 11 environments, it learns to explore efficiently, achieving roughly twice the returns with far fewer interactions than reinforcement learning baselines.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":78,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.30097","arxivId":"2610.30097","doi":"10.48550/arXiv.2610.30097","title":"AtomicRL: Atomic Reinforcement Learning for Long-Horizon Robotic Manipulation","abstract":"Vision-language-action (VLA) models have shown promise in robotic manipulation, but imitation learning (IL) alone often struggles to master complex long-horizon skills. Reinforcement learning (RL) offers an effective avenue for further improving VLA policies. However, existing VLA-RL methods typically suffer from delayed and biased credit assignment due to trajectory-level sparse rewards, while imbalanced subtask mastery creates bottlenecks that limit overall performance. To address these challenges, we propose AtomicRL, an RL post-training framework that leverages atomic task structures for reliable credit assignment and adaptive exploration. AtomicRL introduces Semantic-Aware Policy Optimization (SAPO), which uses semantic-aware gating to constrain TD residuals from failed atomic subtasks from propagating across task boundaries, mitigating credit contamination while preserving intra-subtask credit assignment and value bootstrapping. In addition, Bottleneck-Adaptive Target Exploration (BATE) dynamically coordinates task sampling and local exploration incentives based on the mastery of individual atomic subtasks, directing learning toward the remaining bottlenecks. Finally, we develop a Real2Sim2Real pipeline that bridges online RL in simulation with real-world deployment, facilitating practical transfer to physical robots. On LIBERO-LONG, AtomicRL achieves absolute performance gains of 3.6% and 4.0% over PPO across two VLA architectures. On CALVIN ABC-D, it improves full-task performance by up to 9.5% over PPO. Combined with the Real2Sim2Real pipeline, AtomicRL further improves real-world task performance by an average of 21.3%. These results demonstrate that AtomicRL effectively improves the learning efficiency and manipulation performance of VLA policies on long-horizon tasks.","authors":[{"name":"Likui Zhang","position":0},{"name":"Zhe Zhu","position":1},{"name":"Zhihao Zhan","position":2},{"name":"Yibo Bai","position":3},{"name":"Hanlin Wang","position":4},{"name":"Dongwei Li","position":5},{"name":"Bin Wang","position":6},{"name":"Tianle Zhang","position":7},{"name":"Xiaodan Liang","position":8},{"name":"Liang Lin","position":9}],"categories":["cs.RO","cs.AI","cs.CV"],"primaryCategory":"cs.RO","publishedAt":"2026-10-21T17:45:24Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30097"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30097"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.30097.jpg","label":"thumbnail"}],"brief":{"text":"AtomicRL uses atomic subtasks to improve credit assignment and focus exploration on weak points in long-horizon robotic manipulation. It outperforms PPO in simulation and improves real-world task performance by an average of 21.3%.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":79,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.30077","arxivId":"2610.30077","doi":"10.48550/arXiv.2610.30077","title":"A linear, second-order, unconditionally energy-stable scalar-auxiliary-variable scheme for the CHQ viscoelastic phase-separation model","abstract":"The CHQ model describes viscoelastic phase separation in a polymer--solvent mixture: a Cahn--Hilliard equation for the polymer volume fraction $ϕ$ is two-way coupled, through a phase-dependent cross-diffusion flux, to a damped relaxation equation for a scalar bulk elastic stress $q$. We propose a scalar-auxiliary-variable (SAV) time discretisation of this system, combining the SAV reformulation of the double-well potential with a second-order backward differentiation formula (BDF2) and a second-order extrapolation of all phase-dependent mobilities: a linear, computationally cheaper complement to the reference scheme's fully nonlinear Newton solve. The resulting method is (i) linear at every time step, with a coupling structure we prove has exactly one solution at every step, for every choice of time-step size; (ii) mass conserving, exactly, as an algebraic identity of the discrete equations themselves (the only drift ever seen numerically traces to the finite tolerance of the iterative linear solver, not a failure of the conservation property); and (iii) unconditionally stable for a modified discrete energy, via an exact one-step energy--dissipation identity. We show the energy identity extends to this filtered variant as well, with two explicit extra non-negative terms. Hence, the scheme actually producing the physics results is itself provably stable for the same energy. A Fourier--Galerkin implementation verifies that the energy identity holds at every time step to within floating-point round-off. It also confirms second-order temporal accuracy, including in the steep-interface regime, and agreement with an independently implemented nonlinear Newton--BDF2 solver. Compared with that independent nonlinear Newton--BDF2 solver, at equal grid resolution, the SAV method uses $2.5$--$3.0\\times$ fewer linear-solver iterations and $1.7$--$2.0\\times$ less wall-clock time.","authors":[{"name":"Ritesh Bhola","position":0},{"name":"Michael te Vrugt","position":1}],"categories":["math.NA","cond-mat.soft","cs.NA"],"primaryCategory":"math.NA","publishedAt":"2026-10-21T17:27:54Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30077"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30077"}],"brief":{"text":"A linear SAV scheme solves the CHQ viscoelastic phase-separation model with second-order accuracy and exact mass conservation. It is unconditionally stable for a modified energy and runs faster than a nonlinear Newton-based method.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":80,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.30074","arxivId":"2610.30074","doi":"10.48550/arXiv.2610.30074","title":"Parallel AI Agent Collaboration as Graph-Structured Distributed Block Optimization","abstract":"Parallel AI agents can produce useful local edits yet degrade a shared solution when their edits are combined. We model this coordination problem as discrete block optimization with an exactly evaluable objective and a known dependency graph. Independent sets certify compatible edits whose verified gains, measured against the same current solution, add exactly. A coordinator constructs a safe compatible candidate and can also evaluate interacting edits to capture favorable interactions. We prove that an optimal fractional coloring yields a merger that preserves a $1/χ_f(G)$ fraction of the total verified local improvement, where $χ_f(G)$ is the graph's fractional chromatic number. This guarantee is tight among mergers that select a subset of the same proposals. Under conditional local-improvement guarantees and a block error-bound condition, repeated collaboration converges geometrically in the objective gap. The rate separates task structure, agent allocation, and computation cost, yielding a linear program for agent assignment and an explicit time-cost trade-off. Code-repair experiments show that verified merging resolves conflicts between individually useful edits that cause naive merging to fail. On larger conflict-free tasks, parallel agents reduce modeled completion time at higher input-token-equivalent cost. On three public road networks, fractional integration retains 922 units of verified graph-cut improvement across 90 paired runs. For route-search agents sharing exclusive access to vehicle routes, fractional scheduling reduces measured wall time by a median of 15% across twelve paired runs, with route-cost changes varying across runs.","authors":[{"name":"Hanmin Li","position":0},{"name":"Yury Demidovich","position":1},{"name":"Peter Richtárik","position":2}],"categories":["cs.MA","cs.AI","math.OC"],"primaryCategory":"cs.MA","publishedAt":"2026-10-21T17:27:17Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30074"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30074"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.30074.jpg","label":"thumbnail"}],"brief":{"text":"A graph-based framework coordinates parallel agents by verifying edits and merging compatible proposals. It proves a tight guarantee on retained improvement and geometric convergence under stated conditions; experiments show conflict resolution and faster route searches, with route-cost changes varying across runs.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":81,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.30073","arxivId":"2610.30073","doi":"10.48550/arXiv.2610.30073","title":"When Windowed Anderson Acceleration Converges Faster","abstract":"Windowed Anderson acceleration AA($m$) combines at most $m+1$ recent iterates by least squares. For a real affine contraction $x\\mapsto Wx+b$ with $q=\\|W\\| \\in(0,1)$, we determine exactly when AA($m$) converges from every starting point with an asymptotic factor below $q$. We give a polynomial criterion for this improvement and an equivalent spectral condition: there is no probability distribution on the phases of the eigenvalues of modulus $q$ whose first $m$ cosine moments are $q,q^2,\\ldots,q^m$. If such a distribution exists, we construct a starting point from which AA($m$) reproduces fixed-point iteration at every step. Otherwise, over a fixed number of consecutive steps, every run contracts the residual norm by a common factor strictly smaller than the corresponding power of $q$. The polynomial criterion also gives explicit convergence-rate bounds with constants independent of the dimension. When $q$ equals the spectral radius, this characterizes improvement over the asymptotic convergence factor of fixed-point iteration. For general affine contractions, the asymptotic factor never exceeds the numerical radius of $W$, and a related moment condition gives a uniform upper bound strictly below that radius.","authors":[{"name":"Gilad Lerman","position":0},{"name":"Teng Zhang","position":1}],"categories":["math.NA","cs.NA","stat.CO"],"primaryCategory":"math.NA","publishedAt":"2026-10-21T17:26:58Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30073"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30073"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.30073.jpg","label":"thumbnail"}],"brief":{"text":"Windowed Anderson acceleration is characterized by a polynomial and an equivalent condition on eigenvalue phases that determine when it beats fixed-point iteration. The criterion also gives dimension-independent rate bounds and shows when some starting point prevents any improvement.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":82,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.30054","arxivId":"2610.30054","doi":"10.48550/arXiv.2610.30054","title":"Multiplying matrices via polynomials: exposition and barriers","abstract":"We give an exposition of OpenAI's recent proof that the exponent $ω_\\mathbb{C}$ of matrix multiplication over the complex numbers satisfies $ω_\\mathbb{C}\\leq9/4$. Our version of the proof gives $ω_\\mathbb{F}\\leq9/4$ over every field $\\mathbb{F}$. We also treat the rectangular matrix multiplication exponent $ω_\\mathbb{F}(a,b,c)$. Here we have $ω_\\mathbb{F}(1,k,1)=2$ for $0\\leq k\\leq1/2$ and $ω_\\mathbb{F}(1,k,1)\\leq1+k+1/(4k)$ for $k\\geq1/2$, and, for real numbers $a\\geq b\\geq c\\geq0$, $ω_\\mathbb{F}(a,b,c)=a+b$ if $c(a+b)\\leq ab$ and $ω_\\mathbb{F}(a,b,c)\\leq(a+b+c)/2+(ab/c+ac/b+bc/a)/4$ otherwise. Our exposition is in the language of degeneration inequalities among polynomial multiplication tensors and Strassen's asymptotic spectrum duality. We moreover prove an operational barrier (which can be thought of as coming from an operational asymptotic spectrum) showing that the current analysis of all these upper bounds on $ω_\\mathbb{F}$ and $ω_\\mathbb{F}(a,b,c)$ is optimal and that in fact the current degeneration ingredients cannot improve the bounds.","authors":[{"name":"Alimzhan Amanov","position":0},{"name":"Maxim van den Berg","position":1},{"name":"Jeroen Zuiddam","position":2}],"categories":["cs.CC","math.RA","math.RT"],"primaryCategory":"cs.CC","publishedAt":"2026-10-21T17:15:05Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30054"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30054"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.30054.jpg","label":"thumbnail"}],"brief":{"text":"Polynomial degeneration methods yield matrix multiplication exponent bounds of 9/4 over every field and improved rectangular bounds. An operational barrier shows these bounds cannot be improved using the current degeneration ingredients.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":83,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.30042","arxivId":"2610.30042","doi":"10.48550/arXiv.2610.30042","title":"HGM-Dual: Lineage-Aware Guided Agent Harness Optimization for Long-Horizon Planning","abstract":"Harness optimization, which improves an agent's prompts, tools, and workflows, is an increasingly effective way to enhance agents built on frozen or closed-source LLMs. Leading approaches such as the Darwin--Godel Machine and the Huxley--Godel Machine (HGM) rely on recursive self-improvement and were developed mainly for coding, where the competence needed to improve the agent coincides with the competence being improved. Planning tasks, in which an agent must produce a sequence of tool calls that jointly satisfies interacting constraints, break this alignment. There, open-ended self-modification repeatedly explores ineffective changes and consumes substantial evaluation budgets. We introduce HGM-Dual, a diagnostic-guided framework for budget-efficient harness evolution for planning agents. First, HGM-Dual represents agent evolution as lineage-aware context. This context records what each ancestor changed, which behaviors improved or regressed, and which modifications were ineffective, giving the editor explicit edit-outcome attribution instead of an unstructured archive of past agents. Second, HGM-Dual introduces error-conditioned dual-path child generation. For each selected parent, it generates a general child that preserves open-ended exploration and specialized children that target the harness components most likely responsible for the parent's highest-ranked constraint violations. HGM-Dual outperforms the strongest baseline by up to 14.45% relative improvement in case accuracy while matching HGM variants' final performance with 2x-4x fewer evaluation budget.","authors":[{"name":"Sudipta Paul","position":0},{"name":"Pengrui Quan","position":1},{"name":"Aounon Kumar","position":2},{"name":"Nicholas Tzou","position":3},{"name":"Vivek Kulkarni","position":4},{"name":"Srinivas Chappidi","position":5}],"categories":["cs.AI","cs.LG","cs.NE"],"primaryCategory":"cs.AI","publishedAt":"2026-10-21T17:02:28Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30042"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30042"}],"brief":{"text":"HGM-Dual uses ancestor change histories and constraint-targeted edits to improve planning agents’ prompts, tools, and workflows. It beats the strongest baseline by up to 14.45% in case accuracy and matches HGM variants’ final performance with 2–4 times fewer evaluations.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":84,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.30034","arxivId":"2610.30034","doi":"10.48550/arXiv.2610.30034","title":"Dual Effect and Certainty Equivalence Revisited for a Linear-Gaussian Dual Control Problem","abstract":"This paper revisits the relation between dual effect and certainty equivalence (CE) through a scalar linear-Gaussian model. In dual control, the current input affects not only the immediate cost but also the information available for future decisions, and hence the future cost. Dynamic programming (DP) provides a natural framework for formulating dual control problems. At each step of the Bellman recursion, an optimal input minimizes the sum of the stage cost and the expected cost-to-go. We restrict attention to stage costs minimized by the CE input. However, CE may not minimize the Bellman objective because of the input-dependent expected cost-to-go. Classical no-dualeffect (NDE) conditions guarantee CE optimality in certain problem classes by requiring the conditional distribution of the estimation error to be independent of past controls. In the present model, we observe that CE can remain optimal at the level of a single Bellman recursion even when the classical NDE conditions fail, provided that the expected cost-to-go is independent of the current input. Motivated by this observation, we introduce a notion of NDE formulated directly at the level of the Bellman objective, which we call Bellman-side NDE. Under some mild regularity conditions, we show that Bellman-side NDE can be characterized by a heat equation. This characterization identifies the class of Bellman-side NDE functions, and further yields a decomposition of a generic cost-to-go function into a Bellman-side NDE component and a dual-effect component.","authors":[{"name":"Ying Wang","position":0},{"name":"Yue Ju","position":1},{"name":"Håkan Hjalmarsson","position":2}],"categories":["math.OC","cs.SY","eess.SY"],"primaryCategory":"math.OC","publishedAt":"2026-10-21T16:58:17Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30034"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30034"}],"brief":{"text":"A scalar linear-Gaussian dual control model shows that certainty-equivalent inputs can minimize a Bellman objective even when classical no-dual-effect conditions fail; a heat equation characterizes this broader condition and separates cost-to-go into no-dual-effect and dual-effect components.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":85,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.30026","arxivId":"2610.30026","doi":"10.48550/arXiv.2610.30026","title":"What the Author Wrote, What the Agent Reads: Silent Cue Loss in Enterprise Document Ingestion","abstract":"Enterprise agents act on documents. Before an agent can follow a policy, answer a compliance question or file a ticket, some ingestion layer must turn a Word file or PDF into tokens - and that layer is where an agent deployed in the wild silently loses information. We show that one layer is lost by every mainstream converter - the meaning an author encodes in how text looks - and that the loss makes agents confidently wrong rather than visibly incomplete. We build a task from 76 real policy documents: recover the requirements the author singled out, scored against the file's own OOXML properties by a deterministic matcher with no LLM judge. Formatting-blind ingestion collapses from micro F1 0.690 to 0.150 (docling) and 0.075 (plain text), a gap that replicates across six answering models from two vendors - five Claude models spanning the Opus, Sonnet and Haiku tiers, and OpenAI's GPT-5.6. An agent that cannot see the cue guesses, and docling's false-positive rate on requirements the author did not flag is four times ours. A span-level leave-one-out localises the loss to two cues - highlight (+0.302) and colour (+0.262), both Holm-adjusted p < 1e-4 - while bold, italic, monospace and strikethrough each sit within +/-0.003 of zero. The same asymmetry appears in PDF. Colour is also the cue no mainstream converter preserves and no parsing benchmark scores, and it is common: 32.8% of 6,460 documents carry a deliberate colour or highlight, 10.3% on an obligation. A pre-registered test shows that typing a cue for the agent adds +0.085 beyond merely preserving it (Holm-adjusted p = 0.049). We validate the labels against rendered page images through an independent modality (kappa = 0.72) and bound what wording alone can achieve (AUC 0.664). For agent builders the practical finding is that ingestion, not reasoning, is where this reliability is won or lost, and that the fix is cheap and measurable.","authors":[{"name":"Yikai Zhao","position":0},{"name":"Saurabh Pandey","position":1},{"name":"Pradeep Kumar Misra","position":2}],"categories":["cs.CL","cs.AI","cs.IR"],"primaryCategory":"cs.CL","publishedAt":"2026-10-21T16:51:25Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30026"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30026"}],"brief":{"text":"Document converters discard visual cues such as color and highlighting that authors use to mark policy requirements, making agents less accurate and more prone to false positives. Preserving and explicitly describing those cues improves performance, locating a practical reliability fix in document ingestion rather than reasoning.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":86,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.30016","arxivId":"2610.30016","doi":"10.48550/arXiv.2610.30016","title":"Code Comprehensor: Hierarchical Documentation and Cross-Modal Retrieval for Large Repositories","abstract":"Understanding large, complex, and rapidly evolving code bases remains a core challenge for software engineering at scale, hindering onboarding, maintenance, refactoring, and impact analysis. Providing this information to a Large-Language Model (LLM) is another hurdle, as the LLM must understand not only the required code method but also its associated methods in the grand scheme of the repository. We propose Code Comprehensor, a hybrid retrieval-reasoning framework combining (i) a code dependency callgraph, (ii) CodeScribe, which uses code-aware LLMs to generate hierarchical documentation across repository call graphs, and (iii) CodeContrast, a CLIP-style dual-encoder model trained on CodeScribe output to learn aligned text-code embeddings. CodeContrast enables fast, semantically precise retrieval of relevant code-context neighborhoods, which serve as grounded evidence for an MCP-based LLM orchestrator answering domain-specific queries. Evaluation on OpenCV and a proprietary 5M-LOC repository shows that CodeScribe-derived supervision significantly improves cross-modal retrieval in text-to-code accuracy over baseline approaches. Richer textual supervision (hierarchical documentation, parent/sibling context, and LLM-generated synthetic questions) consistently enhances performance across both repositories. LLM-as-a-judge evaluation using the popular RAGAS of the Code Comprehensor output yields above 90% faithfulness scores.","authors":[{"name":"Sujoy Roychowdhury","position":0},{"name":"Joseph H. R. Isaac","position":1},{"name":"Jiangning Gao","position":2},{"name":"Martin Klitte","position":3},{"name":"Diarmuid Corcoran","position":4}],"categories":["cs.SE","cs.AI","cs.LG"],"primaryCategory":"cs.SE","publishedAt":"2026-10-21T16:45:43Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30016"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30016"},{"kind":"other","url":"https://science-os.github.io/scienceos-atlas/thumbs/2610.30016.jpg","label":"thumbnail"}],"brief":{"text":"Code Comprehensor combines call graphs, hierarchical code documentation, and cross-modal retrieval to find relevant code context for LLM answers. Tests on OpenCV and a 5-million-line repository show improved text-to-code retrieval, with answers scoring above 90% for faithfulness.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":87,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.30014","arxivId":"2610.30014","doi":"10.48550/arXiv.2610.30014","title":"Agent Behavior Maps: Locating Agents by What They Would Do, Without Running Them","abstract":"LLM agents with similar success rates can follow different action sequences. An Agent Behavior Map represents LLM-harness combinations using centered action log-likelihoods on shared reference trajectories, without requiring new environment rollouts from the target agent. We derive error bounds relating squared coordinate distances to trajectory KL under explicit assumptions, and across eight cross-scored models the distances track the directly measured trajectory KL in rank order. Across 100 LLMs and four environments, ALFWorld-based coordinates support performance prediction and identify behavioral similarities. In ALFWorld, these similarities persist after controlling for success rates and model size, including across model families. As an extension, harness-component displacements approximately compose, with accuracy varying across components and environments.","authors":[{"name":"Shota Fujikawa","position":0},{"name":"Takuya Tamura","position":1},{"name":"Kosuke Akimoto","position":2},{"name":"Daichi Haraguchi","position":3},{"name":"Masafumi Oyamada","position":4}],"categories":["cs.LG","cs.AI","cs.CL"],"primaryCategory":"cs.LG","publishedAt":"2026-10-21T16:44:02Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30014"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30014"}],"brief":{"text":"Agent Behavior Maps locate LLM agents from their action likelihoods on shared trajectories, without new rollouts. Across 100 models and four environments, their coordinates predict performance and reveal behavioral similarities beyond success rates and model size.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":88,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.30009","arxivId":"2610.30009","doi":"10.48550/arXiv.2610.30009","title":"Mitigating Model Collapse through Dreaming Learning","abstract":"Training generative models on synthetic data causes loss of diversity and a decrease in performance, a phenomenon known as model collapse. We study this degeneration spectrally in a controlled Markov setting. The eigenvalue spectrum of the learned transition operator offers a direct view of the process, since the subdominant eigenvalues set how fast the chain mixes and therefore how much of the state space it continues to explore. We find that sampling at temperature $T_s < 1$ acts on the operator like repeated sharpening, pushing it toward a deterministic system: the subdominant eigenvalues approach the unit circle and the spectral gap narrows. When $T_s < 1$, even data accumulation does not appear to arrest this drift. We then analyze Dreaming Learning, a regulariser that, in parallel with regular training, adds a loss term on the model's own samples drawn at a raised temperature $T_d > 1$. Because high-temperature samples flatten the predicted distribution, training on them keeps the entries of the learned operator from becoming vanishingly small, which in turn keeps the eigenvalues bounded away from the unit circle. Measurements of the spectra over 5000 generations are consistent with this picture, and the mitigation effect persists in more complex settings such as LSTM language models and a 124M-parameter GPT-2.","authors":[{"name":"Matteo Benati","position":0},{"name":"Alessandro Londei","position":1},{"name":"Denise Lanzieri","position":2},{"name":"Raffaello Mastromarino","position":3},{"name":"Vittorio Loreto","position":4}],"categories":["cs.LG","cond-mat.dis-nn","stat.ML"],"primaryCategory":"cs.LG","publishedAt":"2026-10-21T16:41:53Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30009"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30009"}],"brief":{"text":"Dreaming Learning trains models on their own higher-temperature samples to counter the loss of diversity caused by repeated synthetic-data training. Spectral measurements over 5,000 generations show that it mitigates model collapse, including in LSTM language models and a 124M-parameter GPT-2.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":89,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.30003","arxivId":"2610.30003","doi":"10.48550/arXiv.2610.30003","title":"On optimal (v,5,3,1) optical orthogonal codes","abstract":"In this paper, we establish a tight upper bound for the size of a $(v,5,3,1)$ optical orthogonal code (OOC) and call such a code optimal when it achieves this upper bound. In addition, if an optimal OOC has the property that all non-identity elements of $\\mathbb{Z}_v$ are covered by the list of differences of its codeword-sets then it is called perfect. In this paper, we present direct and recursive constructions for infinite classes of optimal and perfect $(v,5,3,1)$-OOCs.","authors":[{"name":"Laura Johnson","position":0},{"name":"Lorenzo Mella","position":1},{"name":"Anita Pasotti","position":2},{"name":"Emma Smith","position":3}],"categories":["math.CO","cs.IT","math.IT"],"primaryCategory":"math.CO","publishedAt":"2026-10-21T16:40:00Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30003"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30003"}],"brief":{"text":"The work establishes a tight upper bound for the size of these optical codes and gives direct and recursive constructions of infinite classes that meet the bound, including perfect codes.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":90,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29996","arxivId":"2610.29996","doi":"10.48550/arXiv.2610.29996","title":"HiddenLens: A Multi-Layer LLM Framework for Detecting and Resolving Hidden Assumptions in Software Requirements","abstract":"Natural-language software requirements often omit assumptions about performance, operating conditions, user behaviour, and domain constraints, affecting architecture, testing, and deployment. If undiscovered until implementation, resolving them is far more costly than addressing them early. Although ambiguity and vagueness are widely studied, less attention has been given to assumptions embedded in requirements that appear clear and reasonable. We propose HiddenLens, a novel six-layer BERT-LLM pipeline and, to our knowledge, the first framework engineered to detect, classify, question, and resolve hidden assumptions rather than merely flag ambiguity. Layer 1 uses BERT attention entropy to localise unresolved spans; Layer 2 uses CoT-prompted LLM reasoning to articulate each assumption and its failure risk; Layer 3 uses fine-tuned BERT to classify assumptions into six non-functional requirement categories: performance, security, scalability, usability, reliability, and domain; Layer 4 generates ranked clarification options using FS LLM prompting; Layer 5 enables human validation through confirmation, editing, or rephrasing; and Layer 6 rewrites the requirement using the confirmed resolution, producing an explicit, testable, and traceable requirement. We also introduce the HiddenLens Corpus (HLC), comprising 312 requirements annotated with 847 hidden assumptions across six NFR categories. Inspired by PROMISE and PURE, HLC addresses the lack of annotated data for this task and is, to our knowledge, the first dataset developed specifically for it. HiddenLens outperforms keyword-based and QuARS-Adapted baselines by 71% and 61%, respectively, achieving an end-to-end F1-score of 0.79; its BERT-based NFR classifier achieves an F1-score of 0.847. These results indicate that HiddenLens can help analysts resolve hidden assumptions before they propagate through software development.","authors":[{"name":"Tanu Singh","position":0},{"name":"Aashna Suman","position":1},{"name":"Vinod Patidar","position":2}],"categories":["cs.SE","cs.AI","cs.CL"],"primaryCategory":"cs.SE","publishedAt":"2026-10-21T16:35:35Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29996"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29996"}],"brief":{"text":"HiddenLens detects, classifies, and helps resolve unstated assumptions in software requirements, then rewrites them as explicit, testable statements. On a new annotated dataset, it achieved an end-to-end F1-score of 0.79, outperforming keyword-based and QuARS-Adapted baselines.