arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~

期刊&会议

NeurIPS

Conference on Neural Information Processing Systems · 会议 · Machine Learning

至 收录 73
2507.11688 2026-06-11 cs.LG 版本更新

Composing Linear Layers from Irreducibles

从不可约元组合线性层

Travis Pence, Daisuke Yamada, Vikas Singh

机构 * University of Wisconsin-Madison(威斯康星大学麦迪逊分校)

AI总结 提出用Clifford代数将线性层分解为双向量(几何基元)的组合,仅需O(log^2 d)参数,在LLM注意力投影中匹配强基线性能。

Comments 35 Pages, 11 Tables, 6 Figures, Appearing in NeurIPS 2025

Journal ref Advances in Neural Information Processing Systems 38 (2025)

URL PDF HTML 收藏
2512.03077 2026-06-11 cs.CY cs.AI 版本更新

Irresponsible AI: big tech's influence on AI research and associated impacts

不负责任的人工智能:大型科技公司对AI研究的影响及相关影响

Alex Hernandez-Garcia, Alexandra Volokhova, Ezekiel Williams, Dounia Shaaban Kabakibo, Mélisande Teng

机构 * Big Tech(大科技公司)

AI总结 本文指出大型科技公司对AI研究的不成比例影响推动了不负责任的AI发展,并加剧了环境和社会负面影响,呼吁研究者通过集体行动加以抵制。

Comments Presented as a spotlight oral at the International Conference on Machine Learning 2026 (Position Paper Track). First version presented at NeurIPS 2025 Workshop on Algorithmic Collective Action

URL PDF HTML 收藏
2509.16456 2026-06-11 cs.AI 版本更新

GPO: Learning from Critical Steps to Improve LLM Reasoning

GPO:从关键步骤中学习以改进大语言模型推理

Jiahao Yu, Zelei Cheng, Xian Wu, Xinyu Xing

机构 * Department of Computer Science Northwestern University(计算机科学系西北大学) AI Foundations Capital One(人工智能基础资本 one) Meta AI

AI总结 提出引导式关键优化(GPO)微调策略,通过识别推理轨迹中的关键步骤并优先学习,显著提升大语言模型的多步推理能力。

Comments 39th Conference on Neural Information Processing Systems (NeurIPS 2025)

URL PDF HTML 收藏
2510.08073 2026-06-11 cs.CV cs.LG 版本更新

Physics-Driven Spatiotemporal Modeling for AI-Generated Video Detection

物理驱动的时空建模用于AI生成视频检测

Shuhai Zhang, ZiHao Lian, Jiahao Yang, Daiyuan Li, Guoxuan Pang, Feng Liu, Bo Han, Shutao Li, Mingkui Tan

机构 * South China University of Technology(华南理工大学) University of Science and Technology of China(中国科学技术大学) Key Laboratory of Big Data and Intelligent Robot, Ministry of Education(教育部大数据与智能机器人重点实验室) Pazhou Lab(琶洲实验室) University of Melbourne(墨尔本大学) Hunan University(湖南大学) Hong Kong Baptist University(香港 Baptist大学)

AI总结 提出基于概率流守恒的物理驱动AI生成视频检测范式,通过归一化时空梯度(NSG)统计量捕捉物理异常,结合预训练扩散模型估计NSG,并利用最大均值差异(MMD)进行检测,在Recall和F1-Score上分别提升16.00%和10.75%。

Comments Accepted at NeurIPS 2025 spotlight

URL PDF HTML 收藏
2510.02660 2026-06-11 cs.HC cs.AI 版本更新

When Researchers Say Mental Model/Theory of Mind of AI, What Are They Really Talking About?

当研究人员谈论AI的心理模型/心智理论时,他们究竟在说什么?

