arXivDaily arXiv每日学术速递 周一至周五更新

期刊&会议

International Conference on Machine Learning · 会议 · Machine Learning

共收录 11797
2602.16400 2026-02-19 cs.LG

Easy Data Unlearning Bench

易于数据删除的基准

Roy Rinberg, Pol Puigdemont, Martin Pawelczyk, Volkan Cevher

机构 * epfl(苏黎世联邦理工学院) harvard(哈佛大学)

AI总结 本文提出了一种统一的基准测试套件,通过KLoM度量简化机器删除算法的评估,促进研究和最佳实践。

Comments ICML 2025 Workshop on Machine Unlearning for Generative AI

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.15183 2026-02-18 cs.LG cs.CL

Seeing to Generalize: How Visual Data Corrects Binding Shortcuts

看见以泛化:视觉数据如何纠正绑定捷径

Nicolas Buzeta, Felipe del Rio, Cristian Hinostroza, Denis Parra, Hans Lobel, Rodrigo Toro Icarte

机构 * Department of Computer Science, Pontificia Universidad Católica, Santiago, Chile(计算机科学系,天主教大学,圣地亚哥,智利)

AI总结 视觉数据训练可增强模型在单模态任务中的推理与泛化能力,通过改变绑定策略提升分布外性能。

Comments Submitted to ICML 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2305.03571 2026-02-18 eess.SP cs.IT cs.LG math.IT stat.ML

Model-free Reinforcement Learning of Semantic Communication by Stochastic Policy Gradient

无模型强化学习的语义通信

Edgar Beck, Carsten Bockelmann, Armin Dekorsy

AI总结 本文提出利用随机策略梯度方法设计语义通信系统,通过强化学习实现发射端与接收端分离,无需已知信道模型,实现信息率节省。

Comments Accepted for publication in IEEE International Conference on Machine Learning for Communication and Networking (ICMLCN 2024), Source Code: https://github.com/ant-uni-bremen/SINFONY

详情

展开后加载摘要…

URL PDF HTML 收藏
2601.21812 2026-02-17 stat.ML cs.AI cs.LG

A Decomposable Forward Process in Diffusion Models for Time-Series Forecasting

扩散模型中用于时间序列预测的可分解前向过程

Francisco Caldas, Sahil Kumar, Cláudia Soares

机构 * Department of Informatics, NOVA University of Lisbon, Caparica, Portugal(信息学院,里斯本NOVA大学,葡萄牙卡帕里卡)

AI总结 本文提出了一种可分解的前向扩散过程,通过频谱分解提升时间序列预测的准确性,且计算开销低。

Comments submitted to ICML'26

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.05354 2026-02-17 cs.AI

PATHWAYS: Evaluating Investigation and Context Discovery in AI Web Agents

PATHWAYS:评估人工智能网络代理中的调查与上下文发现

Shifat E. Arman, Syed Nazmus Sakib, Tapodhir Karmakar Taton, Nafiul Haque, Shahrear Bin Amin

机构 * Robotics and Mechatronics Engineering, University of Dhaka, Dhaka, Bangladesh(机器人与机电工程系,达卡大学,达卡,孟加拉国)

AI总结 PATHWAYS评估了人工智能网络代理在多步骤任务中发现和利用隐藏上下文的能力,揭示了当前架构在适应性调查和判断覆盖方面的不足。

Comments 35 pages, 13 figures

Journal ref Under Review in ICML 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2510.03669 2026-02-17 cs.LG cs.CL

Token Hidden Reward: Steering Exploration-Exploitation in Group Relative Deep Reinforcement Learning

令牌隐藏奖励:在组相对深度强化学习中引导探索-利用

Wenlong Deng, Yi Ren, Yushu Li, Boying Gong, Danica J. Sutherland, Xiaoxiao Li, Christos Thrampoulidis

机构 * University of British Columbia(不列颠哥伦比亚大学) Vector Institute(向量研究所) Amii(阿米人工智能研究所) UC Berkeley(加州大学伯克利分校)

AI总结 令牌隐藏奖励通过调整组相对策略优化的学习信号,引导探索-利用平衡,提升大语言模型在推理任务中的表现。

Comments Full version of submission to 2nd AI for Math Workshop@ ICML 2025 (best paper)

详情

展开后加载摘要…

URL PDF HTML 收藏
2512.09407 2026-02-17 cs.CV

Geometry-to-Image Synthesis-Driven Generative Point Cloud Registration

基于几何到图像合成的生成式点云配准

Haobo Jiang, Jin Xie, Jian Yang, Liang Yu, Jianmin Zheng

机构 * ANGEL CorpLab and College of Computing and Data Science, Nanyang Technological University, Singapore(ANGEL CorpLab和计算与数据科学学院,南洋理工大学,新加坡) Alibaba Cloud, Alibaba Group, China(阿里巴巴云,阿里巴巴集团,中国) PCA Lab, VCIP, College of Computer Science, Nankai University, China(PCA实验室,VCIP,计算机科学学院,南开大学,中国) State Key Laboratory for Novel Software Technology & Schoolof Intelligence Science and Technology, Nanjing University, China(新型软件技术国家重点实验室与智能科学与技术学校,南京大学,中国)

AI总结 本文提出生成式点云配准方法,通过生成跨视角一致的图像对,结合几何与颜色特征融合,提升3D配准性能。

Comments Journal extension of the ICML 2025 paper "Generative Point Cloud Registration". This version adopts a new title, and includes substantial methodological improvements, additional experiments, and extended analysis. Under review at IEEE TPAMI