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":91,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29961","arxivId":"2610.29961","doi":"10.48550/arXiv.2610.29961","title":"Commit or Abstain? System One Decision Models, LLMs, and Missing Legal Authority","abstract":"Given the enormous number of claims to be checked, verifiable legal AI requires scalable, cost-effective models that answer the practical needs of law. System One decision models are built on large language models but, instead of generating text, return a calibrated probability for each option. This suits the verification of legal authority, where the question is closed: a given holding, or none of them. We test whether such models abstain when the correct authority is missing. We extend CaseHOLD with an abstention condition in which the correct holding is replaced in 1,670 of 5,314 items, and compare four decision models with six generative LLMs. When no option is correct, generative LLMs still pick one in 38-86% of items; the most cautious decision model does so in 30%. On standard CaseHOLD, a decision model matches the best generative LLM (88.1% vs. 88.2%) at 1/55 of the cost. Accepting that LLM's answer only when a decision model agrees halves its wrong citations, and distilling a decision model's probabilities into Qwen3-8B, without labels, yields a well-calibrated local model. Legal AI should be judged by its wrong citations, not by accuracy alone.","authors":[{"name":"M. Mikail Demir","position":0}],"categories":["cs.AI","cs.CL","cs.CY"],"primaryCategory":"cs.AI","publishedAt":"2026-10-21T16:17:49Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29961"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29961"}],"brief":{"text":"System One decision models can match the best generative LLM on CaseHOLD at far lower cost, and agreeing with a decision model halves wrong citations. Yet they also choose unsupported authorities when the correct option is missing, underscoring the need to measure abstention and citation errors.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":92,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29960","arxivId":"2610.29960","doi":"10.48550/arXiv.2610.29960","title":"Mastering Model Fit: Applications of the InterModel Vigorish for Interpretable Machine Learning","abstract":"Accuracy alone gives little information about which features or model components improve probabilistic fit, which outcomes benefit, or how fit changes during training. We introduce the InterModel Vigorish (IMV) into the Machine Intelligence space in order to express differences in fit as an expected betting advantage over a reference model. SHAP-IMV attributes gains to features through Shapley-weighted subset comparisons, Multi-IMV evaluates one-vs-rest and pairwise class discrimination, and Ablate-IMV assesses architectural components and feature groups by ablation. We evaluate these applications over ten random seeds using tabular data, images, sensor sequences, and graphs. Feature rankings vary across estimators. High one-vs-rest accuracy can accompany negligible improvement over a null, while ablation can preserve or improve fit under a fixed training budget. On six graph-classification benchmarks, we time-profile IMV over 200 training epochs, reporting seed means and ranges. The open-source imvpy library and accompanying notebooks provide code to replicate our work and to use the IMV more widely.","authors":[{"name":"Jiani Yan","position":0},{"name":"Pengyun Wang","position":1},{"name":"Klint Kanopka","position":2},{"name":"Ben Domingue","position":3},{"name":"Charles Rahal","position":4}],"categories":["stat.ME","cs.LG","stat.ML"],"primaryCategory":"stat.ME","publishedAt":"2026-10-21T16:17:17Z","version":1,"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29960"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29960"}],"brief":{"text":"InterModel Vigorish measures probabilistic fit gains as an expected betting advantage, with extensions for ranking features, comparing classes, and testing model components. Across varied data, it reveals that feature rankings depend on the estimator and high accuracy may offer little improvement over a null model.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":93,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29949","arxivId":"2610.29949","doi":"10.48550/arXiv.2610.29949","title":"Existence of Minimizers for the Gaussian Mumford-Shah Model via Covariance Floors on the Value Distribution","abstract":"The Gaussian Mumford-Shah model segments an image into regions by minimizing a continuous multivariate Gaussian negative log-likelihood term plus a perimeter penalty. It is used for scalar- and vector-valued images, in particular hyperspectral images, but the existence of a minimizer in general is not guaranteed. The energy is unbounded below whenever the covariance of a single region degenerates, regardless of the perimeter of the regions. A quantitative covariance floor on the covariances of all regions, or equivalently uniform Hölder continuity with exponent at most one of all one-dimensional marginals of the histogram, guarantees a lower bound on the energy and the existence of a minimizer. Our approach therefore shows that existence follows from a property of the histogram of the image, and we furthermore argue that scale-independent covariance floors cannot exist. Admitting a quantitative covariance floor is generic: it holds almost surely for all images with suitable noise. As applications of our theory, we show the convergence of regularized models used in practice and of numerical methods.","authors":[{"name":"Faian Conrad","position":0}],"categories":["math.ST","cs.CV","stat.TH"],"primaryCategory":"math.ST","publishedAt":"2026-10-21T16:12:20Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29949"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29949"}],"brief":{"text":"A covariance floor on each region’s values guarantees that the Gaussian Mumford–Shah energy has a minimizer; suitable noise makes this condition hold almost surely, and the result supports convergence of practical regularized models and numerical methods.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":94,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29938","arxivId":"2610.29938","doi":"10.48550/arXiv.2610.29938","title":"A PDE-theory-guided neural operator and the long-time relaxation of the two-dimensional anisotropic Boussinesq system","abstract":"We design a neural operator for the two-dimensional Boussinesq system with anisotropic dissipation on the periodic box, guided by a rigorous analysis of its long-time behavior. The operator maps a small perturbation of hydrostatic equilibrium, of amplitude $\\varepsilon$, to the vorticity at a late time. We prove that the flow relaxes to a stratified shear flow, with the horizontally averaged vorticity staying within $O(\\varepsilon^2)$ of its initial value $\\barω_0$ and the oscillation about it of order $\\varepsilon$. Accordingly, the network outputs $\\barω_0$ plus these two parts, each scaled to its size, which grows as $\\varepsilon^2$ and $\\varepsilon$, respectively. Without this guidance the training either does not converge or misses the $\\varepsilon^2$ growth. With it, one network reproduces, sample by sample on $2\\times10^3$ test runs over a factor of $23$ in amplitude, the quantity bounded by the theorem to a median error of $2.2$ percent. Being fast and differentiable with respect to the initial data, the network is also used to search the space of initial data, and with every candidate checked by the solver the search finds initial data for which this quantity is $4.4$ times the largest value among the test runs, raising the lower bound for the constant of the theorem by that factor.","authors":[{"name":"Akram Moustafa","position":0},{"name":"Jiahong Wu","position":1},{"name":"Xiaoming Zheng","position":2}],"categories":["math.AP","cs.NA","math.NA"],"primaryCategory":"math.AP","publishedAt":"2026-10-21T16:08:51Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29938"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29938"}],"brief":{"text":"A PDE-guided neural operator predicts late-time vorticity in the anisotropic Boussinesq system, matching theorem-scaled behavior with a median error of 2.2% across test runs. Searching with the network also finds initial data that raises the theorem’s lower bound by a factor of 4.4.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":95,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29934","arxivId":"2610.29934","doi":"10.48550/arXiv.2610.29934","title":"ToM2Vec: Continuous Theory-of-Mind Tokens for Multi-Agent Coordination","abstract":"Forming an accurate theory-of-mind (ToM) is key to successful multi-agent collaboration and requires reasoning under uncertainty to capture multiple possible partner states. This is hard to do in the text representation space native to large language model (LLM) agents, which more easily describe a single partner state. To address this challenge, we introduce ToM2Vec, which learns to form abstract theories-of-mind in a rich latent space via a fixed budget of continuous reasoning tokens. During training, supervision for these tokens comes from the partner's actual thoughts, which the ToM tokens are trained to abstract and match. At test-time, ToM tokens are generated by each agent and act as conditioners for that agent's thoughts and actions. Across several environments covering collaborative, competitive, and mixed-motive settings (Overcooked-AI, multiparty negotiation, Decrypto, Hanabi, and Collab Escape/Capture), we evaluate ToM2Vec's ability to support multi-agent interactions and collaboration, finding consistent improvements over the base model and a variety of competitive trained and prompted baselines. We show that latent reasoning is critical, with performance dropping when reasoning is expressed in text. Moreover, we show that ToM2Vec training generalizes to partner model families not seen at train-time, including a proprietary model (GPT-5.4). In our analysis, we provide evidence that the latent tokens encode a ToM, rather than task-specific information: ToM2Vec's tokens improve the agent's ability to directly answer questions about partner beliefs, desires, and next actions.","authors":[{"name":"Nithin Sivakumaran","position":0},{"name":"Ali Payani","position":1},{"name":"Hyunji Lee","position":2},{"name":"Mohit Bansal","position":3},{"name":"Elias Stengel-Eskin","position":4}],"categories":["cs.AI","cs.CL","cs.MA"],"primaryCategory":"cs.AI","publishedAt":"2026-10-21T16:05:54Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29934"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29934"}],"brief":{"text":"ToM2Vec trains agents to represent possible partner thoughts as continuous tokens, improving performance across collaborative, competitive, and mixed-motive tasks. Its tokens generalize to unseen partner models and help agents predict partners’ beliefs, desires, and actions.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":96,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29927","arxivId":"2610.29927","doi":"10.48550/arXiv.2610.29927","title":"Your Benchmark Is an Empirical Measure Over Difficulty","abstract":"A benchmark is not merely a collection of test questions; it naturally induces an empirical measure over difficulty. In this work, we develop this measure-based view of benchmark evaluation. This perspective allows us to identify and formalize two central failure modes: measure mismatch, where the benchmark difficulty distribution differs from the intended difficulty distribution, and coverage gaps, where uncovered regions along the difficulty axis prevent benchmarks from detecting meaningful localized training progress. We propose two diagnostic metrics that capture these two failure modes and provide rigorous theoretical guarantees. Empirically, we analyze 25 popular benchmarks spanning 7 domains using carefully estimated item difficulties. Our results show that many benchmarks implicitly overweight certain difficulty regions, so their leaderboard conclusions should be interpreted with caution; others leave substantial portions of the difficulty axis uncovered, limiting their ability to monitor the model improvement process. Finally, we turn diagnosis into action by introducing BenchFill, a simple difficulty-targeted item rewriting algorithm that effectively reduces local coverage gaps.","authors":[{"name":"Yifan Sun","position":0},{"name":"Naicheng Yu","position":1},{"name":"Jiawen Gong","position":2},{"name":"Jingyan Shen","position":3},{"name":"Huan Zhang","position":4}],"categories":["cs.LG","cs.AI","cs.CL"],"primaryCategory":"cs.LG","publishedAt":"2026-10-21T16:02:37Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29927"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29927"}],"brief":{"text":"Benchmark evaluation can be viewed as measuring performance across a distribution of difficulty. Analysis of 25 benchmarks finds that many overemphasize some difficulty levels or leave gaps, while BenchFill reduces local gaps by rewriting items to target specific difficulties.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":97,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29922","arxivId":"2610.29922","doi":"10.48550/arXiv.2610.29922","title":"Unintended Side Effects of Value Training in LLMs","abstract":"As LLMs are increasingly deployed in value-sensitive contexts, aligning them with human values has emerged as a post-training objective in NLP, aimed at instilling additional values and guiding model behaviour. However, current evaluations remain narrow. Because values are interconnected, it is unclear how fine-tuning for specific values alters unrelated capabilities and behaviours. In this work, we adapt four instruction-tuned LLMs across individual Hofstede's value dimensions and evaluate them on both cross-value transfer and broader behaviours. We find that value-specific training produces unintended side effects on untrained values, risk preferences, social cognition, and safety traits. Consequently, we suggest that value alignment after post-training should incorporate broader behavioral evaluations, and we highlight the need to investigate where additional value training might best fit into the pipeline.","authors":[{"name":"Siddhesh Pawar","position":0},{"name":"Chen Cecilia Liu","position":1},{"name":"Anna Korhonen","position":2},{"name":"Isabelle Augenstein","position":3}],"categories":["cs.CL","cs.AI","cs.LG"],"primaryCategory":"cs.CL","publishedAt":"2026-10-21T15:58:53Z","version":1,"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29922"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29922"}],"brief":{"text":"Value-specific fine-tuning across Hofstede’s dimensions altered untrained values, risk preferences, social cognition, and safety traits in four instruction-tuned LLMs, showing why alignment needs broader behavioral evaluations.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":98,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29910","arxivId":"2610.29910","doi":"10.48550/arXiv.2610.29910","title":"Worst-Case Finite-Horizon Analysis of Perturbed LQR via Performance Estimation Problem","abstract":"We propose an exact formulation to quantify the finite-horizon robustness of a Linear Quadratic Regulator (LQR) subject to norm-bounded Linear Time-Invariant (LTI) parametric uncertainties, using the Performance Estimation Problem (PEP) framework. Originally developed for optimization, this method allows us to reformulate the worst-case performance search as a rank-constrained Semidefinite Program (SDP). By applying operator interpolation, this formulation retains the sequence variables, allowing the worst-case to be explicitly reconstructed when a rank-one solution is found. This approach also opens new possibilities for the finite-horizon certification of broader system classes, such as data-driven models or nonlinear components. Using this framework, we observe that finite-horizon LTI perturbations do not simply follow asymptotic strategies. Instead, they align slowly decaying invariant manifolds with specific initial conditions to drive transient growth, causing a bifurcation in the worst-case behavior as the horizon length changes. Finally, we show numerically that our performance bounds sit between the upper bounds of Integral Quadratic Constraint (IQC) relaxations and a lower bound derived from a pseudospectral heuristic.","authors":[{"name":"Anthony Couthures","position":0},{"name":"Gianluca Bianchin","position":1},{"name":"Julien M. Hendrickx","position":2}],"categories":["math.OC","cs.SY","eess.SY"],"primaryCategory":"math.OC","publishedAt":"2026-10-21T15:49:58Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29910"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29910"}],"brief":{"text":"A Performance Estimation Problem formulation exactly characterizes worst-case finite-horizon LQR performance under bounded LTI uncertainty. It reveals horizon-dependent worst-case behavior driven by transient growth and yields bounds between IQC upper bounds and a pseudospectral lower bound.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":99,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29889","arxivId":"2610.29889","doi":"10.48550/arXiv.2610.29889","title":"Nashtopes and their A-discriminants","abstract":"An $n$-player game in normal form is said to be of format $\\boldsymbol{d} = (d_1, \\dots, d_n)$ if the $i$-th player has $d_i$ options for $1 \\le i \\le n$, and is said to be balanced if $d_n - 1 \\le \\sum_{i=1}^{n-1}(d_i - 1)$. The Nash equilibria of the game correspond to solutions of the system of indifference equations for the game, with payoff differences as coefficients and strategy probabilities as variables. As such, the mixed discriminant of this system gives some information about which games could have an unusual number of totally mixed Nash equilibria. This mixed discriminant can be regarded as an $\\mathcal{A}$-discriminant for some matrix $\\mathcal{A}$ in the sense of [Gelfand--Kapranov--Zelevinsky 1994], and the convex hull of the columns of $\\mathcal{A}$ is a polytope, which we call the Nashtope. We study this $\\mathcal{A}$-discriminant for each balanced game format with at least three players. This discriminant is a subset of the Nash discriminant variety introduced in [Abo--Portakal--Sodomaco 2026]. We compute the codimension and degree of this discriminant for all balanced game formats, including on-boundary formats, using the Euler obstruction formula in [Matsui--Takeuchi 2011]. Our $\\mathcal{A}$-discriminant locus corresponds to a special multigraded Hurwitz locus studied in [Pratt--Sodomaco--Sturmfels 2026]. As a consequence of [Pokidkin 2026], we obtain the codimensions and degrees of the irreducible components of the Nash discriminant variety for each balanced game format with at least three players, fully resolving Problem 3.8 of [Abo--Portakal--Sodomaco 2026].","authors":[{"name":"Elke Neuhaus","position":0},{"name":"Irem Portakal","position":1},{"name":"Luca Sodomaco","position":2},{"name":"Katherine Tung","position":3}],"categories":["math.CO","cs.GT","math.AG"],"primaryCategory":"math.CO","publishedAt":"2026-10-21T15:31:51Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29889"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29889"}],"brief":{"text":"For balanced games with at least three players, the study computes the codimensions and degrees of a discriminant linked to unusual numbers of mixed Nash equilibria, resolving a problem about the Nash discriminant variety.","source":"llm","model":"gpt-6-luna","generatedAt":"2026-10-22T08:36:18Z"}},"rank":100,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29872","arxivId":"2610.29872","doi":"10.48550/arXiv.2610.29872","title":"Recovering Plausible Transmission Outage Scenarios","abstract":"Expansion planning for extreme weather needs outage scenarios that specify which transmission branches fail. However, public outage records often only report customers without power by county, making it unclear how to exploit this information to create outage scenarios. We propose a method to recover sets of outaged branches that most closely match regional load shed data. This gives us sets of plausible outages for a given hazard event. We formulate a bilevel problem with a load shedding DC optimal power flow in the lower level. We relax the binary branch states to differentiate the lower level problem, replace it with its sensitivity matrix, and use gradient descent to update the upper level. We treat the relaxed branch states as probabilities and perform randomized rounding to restore integrality. On three test systems with synthetic outages, the recovered sets reduce the load shed error to $0.02-4.5\\%$. On the California Test System with storm data from December 2025, the error falls to $23\\%$. These scenarios give expansion planning a reliable source of outage data tied to observed hazards.","authors":[{"name":"Karen Ji","position":0},{"name":"Cameron Khanpour","position":1},{"name":"Samuel Talkington","position":2},{"name":"J. Kyle Skolfield","position":3},{"name":"Daniel K. Molzahn","position":4}],"categories":["eess.SY","cs.SY","math.OC"],"primaryCategory":"eess.SY","publishedAt":"2026-10-21T15:25:08Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29872"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29872"}]},"rank":101,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29871","arxivId":"2610.29871","doi":"10.48550/arXiv.2610.29871","title":"Rigorous integration of semilinear parabolic PDEs: forward evolution bounds and projected boundary value problems","abstract":"In this work, we present a rigorous integrator for initial and boundary value problems for semilinear parabolic PDEs. The method builds on earlier semigroup-based approaches while improving the bounds used in the fixed-point argument and extending the class of problems that can be treated. In particular, it controls the solution operator without requiring backward integration of the variational equation and allows projected boundary value problems to be studied through a simplified construction of an approximate inverse for the linearized operator. To illustrate the scope of the method, we present three applications to the two-dimensional Ohta--Kawasaki equation. For two initial value problems, we prove global existence and obtain explicit enclosures valid for all sufficiently large times around numerical approximations of stripe and spot equilibria. These results use a fixed-point formulation on an unbounded time interval, avoiding the construction of trapping regions. We also rigorously solve a projected boundary value problem with endpoint conditions chosen from linear approximations of the unstable and stable spaces of two saddle equilibria. This provides a step toward the rigorous study of heteroclinic saddle-to-saddle connections.","authors":[{"name":"Jean-Philippe Lessard","position":0},{"name":"Akitoshi Takayasu","position":1}],"categories":["math.NA","cs.NA","math.DS"],"primaryCategory":"math.NA","publishedAt":"2026-10-21T15:25:03Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29871"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29871"}]},"rank":102,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29867","arxivId":"2610.29867","doi":"10.48550/arXiv.2610.29867","title":"Swarmlet: a GPU-resident reformulated vortex particle method for coupled multirotor aerodynamics and flight dynamics","abstract":"We present Swarmlet, a GPU solver for the unsteady aerodynamics and flight dynamics of one or many multirotor vehicles. The wake is a meshless, evolving-core vortex particle field in the reformulated (rVPM) large-eddy formulation, coupled to rotor models of three fidelities, six-degree-of-freedom vehicles and flight control. Particles carry a smooth core of compact support whose field outside four core radii is exactly the singular Biot-Savart field, so the multipole far field of the treecode carries no regularization error. The subfilter closure is a dynamic eddy viscosity on a fixed filter width, applied by core spreading with a Lagrangian exchange of its local excess under an explicit stability bound, and the strengths are relaxed by projection; global corrections make both the exchange and the relaxation conserve total vorticity and linear impulse. All rotor models shed connected bound circulation, the disks along the arcs swept in each step however long, so the wake receives the lift impulse. The solver agrees with direct summation to $5\\times 10^{-7}$ and converges at third order in time. Without calibration, an actuator-line model that includes its own near sheet predicts the hover thrust of a DJI 9443 propeller within 2%. With calibrated section polars it reproduces a Crazyflie propeller's thrust and torque within 3.2% at three speeds and the near wake of a hovering quadrotor with relative $L^2$ differences of 0.08-0.21 against particle image velocimetry, while the wake keeps the delivered impulse. A closed-loop formation of 100 quadrotors, run at a coarser setting on one GPU, illustrates the coupled method at swarm scale.","authors":[{"name":"Wyatt James Horne","position":0}],"categories":["math.NA","cs.NA","physics.flu-dyn"],"primaryCategory":"math.NA","publishedAt":"2026-10-21T15:23:22Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29867"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29867"}]},"rank":103,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29818","arxivId":"2610.29818","doi":"10.48550/arXiv.2610.29818","title":"A Super-Convergent Quasi-Discontinuous Galerkin Method for Advection-Diffusion Problems","abstract":"The aim of the present study is to develop discontinuous Galerkin and nonconforming finite-element methods for solving the 2D advection-diffusion equation. The discrete formulations are stabilized by means of a family of numerical fluxes and the dispersion relations are constructed via a Fourier/stability analysis. The linear discontinuous P1DG and non conforming P1NC Galerkin discretizations are studied. A long time stability result is proven for the P1NC and P1DG schemes and fluxes examined here by inspecting the discrete frequencies, solutions of the dispersion relations. A super-convergent result is also demonstrated for the P1NC discrete frequency (O(h 4 ) accuracy) compared to the P1DG one (O(h 2 ) accuracy). These results are in line with those obtained in a previous paper for the shallow-water system.","authors":[{"name":"Daniel Y Le Roux","position":0}],"categories":["math.AP","cs.NA","math.NA"],"primaryCategory":"math.AP","publishedAt":"2026-10-21T14:44:06Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29818"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29818"}]},"rank":104,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29771","arxivId":"2610.29771","doi":"10.48550/arXiv.2610.29771","title":"Logic vs doctrines, a mirror symmetry","abstract":"This paper studies doctrines in categorical logic and their relations to fragments of geometric logic. After a brief account of the historical context for the notion of doctrine in categorical logic, we propose our working definition. We then show that there is a Galois connection between the poset of doctrines and the poset of fragments of geometric logic as defined in the framework of Kan injectivity in the 2-category of topoi, recently developed by the first and third authors. Concretely, the connection is realised by a general construction of syntactic categories and classifying topoi which encompasses all well-known examples in categorical logic. This yields a mirror symmetry, between the algebraic world of doctrines and the geometric world of topoi: to provide evidence of this phenomenon, we prove a few symmetric results relating doctrines and logics by means of syntactic categories and classifying topoi. We conclude with a discussion on the theme of conceptual completeness.","authors":[{"name":"Ivan Di Liberti","position":0},{"name":"Umberto Tarantino","position":1},{"name":"Lingyuan Ye","position":2}],"categories":["math.CT","cs.LO","math.LO"],"primaryCategory":"math.CT","publishedAt":"2026-10-21T14:02:54Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29771"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29771"}]},"rank":105,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29744","arxivId":"2610.29744","doi":"10.48550/arXiv.2610.29744","title":"On Almost MDS codes holding $(n-k-1)$-designs and an open problem on Witt design","abstract":"Let $\\mathcal{C}$ be an almost maximum distance separable (AMDS) code over the finite fields with parameters $[n,k,n-k]$ whose minimum-weight codewords support an $(n-k-1)$-$(n,n-k,λ)$ design. We first determine the parameters of the dual code $\\mathcal{C}^{\\bot}$. We also characterize the design and its complementary design supported by the minimum weight codewords of $\\mathcal{C}^{\\bot}$. In particular, we present explicit formulas for the weight enumerators of all subset codes of $\\mathcal{C}^{\\bot}$ with length at least $k+1$. These formulas are completely determined by the cardinality of the subset, the dimension, together with the parameter $λ$, and are independent of the specific choice of subsets. As an application, we answer an open problem posed by Wang, Tang and Ding in 2023 by proving that the $3$-design supported by the minimum-weight codewords in their proposed cyclic dual code is isomorphic to the complementary design of the Witt spherical geometry design.","authors":[{"name":"Huakai Wei","position":0},{"name":"Shiyuan Qiang","position":1},{"name":"Shaofang Hong","position":2}],"categories":["cs.IT","math.IT","math.NT"],"primaryCategory":"cs.IT","publishedAt":"2026-10-21T13:44:32Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29744"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29744"}]},"rank":106,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29743","arxivId":"2610.29743","doi":"10.48550/arXiv.2610.29743","title":"Weighted Risk Aggregation under Long-Memory Dependence: Pathwise Fluctuations of Exposure-Weighted Losses","abstract":"We model insurance loss in each period as deterministic exposure multiplied by a nonnegative function of a long-memory Gaussian sequence. A weighted law of the iterated logarithm yields the exact almost-sure scale and sharp constant of centered cumulative loss. Loss per unit of cumulative exposure converges almost surely to its mean, with an exact error scale common to all admissible exposure profiles; the sharp constant depends on the exposure exponent. Lognormal and quadratic examples, of Hermite ranks one and two, and a simulation illustrate the results. A known spectral formula expresses the quadratic coefficient through a leading eigenvalue.","authors":[{"name":"Elina Moldavskaya","position":0}],"categories":["math.PR","math.ST","stat.TH"],"primaryCategory":"math.PR","publishedAt":"2026-10-21T13:41:45Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29743"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29743"}]},"rank":107,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29733","arxivId":"2610.29733","doi":"10.48550/arXiv.2610.29733","title":"SAKURA: Serving Attention Key-Value State Under Rate Adaptation for Private Image Editing on Phones","abstract":"Diffusion transformers are widely used in image-editing services. To preserve both efficiency and privacy, existing systems commonly rely on cloud-device collaboration. Keeping request-specific inference on the device avoids sending personal images and prompts to the cloud. These workloads often use common templates, such as in face swapping and virtual try-on, enabling KV state to be precomputed and reused across requests. However, this reusable state is large and scales with the number of templates, making it impractical to store all of them on mobile devices. Storing them in the cloud addresses this capacity issue, but cache transfer is placed onto the inference path and can even be slower than on-device computation, especially under fluctuating mobile networks. We introduce SAKURA, a cloud-edge architecture that delivers reusable attention state to mobile devices under a bandwidth budget. It first uses a fine-grained token-importance-aware compression method to preserve image quality under limited transfer budgets. At runtime, SAKURA adaptively selects the cache representation for each upcoming shard based on network bandwidth. To enable efficient execution, we develop a pipelined runtime that overlaps cache transfer and reconstruction with on-device execution. We evaluate SAKURA using three editing tasks and two diffusion models, using a Snapdragon 8 Elite Gen 5 phone. SAKURA reduces latency by 1.82x and 3.27x over CacheGen adapted to DiT and on-device generation, respectively. It