Xiaoyun Yin, Elmira Zahmat Doost, Shiwen Zhou, Garima Arya Yadav, Jamie C. Gorman

机构 * Center for Human, Artificial Intelligence, and Robot Teaming(人类、人工智能与机器人协同中心)

AI总结 本文指出当前AI心智理论研究混淆了行为预测与真实认知,提出应转向人机交互中的互惠心智理论框架。

Comments This work have been accepted in CogInterp @ NeurIPS 2025

URL PDF HTML 收藏
2510.04514 2026-06-10 cs.AI cs.CE cs.CL cs.CV stat.ME 版本更新

ChartAgent: A Multimodal Agent for Visually Grounded Reasoning in Complex Chart Question Answering

ChartAgent: 一种用于复杂图表问答中视觉基础推理的多模态智能体

Rachneet Kaur, Nishan Srishankar, Zhen Zeng, Sumitra Ganesh, Manuela Veloso

机构 * J.P. Morgan AI Research(摩根大通人工智能研究)

AI总结 提出ChartAgent框架,通过迭代分解查询为视觉子任务并利用图表专用视觉工具(如绘制注释、裁剪区域)进行空间域推理,在ChartBench和ChartX上取得最先进性能,尤其对无标注图表提升显著。

Comments Accepted at ACL 2026 (Main Conference). Also presented as an oral paper at the NeurIPS 2025 Multimodal Algorithmic Reasoning Workshop (https://marworkshop.github.io/neurips25/)

URL PDF HTML 收藏
2511.02603 2026-06-10 cs.CL 版本更新

CGES: Confidence-Guided Early Stopping for Efficient and Accurate Self-Consistency

CGES:面向高效准确自一致性的置信引导早停方法

Ehsan Aghazadeh, Ahmad Ghasemi, Hedyeh Beyhaghi, Hossein Pishro-Nik

机构 * University of Massachusetts Amherst(马萨诸塞大学阿姆赫斯特分校)

AI总结 提出贝叶斯框架CGES,通过自适应停止采样减少自一致性推理调用次数,在5个推理基准上平均减少58%调用且精度损失仅0.4个百分点。

Comments Extended version. A preliminary version was accepted at the Efficient Reasoning Workshop @ NeurIPS 2025. Code: https://github.com/EhsanAghazadeh/cges

URL PDF HTML 收藏
2310.05264 2026-06-10 cs.LG cs.CV 版本更新

The Emergence of Reproducibility and Generalizability in Diffusion Models

扩散模型中可重复性与泛化性的出现

Huijie Zhang, Jinfan Zhou, Yifu Lu, Minzhe Guo, Peng Wang, Liyue Shen, Qing Qu

机构 * CIFAR-10 dataset(CIFAR-10数据集)

AI总结 研究发现扩散模型在相同初始噪声和确定性采样器下,不同模型输出高度相似,且这种可重复性在记忆和泛化两种训练模式下均存在,对训练效率、模型隐私等有重要启示。

Comments NeurIPS Diffusion Model Workshop 2023 (best paper award), the Forty-first International Conference on Machine Learning (ICML 2024)

URL PDF HTML 收藏
2604.04251 2026-06-09 cs.AI cs.CY cs.LG 版本更新

MC-CPO: Mastery-Conditioned Constrained Policy Optimization for Pedagogically Safe Intelligent Tutoring Systems

MC-CPO:基于 mastery 的约束策略优化用于教学安全的智能辅导系统

Oluseyi Olukola, Nick Rahimi

机构 * School of Computing Sciences(计算科学学院) Computer Engineering, University of Southern Mississippi, Hattiesburg, MS 39406, USA(计算机工程,密西西比大学,哈特斯伯格,MS 39406,USA)

AI总结 本文提出 MC-CPO 框架,通过结构化约束解决教学安全问题,提升学习者知识掌握率,实验证明其在两个平台上的效果显著。

Comments 35 pages, 8 figures. v2: Major revision adding real-world validation on Junyi Academy (16.2M interactions, 72,758 students) and XES3G5M (NeurIPS 2023, 5.1M interactions, 14,453 students). Revised title and abstract. Submitted to Computers and Education: Artificial Intelligence