also reduces cache storage by 3.06x over uncompressed cache, while preserving image quality.","authors":[{"name":"Shuncheng Jie","position":0},{"name":"Desen Sun","position":1},{"name":"Ri Hong","position":2},{"name":"Yuyang Tian","position":3},{"name":"Hong Zhang","position":4},{"name":"Sihang Liu","position":5}],"categories":["cs.SE","cs.AI","cs.NI"],"primaryCategory":"cs.SE","publishedAt":"2026-10-21T13:35:39Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29733"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29733"}]},"rank":108,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29730","arxivId":"2610.29730","doi":"10.48550/arXiv.2610.29730","title":"Post-hoc Epistemic Uncertainty for Vision-Language-Action Policies","abstract":"Vision-language-action (VLA) policies need to recognize gaps in their training knowledge to request useful demonstrations or assistance in new situations, especially when failures are costly. Disagreement within an ensemble of independently trained policies can guide these decisions, but requires training and running multiple policies. We propose post-hoc kernel-variance heads that estimate epistemic uncertainty from the embeddings of a single policy. Each head computes the conditional variance of a Gaussian process conditioned on embeddings from the training demonstrations. Fitting requires no regression targets or gradient updates and leaves the policy unchanged. In experiments with SmolVLA on LIBERO-10, the kernel variance heads perform competitively with two-policy ensemble scores in task-level correlation with task success, in demonstration acquisition during active learning, and in referral to a stronger supervisor policy as a proxy for human assistance. For SmolVLA, the head on the vision-language backbone reduces uncertainty-scoring time by up to approximately a factor of 50 compared with ensemble scores, while also substantially reducing setup costs. We also show that the established evaluation protocol on LIBERO-10 primarily rewards recognition of coverage differences between tasks, while execution variability weakens assessment of within-task rankings. Thus, controlled perturbations provide a complementary evaluation: modality-specific heads identify unfamiliar camera, instruction, or state inputs, while fused heads can detect unfamiliar combinations of individually familiar inputs. With a supervisor proxy trained on the evaluated perturbation families, the decoder head improves within-task referral over ensemble disagreement. These findings show that useful uncertainty estimates for adaptive robot learning can be obtained from a single policy.","authors":[{"name":"Johannes Keller","position":0},{"name":"Albrecht M. Bloße","position":1},{"name":"Adrian Lindenmeyer","position":2},{"name":"Thomas Neumuth","position":3},{"name":"Daniel Schneider","position":4}],"categories":["cs.RO","cs.LG","stat.ML"],"primaryCategory":"cs.RO","publishedAt":"2026-10-21T13:34:03Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29730"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29730"}]},"rank":109,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29700","arxivId":"2610.29700","doi":"10.48550/arXiv.2610.29700","title":"S-PASS: Star-Topology Pinching-Antenna Systems with Pipelined Antenna Pre-Positioning","abstract":"Pinching-antenna systems (PASS) flexibly reconfigure the wireless channel by positioning dielectric antennas along a waveguide near each user. Releasing and re-attaching an antenna, however, is a mechanical operation orders of magnitude slower than electronic switching, wasting much transmission time. Existing designs generally assume instant reconfiguration, neglect the mechanical latency, or keep the antenna fixed. This paper proposes S-PASS, a star architecture of waveguides sharing one radio-frequency chain through a switch, and pipelined antenna pre-positioning (PAP), a time-division protocol in which idle waveguides pre-position their antennas for their next users while one transmits. It derives in closed form the scheduling efficiency, the number of waveguides beyond which the latency costs no transmission time, and the latency up to which PAP outperforms fixed placement. Analytical and numerical results show that this latency grows with each added waveguide, and users at the outer ends of the waveguides gain most.","authors":[{"name":"Wei Jiang","position":0},{"name":"Hans D. Schotten","position":1}],"categories":["cs.IT","eess.SP","math.IT"],"primaryCategory":"cs.IT","publishedAt":"2026-10-21T13:13:34Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29700"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29700"}]},"rank":110,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29687","arxivId":"2610.29687","doi":"10.48550/arXiv.2610.29687","title":"Wave Processing Unit: Closed-Form Matrix Synthesis for Wave Computing in Multiuser MIMO Communications","abstract":"Wave computing offers a compelling route to bypass the latency and energy bottlenecks of digital baseband processing. However, existing wave-domain processors rely on task-specific, non-convex iterative optimization, which incurs substantial configuration overhead and limits real-time applications. In this paper, we propose the wave processing unit (WPU), a programmable planar Mach--Zehnder interferometer (MZI) network that accelerates matrix--vector multiplications in MIMO communications. First, we model the tunable MZI element as a parameterized $2\\times2$ unitary transformation, and establish a closed-form synthesis framework for the WPU that decomposes an arbitrary unitary matrix into a cascade of MZI elements arranged in a compact rectangular topology. To handle arbitrary complex-valued operators, we further extend the method through singular value decomposition (SVD) and a hardware-efficient semi-unitary synthesis, avoiding redundant hardware and reducing the compilation complexity. We then apply the WPU to multiuser downlink precoding by an efficient WPU synthesis method, and compare the proposed architecture with hybrid and wave-domain ones. Numerical results demonstrate machine-precision synthesis with negligible residuals for up to $64\\times64$ operators, with approximately $380\\times$ faster configuration than iterative wave computing baselines. For 64-antenna, 8-user downlink precoding, WPU requires 87.5% fewer RF chains than the fully digital architecture while achieving the same sum rate under ideal hardware. These results highlight the potential of the WPU for precise, hardware-efficient wave-domain processing with rapid configuration.","authors":[{"name":"Xiaokun Teng","position":0},{"name":"Yanqing Ren","position":1},{"name":"Weicong Chen","position":2},{"name":"Wankai Tang","position":3},{"name":"Xiao Li","position":4},{"name":"Shi Jin","position":5}],"categories":["eess.SP","cs.IT","math.IT"],"primaryCategory":"eess.SP","publishedAt":"2026-10-21T13:00:16Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29687"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29687"}]},"rank":111,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29685","arxivId":"2610.29685","doi":"10.48550/arXiv.2610.29685","title":"Optimally Emulating Smooth Functions for Bayesian Analyses","abstract":"For Bayesian analyses aimed at constraining the parameter space of computationally expensive models by comparison with heterogeneous data sets, model emulators play a pivotal role. Model emulators enable searches through high-dimensional parameter spaces, searches which might otherwise require millions of full model runs. Here, an emulation strategy is presented based on the assumption that the model output can be well described by a polynomial expansion with all terms being independent of one another and with higher order terms being given decreasing weights according to a convergence parameter. It is found that if the expansion's form is constrained to being invariant to rotation and translation of the parameter space that the polynomial form becomes equivalent to that of a Gaussian process emulator using a correlation kernel with a Gaussian form. In the limit that the function is smooth, quantified by a convergence parameter, $Λ$, it is found that rather few training points are required to reproduce the full model and provide a good estimate of its uncertainty. An accuracy metric is defined based on an average of the expected uncertainty over the prior before training data are actually gathered. Analytic expressions for the expected accuracy are provided based solely on an expected value of $Λ$ and the location of the training points. This enables a tenable optimized placement of training points, which is shown to be significantly superior to random placement or that from Latin hyper-cube sampling. Examples of optimized placement are provided, and the accuracy is studied as a function of various variables, such as the number of model parameters, the number of training points, and the convergence parameter.","authors":[{"name":"Scott Pratt","position":0},{"name":"Oleh Savchuk","position":1},{"name":"Eren Erdogan","position":2},{"name":"Ekaksh Kataria","position":3},{"name":"Christal Martin","position":4},{"name":"Syed Afrid Jahan","position":5},{"name":"Frederi Viens","position":6}],"categories":["math.ST","stat.ME","stat.TH"],"primaryCategory":"math.ST","publishedAt":"2026-10-21T12:58:07Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29685"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29685"}]},"rank":112,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29679","arxivId":"2610.29679","doi":"10.48550/arXiv.2610.29679","title":"Lifting PCA in the Wasserstein Space","abstract":"Principal Component Analysis (PCA) is a fundamental tool for statistical dimension reduction, yet no canonical extension to distribution-valued data has emerged within the framework of optimal transport. While Wasserstein barycenters and Wasserstein least-squares regression provide canonical liftings of Euclidean averaging and linear regression in the Wasserstein space, an analogous principle for PCA has remained unexplored. In this paper, we introduce Lifted PCA, a canonical extension of Euclidean PCA to Wasserstein space, obtained through the framework of lifting functionals in the space of probability measures by convex extension. Starting from the classical variational formulations of PCA on the Stiefel and Grassmann manifolds, we derive two equivalent minimization problems over probability measures supported on these manifolds that correspond to Wasserstein barycenters with generalized costs. We establish an exact analogue of the equivalence between the Stiefel and Grassmann formulations of Euclidean PCA, preserving PCA's original variational structure under lifting. We further derive dual formulations and optimality conditions, and establish equivalences with multimarginal optimal transport and matching-for-teams problems. We also establish statistical consistency for empirical estimators and develop computational algorithms based on the Wasserstein barycenter formulation over the Stiefel manifold. Whereas Euclidean PCA summarizes vectors through projections onto an optimal low-dimensional linear subspace, Lifted PCA summarizes probability distributions through an optimal probability measure supported on multiple low-dimensional subspaces. Numerical experiments on distributional datasets illustrate the practical performance of the proposed methodology and its benefits for distributional data analysis.","authors":[{"name":"Jérémie Bigot","position":0},{"name":"Clément Bonet","position":1}],"categories":["stat.ML","cs.LG","stat.ME"],"primaryCategory":"stat.ML","publishedAt":"2026-10-21T12:57:22Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29679"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29679"}]},"rank":113,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29663","arxivId":"2610.29663","doi":"10.48550/arXiv.2610.29663","title":"Lineage-Aware Search Control for LLM-Based Evolutionary Factor Mining","abstract":"Formulaic alpha factor mining searches for symbolic expressions that transform market data into predictive signals. Recent LLM-based evolutionary factor mining methods use large language models as semantic variation operators. Under the conventional LLM-EA paradigm, stochastic variation and fitness-based selection can leave the search weakly directed and concentrated around a narrow set of structurally similar factors. We argue that these limitations arise because current methods mainly organize search at the individual and population levels, leaving lineage-level control underused. A lineage connects successive factor variants through evolutionary descent. We propose LineageEvo, which uses lineage in two coupled forms of search control: (1) lineage-conditioned LLM variation, which distills Lineage-Scoped Search Experience into operator-specific priors and fuses them with global operator priors, providing persistent, lineage-relevant guidance for LLM generation; and (2) lineage-stratified search allocation, which balances factor quality against lineage concentration by distributing the search budget across lineages during evolutionary search. Together, these mechanisms make semantic variation more direction aware and prevent branches from dominating the evolution, thereby promoting the discovery of predictive and diverse factors. We evaluate LineageEvo on CSI 500 and S&P 500 in terms of factor predictiveness, portfolio performance, search efficiency, and lineage- and expression-level diversity. LineageEvo achieves the strongest predictive performance on both markets, while LSPS produces a more balanced lineage distribution and a structurally broader factor pool. The results support lineage as a useful intermediate unit for controlling LLM-based evolutionary factor search.","authors":[{"name":"Xintong Liu","position":0},{"name":"Chenkai Zhang","position":1},{"name":"Yanzhao Ma","position":2},{"name":"Peng Yang","position":3}],"categories":["q-fin.CP","cs.AI","cs.LG"],"primaryCategory":"q-fin.CP","publishedAt":"2026-10-21T12:40:52Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29663"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29663"}]},"rank":114,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29630","arxivId":"2610.29630","doi":"10.48550/arXiv.2610.29630","title":"Space weather dependence of decameter-scale ionospheric irregularities spatial distribution","abstract":"This paper describes an approach to constructing nowcast and forecast predictive models for the occurrence of high-latitude decameter ionospheric irregularities based on SuperDARN and SECIRA radar data and space weather parameters. The method relies on self-learning pattern recognition in the dynamics of space weather parameters, that influence the irregularities formation. The proposed approach is a data-driven, multiscale method in which the spatial distribution of irregularities is approximated using a neural network of an original architecture, inspired by graph neural networks. The model was trained and tested using data from 11 radars during the period 2010-2025 (SuperDARN: CVW, ICE, ICW, INV, PGR, PYK, RKN, SAS, STO; SECIRA: EKB, MAGW). It is shown that the model achieves optimal performance when using geomagnetic coordinates MLT,MLAT. The optimal set of parameters for the best nowcast and forecast over studied set of parameters is a combination of geomagnetic and solar disturbance indices ($a_{p}$, $K_{p}$, $AE$, F10.7) from ground-based observations, and satellite-measured solar wind and interplanetary magnetic field parameters ($P_{sw}$, $E_{sw}$, $\\widetilde{B_{z}}$, $T_{sw}$). Hourly mean values for all the parameters were used in the analysis. The results indicate that acceptable forecast quality is achieved with a lead time of up to 7-9 hours, with the highest quality reached for nowcast (zero lead time). It is shown that the parameters $\\widetilde{B_{z}}$ and F10.7 are significant for nowcasting only and can be neglected for other lead times. It is shown that $AE$ and $T_{sw}$ are significant for mid-term forecast (\\textgreater 2hours ahead) and can be neglected for short-term predictions. Statistical relationships between the maximum latitude of the equatorial boundary of the irregularities and space weather parameters are also discussed.","authors":[{"name":"Oleg I. Berngardt","position":0},{"name":"Ivan A. Lavygin","position":1}],"categories":["physics.ao-ph","cs.LG","physics.geo-ph"],"primaryCategory":"physics.ao-ph","publishedAt":"2026-10-21T12:12:15Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29630"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29630"}]},"rank":115,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29617","arxivId":"2610.29617","doi":"10.48550/arXiv.2610.29617","title":"TernScope: Model-Hardware Co-Design for Efficient Ternary Endoscopic Segmentation","abstract":"Edge medical segmentation must reconcile small model size, high accuracy, and fast inference. TernScope addresses this tradeoff through model--hardware co-design, learning a compact segmentation network whose low-bit operators also simplify hardware execution. Split-channel large-kernel blocks preserve spatial context with limited computation, while branch fusion removes redundant inference paths. Ternary Progressive Quantization-aware Training (TPQ) then adapts the fused network from deep to shallow layers, through the decoder, and finally to the input and prediction layers. All 25 convolutions use ternary weights and 8-bit activations, enabling packed storage and multiplier-free computation. The resulting model uses 0.2483 M parameters and 0.0949 GMACs while retaining competitive segmentation accuracy on in-domain and external datasets. Under the same 45 nm technology, TernScope reduces compute-array area by 88.46% compared with the original Gemmini, a systolic-array accelerator with 8-bit $\\times$ 8-bit multipliers. These results demonstrate a path from accurate lightweight medical segmentation to compact low-bit hardware.","authors":[{"name":"Zhuoyu Wu","position":0},{"name":"Wenhui Ou","position":1},{"name":"Qiawei Zheng","position":2},{"name":"Xuhong Qiu","position":3},{"name":"Wenqi Fang","position":4},{"name":"Raphaël C. -W. Phan","position":5}],"categories":["eess.IV","cs.AR","cs.CV"],"primaryCategory":"eess.IV","publishedAt":"2026-10-21T11:56:23Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29617"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29617"}]},"rank":116,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29605","arxivId":"2610.29605","doi":"10.48550/arXiv.2610.29605","title":"Improved Training of Strong Stochastic Flow Maps","abstract":"Stochastic flow maps promise few-step sampling of stochastic differential equations (SDEs) by learning to directly jump between times along the solution path, yet training these models to learn this solution map remains difficult. The primary obstacle is the Brownian path itself: it is an infinite-dimensional object which needs to be passed to the neural network and is nowhere differentiable, which makes the implementation of derivative-based losses that power the state-of-the-art deterministic flow maps just out of reach. We identify two core design principles which address these issues: a) represent the Brownian path with a basis which carries higher-order local information, and b) train with an objective that forces the network to use the full representation of the path, not just the first order. To address this, we leverage the shifted Legendre polynomial expansion with a Wong-Zakai styled objective, unlocking derivative-based losses. Finally, we propose a recipe for fine-tuning large pre-trained flow models into strong stochastic flow maps. We ablate through this design space on several toy problems and use our findings to transform a large flow model into a strong stochastic flow map on ImageNet, achieving roughly a three-fold improvement in FD-DINOv2 over SSFM at four NFE.","authors":[{"name":"Zander W. Blasingame","position":0},{"name":"Antonio Franca","position":1},{"name":"Raul Minan","position":2},{"name":"Christian Koke","position":3},{"name":"Niklas Rindtorff","position":4},{"name":"Alexander Tong","position":5}],"categories":["cs.LG","cs.AI","math.DS"],"primaryCategory":"cs.LG","publishedAt":"2026-10-21T11:43:19Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29605"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29605"}]},"rank":117,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29559","arxivId":"2610.29559","doi":"10.48550/arXiv.2610.29559","title":"Certified Neural Approximation of Robust Nonlinear Model Predictive Control","abstract":"Model predictive control is a high-performance control method, but it incurs comparatively high online computational costs across many control tasks and is challenging to realize in low-level hardware settings. To address this, we present a framework for generating neural network controllers that provably inherit the safety and stability guarantees of a robust model predictive policy. In contrast to approaches based on post-verification or additional online safety filters, our method provides constructive, a priori design conditions that ensure preservation of recursive feasibility and stability before deployment. We treat the approximation error as a bounded input disturbance, which is addressed in the MPC design. This allows the neural network controller to directly inherit robustness from the underlying model predictive controller, which enables the transfer of guarantees to the approximation. The design process is an offline, constructive synthesis pipeline and can be implemented once Lipschitz bounds are specified. We further demonstrate in simulation that the learned controller reproduces the closed-loop behavior of the original controller while reducing online computation.","authors":[{"name":"Hendrik Alsmeier","position":0},{"name":"Maik Pfefferkorn","position":1},{"name":"Rolf Findeisen","position":2}],"categories":["eess.SY","cs.LG","cs.SY"],"primaryCategory":"eess.SY","publishedAt":"2026-10-21T10:46:55Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29559"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29559"}]},"rank":118,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29554","arxivId":"2610.29554","doi":"10.48550/arXiv.2610.29554","title":"Dynamical low-rank approximations for the optimal control of matrix ODEs","abstract":"We consider the open-loop optimal control of a dynamical system whose state is described by a time-dependent matrix $X(t)\\in\\mathbb{R}^{m\\times n}$. Gradient-based optimization methods may incur substantial computational and memory costs, particularly for large-scale systems and fine temporal discretizations. To alleviate this computational burden, we develop a mathematical framework to use Dynamical Low Rank Approximations (DLRA) in optimal control problems. We study two distinct paradigms: the first one, called Optimize-then-Project (OtP), first derives the optimality system and then approximates both the state and adjoint equations using DLRA. The second one, called Project-then-Optimize (PtO), first replaces the state equation with its DLR approximation, and then derives the optimality conditions for the resulting optimization problem. The two approaches differ fundamentally concerning the role of the adjoint variable. In OtP, the adjoint variable evolves on the manifold of fixed rank matrices, whereas in PtO, the adjoint variable belongs to the tangent space to the fixed-rank manifold at the evolving state variable. This latter formulation leads to a covariant ODE describing the dynamics of the adjoint variable as the underlying tangent space evolves in time. For both approaches we establish rigorous error estimates for the approximated gradients, identifying how their accuracy depends on the low-rank approximations of initial and terminal conditions, the components of the state and adjoint vector fields normal to the relevant tangent spaces, the stability properties of the dynamics, and the manifold curvature. PtO further requires tailored time discretization strategies for the adjoint covariant ODE, which we show to be robust against small singular values of the state variable. To illustrate the framework, we apply these methods to the optimal control of a random parabolic PDE.","authors":[{"name":"Benjamin Carrel","position":0},{"name":"Tommaso Vanzan","position":1}],"categories":["math.NA","cs.NA","math.OC"],"primaryCategory":"math.NA","publishedAt":"2026-10-21T10:41:40Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29554"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29554"}]},"rank":119,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29546","arxivId":"2610.29546","doi":"10.48550/arXiv.2610.29546","title":"Sequential Entropic Optimal Transport for Bayesian Inverse Problems via Multilevel Monte Carlo","abstract":"We consider the problem of Bayesian estimation associated to inverse problems governed by partial differential equations (PDEs). In most cases of practical interest, one must use numerical solvers to implement computational methods based on Monte Carlo methods. This inevitably leads to an adaptation of multilevel Monte Carlo (MLMC) methods to reduce computational efforts to achieve a pre-specified mean square error (MSE). MLMC is based on sampling coupled versions of Bayesian posteriors associated to increasingly precise numerical solves of the PDE. In this article we present a novel method to implement MLMC based upon sequential entropic optimal transport couplings learned with light Schrödinger bridges, and correct the transport with Metropolis--Hastings steps that preserve the pairing. We bound the mean square error associated to a version of our new estimation method and show, for an elliptic inverse problem, that to achieve a MSE of $\\mathcal{O}(δ^2)$ ($δ>0$) one can have an optimal cost of $\\mathcal{O}(δ^{-2})$. The results feature explicit bounds on the effect of the fitted transport and of the Metropolis--Hastings correction. From a numerical perspective, on elliptic, Darcy-flow and Navier-Stokes problems, the method reduces the variance of the most precise solver by a factor of five to ten and is faster than multilevel sequential Monte Carlo [arXiv:1503.07259] at tight accuracies.","authors":[{"name":"Jason Beh","position":0},{"name":"Pierre Del Moral","position":1},{"name":"Ajay Jasra","position":2}],"categories":["math.NA","cs.NA","stat.CO"],"primaryCategory":"math.NA","publishedAt":"2026-10-21T10:34:24Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29546"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29546"}]},"rank":120,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29531","arxivId":"2610.29531","doi":"10.48550/arXiv.2610.29531","title":"Phase Transitions in Learning Driven by Extremal Data","abstract":"We study an extremal regression problem, an example of multiple instance learning (MIL), where the predictors are given as multisets, called bags. The bag elements (called instances) possess hidden labels which are pooled to produce an observed bag-level label. Max pooling is common in MIL applications, but gives rise to vanishing gradients. To handle this, modern MIL pipelines approximate the max with a softmax, treating the sharpness as a learnable parameter, denoted $λ$. When the max pooling model is the true data generating process, learning will be successful when $λ$ is increased during gradient descent, thereby driving the dynamics into regions of parameter space with vanishing gradients. To understand the conditions under which gradient descent breaks down, we study a minimal model and derive a low dimensional projection of the gradient flow in the joint limit of large bags $n$ and high dimension $d$ with $α=\\ln n/d$ fixed. The flow has two distinct phase transitions: an intrinsic freezing transition related to the random energy model of disordered systems, and a retrieval transition occurring in modern Hopfield networks. Numerical simulation confirms both the phase diagram and the convergence of the finite-size training gradient to its predicted asymptotic form.","authors":[{"name":"Ethan Levien","position":0},{"name":"Trevor GrandPre","position":1}],"categories":["cs.LG","cond-mat.stat-mech","physics.data-an"],"primaryCategory":"cs.LG","publishedAt":"2026-10-21T10:22:18Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29531"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29531"}]},"rank":121,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29526","arxivId":"2610.29526","doi":"10.48550/arXiv.2610.29526","title":"Geometric knowledge spaces, Attention pairing and Architectures","abstract":"In this paper, we propose a geometric approach to deriving attention mechanisms and exploring potential new transformer architectures. We begin by formalizing the notions of knowledge worlds and attention pairings, then demonstrate how this framework naturally leads to the architectures of standard attention, causal attention, linear attention, tensor attention, and recurrent neural networks. Finally, we introduce several new geometric spaces that may inspire novel neural network architectures.","authors":[{"name":"Shang Xu","position":0}],"categories":["cs.LG","math.AG","math.GN"],"primaryCategory":"cs.LG","publishedAt":"2026-10-21T10:14:27Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29526"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29526"}]},"rank":122,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29523","arxivId":"2610.29523","doi":"10.48550/arXiv.2610.29523","title":"pplx-embed-v2-late Technical Report","abstract":"In this report, we introduce pplx-embed-v2-late, a pair of 0.6B and 9B multilingual and multimodal late-interaction models. Both models achieve state-of-the-art results within their respective size classes across a range of text, image, and visual document retrieval benchmarks. On visual document retrieval tasks, pplx-embed-v2-late-0.6b is competitive with models several times larger than itself. The models are trained via representation distillation from an 18B teacher. Consequently, they share an embedding space that supports asymmetric deployment: documents can be indexed with the 9B model and queried with the low-latency 0.6B model. Both models are available at https://huggingface.co/collections/perplexity-ai/pplx-embed-v2.","authors":[{"name":"Markus Krimmel","position":0},{"name":"Louis Milliken","position":1},{"name":"Ekaterina Glazkova","position":2},{"name":"Maximilian Schall","position":3},{"name":"Antoine Chaffin","position":4},{"name":"Sedigheh Eslami","position":5},{"name":"Jonas Golde","position":6},{"name":"Denis Bykov","position":7},{"name":"Claire Beamer","position":8},{"name":"Jeremy Yang","position":9},{"name":"Bo Wang","position":10}],"categories":["cs.IR","cs.CL","cs.LG"],"primaryCategory":"cs.IR","publishedAt":"2026-10-21T10:10:26Z","version":1,"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29523"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29523"}]},"rank":123,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29518","arxivId":"2610.29518","doi":"10.48550/arXiv.2610.29518","title":"Mitigating Query Shortcuts in Chain-of-Thought with Context-Aware Decoding","abstract":"While Chain-of-Thought (CoT) prompting improves the reasoning performance of large language models (LLMs), recent studies have shown that final answers do not always depend on the generated rationales. In particular, LLMs may rely on query-specific shortcuts rather than the information provided by their reasoning steps. In this paper, we investigate whether context-aware decoding (CAD), an existing contrastive decoding method, can mitigate such shortcut behavior when applied to CoT reasoning. We study this application as Rationale Contrastive Decoding (RCD), which contrasts the next-token distributions conditioned on the query with and without its rationale, amplifying tokens whose likelihood increases due to the rationale. We provide a theoretical interpretation of this decoding rule via a KL-regularized reward-maximization objective. Experiments across five reasoning benchmarks demonstrate that RCD improves answer accuracy over standard CoT. Furthermore, intervention analyses using ground-truth and random rationales reveal that RCD increases the behavioral dependence of final answers on the rationale content. Our findings suggest that contrastive decoding offers a simple, training-free approach to mitigating query shortcuts.","authors":[{"name":"Shin'ya Yamaguchi","position":0}],"categories":["cs.CL","cs.AI","cs.LG"],"primaryCategory":"cs.CL","publishedAt":"2026-10-21T10:05:25Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29518"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29518"}]},"rank":124,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29489","arxivId":"2610.29489","doi":"10.48550/arXiv.2610.29489","title":"Reorienting the Vector Axis: Restructuring FlashAttention for Long-Vector Processors","abstract":"A Transformer is built around attention, whose cost grows with the square of the sequence length and dominates the layer on long sequences. FlashAttention made that cost manageable by tiling the attention matrix and updating the softmax as it goes, but almost all work since has targeted GPUs. On long-vector processors one vector register can hold up to 512 single-precision numbers. The savings of a vector instruction are proportional to how much of the register it uses, so the choice of which loop to vectorize is a first-order decision. We port PyTorch's CPU FlashAttention kernel to a RISC-V CPU with 2KB vector registers and find that it reaches 16.2% of peak floating-point throughput. The kernel walks each tile one query row at a time: the product against V is vectorized along the 64-element head dimension, the softmax's per-row maximum and sum are reductions that collapse a full register to a scalar, and the software exponential's polynomial coefficients compete with tile data for vector registers and cause spilling. We propose two techniques. The first transposes the score tile and the running output so that queries lie along the contiguous axis. The algorithm and its values are unchanged. A single parameter, the number of queries in a tile, now sets the vector length of every part of the kernel and can be matched to the machine. Both reductions also disappear. The second moves the polynomial's coefficients into scalar registers, removing the spills and cutting the softmax's vector memory instructions by 78.7%. On silicon the two together cut whole-kernel cycles by 1.84 to 2.08x at a head dimension of 64, raising floating-point utilization from 16.2% to 29.7% of peak. An FPGA of the same design ranks the changes identically and scores them higher, at 2.71x.","authors":[{"name":"Emmanuel Irabor","position":0},{"name":"Marc Casas","position":1}],"categories":["cs.AR","cs.DC","cs.PF"],"primaryCategory":"cs.AR","publishedAt":"2026-10-21T09:45:14Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29489"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29489"}]},"rank":125,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29479","arxivId":"2610.29479","doi":"10.48550/arXiv.2610.29479","title":"Point Cloud Transmission With Patch-Based Rate Adaptation","abstract":"We propose a joint source-channel coding (JSCC) framework for efficient wireless point cloud transmission. In the proposed scheme, the point cloud is first divided into local patches and transformed into compact latent representations, which are then mapped to symbols for transmission over a noisy channel. Our model incorporates an entropy model that exploits inter-patch spatial dependencies to estimate patch-level coding costs and allocate different numbers of symbols accordingly. Unlike existing patch-based point cloud JSCC methods that use the same transmission rate between patches, we adapt the channel usage of each patch latent representation. To exploit inter-patch dependencies without patch-by-patch autoregressive processing, we select a group of anchor patches whose features provide contextual information for entropy estimation of non-anchor ones. The framework is trained using a reconstruction loss and an entropy-based rate surrogate. Experiments on ModelNet40 demonstrate overall rate-distortion improvements over existing baselines across channel conditions.","authors":[{"name":"Shipeng Liu","position":0},{"name":"Garima Bazard","position":1},{"name":"Khac-Hoang Ngo","position":2},{"name":"Per-Erik Forssén","position":3}],"categories":["cs.IT","eess.SP","math.IT"],"primaryCategory":"cs.IT","publishedAt":"2026-10-21T09:34:27Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29479"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29479"}]},"rank":126,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29476","arxivId":"2610.29476","doi":"10.48550/arXiv.2610.29476","title":"Generative Modeling via Nonlinear Schrödinger Bridges and Branching Diffusions","abstract":"We propose a new generative modeling framework grounded in forward-backward partial differential equations (PDEs). These equations arise naturally in mean-field game (MFG) theory, and are closely related to nonlinear Schrödinger systems. Our main contribution is to reinterpret this connection within the setting of generative modeling and to develop an efficient numerical methodology for its practical implementation. Specifically, we derive and characterize a nonlinear extension of the classical Schrödinger bridge problem through a coupled system of PDEs, yielding a model we call SBNL (Schrödinger Bridge NonLinear). To compute its solution, we design a simulation algorithm based on branching diffusions and the probabilistic representation of semilinear PDEs. The resulting method is embarrassingly parallel, scalable to high dimensions, and applicable to a broad class of nonlinear interaction functionals. We validate the approach through numerical experiments, including recovery of the classical Schrödinger bridge, nonlinear extensions with convex penalties (Gross-Pitaevskii) as well as potential interaction terms, and high-dimensional generative tasks such as image synthesis on the MNIST and FFHQ human face dataset. Furthermore, the same framework embeds score-based generative modeling, diffusion models, and tractable gaussian Schrödinger bridges.","authors":[{"name":"Mathias Grau","position":0},{"name":"Pierre Henry-Labordère","position":1}],"categories":["cs.LG","cs.AI","stat.ML"],"primaryCategory":"cs.LG","publishedAt":"2026-10-21T09:30:53Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29476"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29476"}]},"rank":127,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29472","arxivId":"2610.29472","doi":"10.48550/arXiv.2610.29472","title":"OmniSimulator: Aligning Small Language Models for Authentic Heterogeneous Behavior Modeling","abstract":"Large language models (LLMs) have shown great potential as general-purpose user simulators for interactive systems. Prior studies have demonstrated that training on high-quality data can substantially improve model reasoning and downstream task performance. However, the upper bound of user simulation is still constrained by the model's reasoning paradigm: native chain-of-thought often degenerates into shallow pattern matching and fails to adequately model the behavioral factors underlying real user decisions. In this paper, we propose OmniSimulator, a novel training framework designed to improve the predictive ability of small LLMs for realistic user simulation on short-video platform environments. Specifically, OmniSimulator organizes behavior-oriented reasoning through a structured schema with five dimensions, carefully selects data from real user-behavior scenarios, and introduces a teacher model to generate high-quality reasoning traces under the proposed schema, helping small models learn behavior-relevant reasoning patterns rather than relying on superficial behavioral correlations. Based on this process, we construct 9,881 high-quality training instances from three months of historical interaction trajectories of 200 real users, averaging about 50 distilled examples per user. Experimental results show that OmniSimulator improves the average score of Qwen3-8B by 34.33%, achieves performance competitive with Claude-Sonnet-4.5, and supports smaller-scale, self-hosted deployment. These results suggest that learning structured behavior-oriented reasoning patterns is a promising direction towards scalable and high-fidelity LLM-based user simulation.","authors":[{"name":"Jiawei Chen","position":0},{"name":"Ruoxi Xu","position":1},{"name":"Boxi Cao","position":2},{"name":"Zhuoqun Li","position":3},{"name":"Ruotong Pan","position":4},{"name":"Yunfei Zhang","position":5},{"name":"Zhirui Yang","position":6},{"name":"Jingyan Chen","position":7},{"name":"Tingting Gao","position":8},{"name":"Han Li","position":9},{"name":"Yaojie Lu","position":10},{"name":"Xianpei Han","position":11},{"name":"Le Sun","position":12},{"name":"Xiangyu Wu","position":13},{"name":"Hongyu Lin","position":14}],"categories":["cs.CL","cs.AI","cs.LG"],"primaryCategory":"cs.CL","publishedAt":"2026-10-21T09:30:12Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29472"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29472"}]},"rank":128,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29466","arxivId":"2610.29466","doi":"10.48550/arXiv.2610.29466","title":"Approximate Matrix Multiplication through Joint Leverage Score Sampling","abstract":"In this paper, we investigate approximate matrix multiplication through random sampling. Specifically, we propose using a non-uniform sampling distribution constructed from the leverage scores of the two matrices, and characterize the associated approximation error in terms of the matrix spectral norm; which is well-known to better characterize the downstream error when the matrix product operator is applied to an input vector. We provide the probability error bounds for the proposed algorithm are also given, and discuss connections to data-oblivious approximate matrix multiplication, and graph sparsification algorithms.","authors":[{"name":"Neophytos Charalambides","position":0},{"name":"Mert Pilanci","position":1},{"name":"Alfred O. Hero III","position":2}],"categories":["math.NA","cs.NA","eess.SP"],"primaryCategory":"math.NA","publishedAt":"2026-10-21T09:21:31Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29466"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29466"}]},"rank":129,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29456","arxivId":"2610.29456","doi":"10.48550/arXiv.2610.29456","title":"RegoMender: A Multi-agent Framework for Automated Rego Policy Repair","abstract":"Policy-as-Code (PaC) has emerged as a foundational paradigm in cloud-native systems. However, repairing Rego policies remains challenging due to their strict syntax, declarative semantics, and specific verification requirements. Existing automated program repair (APR) methods are mainly designed for mainstream languages and struggle to accommodate the unique policy logic and repair workflows of Rego. Thus, Rego policy repair remains a manual process driven by trial and error. To address this, we introduce RegoBench, the first automated Rego policy benchmark comprising 1,159 real-world defects, and propose RegoMender, a feedback-guided multi-agent framework. Collaborating via a shared blackboard and an atomic toolkit, RegoMender dynamically executes patching, rewriting, or rolling back, to resolve complex repair scenarios. Evaluated on RegoBench, RegoMender achieves syntactic and test-based semantic success rates of 82.9\\% and 77.9\\%, respectively, outperforming the strongest baseline by 13.4\\%. Unlike other approaches whose performance drops in complex scenarios, RegoMender maintains effectiveness across single-step and chained repair tasks. Furthermore, it generalizes strongly across HCL and SQL and achieves an 86.4\\% repair rate on 132 real-world CVE vulnerabilities, proving its robustness and practical utility for security.","authors":[{"name":"Ruijie Zhang","position":0},{"name":"Huaxiao Yin","position":1},{"name":"Li Hu","position":2},{"name":"Fanqi Kong","position":3},{"name":"Haozhe Liang","position":4},{"name":"Shaoyin Cheng","position":5},{"name":"Kun Zhang","position":6},{"name":"Bibo Tu","position":7}],"categories":["cs.SE","cs.AI","cs.PL"],"primaryCategory":"cs.SE","publishedAt":"2026-10-21T09:14:12Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29456"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29456"}]},"rank":130,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29445","arxivId":"2610.29445","doi":"10.48550/arXiv.2610.29445","title":"Deciding Scale Placement in MXFP4 Dot Products with Integer Arithmetic","abstract":"The Open Compute Project's microscaling (MX) family includes MXFP4, which stores 32 four-bit elements with a shared power-of-two scale. When lowering a dot-product reduction, a compiler can apply these scales before the reduction, or defer scaling until afterwards. Although equivalent over real numbers, these formulations can produce different IEEE binary32 results through overflow or underflow, with equivalence also depending on the reduction tree. We study the legality of compiler rewrites in either direction by deriving conditions under which the two formulations are bitwise equivalent. To decide rewrite legality, we construct an integer state machine that exploits the finite element alphabet and shared power-of-two scales, modeling overflow, NaN, and subnormal rounding. We prove its faithfulness under bounded scale codes and integer coefficients, yielding an exact encoding of the equivalence query in quantifier-free linear integer arithmetic (QF_LIA). For MXFP4, we derive a closed-form sufficient upper bound on the sum of scale exponents for any fixed binary reduction tree within the supported coefficient range. This bound depends on the size of the largest proper subtree. For four tree shapes at block lengths $k\\in\\{2,3,4,8,16,32\\}$ and two shapes at $k=64$, SMT finds counterexamples when this upper bound is increased by one exponent step. Against the standard direct encoding in quantifier-free floating-point arithmetic (QF_FP), on matched queries cvc5 proves the integer version in seconds but reaches a configured one-CPU-hour limit on the floating-point proof even at $k=2$. The implementation was cross-checked using an independent floating-point encoding, exhaustive and stratified binary32 execution, and GPU replay of counterexamples. An MLIR lowering prototype demonstrates selection between the two formulations using the certified conditions as static or runtime guards.","authors":[{"name":"Shreyas Vinayakumar","position":0}],"categories":["cs.LO","cs.AR","cs.PL"],"primaryCategory":"cs.LO","publishedAt":"2026-10-21T09:05:50Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29445"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29445"}]},"rank":131,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29425","arxivId":"2610.29425","doi":"10.48550/arXiv.2610.29425","title":"Generalization Bounds via Vector-Valued $L_p$ Correlation $V_α$","abstract":"Understanding generalization of randomized learning algorithms remains a central challenge. Mutual-information and total-variation bounds provide complementary ways to quantify data--hypothesis dependence, but they can behave quite differently in high-dimensional regimes. We study a one-parameter family of dependence measures, the vector-valued $L_p$-norm correlation $V_α$, and derive generalization bounds whose dependence terms interpolate between total variation (TV) at $α=1$ and variance and $χ^2$ related behavior at $α=2$. For $α>2$, the Hölder-conjugate loss moment is below two, yielding heavy-tailed specializations that can remain finite even when the loss variance is infinite. A particularly interpretable specialization occurs at $α=2$, where the bound can be expressed in terms of the variability of the algorithm's output density across datasets, $\\int \\sqrt{\\mathrm{Var}_S[p(w\\mid S)]}~ dw$. We show that the entire $V_α$ family is invariant under smooth bijective reparameterizations of the hypothesis space. We also introduce an adaptive density stability (ADS) condition that implies explicit control of $V_2$, and we verify it for Gibbs-type posteriors under bounded losses (yielding rates that depend on the inverse temperature and sample size). To address scaling, we extend the framework to a conditional formulation using supersamples, analogous in spirit to conditional mutual information (CMI), and discuss evaluations in the output space as a practical surrogate for estimating dependence. Simulations on Bayesian linear regression illustrate how tuning $α$ can change the tightness of the resulting bounds across regimes. Additional supersample experiments on noisy MNIST and on a pretrained ResNet-50 for CIFAR-10 show similar behavior of the projected loss-space bounds beyond the original clean-MNIST setting.","authors":[{"name":"Mohammad Hossein Momeni Hamaneh","position":0},{"name":"Mohammad Mahdi Mojahedian","position":1},{"name":"Mohammad Hossein Yassaee","position":2}],"categories":["cs.LG","cs.IT","math.IT"],"primaryCategory":"cs.LG","publishedAt":"2026-10-21T08:49:33Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29425"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29425"}]},"rank":132,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29413","arxivId":"2610.29413","doi":"10.48550/arXiv.2610.29413","title":"Beyond Small Talk: Modeling Disclosure Trajectories in Human-AI Conversation","abstract":"Recently, we see a rise in conversational agents as a form of social and emotional support, where disclosure will unfold over time. Yet, current evaluations of these agents typically examine isolated conversational states. In analyses of naturally occurring conversations, disclosure history and current disclosure are confounded which tends to make their effects difficult to isolate. We thus introduce Trajectory-based Disclosure Evaluation (TraDE), an evaluation framework grounded in Social Penetration Theory that utilizes predefined disclosure trajectories and evaluates how preceding disclosure history shapes model behavior under matched current disclosure. Through the use of TraDE, we demonstrate across multiple models and relational behavior benchmarks that responses to disclosures matched on depth can differ depending on the preceding trajectory, particularly in whether models accept or challenge the user's framing. We also observed that these differences track accumulated exposure to deeper disclosure rather than any single moment of vulnerability or the order of disclosures. It also tends to persists for several turns after disclosure returns to shallow levels. For several behaviors, we also demonstrate that models diverge most consistently from human supporters in the Emotional Support Conversation corpus (ESConv) not when users open up but rather when they step back in their disclosure. TraDE thus provides a systematic way to expose history-dependent relational behavior that evaluations of isolated conversational states would otherwise miss.","authors":[{"name":"Keith Kwok","position":0},{"name":"Wanting Ning","position":1},{"name":"Tej Deep Pala","position":2},{"name":"Desmond C. Ong","position":3},{"name":"S. Supraja","position":4},{"name":"Renwen Zhang","position":5},{"name":"Soujanya Poria","position":6}],"categories":["cs.CL","cs.AI","cs.HC"],"primaryCategory":"cs.CL","publishedAt":"2026-10-21T08:36:47Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29413"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29413"}]},"rank":133,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29398","arxivId":"2610.29398","doi":"10.48550/arXiv.2610.29398","title":"Region-Aware Multi-View Inspection Using Deep Learning and Domain-Informed Decision Fusion","abstract":"Automated optical inspection of precision components with tiny surface defects remains challenging when different regions of the same component exhibit different defect characteristics and, therefore, different acceptance criteria. This paper presents a region-aware, dual-model inspection framework that combines instance segmentation of localized seat-region defects with image-level classification of global shaft-region anomalies. Rather than treating these two model detections as sufficient evidence for component rejection, a deterministic decision fusion layer that formalizes industrial Standard Operating Procedures is applied to the model outputs to make the final component level decision. A PLC-synchronized imaging system acquires 18 seat and 18 shaft images over a full 360-degree rotation to provide multi-view circumferential coverage. The framework was evaluated on 380 automotive fuel injector needles, covering representative operational defect types, using identical model predictions in all configurations. Direct model-output aggregation via simple union logic (where any single-view detection flags a rejection) identified all 104 non-conforming components but produced a component level false rejection rate of 57.6%. Inclusion of a complete decision layer maintained the observed non-conforming (NOK) recall of 100% while reducing the false rejection rate to 10.9%, corresponding to an 81.1% reduction in false-positive component decisions. The results demonstrate that formalized manufacturing acceptance criteria can complement sensitive visual models by translating frame-level detections into practical and traceable component-level decisions.","authors":[{"name":"Srivishnu Ethirajulu Krishnaraj","position":0},{"name":"Rishikesh R. Chopade","position":1}],"categories":["cs.CV","cs.LG","eess.IV"],"primaryCategory":"cs.CV","publishedAt":"2026-10-21T08:29:12Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29398"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29398"}]},"rank":134,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29377","arxivId":"2610.29377","doi":"10.48550/arXiv.2610.29377","title":"Integrating Flow Matching and Model Predictive Control for Safe Robot Navigation","abstract":"Learning-based path planners based on diffusion and flow matching can generate diverse trajectories from demonstrations but typically lack mechanisms for constraint satisfaction and collision avoidance during deployment. This work proposes a framework that combines flow-matching-based path planning with model predictive path-following control to integrate data-driven path generation with real-time, constraint-aware execution. The flow-matching model captures multimodal path distributions, while the predictive controller adapts motion online to satisfy state and input constraints and avoid obstacles using updated local sensor information. An event-triggered re-planning scheme incorporates predictive control solutions into subsequent path generation, promoting collision-free navigation in environments with previously unseen obstacles. Simulation results for mobile robot navigation demonstrate robust performance across cluttered environments. The framework provides a systematic approach to combining flow matching with predictive control for adaptive, data-driven navigation.","authors":[{"name":"Philipp Holzmann","position":0},{"name":"Maik Pfefferkorn","position":1},{"name":"Joao Carvalho","position":2},{"name":"Ali Younes","position":3},{"name":"An Thai Le","position":4},{"name":"Georgia Chalvatzaki","position":5},{"name":"Jan Peters","position":6},{"name":"Rolf Findeisen","position":7}],"categories":["cs.RO","cs.SY","eess.SY"],"primaryCategory":"cs.RO","publishedAt":"2026-10-21T08:03:15Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29377"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29377"}]},"rank":135,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29362","arxivId":"2610.29362","doi":"10.48550/arXiv.2610.29362","title":"Compositional neural networks for multimodal X-ray and neutron scattering inversion","abstract":"X-ray and neutron scattering are indispensable for structural characterization across condensed-matter physics and materials science. Yet, because scattering experiments generally record intensities without phase information, distinct structures can yield similar scattering patterns, rendering the corresponding inverse problems non-unique. This ambiguity can be reduced by combining complementary measurements, but their integration into machine-learning frameworks remains challenging. Here we introduce a modular framework for probabilistic multi-measurement analysis based on prior-amortized flow matching. Single-observation posterior estimators are trained separately for individual observation types or physical probes and combined only at inference time through a compositional sampler. This late fusion approach allows pretrained single-observation models to be reused without dedicated joint retraining. As a case in point, we consider reflectometry, which provides depth-resolved, nanometer-scale information about thin films and interfaces. Our single-observation demonstrations span X-ray, time-of-flight neutron and polarized neutron reflectometry. In separate co-refinement studies, task-matched estimators enable homogeneous composition across multiple neutron contrasts and heterogeneous multimodal composition of X-ray and neutron measurements. Together, these results establish compositional flow matching as a modular approach to uncertainty-aware inversion and co-refinement, with potential applicability to other scattering problems in which multiple observations constrain a common physical structure while retaining measurement-specific parameters.","authors":[{"name":"Valentin Munteanu","position":0},{"name":"Alexander Hinderhofer","position":1},{"name":"Dmitry Lapkin","position":2},{"name":"Vladimir Starostin","position":3},{"name":"Frank Schreiber","position":4}],"categories":["cs.LG","cond-mat.mtrl-sci","physics.app-ph"],"primaryCategory":"cs.LG","publishedAt":"2026-10-21T07:54:26Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29362"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29362"}]},"rank":136,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29356","arxivId":"2610.29356","doi":"10.1016/j.conengprac.2026.107248","title":"Payload generalization on feedforward control actions for trajectory tracking in robots with rigid and elastic joints","abstract":"Industries employ robots for repetitive tasks due to their high precision and performance, typically achieved through feedforward control actions based on an accurate robot description. However, an exact dynamic model can be difficult to obtain, especially for robots with elastic elements or when the payload changes. Learning-based feedforward approaches are promising for trajectory tracking under model uncertainties, but typically face generalization issues when task parameters vary. This paper proposes a novel method to generalize the feedforward control action with respect to payload parameters, independently of the underlying controller. Firstly, the minimal number of example objects needed for payload generalization is derived. Then, an algorithm is proposed to execute the same trajectory with a new payload, given only its parameters, without knowledge of the robot's dynamic model and even in the presence of elastic joints. The effectiveness of the proposed algorithm is validated in simulations and experiments on tracking tasks with different robots and controllers. Two sets of experiments are performed: on a standard Franka Emika Panda robot, and on a Franka Emika Panda robot equipped with an elastic wrist.","authors":[{"name":"Giorgia Pazzi","position":0},{"name":"Giorgio Simonini","position":1},{"name":"Michele Pierallini","position":2},{"name":"Paolo Salaris","position":3},{"name":"Giorgio Grioli","position":4},{"name":"Manolo Garabini","position":5},{"name":"Franco Angelini","position":6}],"categories":["cs.RO","cs.SY","eess.SY"],"primaryCategory":"cs.RO","publishedAt":"2026-10-21T07:49:11Z","version":1,"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","venue":"Control Engineering Practice, Volume 178, January 2027, 107248","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29356"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29356"}]},"rank":137,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29303","arxivId":"2610.29303","doi":"10.48550/arXiv.2610.29303","title":"ExpEvolve: Structuring and Reusing Experience for Open-Ended Discovery","abstract":"Evolutionary coding agents such as AlphaEvolve are one of the notable successes in using Large Language Models (LLMs) to build AI Scientists. These agents tackle open-ended scientific problems by iteratively improving and evolving programs, leveraging the prior knowledge and reasoning capabilities of LLMs. Despite their success, mutations over programs are mostly low-level upon surface heuristics, without a persistent unit of control. In contrast, science compounds through ideas. In this work, we propose ExpEvolve, which injects idea-level extraction and control into evolution: it discovers and leverages the high-level factors from programs that explain scoring differences among different programs. We characterize theoretical conditions on when discovery with idea factors improves sample efficiency. We also build a practical framework to show that the extracted high-level idea factors are helpful for guiding exploitation from experience and exploration when discovery comes to a plateau. Across 11 open-ended mathematical tasks, ExpEvolve significantly improves the final performance and sample efficiency up to 5.0x. The improvements also transfer to 4 other LLMs and to 6 machine learning tasks from MLS-Bench. Analysis shows that ExpEvolve can discover idea-level factors to provide better exploration and exploitation for evolutionary discovery.","authors":[{"name":"Yongqiang Chen","position":0},{"name":"Chenxi Liu","position":1},{"name":"Deyu Zou","position":2},{"name":"Zhenhao Chen","position":3},{"name":"Tongliang Liu","position":4},{"name":"Bo Han","position":5},{"name":"Kun Zhang","position":6}],"categories":["cs.LG","cs.CL","stat.ML"],"primaryCategory":"cs.LG","publishedAt":"2026-10-21T07:00:23Z","version":1,"license":"http://creativecommons.org/licenses/by-sa/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29303"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29303"}]},"rank":138,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29300","arxivId":"2610.29300","doi":"10.48550/arXiv.2610.29300","title":"When Are We Near Nash? Certifying Solutions in General-Sum Games","abstract":"Heuristic solvers for general-sum extensive-form games produce a stream of iterates, and practice reports the NashGap of the returned profile. We ask what a solver can know about its own iterates from the regret accumulators and the terminal-distribution accumulators its updates already maintain. In two-player games the NashGap of a weighted average strategy equals, at the maximizing player, the weighted external regret plus a signed value correction. Bounding the regret by the running counterfactual-regret accumulator gives an online certificate whose cost per checkpoint is linear in the number of terminal histories. The certificate stays sound under alternating updates and in games with three or more players. The mismatch between what alternation evaluates and what the averages record enters as one signed comparator term, maximized in a single backward pass per player whose comparator weight is nonzero. The bound is read off the regret state, with no best response against the average. Run a portfolio of training-free regret-matching solvers, output the iterate with the smallest certificate, and verify that iterate once. On the benchmark's biased Leduc family a pool of sixteen training-free solvers, seven original and nine recent variants, solves $89\\%$ of the $64$ instances at threshold $10^{-5}$ this way, matching the trajectory-minimum union of the solvers. The same protocol solves $92\\%$ of the $64$ pre-registered held-out instances and $89\\%$ again on a third set of $64$ instances that no earlier stage used. The zero-best-response certificate certifies all $57$ in-sample solves and all $57$ fresh ones, and $58$ of the $59$ held-out ones. Its exact-comparator refinement certifies every solve in the three sets. The certificate states both and attributes its bound. Code is available at https://github.com/lbn187/CertNE.","authors":[{"name":"Boning Li","position":0},{"name":"Longbo