URL PDF HTML 收藏
2502.19049 2026-06-09 cs.LG 版本更新

In-Context Learning of Stochastic Differential Equations with Foundation Inference Models

基于基础推理模型的随机微分方程上下文学习

Patrick Seifner, Kostadin Cvejoski, David Berghaus, Cesar Ojeda, Ramses J. Sanchez

机构 * Lamarr Institute(拉马尔研究所) University of Bonn(波恩大学) Fraunhofer IAIS(弗劳恩霍夫智能系统研究所) University of Potsdam(波茨坦大学)

AI总结 提出FIM-SDE,一种预训练识别模型,通过上下文学习从噪声时间序列中零样本估计低维SDE的漂移和扩散函数,并支持快速微调,在合成和真实数据上表现鲁棒。

Comments Accepted at NeurIPS 2025. The previous version appeared under the title "Foundation Inference Models for Stochastic Differential Equations: A Transformer-based Approach for Zero-shot Function Estimation.";

Journal ref 39th Conference on Neural Information Processing Systems (NeurIPS 2025)

URL PDF HTML 收藏
2507.00322 2026-06-09 cs.CL cs.AI cs.SE 版本更新

Failure by Interference: Language Models Make Balanced Parentheses Errors When Faulty Mechanisms Overshadow Sound Ones

干扰导致的失败:当有缺陷机制掩盖健全机制时,语言模型在平衡括号任务中出错

Daking Rai, Samuel Miller, Kevin Moran, Ziyu Yao

机构 * George Mason University(乔治·马歇尔大学) University of Central Florida(中央佛罗里达大学) Department of Computer Science(计算机科学系)

AI总结 研究揭示语言模型在平衡括号任务中出错的原因:部分组件实现可靠机制,而其他组件引入噪声,当噪声机制主导时导致错误。提出RASteer方法,通过增强可靠组件贡献,将部分模型准确率从0%提升至近100%,并在算术推理任务中取得约20%的性能提升。

Comments 23 pages, 10 figures, accepted for NeurIPS 2025

URL PDF HTML 收藏
2411.03253 2026-06-09 cs.LG cs.AI cs.DS 版本更新

Discovering Data Structures: Nearest Neighbor Search and Beyond

发现数据结构:最近邻搜索及其他

Omar Salemohamed, Laurent Charlin, Shivam Garg, Vatsal Sharan, Gregory Valiant

机构 * Université de Montréal(蒙特利尔大学) Mila HEC Montréal(蒙特利尔高等商学院) Microsoft Research(微软研究院) University of Southern California(南加州大学) Stanford University(斯坦福大学)

AI总结 提出一个端到端学习数据结构的通用框架,自动适应数据分布并控制查询与空间复杂度,在最近邻搜索中逆向工程出二分搜索、插值搜索、k-d树和局部敏感哈希等算法。

Comments Neurips 2025 Version

URL PDF HTML 收藏
2502.00527 2026-06-08 cs.LG cs.CL 版本更新

PolarQuant: Leveraging Polar Transformation for Efficient Key Cache Quantization and Decoding Acceleration

PolarQuant: 利用极坐标变换实现高效键缓存量化和解码加速

Songhao Wu, Ang Lv, Xiao Feng, Yufei Zhang, Xun Zhang, Guojun Yin, Wei Lin, Rui Yan

机构 * Gaoling School of Artificial Intelligence, Renmin University of China(中国人民大学北京校区人工智能学院) ShanghaiTech University(上海科技大学) Meituan(美团)

AI总结 提出PolarQuant方法,通过将键向量分组为二维子向量并编码为量化半径和极角,解决键缓存量化中的异常值问题,同时通过查表加速解码,保持全精度模型性能。

Comments NeurIPS 2025 version with minor revisions to the methodology

URL PDF HTML 收藏