Huang","position":1}],"categories":["cs.GT","cs.AI","cs.MA"],"primaryCategory":"cs.GT","publishedAt":"2026-10-21T06:53:56Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29300"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29300"}]},"rank":139,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29288","arxivId":"2610.29288","doi":"10.48550/arXiv.2610.29288","title":"Time-Uniform Dependent Central Limit Theory","abstract":"We establish a near-optimal almost sure invariance principle for nonstationary time series under the functional dependence measure (FDM) framework. This result yields a universal time-uniform dependent central limit theorem, which serves as the basis for anytime-valid inference. In particular, we instantiate the theorem through asymptotic confidence sequences (AsympCSs), time-uniform counterparts of asymptotic confidence intervals obtained via central limit theorems, extending those originally proposed in Waudby-Smith et al. (2024) to time series with serial dependence. We further refine their definition because, as we show, it can admit AsympCSs that fail to achieve the desired asymptotic time-uniform coverage for a broad class of data-generating processes, including independent and identically distributed cases. We address this issue by adding a simple eventual-inclusion condition, which in turn permits more flexible variance estimation, particularly useful in time-series settings. For feasible implementation of AsympCSs, we establish strong consistency of HAC variance estimators under the FDM framework, accommodating a wider range of commonly used bandwidth choices than existing results under mixing conditions. As illustrations, we construct AsympCSs for algorithms with Markovian structure and time-series-based causal estimands.","authors":[{"name":"Mingxuan Ge","position":0},{"name":"Yanlong Liu","position":1}],"categories":["math.ST","stat.ME","stat.TH"],"primaryCategory":"math.ST","publishedAt":"2026-10-21T06:41:56Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29288"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29288"}]},"rank":140,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29286","arxivId":"2610.29286","doi":"10.48550/arXiv.2610.29286","title":"Binder migration across film consolidation and pore emptying in electrode drying: a continuous model validated against drying-rate, thickness, and schedule experiments","abstract":"Drying battery electrodes faster saves energy but drives dissolved binder toward the film surface, depleting the current-collector interface and weakening adhesion. Models of this migration stop convecting the binder when film consolidation ends, and models of the subsequent pore emptying omit the binder or start from an assumed binder field, so no simulation has followed binder convection through both stages that a drying schedule acts on. This study presents a continuous simulation of the dissolved binder field across both stages, driven by the published drying rates and completed by effective descriptions of the two questions the base transport physics leaves open: where the liquid of the consolidating film carries the binder, and when the binder stops moving as the pores empty. With a single parameter set, the simulation reproduces three types of published measurements: the change of the surface-to-interface binder ratio with drying rate over a factor of 5.6 and the surface and interface binder levels of films six times thicker; concentration profiles through the film interior; and the response of the interface binder to the time at which the drying rate is switched mid-process. Applied without further adjustment to a published three-stage schedule, the model reproduces the shortened drying and the full interface recovery when the rate is lowered early, and shows that its final high-rate stage begins after the binder has stopped. Combined with a coater-scale process model, the model can be used to identify operating points for electrode production that improve energy efficiency and product quality together.","authors":[{"name":"Younghwan Joo","position":0},{"name":"Sanghyun Oh","position":1},{"name":"Sung-il Kim","position":2}],"categories":["cs.CE","cond-mat.mtrl-sci","physics.app-ph"],"primaryCategory":"cs.CE","publishedAt":"2026-10-21T06:37:13Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29286"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29286"}]},"rank":141,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29284","arxivId":"2610.29284","doi":"10.48550/arXiv.2610.29284","title":"Support periods of polynomials and support-primitive decomposition of constacyclic codes over Galois rings","abstract":"Let \\(A=\\GR(p^e,r)\\) have residue field \\(\\F_q\\), and let \\(λ\\in A^\\times\\). We relate coefficient compression of monic divisors of \\(X^N-λ\\) to support-primitive decomposition of constacyclic codes. At a simple-root length, the support period of every nonconstant monic divisor is preserved at each coefficient precision and equals the order of the multiplicative stabilizer of its root set. For a cyclic Frobenius-orbit polynomial, the multiple equal-difference structure determines its support period, core degree, and least relative degree of an unramified extension over which all its basic-irreducible factors are binomials. Coordinate projections give every constacyclic code over a finite commutative ring a unique maximal interleaving into copies of one support-primitive code. At simple-root lengths over \\(A\\), the intrinsic period of a code is the greatest common divisor of the periods of its torsion codes. The decomposition yields distance and duality formulas and the arithmetic Singleton bound for free codes. At every length, a nonzero code over \\(A\\) has the same minimum Hamming distance as its highest torsion code, which gives a refined bound. At repeated-root lengths, identical torsion filtrations can occur with different intrinsic periods, while a class of nonzero codes obtained by substitution from simple-root generators retains the same core and minimum distance. Trace matrices recover the coefficients of basic-irreducible cores using unit pivots, and local Hensel corrections lift compressed factors directly. For \\(X^n-1\\) with \\(p\\nmid n\\), lifting and output have quasi-linear bit cost in the compressed representation when the arithmetic data and residue factorizations are given.","authors":[{"name":"Li Zhu","position":0},{"name":"Hongfeng Wu","position":1}],"categories":["cs.IT","math.IT","math.NT"],"primaryCategory":"cs.IT","publishedAt":"2026-10-21T06:37:01Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29284"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29284"}]},"rank":142,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29275","arxivId":"2610.29275","doi":"10.48550/arXiv.2610.29275","title":"A proof of Ward's conjecture on Griesmer codes","abstract":"Let $C$ be an $[n,k,d]_q$ Griesmer code, where $q=p^f$ with $p$ prime, and write $e=ν_p(d)$. Ward conjectured that $C$ is $p^{\\max\\{0,e-f+1\\}}$-divisible. In this paper, we prove this conjecture.","authors":[{"name":"Haihua Deng","position":0},{"name":"Weiye Zhang","position":1}],"categories":["math.CO","cs.IT","math.IT"],"primaryCategory":"math.CO","publishedAt":"2026-10-21T06:26:54Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29275"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29275"}]},"rank":143,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29225","arxivId":"2610.29225","doi":"10.48550/arXiv.2610.29225","title":"Beyond Binary Labels: Diagnostic Evaluation of LLM Safety with Prompt-Level Rubrics","abstract":"Current LLM safety benchmarks predominantly rely on binary safe or unsafe labels, lacking explanatory depth and suffering from low consistency across automated judges. To address this, we introduce DiagSafe, a diagnostic evaluation framework that moves beyond binary classification by employing prompt-level rubrics. Grounded in globally applicable AI safety policies, each rubric provides explicit criteria along two dimensions: safety, which targets concrete harms with specific exemptions to suppress false positives, and usefulness, which independently evaluates instruction fulfillment. To ensure a robust and realistic evaluation, we construct a bilingual (English-Chinese) benchmark grounded in a comprehensive 84-label safety taxonomy, consisting of two subsets: DiagBase, which covers single- and multi-turn prompts with direct harmful queries and template-based jailbreaks, and DiagAttack, which embeds harmful intents into complex instruction-following tasks to form challenging instruction attacks. Extensive experiments demonstrate that DiagSafe significantly enhances the reliability of LLM-as-a-judge evaluation, achieving 95.34% agreement with human annotations -- substantially higher than the 69%-87% achieved by the baselines -- and high inter-judge consistency (Cohen's $κ$ = 0.864). Furthermore, applying our framework reveals that leading LLMs differ markedly in their unsafe rates, and each model also exhibits unique vulnerabilities along different dimensions. In addition, we observe that models are particularly susceptible in Chinese and multi-turn contexts. Ultimately, DiagSafe provides a reliable and explainable tool for auditing and optimizing LLM alignment.","authors":[{"name":"Hongyu Li","position":0},{"name":"Xinlong Huang","position":1},{"name":"Zhan Yu","position":2},{"name":"Jing Zhao","position":3},{"name":"Zhe Zhao","position":4},{"name":"Juntao Zhong","position":5},{"name":"Xiong Kuang","position":6},{"name":"Rui Yuan","position":7},{"name":"ZhenXiang Yan","position":8},{"name":"Huilin Xu","position":9},{"name":"Haitao Yu","position":10},{"name":"Lilin Wang","position":11},{"name":"Rui Chen","position":12},{"name":"Blaze Chen","position":13},{"name":"Fengzong Lian","position":14},{"name":"Renfen Hu","position":15}],"categories":["cs.CL","cs.AI","cs.CR"],"primaryCategory":"cs.CL","publishedAt":"2026-10-21T05:33:50Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29225"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29225"}]},"rank":144,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29196","arxivId":"2610.29196","doi":"10.48550/arXiv.2610.29196","title":"A Unified Evaluation Framework of Physical Adversarial Robustness in the Edge VLM Era","abstract":"Recent advancements in Vision-Language Models (VLMs) enable end-to-end visual grounding via language prompts and generative decoding, demonstrating largely improved zero-shot and open-vocabulary detection capabilities. As a result, VLMs are rapidly being deployed at the edge in safety-critical real-world systems. However, the physical adversarial robustness of edge VLMs has not been systematically studied, and the fundamental differences in output with modern object detectors create methodological challenges for fair evaluation. To compare the two families directly, we evaluate every selected representative model on one outcome: whether an adversarial patch worn on a person evades detection of the model. We report this by introducing a unified metric, conditional attack success rate (cASR), counted only on images the model still localizes under a matched gray occluder, and partition every failure into omission (no box), substitution (a box on the patch), and displacement (a box on neither). We benchmark the physical adversarial vulnerability of compact grounding VLMs against a range of object detectors, and find that the VLMs can be as vulnerable as the detectors. We disclose that failure mode is largely associated with the output format of a model rather than its architecture family. Of the three, only omission is visible to a downstream system; displacement and substitution return confident, well-formed boxes, making them potentially more dangerous failures. Further, we find that, under distribution shift, VLMs that primarily failed by omission instead shift to failing more by displacement/substitution. Thus, a grounding VLM can inherit the detector's vulnerability while showing the possibility to adopt more dangerous failure modes. Our code and data are available at https://github.com/ericliang1/physical-adversarial-robustness.","authors":[{"name":"Eric Liang","position":0},{"name":"Alan Luo","position":1},{"name":"Kaiwen Yuan","position":2}],"categories":["cs.CV","cs.CR","cs.LG"],"primaryCategory":"cs.CV","publishedAt":"2026-10-21T04:38:02Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29196"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29196"}]},"rank":145,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29192","arxivId":"2610.29192","doi":"10.48550/arXiv.2610.29192","title":"Curvature-Scaled Sharpness-Aware Minimization: Convergence Beyond Smoothness","abstract":"Sharpness-Aware Minimization (SAM) has established itself as a widely used method for training deep neural networks. Despite its popularity, existing convergence guarantees for SAM-type methods require smoothness, a restrictive assumption. Generalized smoothness conditions have recently been proposed in the analysis of gradient methods but remain underexplored for SAM. To analyze the effect of SAM's perturbation under generalized smoothness, we use relative gradient distortion to quantify how the sharpness-aware direction departs from the original gradient. Under $(L_0,L_1)$-smoothness, we show that this distortion can become unbounded near stationarity for SAM and at large gradients for its unnormalized variant (USAM), with examples where SAM cycles and USAM diverges. We then establish a necessary condition for uniform distortion control among perturbations along the gradient whose nonnegative multipliers depend only on its norm. Motivated by this condition, we propose Curvature-Scaled SAM (CS-SAM), which scales both the perturbation and the descent step by suitable curvature upper bounds, and prove that it uniformly bounds the distortion. Under both $(L_0,L_1)$- and $(H_0,H_1)$-smoothness, we establish deterministic $\\mathcal{O}(1/T)$ rates for the best squared gradient norm on nonconvex objectives, last-iterate $\\mathcal{O}(1/T)$ rates for the function gap on convex objectives with a minimizer, and linear convergence under the Polyak-Lojasiewicz condition. With zero perturbation, our guarantees recover the rates for gradient descent with $(L_0,L_1)$- or $(H_0,H_1)$-based step-sizes. Minibatch experiments on computer vision tasks demonstrate competitive performance for both variants, with the $(H_0,H_1)$ variant achieving the highest mean best-test accuracy in all four settings against tuned SAM and USAM with constant and cosine annealing schedules.","authors":[{"name":"Dimitris Oikonomou","position":0},{"name":"Nicolas Loizou","position":1}],"categories":["math.OC","cs.LG","stat.ML"],"primaryCategory":"math.OC","publishedAt":"2026-10-21T04:26:54Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29192"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29192"}]},"rank":146,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29187","arxivId":"2610.29187","doi":"10.48550/arXiv.2610.29187","title":"A Decade of Learning-Based Control: From Theoretic Foundations to Practical Deployment","abstract":"Motivated by ten years of Learning-Based Control Invited Sessions at the IEEE Conference on Decision and Control (CDC), this tutorial paper reviews key advances at the intersection of machine learning and control. We focus on methods that combine the flexibility of machine learning techniques with the structure and theoretical foundations of control to design autonomous systems that are capable, reliable, and safe. We first introduce core notions for establishing stability and safety guarantees, then review approaches that learn dynamics models to improve control under uncertainty, and finally discuss data-driven methods that directly synthesize or learn policies, including reinforcement learning, imitation learning, and Bayesian optimization. We further discuss representative applications and open challenges. The goal is to provide an accessible entry point to the field and to support continued progress toward reliable learning-based control in real-world systems. To support learning and practical exploration, we share implementations and illustrative examples along with our discussion, together with additional resources on our website: https://learning-based-control.github.io/tutorial/.","authors":[{"name":"Angela P. Schoellig","position":0},{"name":"Matthias A. Müller","position":1},{"name":"Sebastian Trimpe","position":2},{"name":"Melanie N. Zeilinger","position":3},{"name":"Mahathi Anand","position":4},{"name":"Andrea Carron","position":5},{"name":"Oliver Hausdörfer","position":6},{"name":"Victor G. Lopez","position":7},{"name":"Friedrich Solowjow","position":8},{"name":"Haocheng Zhao","position":9},{"name":"Siqi Zhou","position":10}],"categories":["eess.SY","cs.LG","cs.SY"],"primaryCategory":"eess.SY","publishedAt":"2026-10-21T04:18:47Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29187"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29187"}]},"rank":147,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29178","arxivId":"2610.29178","doi":"10.48550/arXiv.2610.29178","title":"Correlated Failures of AI Approval Monitors","abstract":"Before an AI agent takes a consequential action, such as paying, messaging or changing code, several monitors check it. Their verdicts are combined by aggregating scores or by counting approvals, whatever inputs the monitors share. Monitors that share a base model or a poisoned input, however, may fail together. We borrow the fail-prone systems of distributed computing: lists of monitor sets that may fail together. A rule is safe if it executes no bad action and live if it blocks no good one. We measured twelve small open-weights monitors on payment approval under two pre-registered protocols. Without pre-registration, we also measured three closed frontier models (Claude Opus 5.5, GPT-6 Astra, Gemini 4 Argon) and the open DeepSeek-V3.2. Every reliable fooling of a monitor into approving a fraud came through an evidence source and reached only that source's readers. Fooling it into blocking a correct payment depended on injection resistance, a monitor's refusal to follow instructions planted in its data. Small models and DeepSeek obeyed such notes. The three closed models ignored them in all but one case, and Opus also resisted persuasion tailored to it. On new data, a rule that counts sources instead of votes blocked 0.9% of correct payments, against 7.4% for the best counting rule, at equal safety. The pre-registered worst-case test failed. One attack left the approvers on a single source, so no rule over the twelve language-model monitors survives every measured attack. Adding a rule engine, which cannot be persuaded, restored the condition for a safe and live rule on the measured attacks. Diversify the inputs, harden the monitors, count sources, not votes.","authors":[{"name":"Luca Zanolini","position":0}],"categories":["cs.CR","cs.AI","cs.DC"],"primaryCategory":"cs.CR","publishedAt":"2026-10-21T04:09:37Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29178"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29178"}]},"rank":148,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29168","arxivId":"2610.29168","doi":"10.48550/arXiv.2610.29168","title":"How Representational Priors Shape Generalization in World Action Models","abstract":"World action models (WAMs) are shown to substantially diverge under generalization performance based on their generative backbones. We investigate this gap through the representational geometry that WAMs inherit from their backbones and how policy training reorganizes it. Our central hypothesis is that a useful representational prior would structure geometry that separates nuisance information from the in-distribution manifold in advance of training as a policy. We quantify this with nuisance orthogonality; orthogonality of the feature under visual variation outside the empirical in-distribution manifold. Across various image and video backbones, we show that a higher orthogonality from the representational prior is associated with stronger robustness to visual variations after WAM training, enabling policy generalization measurable prospectively for world models. In addition, analyzing the resulting robust WAMs reveals nuisance suppression; relative suppression of feature displacement caused by such variations and robust action generation of the policy. Together, these findings provide a geometric perspective on how the representational prior of the world model shapes the generalization performance of WAMs.","authors":[{"name":"Junha Chun","position":0},{"name":"Youngjoon Jeong","position":1},{"name":"Taesup Kim","position":2}],"categories":["cs.RO","cs.AI","cs.CV"],"primaryCategory":"cs.RO","publishedAt":"2026-10-21T03:59:35Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29168"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29168"}]},"rank":149,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29167","arxivId":"2610.29167","doi":"10.48550/arXiv.2610.29167","title":"TeachBack: Learning to Teach via Bidirectional Preference Optimization for On-Policy Self-Distillation","abstract":"On-policy self-distillation (OPSD) enables a language model to learn from a Teacher instantiated from the same model but given access to privileged information, such as a reference solution. We find that access to sufficient privileged information allows the Teacher to generate highly accurate trajectories; however, accuracy alone does not guarantee that these trajectories are suitable for Student learning. Direct access to a complete solution may encourage shortened, reference-conditioned derivations that omit the exploration, verification, and self-correction required to solve problems without privileged information. We introduce TeachBack, a bidirectional preference-learning framework for on-policy self-distillation that learns not only from the Teacher but also how the Teacher should teach. TeachBack shapes the Teacher's generation behavior by controlling the length of the reference-reasoning suffix injected into the prompt and by varying the prompt style. While maintaining comparable Teacher accuracy, it encourages the Teacher to explore and re-solve problems, thereby reducing its reliance on the reference solution. At each rollout refresh, the shared policy generates both a privileged Teacher trajectory and a problem-only Student trajectory. A pairwise preference objective uses privileged Teacher trajectories to improve Student learning, while the reversed preference uses the Student's behavior to shape future Teacher supervision under the privileged condition. Experiments on mathematical and logical reasoning tasks show that TeachBack consistently improves Student performance over existing OPSD methods and related baselines.","authors":[{"name":"Linfei Ji","position":0},{"name":"Jiayun Li","position":1},{"name":"Haozhao Wang","position":2},{"name":"Siyu Jiang","position":3},{"name":"Zehao Wang","position":4},{"name":"Longtao Xiao","position":5},{"name":"Yu-Jie Yuan","position":6},{"name":"Ruixuan Li","position":7}],"categories":["cs.AI","cs.CL","cs.LG"],"primaryCategory":"cs.AI","publishedAt":"2026-10-21T03:59:23Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29167"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29167"}]},"rank":150,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29152","arxivId":"2610.29152","doi":"10.48550/arXiv.2610.29152","title":"From Augmentation to Synthetic Evidence: Evaluating Generated Chest Radiographs at Dataset Intake","abstract":"A generated chest radiograph can be useful as declared augmentation, but the same image becomes unsupported evidence when submitted as the record of a real examination. We study this problem by separating three questions: whether an image passes an automated quality check, whether an origin classifier flags it as generated, and whether a trusted record supports the claimed acquisition. In our experiments, Qwen selects up to 20 images from pools of 1,000 existing Cheff-identified or SynthCheX candidates. A receiver evaluates quality with MedGemma and predicted origin with a classifier on frozen DINOv3 features. Quality-based selection raises quality passes from 12 to 19 of 20 for Cheff-identified images and from 8 to 17 for SynthCheX. We find no evidence that selection improves passage through both checks: no selected Cheff-identified image passes both, while uniform, quality, and quality-constrained surrogate selection produce 2, 1, and 2 SynthCheX passes. On unprocessed test images, the origin classifier flags 4,677 of 5,000 SynthCheX images, but it also misclassify 110 of 2,297 NIH (real) images and 568 of 957 genuine images from another collection as fake. Our results further show that JPEG compression reduces SynthCheX-versus-NIH AUROC from 0.987 to 0.841. Even without processing, 152 SynthCheX images pass both pixel checks. These checks can filter submissions, but they cannot verify an acquisition claim. We strongly advocate that source and intended use must be recorded and checked separately for admission of chest radiographs in medical datasets.","authors":[{"name":"Shaikh Akib Shahriyar","position":0},{"name":"Zinedine Sehili","position":1},{"name":"Adam Arthur","position":2},{"name":"Christopher Schwartz","position":3}],"categories":["cs.CV","cs.AI","cs.LG"],"primaryCategory":"cs.CV","publishedAt":"2026-10-21T03:39:28Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29152"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29152"}]},"rank":151,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29149","arxivId":"2610.29149","doi":"10.48550/arXiv.2610.29149","title":"SAFTER: Safety-Aware Filtering Through Explicit Recoverability for Robust Reinforcement Learning under Distribution Shift","abstract":"Learning safe policies that generalize to unseen environments is difficult under bounded actuation, where an unsafe decision may become unrecoverable. Conventional control barrier function (CBF) filters enforce only instantaneous safety and ignore future recoverability. We propose SAFTER (Safety-Aware Filtering Through Explicit Recoverability), which embeds a differentiable, closed-form backup-CBF filter into Soft Actor-Critic training. SAFTER unifies current safety, finite-horizon backup safety, terminal recoverability, and input admissibility into a single differentiable certificate. Its CBF condition yields an exact minimum-intervention projection without iterative online optimization. In quadruped navigation, SAFTER reaches 90% training success with 30.6% fewer transitions than Vanilla SAC and 34.1-35.1% fewer than current-state CBF-filtered SAC baselines. Across unseen simulated environments, SAFTER improves success by up to 42.4 and 61.4 percentage points over the state-of-the-art methods CBF-RL and SEA-Nav, respectively, and achieves 73.3% and 76.7% success on two real-world navigation maps. Ablations identify accurate differentiation through the safety filter as a key contributor to robustness, in which replacing the projection Jacobian with the identity approximation reduces success rate in a High-Shift environment from 61.4% to 6.2%. These results show that recoverability-aware filtering serves as both a runtime safety mechanism and a learning mechanism for sample-efficient, robust policies.","authors":[{"name":"Benedictus C. G. Cinun","position":0},{"name":"Bin Hu","position":1}],"categories":["eess.SY","cs.LG","cs.SY"],"primaryCategory":"eess.SY","publishedAt":"2026-10-21T03:33:35Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29149"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29149"}]},"rank":152,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29145","arxivId":"2610.29145","doi":"10.48550/arXiv.2610.29145","title":"ORSMT: Unleashing the Potential of Satisfiability Modulo Theories for Semantic Verification and Repair in LLM-Based Optimization","abstract":"Large language models increasingly automate optimization modeling from natural language problems, producing written mathematics, solver code and a returned report. Verifying one optimization problem across these artifacts requires semantic comparison: unequal models can attain the same objective, equivalent models can use different auxiliary variables, and mutually consistent artifacts can omit the same source requirement. ORSMT provides a typed semantic interface and exact SMT relation queries. It aligns observable decisions, projects auxiliary variables, proves or refutes feasible region and objective relations, localizes refutations to implicated rows or domains and links source texts when available. It then reconstructs one artifact from another, and each adopted candidate passes the required relation checks. Across 3,034 problems and six generator sizes, ORSMT improves the model-code match rate in all 72 benchmark and generator combinations before selection. Protected selection raises the aggregate rate from 43.2% to 88.2% and the share that are both correct and matched from 25.2% to 40.2%. On the 1,905 refuted outputs, certified voting under the shared fallback makes 50.9% of outputs correct and matched, compared with 44.8% for ReLoop and 49.4% for Opt-Verifier. A separate one-round source revision followed by exact reconstruction restores a model-code match for 84.4% of the refuted pairs.","authors":[{"name":"Rui Zhang","position":0},{"name":"Chenyu Zhou","position":1},{"name":"Jianghao Lin","position":2},{"name":"Yaoxin Wu","position":3},{"name":"Dongdong Ge","position":4},{"name":"Wen Song","position":5}],"categories":["math.OC","cs.AI","cs.SE"],"primaryCategory":"math.OC","publishedAt":"2026-10-21T03:27:02Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29145"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29145"}]},"rank":153,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29120","arxivId":"2610.29120","doi":"10.48550/arXiv.2610.29120","title":"An Eco-Friendly Tourism Trip Planner Driven by Z-Number Uncertainty: Evidence from Baku, Dubai, Abu Dhabi, Tashkent, and Istanbul","abstract":"Deployed tourism trip planners ingest travel times and ratings as exact numbers, although routing services return travel times that vary with the hour of departure and a platform rating may rest on a handful of reviews or on tens of thousands. We present an eco-friendly trip planner that represents both quantities as Z-numbers, which pair a fuzzy restriction on a value with a fuzzy reliability on the evidence behind it, and that constructs them directly from routing-service observations at six departure hours and from platform review counts rather than from linguistic elicitation. A hybrid loss combines score, carbon, travel-time, and reliability terms, and an adaptive large neighborhood search returns single-day itineraries with per-leg reliability annotations through a conversational front end. We compare the planner with a crisp-mean baseline under an identical optimizer in Baku, Dubai, Abu Dhabi, Tashkent, and Istanbul, with fifteen points of interest and thirty seeds per city, and we evaluate the selected itineraries on their trip carbon and their total travel time. Pooled over 150 seed-paired runs, the planner lowers trip carbon from 3.24 to 2.61 kg (19.3%, Wilcoxon p = 0.017; city-clustered bootstrap interval 6.4% to 26.7%) and travel time from 123 to 109 minutes per trip (11.5%, p = 2.8e-10; interval 5.0% to 18.2%). The travel-time reductions survive a Holm correction in Baku, Istanbul, and Abu Dhabi, the carbon reduction is significant in Baku and large but uncertain in the dispersed catalogs of Dubai and Abu Dhabi, and neither metric changes in Tashkent. Review counts were measured for Baku and estimated for the other four cities, and the comparison involves a single baseline, so the results establish feasibility and direction rather than a general advantage. Code, routing caches, frozen routing tables, and per-seed results are openly released.","authors":[{"name":"Aziz Nuriyev","position":0},{"name":"Sherzod Turaev","position":1},{"name":"Saja Al-Dabet","position":2},{"name":"Mary John","position":3},{"name":"Kanan Qasimzade","position":4},{"name":"Alihuseyn Orujzada","position":5},{"name":"Alihuseyn Orujzada","position":6},{"name":"Aisha Al Ali","position":7}],"categories":["physics.soc-ph","cs.AI","cs.CE"],"primaryCategory":"physics.soc-ph","publishedAt":"2026-10-21T03:03:19Z","version":1,"license":"http://creativecommons.org/licenses/by-sa/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29120"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29120"}]},"rank":154,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29101","arxivId":"2610.29101","doi":"10.48550/arXiv.2610.29101","title":"Dual-Channel Contrastive Encoding of Sperm Whale Coda","abstract":"Sperm whales (Physeter macrocephalus) communicate through short click sequences called codas, whose structure includes the rhythmic pattern of inter-click intervals (ICI) and the spectral composition of the clicks. In this paper, we propose the Dual-Channel Contrastive Encoder (DCCE), which explicitly encodes rhythm and spectral information with separate encoders before fusing them. We evaluate DCCE on three coda-understanding tasks: social unit, coda type, and individual identification. We compare DCCE against strong baselines from literature under a common linear-probe protocol and analyze the role of each channel and training objective through ablations. On 1,383 codas, shared partitions and validation-only layer selection support three random splits and three recording-date-disjoint folds. DCCE with unit pairing and type supervision achieves mean random-split macro-F1 of 0.901/0.865/0.597 for unit/type/identity. Raw ICI remains strongest for type (0.987), and ICI plus temporal mel statistics is a strong identity reference (0.724). An auxiliary identity objective improves spectral-branch identity prediction but does not improve the mean fused-identity score in this configuration. Date-grouped evaluation and four year holdouts expose generalization limits. The results support task-dependent combinations of timing and spectral information, without establishing independent biological channels or general architectural superiority. Code, notebooks, and complete results are available at https://dcce-research-site.vercel.app.","authors":[{"name":"João Victor Quintanilha","position":0},{"name":"Emma Virnelli","position":1},{"name":"Shuo Zhang","position":2}],"categories":["cs.SD","cs.AI","cs.LG"],"primaryCategory":"cs.SD","publishedAt":"2026-10-21T02:46:56Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","venue":"Proceedings of the 11th Workshop on Detection and Classification of Acoustic Scenes and Events (DCASE 2026)","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29101"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29101"}]},"rank":155,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29045","arxivId":"2610.29045","doi":"10.48550/arXiv.2610.29045","title":"HumanCentric-Scenes-AI: A Human-Curated Multimodal Dataset of AI-Generated Human-Centric Scenes Across Surveillance, Occupational Portraits, AI-Edited Images, and Short-Form Video","abstract":"We introduce HumanCentric-Scenes-AI, a human-curated dataset of 296 AI-generated stimuli that depict people across four contexts in which people encounter AI-generated media. The Surveillance subset contains 85 AI-generated CCTV-style images. The Occupations subset contains 128 AI-generated occupational portraits in a balanced paired design covering 64 occupations with one female and one male version of each. The AI-edited subset contains 27 real photographs in which a region was altered or replaced by a generative model. The Video subset contains 56 AI-generated short videos. We curated the images and videos manually over multiple rounds to make them as photorealistic as possible. For each stimulus we release the file, the generation prompt where recorded, and per-stimulus metadata. We report depicted-gender distributions from an initial set of 134 generations without a specified gender and compare them with U.S. Bureau of Labor Statistics occupational gender shares. We report edit-region size and its association with zero-shot detection difficulty for the AI-edited subset. We report image-quality distributions for the Surveillance subset and subject-category coverage for the Video subset. HumanCentric-Scenes-AI is intended first for human-perception research, and also for calibration of AI-detection classifiers and bias analysis of generative models.","authors":[{"name":"Nima Kamali","position":0},{"name":"Sana Nasiri","position":1},{"name":"Negar Kamali","position":2}],"categories":["cs.CV","cs.AI","cs.HC"],"primaryCategory":"cs.CV","publishedAt":"2026-10-21T01:20:14Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29045"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29045"}]},"rank":156,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29039","arxivId":"2610.29039","doi":"10.48550/arXiv.2610.29039","title":"Neural Vector Field Transformations: Learning the Flow of Autonomous Systems by Conjugation to Explicitly Integrable Reference Flows","abstract":"We represent the flow map of an autonomous ODE $\\dot x=f(x)$ as a conjugate $Φ_t=F_θ^{-1}\\circΨ_t\\circ F_θ$ of an explicitly integrable reference flow $Ψ_t$ by an invertible neural network $F_θ$, the neural vector field transformation. The representation is an exact flow for every parameter value and inherits the phase portrait of the reference. A $C^1$ conjugacy preserves the eigenvalues of equilibria, the periods and Floquet multipliers of periodic orbits and the frequency-amplitude relation of centers, so these invariants must be learnable parameters of $Ψ$. We give parametrized reference flows with explicit integral curves, show that an action-angle reference is Hamiltonian only if its learnable frequency map is a gradient, and show that an orientation-preserving $F_θ$ cannot reverse the rotation sense of the reference, so it must be fixed before training. The main contribution is a training recipe: a push-forward equation residual made relative to $|f|$ and sampled log-uniformly in $|x|$ around an equilibrium, with the anchors $F_θ(0)=0$, $DF_θ(0)=I$; a symplectic transformation with the gradient-form frequency map; and invariants initialized from the field or data. With it the full spectrum of a three- and a six-dimensional nonlinear sink is recovered (eigenvalue deviation $8.3\\times10^{-3}$ and relative error 2.3%, three seeds each), and the action-angle variables of a four-dimensional integrable Hamiltonian system with frequency maps and actions correct to a median of 0.09% and 0.22%, and an exactly symplectic flow. On a dispersive system a time-input network is eight times more accurate inside its horizon, and the data-only identification of the Van der Pol oscillator yields a conjugacy on the data region with the Floquet exponent 17% off; an oracle run with the true invariants frozen attributes this to the expressivity of $F_θ$.","authors":[{"name":"Andrew Qing He","position":0}],"categories":["math.DS","cs.NA","math.NA"],"primaryCategory":"math.DS","publishedAt":"2026-10-21T01:13:44Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29039"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29039"}]},"rank":157,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29026","arxivId":"2610.29026","doi":"10.48550/arXiv.2610.29026","title":"Resonance and Depth: Composing Timbral Identity and Spatial Form in an Artist-in-the-Loop Co-Creative Multi-Agent Percussive Performance","abstract":"This research-creation artistic report presents Resonance and Depth, a 9-minute immersive electroacoustic/acousmatic performance that treats timbre and space as primary compositional materials. Built from live electronic percussion and a personal corpus recorded, curated, and composed by the author, the work investigates how timbral identity and musical form can emerge through real-time recombination, spatialisation, and performer-system interaction rather than conventional harmonic development. The performance is realised as an artist-in-the-loop co-creative multi-agent system integrating MACAT, MACataRT, a MIDI-based musical agent, a spatialisation layer, and a conductor agent. These components recombine and transform the performer's material through concatenative synthesis, audio mosaicing, symbolic mediation, physically inspired synthesis, and object-based spatialisation. The work approaches spatialisation as an active compositional process in which position, trajectory, depth, diffusion, width, and apparent size participate in musical phrasing and form. Through this practice-based framework, Resonance and Depth examines how performer identity can be retained, transformed, and redistributed across human, algorithmic, timbral, and spatial processes within an immersive co-creative performance ecology.","authors":[{"name":"Keon Ju Maverick Lee","position":0}],"categories":["cs.SD","cs.MA","cs.MM"],"primaryCategory":"cs.SD","publishedAt":"2026-10-21T00:59:14Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","venue":"Proceedings of Timbre 2026, 4th International Conference on Timbre, Montréal, Québec, Canada, 2026","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29026"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29026"}]},"rank":158,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29014","arxivId":"2610.29014","doi":"10.48550/arXiv.2610.29014","title":"GenXR-Bench: Benchmarking Coding Agents for Interactive XR Development","abstract":"Coding agents are rapidly advancing from single-file generation to long-horizon repository-scale engineering, yet the evaluation of their capabilities to author interactive extended reality (XR) applications remains an open challenge. Unlike traditional software, XR applications operate within dynamic 3D physical environments where runtime robustness depends jointly on continuous user motion, spatial input, and environmental sensing. We introduce GenXR-Bench, the first benchmark for the end-to-end evaluation of coding agents in XR software development. GenXR-Bench contains 112 expert-authored tasks spanning 24 feature implementations, 21 bug repairs, and 67 application generation tasks. Held-out tests combine scripted and adaptive interactions with programmatic and visual assertions to evaluate core functionality and runtime behavior. GenXR-Bench's testing suite achieves 96.5% accuracy against blinded expert judgments of 124 candidate applications executed on consumer XR hardware. Evaluating 10 models on GenXR-Bench, we find that the strongest configuration achieves only 50.9% overall Pass@1. Failure analysis identifies recurring misunderstandings in asynchronous sensor lifecycles, physical grounding, and embodied interaction semantics. GenXR-Bench establishes a reproducible testbed for evaluating coding agents in the interactive and physically situated conditions of spatial computing.","authors":[{"name":"William Huang","position":0},{"name":"David Li","position":1},{"name":"Xiang 'Anthony' Chen","position":2},{"name":"Xun Qian","position":3},{"name":"Nels Numan","position":4},{"name":"Zhongyi Zhou","position":5},{"name":"Hyunsung Cho","position":6},{"name":"Salman Muin Kayser Chishti","position":7},{"name":"Ding Xia","position":8},{"name":"Xincheng Huang","position":9},{"name":"Jiahao Ren","position":10},{"name":"Faraz Faruqi","position":11},{"name":"Tianyu Xu","position":12},{"name":"Yang Zhang","position":13},{"name":"David Kim","position":14},{"name":"Ruofei Du","position":15}],"categories":["cs.AI","cs.GR","cs.HC"],"primaryCategory":"cs.AI","publishedAt":"2026-10-21T00:28:16Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29014"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29014"}]},"rank":159,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.29005","arxivId":"2610.29005","doi":"10.48550/arXiv.2610.29005","title":"VeriEvolve: Evaluator-First Evolution for Tool-Using Agents","abstract":"A central obstacle to self-improving agents is the absence of a reliable fitness signal for realistic stateful tasks. We introduce \\textsc{VeriEvolve}, an evaluator-first evolution framework that constructs this missing evaluation layer before optimizing the agent. From task instruction, VeriEvolve synthesizes instruction-grounded challenge states and an executable evaluator that checks them. During each round of workflow evolution, the evaluator is frozen and serves as the fitness signal. Between rounds, VeriEvolve revises the evaluator by converting the end states of strong and weak runs into a pair of counterfactual test cases. Across all 30 MCPMark Filesystem tasks, VeriEvolve constructs 30 evaluators with 141 challenge states, achieving 141/141 agreement on the generated suites and 113/122 agreement on held-out states produced by executor agents. Evaluator-guided evolution increases externally verified success from 6/30 to 18/30 for GLM-5.2 and from 27/30 to 30/30 for Claude; refinement removes all three false positives that the weak executor exposes. Cross-environment studies on PostgreSQL, Playwright, Notion, and GitHub further expose complementary limits from observability, state representation, and search/execution capacity. These results show that executable evaluators can be synthesized and used to guide search when task success is observable and an external verifier is available for calibration.","authors":[{"name":"Yueshen Li","position":0},{"name":"Kamer Ali Yuksel","position":1},{"name":"Gokhan Tur","position":2},{"name":"Dilek Hakkani-Tür","position":3},{"name":"João Pedro Maia","position":4}],"categories":["cs.SE","cs.AI","cs.CL"],"primaryCategory":"cs.SE","publishedAt":"2026-10-21T00:13:29Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.29005"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.29005"}]},"rank":160,"reason":"new","score":0.1857},{"paper":{"id":"arxiv:2610.27198","arxivId":"2610.27198","doi":"10.48550/arXiv.2610.27198","title":"DeGroot Dynamics on Multiplex Networks with Directed and Acyclic Interlayer Connectivities","abstract":"In models of opinion dynamics, researchers typically examine how an individual's opinion on one topic is influenced by the opinions of their neighbors in a network. However, people's opinions on a topic are also influenced by their opinions and their neighbors' opinions on related topics. In an election, for example, the opinions of a voter on a variety of issues inform their opinions of electoral candidates. To consider the influence between people's opinions on multiple distinct topics, we study a multiplex extension of the traditional DeGroot consensus model. In our multiplex DeGroot model, different network layers represent distinct topics and directed interlayer edges (which are not allowed to form cycles) encode relationships between distinct topics. In our model, an individual's opinion on one topic is influenced both by the opinions of their neighbors on that topic and by their own opinions on related topics. We prove under mild conditions that our multiplex DeGroot model converges to a limit state. We also prove that strong connectedness in ``source layers'' (which represent fundamental beliefs) eventually yields a consensus opinion in each layer. In particular, we show that consensus on source layers can lead to consensus on other layers even when those layers otherwise would be polarized. By numerically simulating our multiplex DeGroot model on a two-layer network, we examine the effects on opinion dynamics of (1) different network structures and (2) different types of initial opinion data.","authors":[{"name":"Mia Zender","position":0},{"name":"Kye Shi","position":1},{"name":"Abigail Hickok","position":2},{"name":"Weiqi Chu","position":3},{"name":"Mason A. Porter","position":4}],"categories":["cs.SI","cs.SY","econ.TH","eess.SY","math.DS","physics.soc-ph"],"primaryCategory":"cs.SI","publishedAt":"2026-10-19T19:34:07Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.27198"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.27198"}]},"rank":161,"reason":"new","score":0.175},{"paper":{"id":"arxiv:2610.28681","arxivId":"2610.28681","doi":"10.48550/arXiv.2610.28681","title":"Testing the statistical compatibility of data sets with a common prior","abstract":"The Parameter Goodness-of-fit (PG) test [hep-ph/0304176] offers a simple test statistic to study the compatibility of different data sets within a given theoretical model. In its original formulation it assumes that the data sets are statistically independent. We propose an extension of the PG test for the situation when some parameters of the model are constrained by an auxiliary data set called prior here. As the prior is imposed for all data sets, they are no longer statistically independent. We show that under the Gaussian approximation, the prior-constrained PG test statistic follows a generalised $χ^2$ distribution, which can be used to obtain $p$-values in a straight-forward way. In the approximation of a strong prior constraint, the test converges towards the standard PG test for the unconstrained model parameters.","authors":[{"name":"Pilar Coloma","position":0},{"name":"Michele Maltoni","position":1},{"name":"Thomas Schwetz","position":2}],"categories":["hep-ph","astro-ph.CO","hep-ex","physics.data-an","stat.ME"],"primaryCategory":"hep-ph","publishedAt":"2026-10-19T18:00:27Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28681"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28681"}]},"rank":162,"reason":"new","score":0.175},{"paper":{"id":"arxiv:2610.28994","arxivId":"2610.28994","doi":"10.48550/arXiv.2610.28994","title":"The Optimal Sampler for Diverse TSP Solutions: Characterization, Limits, and Learning","abstract":"We address the Diverse Traveling Salesman Problem (D-TSP) that seeks a set of $k$ distinct TSP tours. Each selected tour must not exceed a prescribed multiple of the optimal tour length, while the maximum pairwise Jaccard similarity within the selected set is minimized. This formulation is important for applications requiring both solution quality and fault tolerance, such as logistics planning, robotics, and strategic patrolling. We propose a novel hybrid framework that decomposes D-TSP into a sampling stage and a selection stage. The selection stage is a max-min dispersion problem, for which any approximation better than a factor of two is NP-hard; we therefore focus on the sampling stage. Our contributions are both theoretical and empirical. Theoretically, we give two end-to-end guarantees for the optimal sampling distribution, from two perspectives: including an optimal set in the sampled pool, or having at least $k$ mutually distant tours in the pool. The second reduces the number of samples from exponential to linear in $k$, at the cost of weakening the benchmark from the best $k$ tours to the best $βk$ for some $β> 1$. For practical use, we prove that a pipeline that samples spanning trees and then converts to tours does not hurt the quality of the pool in principle. Empirically, both objectives are learnable, and our experiments not only validate the analysis but also outperform the baselines on both in-distribution and out-of-distribution instances, by a single trained model with no fine-tuning.","authors":[{"name":"Ssu-Yuan Lo","position":0},{"name":"Ching-Kai Wang","position":1},{"name":"Kuan-Lun Chen","position":2},{"name":"Hao-Tsung Yang","position":3}],"categories":["cs.CG","cs.DS","cs.LG"],"primaryCategory":"cs.CG","publishedAt":"2026-10-20T23:48:08Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28994"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28994"}]},"rank":163,"reason":"new","score":0.1612},{"paper":{"id":"arxiv:2610.28991","arxivId":"2610.28991","doi":"10.48550/arXiv.2610.28991","title":"Diffusion Sensitivity Geometry and Strong Duality for Optimal Privacy-Utility Tradeoffs","abstract":"Existing approaches to differential privacy in diffusion models rely on loose sensitivity bounds and isotropic Gaussian noise, leading to unnecessary estimation error. We develop a framework for the optimal privacy-utility tradeoff of ideal diffusion denoisers, replacing conservative bounds with exact sensitivity characterizations and provably optimal noise calibration. First, we show that worst-case sensitivity over datasets of arbitrary size can be characterized exactly using only two records. For a data domain of size $m$, this reduces the analysis to at most $m(m-1)$ constraints, regardless of dataset size. Our characterization can be arbitrarily tighter than conventional diameter-based sensitivity bounds, revealing potentially unbounded improvements in noise calibration. Second, we obtain exact privacy guarantees for Rényi, approximate, and Gaussian differential privacy, as well as mutual-information privacy for binary records, avoiding the looseness of previous generic privacy conversions. Finally, we show that the minimum achievable weighted mean-squared error can be computed through a semidefinite program and prove strong duality, providing certificates of global optimality.","authors":[{"name":"Tue Do","position":0},{"name":"Daniel Alabi","position":1}],"categories":["cs.IT","cs.CR","math.IT"],"primaryCategory":"cs.IT","publishedAt":"2026-10-20T23:37:12Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28991"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28991"}]},"rank":164,"reason":"new","score":0.1612},{"paper":{"id":"arxiv:2610.28989","arxivId":"2610.28989","doi":"10.48550/arXiv.2610.28989","title":"AgentPolicyForge: From Executable Environments to Policy Benchmarks and Adaptive Guards","abstract":"Customized agent deployments define policies over tools, resources, and execution conditions. Providing these policies in the task agent's prompt does not reliably prevent violations. Training a guard requires examples of tool calls that comply with or violate those policies. We introduce AgentPolicyForge, a framework that uses executable environments to construct policy benchmarks and iteratively train policy guards. With this framework, we construct PolicyGym, a dataset across ten policy families using 698 scenarios for training data generation and 254 separate held-out scenarios for human-reviewed test sets. Tool calls are executed and their outcomes recorded to help determine whether they comply with the policy. Every violating case uses a permitted tool, requiring checks of arguments, resources, and context rather than tool identity alone. The case construction process supports iterative guard training: an attacker generates requests using retrieved strategies, and the guard is fine-tuned on balanced, environment-verified examples. Training improves policy enforcement even with a fixed archive. Updating the archive further reduces in-domain attack success compared with keeping it fixed. We compare policy-aware task agent self-regulation with a trained policy guard. On PolicyGym, the final guard-based configuration achieves a 40-44% relative reduction in attack success rate and fewer benign rejections than policy prompting alone.","authors":[{"name":"Chao Shang","position":0},{"name":"Wanyu Du","position":1},{"name":"Yi Zhang","position":2},{"name":"Sandesh Swamy","position":3}],"categories":["cs.AI","cs.CR","cs.LG"],"primaryCategory":"cs.AI","publishedAt":"2026-10-20T23:31:32Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28989"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28989"}]},"rank":165,"reason":"new","score":0.1612},{"paper":{"id":"arxiv:2610.28979","arxivId":"2610.28979","doi":"10.48550/arXiv.2610.28979","title":"Anytime-Valid Inference in Conjoint Experiments","abstract":"Choosing an appropriate sample size is a fundamental challenge in experimental design. This challenge is particularly salient in conjoint experiments, where researchers simultaneously target many estimands whose inferential goals may be reached at different sample sizes. Conventional sample-size calculations address this challenge in an ex ante manner. In large online studies, including many conjoint experiments, data often arrive continuously, allowing researchers to dynamically stop data collection. In such settings, the relevant practical question is not \"what sample size should I choose in advance?\" but rather, as data accumulate, \"have I learned enough yet?\" A naive response is to repeatedly inspect conventional confidence intervals and $p$-values, but doing so invalidates their inferential guarantees. We address this challenge by developing a general anytime-valid framework for conjoint experiments that allows continuous monitoring and data-dependent stopping. This framework equips many common conjoint estimands with asymptotic $e$-processes, confidence sequences, and anytime-valid $p$-values, and provides a general recipe for any estimand that can be expressed as a linear contrast of population OLS coefficients. Across common conjoint settings, we demonstrate that anytime-valid inference can yield substantial sample savings relative to fixed-sample designs. The probability of early stopping frequently exceeds 50%, with mean sample savings often surpassing 25%.","authors":[{"name":"Daniel Molitor","position":0},{"name":"Jennah Gosciak","position":1},{"name":"Michael Lindon","position":2}],"categories":["stat.ME","econ.EM","stat.AP"],"primaryCategory":"stat.ME","publishedAt":"2026-10-20T23:06:51Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28979"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28979"}]},"rank":166,"reason":"new","score":0.1612},{"paper":{"id":"arxiv:2610.28968","arxivId":"2610.28968","doi":"10.48550/arXiv.2610.28968","title":"Unit-Aware Differentially Private Quantum Machine Learning for Clinical and Wearable Health Data","abstract":"Clinical registries and wearable sensors support early health-risk assessment, yet models trained on these records may leak sensitive information when released. Differential privacy (DP) bounds this leakage by injecting noise whose total magnitude grows with the number of trainable parameters, a cost especially severe for small health cohorts. This motivates parameter efficient models including quantum machine learning (QML), whose variational circuits can represent complex decision boundaries with few trainable parameters. It raises a question: at matched input dimensionality and privacy budget, does a quantum model retain more utility than a classical one? We investigate this question on three clinical cohorts and one wearable cohort, treating each clinical record or participant as a privacy unit. We compare capacity-matched quantum, multilayer perceptron, and logistic regression models across five privacy budgets and five seeds. The quantum model leads on three clinical cohorts by up to 0.104 ROC AUC averaged across budgets; however, trails on the wearable cohort, where the small number of participants requires substantial noise. No difference survives Holm correction across 24 tests, so these results indicate a trend rather than a quantum advantage. An analytic gradient bound shows that lower clipping thresholds can improve utility by reducing noise despite clipping bias. Finally, an experiment on IBM quantum computer (ibm_fez, a 156-qubit superconducting processor) using eight fixed test cases yields mean absolute probability errors of 0.046 (private) and 0.063 (non-private) relative to simulation, supporting the feasibility of inference on noisy intermediate-scale quantum (NISQ) hardware. Code is publicly available at https://github.com/TITANS-Lab-Artifacts/DP-QML-Health.","authors":[{"name":"Md Khalid Syfullah","position":0},{"name":"Alvi Ataur Khalil","position":1}],"categories":["cs.LG","cs.AI","cs.NE"],"primaryCategory":"cs.LG","publishedAt":"2026-10-20T22:38:13Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28968"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28968"}]},"rank":167,"reason":"new","score":0.1612},{"paper":{"id":"arxiv:2610.28955","arxivId":"2610.28955","doi":"10.48550/arXiv.2610.28955","title":"SLoSa: Small, Low-Power, and Safe Foundation Models for Robotic Task Planning","abstract":"Safe robotic task planning requires planners that can reliably achieve task goals while satisfying safety constraints. Classical planners search over explicit symbolic task representations and can provide formal guarantees with respect to encoded safety constraints, but require rigid input formats. Their computational cost also increases rapidly with the search space. In contrast, large language models (LLMs) offer flexible task interfaces and strong generalization capabilities, but they are large, power-hungry, and pre-trained with little robot-specific safety data and knowledge. Thus, they can be difficult to deploy locally on robotic systems with limited resources. To address these limitations, we present SLoSa, a neuro-symbolic compression and fine-tuning framework for training foundation-model-based robotic task planners with Small size, Low power demand, and Safety awareness. The framework first constructs a safety-aware dataset using classical planners, an advanced teacher LLM, and formal verifiers, followed by two training stages: supervised fine-tuning and reinforcement learning (RL) fine-tuning guided by a verification-based reward machine to enforce safety alignment. We evaluate SLoSa across four Planning Domain Definition Language (PDDL) domains, two household simulation environments, local edge platforms, and a desktop robotic arm. Across the PDDL tasks, verifier-guided post-training raises the success rate of compact 4B-14B planners to 70.5-76.0%, while INT4 deployment on a Jetson Orin requires 17.85-26.41 W of average power. Results in household simulators and on the physical robot further demonstrate the applicability of SLoSa across diverse robotic task settings.","authors":[{"name":"Weizhe Xu","position":0},{"name":"Jialiang Fan","position":1},{"name":"Essien Taylor","position":2},{"name":"Zijun Wang","position":3},{"name":"Russell Joseph","position":4},{"name":"Fanxin Kong","position":5}],"categories":["cs.RO","cs.AI","cs.LO"],"primaryCategory":"cs.RO","publishedAt":"2026-10-20T22:09:44Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28955"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28955"}]},"rank":168,"reason":"new","score":0.1612},{"paper":{"id":"arxiv:2610.28930","arxivId":"2610.28930","doi":"10.48550/arXiv.2610.28930","title":"Amplified Pruning and Path Selectors for Deterministic Radio Gossiping","abstract":"We study deterministic gossiping in unknown strongly connected directed radio networks with $n$ vertices, $m$ arcs, and distinct identifiers in $\\{1,\\ldots,N\\}$, where $N=n^{\\mathcal O(1)}$. Messages are unrestricted; collision detection is unavailable. Our protocol combines amplified reception certificates, adaptive pruning, and permutation selectors. Each pruning iteration costs $\\mathcal O(n\\log^2n\\log\\log n)$ rounds. A two-step quadratic contraction limits each receiver and each class of identical closed in-neighborhoods to $\\mathcal O(\\log\\log n)$ qualifying selections. Under an explicit degree-concentration condition, our protocol takes $\\mathcal O(m\\log^2n(\\log\\log n)^2)$ rounds. For directed regular graphs with common in- and out-degree $Δ$, our strategy achieves $\\widetilde{\\mathcal O}(\\min\\{n\\sqrtΔ,n^2/Δ\\})$, extending the identifier range of Wu and Chrobak and improving their high-degree bound. At completion entry, let $a\\ge1$ count original sources not yet credited as globally disseminated. Permutation selectors reduce our balanced completion budget by a factor of $\\log^{2/3}(a+1)$ relative to interleaved path selectors. Our geometric protocol attains $\\widetilde{\\mathcal O}(k^2+nτ+\\min\\{u^2,(nu)^{2/3}\\})$ for an original path cover of scale $k$, weight $τ$, and $u$ exceptions. Every graph also admits $\\widetilde{\\mathcal O}(\\min\\{n^{4/3},N+(mn)^{3/5}\\})$ rounds. Only $n,N$ are inputs. Within the stated graph classes and identifier ranges, these results answer the polynomial-identifier and average-degree questions of Wu and Chrobak. The unrestricted general-graph cases remain open.","authors":[{"name":"Qin Xin","position":0}],"categories":["cs.DS","cs.CC","cs.DC"],"primaryCategory":"cs.DS","publishedAt":"2026-10-20T21:07:58Z","version":1,"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28930"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28930"}]},"rank":169,"reason":"new","score":0.1612},{"paper":{"id":"arxiv:2610.28910","arxivId":"2610.28910","doi":"10.48550/arXiv.2610.28910","title":"Backtracking Direct, Indirect and Total Effects","abstract":"Natural direct and indirect effects are defined through cross-world interventional counterfactuals. We propose an alternative construction based on backtracking counterfactuals. Rather than intervening on the treatment or mediator, we use a backtracking conditional distribution to construct an alternative exogenous state satisfying a desired treatment-mediator constraint. This yields the backtracking direct and indirect effects, and also the backtracking total effect. Likewise, we adapt the population intervention effects.","authors":[{"name":"Jose M. Peña","position":0}],"categories":["stat.ME","cs.LG","stat.ML"],"primaryCategory":"stat.ME","publishedAt":"2026-10-20T20:46:56Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28910"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28910"}]},"rank":170,"reason":"new","score":0.1612},{"paper":{"id":"arxiv:2610.28886","arxivId":"2610.28886","doi":"10.48550/arXiv.2610.28886","title":"SANE: Automated Scorer-Specific Metamorphic Testing for LLM Judges","abstract":"LLM judges increasingly mediate benchmark evaluation, model selection, data curation, and training. Their decisions are useful only when each judge responds reliably to changes affecting its scoring criterion. Existing methods for reliability testing perturb inputs with known expected effects on judge scores. Prominent LLM-judge testing harnesses combine human-authored perturbation recipes with synthetic test generation, leaving scorer-specific recipe ideation outside the automated pipeline. This limits their use to fixed perturbations and pre-assumed scorer types. We introduce SANE, an agent-based framework that encodes behavioral requirements for stochastic evaluators as reusable agent instructions. Given a scorer specification, implementation, and representative inputs, the pipeline first identifies scorer-specific capabilities, then automatically derives perturbation recipes and expected score relations for each. SANE also defines a scorer-independent validation framework for generated judge tests. It determines whether each recipe agrees with scorer intent and whether each instantiated case realizes the recipe and supports its expected relation. Because SANE generates a recipe inventory tailored to each scorer, the number of aligned recipes grows as scorers are added. Across five domains, SANE produces 3.7 times as many validated test families as the Judge Reliability Harness (JRH) at comparable end-to-end validity and detects scorer deficiencies in every domain, including professionalism and toxicity, where JRH detects none. On faithfulness, SANE achieves strong human alignment in judge ranking: its test-suite scores rank 11 judge models consistently with human-labeled FaithBench performance (Spearman rho = 0.852), while JRH does not (rho = 0.100).","authors":[{"name":"Ethan Ludwick","position":0},{"name":"Yu-Chung Hsiao","position":1}],"categories":["cs.CL","cs.AI","cs.SE"],"primaryCategory":"cs.CL","publishedAt":"2026-10-20T20:04:40Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28886"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28886"}]},"rank":171,"reason":"new","score":0.1612},{"paper":{"id":"arxiv:2610.28858","arxivId":"2610.28858","doi":"10.48550/arXiv.2610.28858","title":"DualHeads: Reward-tilted few-step diffusion made simple","abstract":"Distilled diffusion models generate high-quality images in a few sampling steps, but aligning them with task requirements and human preferences is difficult. Methods developed for multi-step diffusion models rely on differentiable rewards, or on score fields and stochastic trajectories that distilled generators may lack. We ask whether a few-step generator can be adapted using only external reward evaluations and reference samples, and show that two binary classifiers suffice. Our method, DualHeads, implements them as lightweight heads on a frozen diffusion backbone. The reward head learns which samples of the same prompt score higher, the reference head separates reference images from generated ones, and the generator maximizes a weighted sum of their logits. When the heads are optimal, this objective is KL-regularized maximization of the learned reward, and the ratio of head weights sets the regularization strength. Empirically, with both rule-based and model-based rewards, a few-step DualHeads generator reaches task rewards close to those of 40-step reinforcement learning methods while preserving image quality, at no extra inference cost. We hope that this simple and elegant dual-head formulation will help advance few-step reward-tilted generation.","authors":[{"name":"Yuanzhi Zhu","position":0},{"name":"Eleftherios Tsonis","position":1},{"name":"Simo Alami Chehboune","position":2},{"name":"Xi Wang","position":3},{"name":"Stéphane Lathuilière","position":4},{"name":"Vicky Kalogeiton","position":5}],"categories":["cs.CV","cs.AI","cs.LG"],"primaryCategory":"cs.CV","publishedAt":"2026-10-20T19:31:03Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28858"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28858"}]},"rank":172,"reason":"new","score":0.1612},{"paper":{"id":"arxiv:2610.28836","arxivId":"2610.28836","doi":"10.7717/peerj-cs.4103","title":"Sketch2Sim: an end-to-end modular framework for hand-drawn circuit analysis","abstract":"This study proposes a sketch-to-simulation system to solve handwritten circuit diagrams. The system converts raw images into a Simulation Program with Integrated Circuit Emphasis (SPICE)-compatible netlist and automatically calculates the circuit's nodal voltages and currents. Another objective is to propose a holistic framework that benefits students and educators by enhancing the teaching and learning experience of electric circuits and expediting engineering design reviews. This article presents a multistage pipeline that converts raw photographs of handwritten electric circuits into executable SPICE-like netlists, providing a solution. A dataset of 1,000 hand-drawn Resistor-Inductor-Capacitor (RLC) circuits was curated and annotated for components, text regions, and character recognition. The proposed system decomposes a circuit diagram into three stages: (i) dual You Only Look Once (YOLO)-based detectors for components and text regions, (ii) a Convolutional Recurrent Neural Network (CRNN) with Connectionist Temporal Classification (CTC) and Optical Character Recognition (OCR)-based domain-aware symbol value extraction, and (iii) image processing with thresholding, skeletonization, and Density-Based Spatial Clustering of Applications with Noise (DBSCAN) to automatically identify nodes and map them with corresponding components. The proposed system produces the netlist and solves the circuit. The qualitative and quantitative ablation studies on modular stages, as well as end-to-end tests, demonstrate robust performance across various handwriting styles, lighting conditions, and clutter.This study serves as an interactive learning aid and practical tool for professionals, providing a comprehensive user interface (UI), enabling rapid digitization of circuit sketches for study, design, prototyping, or troubleshooting.","authors":[{"name":"Mohammad Kawsar","position":0},{"name":"Al Amin Islam Ridoy","position":1},{"name":"Md Rakib Hossain","position":2},{"name":"Farzana Akter","position":3},{"name":"AKM Bahalul Haque","position":4},{"name":"Rejwan Bin Sulaiman","position":5},{"name":"Mohammed Amin Almaiah","position":6}],"categories":["cs.AI","cs.CV","cs.LG"],"primaryCategory":"cs.AI","publishedAt":"2026-10-20T19:04:12Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28836"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28836"}]},"rank":173,"reason":"new","score":0.1612},{"paper":{"id":"arxiv:2610.28834","arxivId":"2610.28834","doi":"10.48550/arXiv.2610.28834","title":"Where Should a Stuck Agent Spend Its Next Tokens? Causal Attribution of Inference-Time Resources","abstract":"Inference-time scaling is often treated as a token-budget problem, but a failed coding agent can seek qualitatively different resources: computation, information, or feedback. We introduce Banzhaf Resource Attribution (BRA), a controlled factorial framework that measures each resource's marginal contribution to repair and its interactions while holding the failed state and revision protocol fixed. We evaluate six models on 194 function-level failures from MBPP+ and HumanEval+ and 126 repository-level failures from SWE-bench Verified. Task-relevant specifications consistently improve repair, whereas public-test feedback provides small gains and retrieved code little to none. The value of additional computation declines with model capability, while information remains useful; resource channels largely exhibit diminishing returns, although complementary information can reinforce itself. Yet three of four frontier models select feedback in 95-100% of cases and obtain lower gains than uniform random resource selection. BRA thus exposes a gap between the resources agents prefer and those that measurably improve repair.","authors":[{"name":"Saiyue Lyu","position":0},{"name":"Jerrold Huang","position":1},{"name":"Yunbei Zhang","position":2},{"name":"Janet Wang","position":3},{"name":"Zhenting Wang","position":4},{"name":"Yingqiang Ge","position":5},{"name":"Haibo Zhang","position":6}],"categories":["cs.SE","cs.AI","cs.CL"],"primaryCategory":"cs.SE","publishedAt":"2026-10-20T18:59:01Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28834"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28834"}]},"rank":174,"reason":"new","score":0.1612},{"paper":{"id":"arxiv:2610.28820","arxivId":"2610.28820","doi":"10.48550/arXiv.2610.28820","title":"A Matroidal Framework for Random Access in Linear Codes","abstract":"We study random access in linear codes when codeword symbols are sampled independently and uniformly with replacement. A code is recovery balanced if the expected number of samples required to recover a codeword coordinate is the same for every coordinate. We express the recovery expectations in terms of the associated matroid and derive a coordinatewise identity relating these values for a code and its dual. This proves that a linear code is recovery balanced if and only if its dual is recovery balanced, establishing a conjecture of Gruica et al. We further define the recovery balance for arbitrary matroids and show that asymptotically almost every sparse paving matroid is not recovery balanced.","authors":[{"name":"Daniella Bar-Lev","position":0},{"name":"Hiram H. López","position":1},{"name":"Gretchen L. Matthews","position":2},{"name":"Flavio Salizzoni","position":3}],"categories":["cs.IT","math.CO","math.IT"],"primaryCategory":"cs.IT","publishedAt":"2026-10-20T18:40:28Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28820"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28820"}]},"rank":175,"reason":"new","score":0.1612},{"paper":{"id":"arxiv:2610.28808","arxivId":"2610.28808","doi":"10.48550/arXiv.2610.28808","title":"Deflation-Based Spectral Preconditioning for the Conjugate Gradient Method on Perturbed Systems","abstract":"We consider sequences of large symmetric positive-definite linear systems in which spectral information from a reference matrix $A$ is reused to accelerate the solution of a nearby system with perturbed matrix $A(t)=A+tE$, where $E$ is symmetric, $\\|E\\|_F=1$, and $t\\geq0$ is the perturbation size. We compare deflated conjugate gradient (deflated CG) with preconditioned CG (PCG) equipped with a scaled spectral preconditioner constructed from the same spectral information of $A$. First, we derive an energy-norm error characterization of deflated CG. We then use perturbation analysis to show that, after one additional PCG iteration, the squared PCG error is bounded, up to terms of order $t^2$, by an explicit multiple of the squared deflated CG error. This result clarifies the role of the scaling parameter in the preconditioner. Experiments show that the scaled spectral preconditioner recovers much of the convergence improvement provided by deflation while requiring fewer products with $A(t)$.","authors":[{"name":"Youssef Diouane","position":0},{"name":"Selime Gürol","position":1},{"name":"Oussama Mouhtal","position":2},{"name":"Dominique Orban","position":3}],"categories":["math.NA","cs.NA","math.OC"],"primaryCategory":"math.NA","publishedAt":"2026-10-20T18:29:29Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28808"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28808"}]},"rank":176,"reason":"new","score":0.1612},{"paper":{"id":"arxiv:2610.28787","arxivId":"2610.28787","doi":"10.48550/arXiv.2610.28787","title":"Is Anyone Enforcing the Cartel? What Reinforcement Learning Teaches LLM Agents in Markets","abstract":"Evaluations of large language model (LLM) agents report that pricing agents collude at others' expense. Yet collusion is defined by enforcement: a cartel, in which competitors share monopoly profits, holds only if a member who cheats is punished. We introduce an audit that tests enforcement by forcing a member to cheat in counterfactual copies of a market, and use it to study what reinforcement learning (RL) teaches agents in pricing and trading. Small LLMs trained by RL learn what their training rivals reward, but only what they readily try. In self-play, they learn fixed rules ignoring their rivals; against a punishing rival, to be deterred; against a cheating rival, to match its cuts, never the costly price war that would deter it, which they rarely try. A war forced into training is learned, but mostly only where it was forced. Against their own copies, deterred and matching agents can earn the same cartel profits, but only matching ones can enforce them, and then thinly: as a by-product and only against large cuts. Q-learners enforce collusion only in pricing. Prompted LLMs reach cartel-like outcomes mostly through rules that ignore the market. Outcome measures cannot tell these cases apart; the audit can.","authors":[{"name":"Hieu Tran","position":0},{"name":"Hong Yu","position":1}],"categories":["cs.CL","cs.GT","cs.LG"],"primaryCategory":"cs.CL","publishedAt":"2026-10-20T18:18:10Z","version":1,"license":"http://creativecommons.org/licenses/by-sa/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28787"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28787"}]},"rank":177,"reason":"new","score":0.1612},{"paper":{"id":"arxiv:2610.28770","arxivId":"2610.28770","doi":"10.48550/arXiv.2610.28770","title":"ReAPO: Retry-Aware Policy Optimization for Efficient Exploration in LLM Reasoning","abstract":"RLVR improves language-model reasoning, but commonly relies on independent stochastic rollouts for exploration, which can repeatedly sample similar failed strategies. Prior work has seen success with approaches that guide exploration by introducing dependence into the sampling process by including incorrect solutions in the context for subsequent attempts. Still, these methods introduce the additional question of how to assign credit to each attempt. We propose to view the credit assignment through the lens of Stochastic Shortest Paths, a framework in classical reinforcement learning that seeks to minimize the expected cumulative cost to reach a terminal state (in our case, a verified successful response). We show that, under suitable simplifying assumptions, this perspective naturally leads to a simple modification to GRPO, which we call Retry-Aware Policy Optimization (ReAPO). ReAPO allocates additional generation to failed trajectories, allowing successful retries to be reinforced without rewarding preceding failures, and assigns credit in a theoretically principled way to each attempt. Across Qwen3 and OLMo-2 models from 0.6B to 4B and eight mathematical and scientific reasoning benchmarks, ReAPO improves first-attempt performance over GRPO as well as natural recent baseline algorithms, while additional ablations demonstrate the utility of both serial sampling and ReAPO's credit assignment. On Qwen3-0.6B, ReAPO also outperforms GRPO with twice the group size or twice the training steps, reaching their peak accuracy with 39% and 31% fewer generated tokens, respectively. Taken together, our results suggest that ReAPO is a simple yet effective approach for improving LM reasoning performance through improved exploration and retry-aware credit assignment.","authors":[{"name":"Nghi Dao","position":0},{"name":"Calder Morton-Ferguson","position":1},{"name":"Noah Golowich","position":2},{"name":"Adam Block","position":3}],"categories":["cs.LG","cs.AI","stat.ML"],"primaryCategory":"cs.LG","publishedAt":"2026-10-20T18:08:29Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28770"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28770"}]},"rank":178,"reason":"new","score":0.1612},{"paper":{"id":"arxiv:2610.28767","arxivId":"2610.28767","doi":"10.48550/arXiv.2610.28767","title":"SkillEval: Learning Interpretable Ability Profiles of LLMs via Cognitive Diagnosis Models","abstract":"Current evaluation of LLMs on benchmarks typically aggregates performance over individual examples into a single scalar score. This approach is overly coarse and ignores the diverse scope of skills covered by benchmark datasets. To address this limitation and better exploit benchmark data, we propose SkillEval, a new evaluation framework that performs cross-model, skill-level analysis. The framework integrates advances from two areas: (1) it leverages the meta-cognitive capabilities of LLMs to automatically discover interpretable skills from questions, and (2) it applies cognitive diagnosis models (CDMs) from psychometrics to estimate per-skill abilities of LLMs. We conduct experiments on 3,811 LLMs evaluated on 9,523 items across five benchmarks spanning diverse domains (MATH, BBH, GPQA, MuSR, IFEval). Compared to the traditional average-based evaluation paradigm, SkillEval produces reliable, more calibrated and informative yet concise model profiles that more accurately predict model behavior on unseen items. We also demonstrate the utility of SkillEval in two downstream tasks: (1) LLM routing, which dynamically assigns each query to a capable but efficient model, reducing inference cost by over 60% while achieving improved performance, and (2) Computerized Adaptive Testing (CAT), where an unseen model can be efficiently evaluated using a trained CDM. Furthermore, we show that these skill profiles provide useful insights, for example, how benchmarks share skills and how instruction-tuned models differ from their base models.","authors":[{"name":"Berke Arda","position":0},{"name":"Peng Cui","position":1},{"name":"Qiaoyuan Zheng","position":2},{"name":"Rudolf Debelak","position":3},{"name":"Mubashara Akhtar","position":4},{"name":"Mrinmaya Sachan","position":5}],"categories":["cs.CL","cs.AI","cs.LG"],"primaryCategory":"cs.CL","publishedAt":"2026-10-20T18:06:47Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28767"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28767"}]},"rank":179,"reason":"new","score":0.1612},{"paper":{"id":"arxiv:2610.28759","arxivId":"2610.28759","doi":"10.48550/arXiv.2610.28759","title":"GPU-Accelerated Construction of the Gabriel Graph: Parallelization Strategies and Applications to Large-Margin Classification and Graph-Based Clustering","abstract":"The Gabriel graph is a proximity graph widely used in computational geometry and geometric machine learning, underlying large-margin classifiers and, as we show in this work, density-aware clustering methods. Its construction, however, has cubic time complexity $O(N^3)$ in the naive formulation, which severely limits its use on datasets beyond a few thousand samples. This paper presents a set of GPU parallelization strategies, implemented in CUDA, that make Gabriel graph construction practical at scale. We introduce (i) an implicit closed-form mapping from a linear thread index to the corresponding vertex pair $(i,j)$, which removes the need for an explicit $O(N^2)$ adjacency structure and lets every thread locate its pair in constant time and constant memory; (ii) a two-pass, bitmask-based, atomically-compacted edge-list construction that produces a dense edge array proportional only to the number of edges actually found; and (iii) a thread-coarsening strategy in which each thread evaluates several candidate pairs, trading launch and scheduling overhead against load balance. We evaluate these strategies on the acceleration of a Gabriel-graph-based large-margin classifier (NN-clas) and on a novel Gabriel-graph clustering algorithm that identifies connected components on a residual graph and performs GPU-parallel Union-Find with hierarchical edge-based merging. On 14 public benchmark datasets, the GPU implementation reduces Gabriel graph construction time by two to nearly four orders of magnitude relative to a sequential CPU baseline (up to 8484x), without loss of classification accuracy, and the proposed clustering method is statistically competitive with DBSCAN on the topological validity index Dunn while eliminating the need for an explicit density radius parameter.","authors":[{"name":"Diego Sanches Nere dos Santos","position":0},{"name":"Luiz Carlos Bambirra Torres","position":1}],"categories":["cs.DC","cs.AI","cs.LG"],"primaryCategory":"cs.DC","publishedAt":"2026-10-20T18:03:08Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28759"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28759"}]},"rank":180,"reason":"new","score":0.1612},{"paper":{"id":"arxiv:2610.28708","arxivId":"2610.28708","doi":"10.48550/arXiv.2610.28708","title":"Borrowing Strength Across Stores: Transfer Learning for Choice-Based Revenue Management","abstract":"Large retail chains must tailor assortment and pricing decisions to local customer preferences, yet even retailers with extensive sales records often have limited data for estimating demand at the store or local market level, particularly in product categories with short selling windows. Local estimation (fitting a demand model for each store on its own) can therefore be unreliable, while pooling data across stores improves estimation precision at the cost of ignoring local demand heterogeneity. We study how retailers can balance these competing effects through transfer learning under a multinomial logit (MNL) demand model. We propose a shrinkage estimator that combines store-specific demand estimates with information pooled from other stores. Unlike the linear or quadratic settings underlying much of the existing theoretical literature on information transfer in operations, the nonlinear MNL likelihood does not admit a straightforward bias-variance decomposition. By exploiting its self-concordant-like properties, we establish finite-sample performance guarantees that characterize when information transfer improves demand estimation and how the appropriate degree of shrinkage depends on local sample size, feature dimensionality, and cross-store heterogeneity. We further derive regret bounds for downstream assortment and pricing decisions. Numerical experiments demonstrate the operational benefits of shrinkage across different data environments. An empirical study using data from several hundred stores of a European apparel retailer shows that shrinkage improves out-of-sample product prioritization and revenue capture relative to both local estimation and full pooling. Our findings demonstrate how retailers can leverage information across their store networks to improve revenue management decisions while preserving local demand differences.","authors":[{"name":"Recep Yusuf Bekci","position":0},{"name":"Stefanus Jasin","position":1},{"name":"Joline Uichanco","position":2}],"categories":["stat.ML","cs.LG","math.OC"],"primaryCategory":"stat.ML","publishedAt":"2026-10-20T16:08:39Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28708"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28708"}]},"rank":181,"reason":"new","score":0.1612},{"paper":{"id":"arxiv:2610.28356","arxivId":"2610.28356","doi":"10.48550/arXiv.2610.28356","title":"CountVideoBench: A Benchmark to Evaluate Prediction Biases on the Object Counting task by Video-Language Models","abstract":"Object counting in videos remains a poorly characterized capability of multimodal large language models. Counting questions are included in many existing VideoQA benchmarks, but the ground-truth count distributions in these benchmarks are heavily skewed toward small values. Across the nine benchmarks we survey, the median ground-truth count is at most five, and the three most frequent values together account for between 45% and 93% of the videos. The resulting mean accuracy is then dominated by the most frequent counts; a value-dependent bias at a rare count contributes negligibly to the average and remains invisible. We introduce CountVideoBench, a benchmark of 2,600 synthetic videos covering 26 cardinalities in the interval [5, 50] (100 videos each), grouped into six clusters of consecutive counts that span the range while enabling near-neighbor comparisons. Each video contains a single class of static objects under a controlled camera motion, removing object motion and reducing recognition as confounds. We evaluate sixteen video-language models from nine architecture families in a zero-shot setting. No model exceeds 27.6% Exact Match. The dependence of accuracy on cardinality varies across models, ranging from monotonic decline to a non-monotonic peak-and-collapse pattern, which rules out increasing difficulty with cardinality as the dominant explanation of counting failure. Every model further concentrates its predictions on its own set of preferred integers, with accuracy peaks coinciding with these values and collapsing at adjacent counts. These findings highlight the need for balanced, diagnostic counting benchmarks to reliably assess this fundamental visual capability.","authors":[{"name":"Marc Chapus","position":0},{"name":"Carlos Crispim-Junior","position":1},{"name":"Véronique Eglin","position":2},{"name":"Atilla Baskurt","position":3}],"categories":["cs.CV","cs.AI","cs.LG"],"primaryCategory":"cs.CV","publishedAt":"2026-10-20T15:20:28Z","updatedAt":"2026-10-21T11:52:51Z","version":2,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28356"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28356"}]},"rank":182,"reason":"new","score":0.1612},{"paper":{"id":"arxiv:2610.28324","arxivId":"2610.28324","doi":"10.48550/arXiv.2610.28324","title":"Latency-Neutral Passive Cyber Monitoring for Grid-Edge Operational Technology Networks","abstract":"Deterministic industrial networks often present an operational trade-off between maintaining microsecond timing of primary control loops and capturing security telemetry. This work investigates the engineering feasibility of the Operational Technology Supervisor (OTSV), a three-tier passive monitoring framework designed to resolve this visibility challenge within timing-sensitive grid-edge infrastructure. By utilizing a Layer-3-unconfigured passive ingestion mode on bare-metal hardware via a physical switched port analyzer (SPAN) loop, the deep packet inspection engine is isolated from the production network, implementing a software-enforced unidirectional export policy. Performance validation conducted over a continuous audit comprising 32,927 active-monitoring round-trip time (RTT) samples yielded stable latency distributions across a laboratory power-monitoring loop running Modbus TCP and IEC 61850 MMS protocols. The measured latency delta between the unmonitored baseline and the active monitoring state was less than 0.12 milliseconds, supporting the feasibility of continuous high-fidelity security observation alongside stringent grid-edge metering timing bounds without measurably degrading the monitored reporting path.","authors":[{"name":"Damian Byrne","position":0},{"name":"Emad Abukhousa","position":1},{"name":"Jason Reed","position":2},{"name":"Saman Zonouz","position":3}],"categories":["eess.SY","cs.CR","cs.SY"],"primaryCategory":"eess.SY","publishedAt":"2026-10-20T14:59:10Z","updatedAt":"2026-10-21T15:52:22Z","version":2,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28324"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28324"}]},"rank":183,"reason":"new","score":0.1612},{"paper":{"id":"arxiv:2610.28695","arxivId":"2610.28695","doi":"10.48550/arXiv.2610.28695","title":"A Catalog of Dispersive Ionospheric and Plasmaspheric Arcs from Cislunar and Lunar-surface GNSS Radio Occultations","abstract":"This dataset is a catalog of 263 dispersive arcs induced by ionospheric and plasmaspheric occultation observed in GNSS signals recorded by the Lunar GNSS Receiver Experiment (LuGRE) payload during Blue Ghost Mission 1 (BGM1), in cislunar space and on the lunar surface. Occultation arcs are identified and classified as ingress, egress or tangent by combining independently GPS and Galileo dual-frequency observables and computing the rate of the total electron content (ROT) and ROT index (ROTI), and are validated through a geometric criterion based on the grazing altitude of the tangent point. No distinction is made between the ionospheric and plasmaspheric contributions to each arc, which is treated as a single dispersive signature. For each arc, the catalog provides timing, grazing altitude, summary ROT/ROTI/$Δ$STEC attributes, the geomagnetic activity level, and time series of leveled slant total electron content (STEC), not corrected for instrumental biases, ROT, ROTI, tangent-point position and local time. The HDF5 (Hierarchical Data Format version 5) file contains a header with dataset-wide metadata and per-class statistics, and one entry per arc. It enables rapid selection of occultation arcs for statistical studies and targeted analyses, and complements ground-based and low Earth orbit radio-occultation data during the declining phase of solar cycle 25.","authors":[{"name":"Francesco Fiorina","position":0},{"name":"Alex Minetto","position":1},{"name":"Rebecca Ghidoni","position":2},{"name":"Marco Guerra","position":3},{"name":"Claudio Cesaroni","position":4},{"name":"Luca Spogli","position":5},{"name":"Andrea Nardin","position":6},{"name":"Fabio Dovis","position":7}],"categories":["physics.space-ph","astro-ph.SR","eess.SP"],"primaryCategory":"physics.space-ph","publishedAt":"2026-10-20T09:39:37Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28695"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28695"}]},"rank":184,"reason":"new","score":0.1612},{"paper":{"id":"arxiv:2610.27888","arxivId":"2610.27888","doi":"10.48550/arXiv.2610.27888","title":"Automatic Differentiation with Constraints in Forward and Inverse Problems","abstract":"Approximate solutions of partial differential equations (PDEs) on unstructured grids can be represented on a graph. Numerical optimization involving stiff evolutionary and static PDEs often requires solving a large system of equations, which can be treated as a constraint. In this work, we present a fully differentiable graph-based implicit solver that incorporates this constraint. We evaluate its performance across four key tasks: a) linear and nonlinear forward modeling of evolutionary and static problems, b) solving inverse problems to estimate equation coefficients, c) solving inverse problems to estimate the right-hand side, and d) learnable graph coarsening to accelerate forward modeling. The stability and differentiability of our solver enable significant reductions in the complexity of forward modeling while efficiently addressing related inverse problems. These capabilities make it a powerful tool for different physics and engineering applications, in particular for geophysics, where inverse problems are commonly solved in order to reduce the model uncertainty.","authors":[{"name":"Nikolay Yavich","position":0},{"name":"Alexander Ryabov","position":1},{"name":"Viacheslav Naumov","position":2},{"name":"Sayan Ranu","position":3},{"name":"N. M. Anoop Krishnan","position":4},{"name":"Evgeny Burnaev","position":5},{"name":"Vladimir Vanovskiy","position":6}],"categories":["cs.AI","cs.NA","math.NA"],"primaryCategory":"cs.AI","publishedAt":"2026-10-20T09:11:38Z","updatedAt":"2026-10-21T05:45:27Z","version":2,"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.27888"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.27888"}]},"rank":185,"reason":"new","score":0.1612},{"paper":{"id":"arxiv:2610.28693","arxivId":"2610.28693","doi":"10.48550/arXiv.2610.28693","title":"Interaction-Gated Time-Frequency Learning for Parameter-Efficient PPG-Based Heart Rate Estimation","abstract":"Photoplethysmography (PPG) enables low-cost and non-invasive heart rate (HR) monitoring but remains sensitive to waveform distortion, physiological variability, and measurement interference. Temporal waveform characteristics and frequency-domain periodicity provide complementary information for HR estimation; however, effectively exploring and integrating these representations within a compact model remains challenging. This paper proposes TF-IG, a parameter-efficient interaction-gated time-frequency learning framework for PPG-based HR estimation. TF-IG employs a dual-branch architecture in which a temporal branch captures pulse morphology via dilated depthwise-separable convolutions, while a frequency branch extracts cardiac periodicity information from bounded log-spectrogram representations. To combine the two domains, we introduce an interaction-gated fusion module that explicitly models feature agreement and discrepancy, while also learning feature-wise weights to balance temporal and spectral information. A parameter-shared regression head further maps the temporal, spectral, and fused representations into the final HR prediction space. Experiments on the benchmark BIDMC PPG and Respiration Dataset using a subject-exclusive training, validation, and test protocol show that TF-IG achieves a mean absolute error (MAE) of 1.227 BPM, a root mean square error (RMSE) of 2.060 BPM, a Pearson correlation coefficient of 0.9934, and 92.28 percent of estimates within 3 BPM, while remaining parameter-efficient with only 84.6K parameters. Ablation results further demonstrate the complementary contributions of the temporal and frequency branches, interaction-gated fusion, and shared regression head. These results indicate that explicitly modeling cross-domain interactions provides an effective accuracy-complexity trade-off for lightweight PPG-based HR estimation.","authors":[{"name":"Anh Khoi Pham","position":0},{"name":"Phat Lam","position":1},{"name":"Thien Truong Nguyen Ly","position":2},{"name":"Lam Pham","position":3}],"categories":["cs.LG","cs.AI","eess.SP"],"primaryCategory":"cs.LG","publishedAt":"2026-10-20T09:10:28Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28693"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28693"}]},"rank":186,"reason":"new","score":0.1612},{"paper":{"id":"arxiv:2610.27710","arxivId":"2610.27710","doi":"10.48550/arXiv.2610.27710","title":"Privacy Accounting for Empirical Denoisers","abstract":"Diffusion models have emerged as the foundation of modern generative systems, yet their high memorization capacity raises privacy concerns. While differentially private (DP) training provides formal guarantees, it comes at a cost in sample quality. In this work, we study a complementary question: how much privacy loss is incurred during the sampling process itself? Directly analyzing modern neural denoisers is challenging because their sensitivity to a single training example is generally intractable. To overcome this challenge, we construct and analyze a clipped privacy-accountable empirical denoiser, a training-free estimator of the Bayes-optimal denoiser, whose per-step sensitivity can be computed explicitly, allowing each denoising step to be a Gaussian mechanism. This perspective enables privacy accounting under Gaussian DP when each training example contributes to each denoising step, and under $f$-DP trade-off functions when each generated sample uses a single randomly subsampled block of the training data. We report every $(\\varepsilon, δ)$ guarantee as a certified upper bound from a privacy loss distribution accountant, and use a cumulant-based Edgeworth approximation as a fast diagnostic, characterizing where it is reliable and where it becomes anti-conservative. We also analyze the privacy loss along the denoising trajectory and show that it grows steeply near the critical window in which semantic content is formed. This observation motivates hybrid samplers that use the private empirical denoiser through the formative steps where it remains a faithful proxy for the neural denoiser, then switch to a public neural denoiser for the remaining refinement. Our results on CIFAR-10 and CelebA-64 show that the privacy of diffusion models depends not only on the training procedure, but also on sampling-time design choices such as subsampling, stopping time, and hybrid denoising.","authors":[{"name":"William Cheng","position":0},{"name":"Dongjun Kim","position":1},{"name":"Mijung Park","position":2}],"categories":["cs.LG","cs.CR","stat.ML"],"primaryCategory":"cs.LG","publishedAt":"2026-10-20T06:30:08Z","updatedAt":"2026-10-21T16:49:44Z","version":2,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.27710"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.27710"}]},"rank":187,"reason":"new","score":0.1612},{"paper":{"id":"arxiv:2610.28690","arxivId":"2610.28690","doi":"10.48550/arXiv.2610.28690","title":"A Duality of the F Distribution","abstract":"The F distribution arises naturally in linear regression or analysis of variance. Elementary textbooks state the result in terms of the standard normal random vector and fixed projection matrices. More advanced textbooks and research articles have extended the result to spherically symmetric random vectors. We complement the literature with a dual result that the F distribution arises with a fixed vector and random projection matrices generated by a left-spherical random matrix.","authors":[{"name":"Zifeng Zhang","position":0},{"name":"Peng Ding","position":1},{"name":"Wen Zhou","position":2},{"name":"Chenlu Shi","position":3}],"categories":["math.ST","stat.ME","stat.TH"],"primaryCategory":"math.ST","publishedAt":"2026-10-20T06:22:39Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28690"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28690"}]},"rank":188,"reason":"new","score":0.1612},{"paper":{"id":"arxiv:2610.28687","arxivId":"2610.28687","doi":"10.48550/arXiv.2610.28687","title":"Solar System Surface Technosignatures: Searching for Signs of Artificial Design on Extraterrestrial Terrain","abstract":"Artificial radio emissions have historically been a major focus of technosignature searches. Here, we apply an anthropological lens to expand this scope to surface technosignatures: non-random, intentional patterning on planetary terrain, similar to long-lasting human-made landscapes and monuments on Earth. On geologically inactive bodies such as the Moon and Mars, artificial traces and patterns could potentially remain visible for hundreds of millions of years. We discuss two recent developments that make this search timely: high-resolution imaging datasets, such as those produced by the HiRISE camera on the Mars Reconnaissance Orbiter (MRO), and ML-based anomaly-detection methods. Finally, we list two possible approaches to developing a model capable of searching for these surface technosignatures: a supervised and an unsupervised one. Complemented by LiDAR, ground-penetrating radar, and reflectance spectroscopy, these methods would enable the first quantitative upper limits on surface technosignatures while yielding ancillary benefits for planetary science.","authors":[{"name":"Ana Henrique","position":0},{"name":"Sofia Sheikh","position":1},{"name":"Benjamin Fields","position":2}],"categories":["astro-ph.IM","astro-ph.EP","cs.LG"],"primaryCategory":"astro-ph.IM","publishedAt":"2026-10-20T03:26:56Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.28687"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.28687"}]},"rank":189,"reason":"new","score":0.1612},{"paper":{"id":"arxiv:2610.27548","arxivId":"2610.27548","doi":"10.48550/arXiv.2610.27548","title":"Act on Your Behalf, Beyond Your Authorization: Benchmarking LLM Agent Authorization Alignment","abstract":"When an LLM agent acts on behalf of a user, it is expected to act within the authorization scope the user grants. Existing agent evaluations largely measure task completion or harmful consequences and can therefore overlook actions that violate this delegation without producing conventional safety or privacy harms. We introduce AuthorizationBench, a benchmark of authorization alignment grounded in documented real-world cases. A scalable construction pipeline produces 286 executable tasks spanning situational triggers, the presence or absence of direct harmful consequences, and ten application domains. We execute every task with four models at two reasoning-effort settings across three agent harnesses in an authorization-aware sandbox that keeps authorized and out-of-authorization paths executable and observable, yielding 6,864 trajectories. Agents act out of authorization in 19.2% of trajectories, with rates ranging from 10.5% to 26.6% across the 24 configurations. Task completion does not imply authorization. Increasing reasoning effort leaves the aggregate rate nearly unchanged, with effects varying across models. On controlled pairs, agents act out of authorization at similar rates with (17.9%) and without (20.8%) direct harmful consequences, and more often under benign pressures than malicious injections (23.5% versus 17.6%, p=0.002). Our results show that task capability, additional reasoning, and consequence-oriented safety evaluation do not by themselves ensure authorization alignment. Moreover, agents that consult the principal before completing a task exhibit lower rates of out-of-authorization behavior (14.0% versus 21.6%, p<0.001), pointing to a tension between agent autonomy and preserving users' decision authority that calls for further human-centered study. Dataset and code are available at https://github.com/PEACH-Research-Lab/AuthorizationBench .","authors":[{"name":"Zhiping Zhang","position":0},{"name":"Franklin Mingzhe Li","position":1},{"name":"Zihao Huang","position":2},{"name":"Tianshi Li","position":3}],"categories":["cs.AI","cs.CR","cs.HC"],"primaryCategory":"cs.AI","publishedAt":"2026-10-20T03:01:39Z","updatedAt":"2026-10-21T06:31:44Z","version":2,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.27548"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.27548"}]},"rank":190,"reason":"new","score":0.1612},{"paper":{"id":"arxiv:2610.25113","arxivId":"2610.25113","doi":"10.48550/arXiv.2610.25113","title":"Self-organizing reinforcement learning: a learner-free swarm approaches the explore-exploit optimum","abstract":"The explore/exploit trade-off is the core problem of reinforcement learning (RL). We show that a minimal foraging swarm approaches its optimum - the ideal-free / Nash allocation of a multi-agent resource-sharing game (an equilibrium, not a return maximum) - with no learner and no reward signal. Agents explore by heavy-tailed (Levy) flights, recruit neighbours to a discovery through a threshold cascade on their proximity network, and disperse as a patch crowds; this cascade cycle balances exploration and exploitation and drives foragers toward the ideal free distribution. The swarm attains the shape of this optimum (committed foragers track richness with slope 0.94) but not its exact equilibrium - it stays measurably exploitable - reaching it dynamically (~150 steps) through the coupling of recruitment to movement: remove the coupling and matching collapses (0.94 -> 0.08). The RL sits at the level of the problem and the allocation reached, not the micro-rule: no step evaluates payoff (reward-free by construction, though the crowd signal is informationally tied to payoff), and the swarm reaches the allocation without learning it (quality plateaus and re-settles after resources move, no cumulative gain). Against a trained baseline the comparison is two-sided: in an easier, disembodied game a learner is more precise, while in the same embodied environment the reward-free swarm reaches ~65% of a validated oracle ceiling at zero training and leads until trained PPO and centralised-critic (CTDE) learners overtake it (~74-75%, 16 seeds, all above the swarm mean) after ~4,000 of 15,000 training steps - its advantage is in budget, not asymptote. In this one setting we read it not as beating RL but as a strong zero-shot, reward-free operating point: a demonstration that the self-organizing physics of a minimal swarm can carry a collective close to the optimum RL seeks, with none of RL's machinery.","authors":[{"name":"Galen J. Wilkerson","position":0}],"categories":["cs.LG","cs.AI","cs.SI","math-ph","math.MP","nlin.AO","physics.bio-ph"],"primaryCategory":"cs.LG","publishedAt":"2026-10-18T08:45:51Z","updatedAt":"2026-10-20T19:20:05Z","version":2,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.25113"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.25113"}]},"rank":191,"reason":"new","score":0.1595},{"paper":{"id":"arxiv:2610.30132","arxivId":"2610.30132","doi":"10.48550/arXiv.2610.30132","title":"LongZero: Enhancing Long-Context Reasoning in LLMs from Zero Data","abstract":"Reinforcement learning (RL) is an effective way to strengthen the long-context reasoning of large language models (LLMs), but it needs training data that are long, diverse, challenging, and verifiable. Human annotation of such data is expensive, naturally occurring data are too scarce to sustain RL, and existing synthesis methods depend on external corpora or stronger teacher models, which become harder to obtain as the trained models get stronger. Recently, RL from zero data, in which a model learns from tasks it proposes itself, has achieved excellent performance, but existing methods target short-context tasks. Therefore, we propose $\\textbf{LongZero}$, a method that lets an LLM improve its own long-context reasoning without any external data. Instead of writing questions and answers over existing documents, LongZero first has the model synthesize the question and answer with its supporting evidence in a compact structured form, and only then separately expands the evidence into a long multi-document context. LongZero combines these self-written multi-hop reasoning tasks with self-written multi-round retrieval tasks and trains the model with GRPO. We evaluate LongZero on two LLMs across five long-context reasoning benchmarks and compare it with RL on data synthesized from external corpora. Experimental results show that LongZero can generalize to real tasks, and outperforms both baselines, suggesting a viable and promising path to improving long-context reasoning of LLMs.","authors":[{"name":"Zhennan Wan","position":0},{"name":"Jianfei Chen","position":1}],"categories":["cs.CL","cs.LG"],"primaryCategory":"cs.CL","publishedAt":"2026-10-21T17:59:14Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30132"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30132"}]},"rank":192,"reason":"new","score":0.1524},{"paper":{"id":"arxiv:2610.30121","arxivId":"2610.30121","doi":"10.48550/arXiv.2610.30121","title":"Geometric Algebra for Neural Networks: A Survey","abstract":"Geometric algebra (GA) unifies geometric primitives and transformations in a single algebraic framework. Neural networks designed around this framework often represent features as multivectors and constrain layer operations to respect the algebraic structure, enabling principled treatment of geometric transformations and, in many cases, built-in equivariance. This survey reviews geometric algebra neural networks, including their building blocks, equivariant properties, application domains, empirical performance, and software support. We provide an algebra-selection guide, an empirical comparison of GA-based and other equivariant architectures, and a failure-mode analysis covering training pathologies, deployment constraints, and performance limitations. Across the surveyed literature, the strongest evidence for GA-based architectures appears in settings with explicit geometric symmetries and heterogeneous geometric entities; evidence for broader advantages in parameter efficiency, interpretability, and dense prediction is comparatively limited and uneven. We therefore conclude that geometric algebra neural networks are not universally superior to well-tuned non-GA equivariant baselines, and outline open problems in algebra selection, numerical stability, scalable implementation, and reproducible evaluation.","authors":[{"name":"Tao Zhong","position":0},{"name":"Kaison Fong","position":1},{"name":"Keqin Wang","position":2},{"name":"Christine Allen-Blanchette","position":3}],"categories":["eess.IV","stat.ML"],"primaryCategory":"eess.IV","publishedAt":"2026-10-21T17:57:13Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30121"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30121"}]},"rank":193,"reason":"new","score":0.1524},{"paper":{"id":"arxiv:2610.30116","arxivId":"2610.30116","doi":"10.48550/arXiv.2610.30116","title":"Environment Duels: A Microcosm of LLM Co-Evolution","abstract":"Through a tangle of real-world interactions, from synthetic training pipelines to data shared on the open web, frontier LLMs absorb what their predecessors and competitors create. How well does each model play its role in this ecosystem, as a creator and as a solver? We introduce Environment Duels, a dynamic agentic benchmark that evaluates both capabilities on interactive environments. Each model writes environments, each with a privileged hint, aiming for tasks that capable solvers fail without the hint and pass with it. Across nine frontier LLMs, solving and authoring prove to be distinct skills: the strongest solvers are not the strongest authors, while smaller models learn more from how others played and write the hardest environments and the most useful hints. Beyond the scores, the pairwise interactions form a microcosm of how models affect one another. Head-to-head outcomes are transitive and so yield a stable ranking; independent authors converge on the same designs; and in a reinforcement-learning study with Qwen3.8-27B, the learner's hint lift on an author's environments tracks how much training on those environments improves it on held-out reasoning and agentic benchmarks, whether the environments are its own or another model's. Each of these patterns has a counterpart in how models shape one another outside the duel. By staging these interactions under controlled conditions, Environment Duels offers a window onto how LLMs teach, challenge and learn from one another. Code is available at https://github.com/LLM-Duels/EnvDuels.","authors":[{"name":"Shibo Jin","position":0},{"name":"Zhiqiu Xu","position":1},{"name":"Yiling Zou","position":2},{"name":"Shangyu Gong","position":3},{"name":"Mayur Naik","position":4}],"categories":["cs.CL","cs.AI"],"primaryCategory":"cs.CL","publishedAt":"2026-10-21T17:55:31Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30116"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30116"}]},"rank":194,"reason":"new","score":0.1524},{"paper":{"id":"arxiv:2610.30114","arxivId":"2610.30114","doi":"10.48550/arXiv.2610.30114","title":"The End of Consensus Research?","abstract":"Consensus is a fundamental problem in distributed computing and at the core of state machine replication (SMR). Despite decades of active human research, discovering practical low-latency solutions to the problem is still open, as it requires combining protocol knowledge, correctness reasoning, implementation, and experimental evaluation. We report on how AI helped us discover A-greement, a new consensus protocol for geo-replicated state machines that significantly outperforms state-of-the-art solutions. Averaged across six deployments with 5 or 9 replicas, A-greement reduces the mean latency of a geo-distributed key-value store over the fastest baseline by 12.2% and 33.3% on write-heavy and read-heavy workloads, respectively, and matches the fastest baseline on read-only workloads. A-greement combines lease-authorized local reads, direct client learning for leader-ordered writes, and a fast fixed-quorum per-key write path, selected according to network topology and workload. At the heart of our discovery lies AI-greement, an agent harness for the \"autonomous\" discovery, implementation, experimentation, and verification of candidate protocols. AI-greement integrates (1) a concept-oriented protocol knowledge base, (2) a modular implementation skeleton, and (3) a Lean-based verification framework with reusable safety definitions, reference specifications, and proofs. AI-greement is of independent interest: its Lean formalization uncovers several previously unreported vulnerabilities in existing consensus protocols, and it provides guidelines on how domain-specific harnesses can turn general-purpose coding agents into effective tools for (distributed) computing research. Yet we argue that human consensus experts remain essential before, during, and after the search: agents reshape the researcher's role rather than end consensus research.","authors":[{"name":"Clément Burgelin","position":0},{"name":"Rachid Guerraoui","position":1},{"name":"Zicong Hong","position":2},{"name":"Yinan Hu","position":3},{"name":"Anne-Marie Kermarrec","position":4},{"name":"Rafael Pires","position":5},{"name":"Rasmus Moorits Veski","position":6},{"name":"Martijn de Vos","position":7}],"categories":["cs.DC","cs.AI"],"primaryCategory":"cs.DC","publishedAt":"2026-10-21T17:55:19Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30114"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30114"}]},"rank":195,"reason":"new","score":0.1524},{"paper":{"id":"arxiv:2610.30104","arxivId":"2610.30104","doi":"10.48550/arXiv.2610.30104","title":"BatchBench: Benchmarking AI Robustness to Batch Effects and Domain Shift in Biomedical Images - A JUMP Cell Painting Leaderboard","abstract":"Despite recent advances, current AI systems remain highly susceptible to domain shifts, particularly in biomedical imaging, where images are usually acquired under substantially different experimental conditions. Although recent domain adaptation methods have made substantial progress, significant performance gaps under domain shift persist. To measure progress in this regard, we introduce BatchBench, a leaderboard that evaluates and compares the robustness of AI systems on biomedical images from unseen experimental batches and data sources. BatchBench is built on JUMP-CP, one of the largest publicly available biomedical imaging datasets. Generated using the Cell Painting assay, JUMP-CP exhibits pronounced batch effects and domain shifts, making it a challenging testbed for evaluating robustness to distributional changes.","authors":[{"name":"Sonja Topf","position":0},{"name":"Ana Sanchez-Fernandez","position":1},{"name":"Günter Klambauer","position":2}],"categories":["cs.LG","cs.AI"],"primaryCategory":"cs.LG","publishedAt":"2026-10-21T17:50:22Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30104"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30104"}]},"rank":196,"reason":"new","score":0.1524},{"paper":{"id":"arxiv:2610.30101","arxivId":"2610.30101","doi":"10.48550/arXiv.2610.30101","title":"Machine Learning-Enabled Wireless Power Transfer: A Task-Oriented Review of Structural Optimization, State and Parameter Estimation, and Adaptive Control","abstract":"Wireless power transfer (WPT) systems must maintain efficient and safe energy delivery under variations in geometry, coupling, load, alignment, and operating conditions. Machine-learning methods are increasingly applied to accelerate structural design, estimate operating parameters and states, and support adaptive control and scheduling. This review develops a task-oriented framework organized around three interconnected domains: structural design and optimization, parameter and state estimation, and control and optimization. Supervised learning, reinforcement learning, and hybrid or physics-informed methods are compared in terms of data requirements, dependence on physical models, information sources, computational requirements, accuracy, online response, application scope, and optimization objectives. Supervised learning provides rapid surrogate modeling, regression, and classification when representative labeled data are available. Reinforcement learning addresses sequential control and scheduling but introduces challenges in reward design, safe exploration, and simulation-to-reality transfer. Hybrid methods integrate physical knowledge with data-driven models to improve data efficiency and physical consistency. The principal barriers to engineering deployment include limited and biased datasets, weak out-of-distribution generalization, insufficient physical consistency, unsafe online adaptation, embedded-resource constraints, and the absence of standardized benchmarks. Future research should prioritize uncertainty-aware and physics-informed learning, safe simulation-to-reality transfer, lightweight edge deployment, distributed coordination, reproducible evaluation, and digital-twin-assisted development.","authors":[{"name":"Denghao Luo","position":0},{"name":"Yiheng Jiang","position":1},{"name":"Jiaqi Zhao","position":2},{"name":"Yuan Gao","position":3},{"name":"Changsong Cai","position":4},{"name":"Weiyi Huang","position":5},{"name":"Yun Zhang","position":6}],"categories":["eess.SY","cs.SY"],"primaryCategory":"eess.SY","publishedAt":"2026-10-21T17:48:23Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30101"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30101"}]},"rank":197,"reason":"new","score":0.1524},{"paper":{"id":"arxiv:2610.30099","arxivId":"2610.30099","doi":"10.48550/arXiv.2610.30099","title":"Differentially Private Sparse Reward Estimation with Preference Feedback","abstract":"Reinforcement Learning with Human Feedback (RLHF) has become a central paradigm for aligning Large Language Models with human values. However, the reliance on sensitive preference data necessitates rigorous differential privacy (DP) guarantees to prevent user information leakage. Existing private alignment methods typically assume dense reward parameterizations, resulting in estimation error bounds that scale polynomially with the ambient {model size $d$}. Such dependence is prohibitive in modern {overparameterized models} where $d$ is vast. In this work, we study \\emph{Private Sparse RLHF} problem to address this limitation, assuming that human preferences (reward model) are governed by a subset of $s^* \\ll d$ relevant features. We provide the first comprehensive theoretical framework for differentially private sparse reward estimation from pairwise preference feedback under the Bradley-Terry-Luce model, and introduce efficient algorithms for both Sample-DP and Label-DP settings. Theoretically, under the squared $\\ell_2$ parameter estimation error, we establish upper bounds of $\\widetilde{O}(s^*/n + {s^*}^2/n^2\\varepsilon^2)$ for $(\\varepsilon,δ)$-Sample-DP and $\\widetilde{O}(s^*/n + s^*/n\\varepsilon^2)$ for $\\varepsilon$-Label-DP based on randomized response, where $n$ is the training data size. For the lower bounds, we show that the Sample-DP rate is nearly optimal, and that the guarantee for our Label-DP method cannot be further improved within the randomized-response framework. We also provide a general lower bound for the problem with Label-DP. Empirically, our framework demonstrates superior efficacy over dense baselines across sentiment generation and safety alignment benchmarks, offering a significantly more favorable privacy-utility trade-off.","authors":[{"name":"Meng Ding","position":0},{"name":"Mingxi Lei","position":1},{"name":"Jie Zhang","position":2},{"name":"Jinyan Liu","position":3},{"name":"Di Wang","position":4}],"categories":["cs.LG","cs.AI"],"primaryCategory":"cs.LG","publishedAt":"2026-10-21T17:45:42Z","version":1,"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30099"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30099"}]},"rank":198,"reason":"new","score":0.1524},{"paper":{"id":"arxiv:2610.30084","arxivId":"2610.30084","doi":"10.48550/arXiv.2610.30084","title":"Why the Swing Parameters Are Sloppy","abstract":"Evangelou, Stanković, Kevrekidis and Transtrum compared analytic and data-driven assessments of parameter identifiability for the infinite-bus synchronous generator. Fisher information with the manifold boundary approximation method (MBAM), and output diffusion maps with geometric harmonics, report the same six identifiable parameters on the late window $t \\in [3, 5]\\text{ s}$, and both place the inertia $H$ and the damping $D$ among the unidentifiable ones. We explain this result from the geometry of the dynamics rather than from that of the parameter-to-data map. The rotor swing is a fast fibre that has collapsed onto a two-dimensional normally attracting invariant manifold (NAIM) of the transient fluxes before the window opens. The zeroth-order NAIM is exactly independent of $H$, $D$, $T_{d0}''$ and $T_{q0}''$, four of the five MBAM limits. We prove, under stated smoothness and rank conditions, that each of the four limits is a boundary of the model manifold at a distance equal, to first order in the parameter, to its current profiled speed. The same analysis predicts how identifiability changes with the experiment: neither a larger initial speed deviation nor a load change restores $H$ or $D$ in the late window, whereas the early window $[0, 2]\\text{ s}$ makes $H$ the best-determined parameter. Evaluated in a physical metric fixed by synchrophasor accuracy limits, the late window identifies five parameters rather than six and the early window nine. The spiked-covariance structure of the linearised trajectory map explains why the principal components of whitened, sampled trajectories reproduce the prior-scaled Fisher spectrum, and regression on those components attains the Fisher-predicted recovery error in 113 of 114 parameter--scenario pairs.","authors":[{"name":"M Parimi","position":0},{"name":"Shashank Varma","position":1},{"name":"Sunny Kumar","position":2},{"name":"S R Wagh","position":3},{"name":"Navdeep Singh","position":4}],"categories":["eess.SY","cs.SY"],"primaryCategory":"eess.SY","publishedAt":"2026-10-21T17:33:25Z","version":1,"license":"http://creativecommons.org/licenses/by/4.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30084"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30084"}]},"rank":199,"reason":"new","score":0.1524},{"paper":{"id":"arxiv:2610.30082","arxivId":"2610.30082","doi":"10.48550/arXiv.2610.30082","title":"When Standardization Spreads Factor Variation Across the Spectrum","abstract":"Standardizing variables by their sample standard deviations is routine in multivariate analysis and removes marginal scale differences before dependence is assessed. The resulting correlation matrix captures both dominant common components and dependence beyond them. A central inferential difficulty arises because the same random sample variances normalize the entire spectrum. In spiked covariance models, the sampling errors of leading and residual eigenvalues can become asymptotically independent. Sample standardization fundamentally changes this behavior and can preserve their dependence, allowing strong factors to remain a major source of uncertainty even after the leading eigenvalues are excluded. To obtain valid inference for both residual and full spectrum statistics, we establish their joint sampling law for sample correlation matrices with diverging spikes. The theory reveals exact trace compensation and statistic dependent transitions between asymptotic independence and dependence as the spikes grow. It leads to tests of an identity correlation matrix under strong factors and a feasible two sample comparison of residual spectral concentration. A sex balanced HapMap analysis illustrates the substantial effect of factor variation on residual spectrum inference.","authors":[{"name":"Tanyang Zhou","position":0},{"name":"Yanqing Yin","position":1}],"categories":["math.ST","stat.TH"],"primaryCategory":"math.ST","publishedAt":"2026-10-21T17:32:01Z","version":1,"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","links":[{"kind":"abs","url":"https://arxiv.org/abs/2610.30082"},{"kind":"pdf","url":"https://arxiv.org/pdf/2610.30082"}]},"rank":200,"reason":"new","score":0.1524}]}