详情

展开后加载摘要…

URL PDF HTML 收藏
2406.04112 2026-02-16 cs.LG cs.AI eess.SP stat.ML

Compressible Dynamics in Deep Overparameterized Low-Rank Learning & Adaptation

深度过参数化低秩学习与适应中的可压缩动力学

Can Yaras, Peng Wang, Laura Balzano, Qing Qu

机构 * Department of Electrical Engineering \& Computer Science, University of Michigan

AI总结 本研究提出Deep LoRA方法,通过深度过参数化低秩学习与适应中的可压缩动力学,提升语言模型微调效率并减少过拟合。

Comments Accepted at ICML'24 (Oral)

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.11169 2026-02-13 cs.CL cs.AI cs.LG

Disentangling Direction and Magnitude in Transformer Representations: A Double Dissociation Through L2-Matched Perturbation Analysis

解构Transformer表示中的方向与幅度:通过L2匹配扰动分析的双重解离

Mangadoddi Srikar Vardhan, Lekkala Sai Teja

AI总结 研究揭示Transformer表示中方向与幅度在语言建模和语法处理中的不同作用,通过L2匹配扰动分析发现方向扰动影响注意力路径,幅度扰动影响语法判断,且解离依赖于架构选择。

Comments 15 pages, 7 figures. will Submit to ICML 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2502.14921 2026-02-13 cs.CL cs.CR cs.LG

The Canary's Echo: Auditing Privacy Risks of LLM-Generated Synthetic Text

知更鸟的回声:审计大语言模型生成合成文本的隐私风险

Matthieu Meeus, Lukas Wutschitz, Santiago Zanella-Béguelin, Shruti Tople, Reza Shokri

机构 * Imperial College London(帝国理工学院伦敦分校) Microsoft(微软公司) National University of Singapore(新加坡国立大学)

AI总结 本文提出通过设计具有分布内前缀和高困惑度后缀的知更鸟,提高基于数据的MIAs的威力,以更准确评估LLM生成合成数据的隐私风险。

Comments 42nd International Conference on Machine Learning (ICML 2025)

Journal ref Proc. Mach. Learn. Res. 267 (2025) 43557-43580

详情

展开后加载摘要…

URL PDF HTML 收藏
2501.02087 2026-02-13 cs.LG stat.ML

Beyond CVaR: Leveraging Static Spectral Risk Measures for Enhanced Decision-Making in Distributional Reinforcement Learning

超越CVaR:利用静态谱风险度量提升分布式强化学习中的决策质量

Mehrdad Moghimi, Hyejin Ku

机构 * Department of Mathematics and Statistics, York University, Toronto, Canada(数学与统计学系,约克大学,多伦多,加拿大)

AI总结 本文提出了一种具有收敛保证的DRL算法,优化更广泛的静态谱风险度量,提升决策质量并优于现有方法。

Comments Accepted at ICML 2025

Journal ref Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:44571-44593, 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2601.22860 2026-02-12 math.NA cs.AI cs.NA

Bayesian Interpolating Neural Network (B-INN): a scalable and reliable Bayesian model for large-scale physical systems

贝叶斯插值神经网络(B-INN):一种可扩展且可靠的贝叶斯模型,用于大规模物理系统

Chanwook Park, Brian Kim, Jiachen Guo, Wing Kam Liu

机构 * Department of Mechanical Engineering, Northwestern University, Evanston, Illinois, USA(机械工程系,西北大学,伊利诺伊州埃文斯顿) Department of Engineering Sciences and Applied Mathematics, Northwestern University, Evanston, Illinois, USA(工程科学与应用数学系,西北大学,伊利诺伊州埃文斯顿) Department of Mathematics, Northwestern University, Evanston, Illinois, USA(数学系,西北大学,伊利诺伊州埃文斯顿) Applied Mechanics Program, Northwestern University, Evanston, Illinois, USA(应用力学项目,西北大学,伊利诺伊州埃文斯顿)

AI总结 B-INN通过结合高阶插值理论与张量分解,提供一种高效且可靠的贝叶斯模型,用于大规模物理系统的不确定性量化与主动学习。

Comments 8 pages, 6 figures, ICML conference full paper submitted

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.09869 2026-02-11 cs.LG

Statistical benchmarking of transformer models in low signal-to-noise time-series forecasting

在低信噪比时间序列预测中对变换器模型进行统计基准测试

Cyril Garcia, Guillaume Remy

机构 * Cyril Garcia Guillaume Remy

AI总结 本文提出了一种在低信噪比时间序列预测中表现更优的双向注意力变换器,并引入了动态稀疏化方法以提升模型在噪声环境下的性能。

Comments Submitted to ICML

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.09784 2026-02-11 cs.LG cs.CL

Circuit Fingerprints: How Answer Tokens Encode Their Geometrical Path

电路指纹:答案标记如何编码其几何路径

Andres Saurez, Neha Sengar, Dongsoo Har

机构 * Korea Advanced Institute of Science and Technology(韩国科学技术院)

AI总结 通过几何对齐实现电路发现,揭示transformer电路本质上是几何结构,实现可控引导并提升情感分类准确率。

Comments Submitted to ICML 2026. 15 pages, 11 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.09783 2026-02-11 cs.LG cs.CL

Why Linear Interpretability Works: Invariant Subspaces as a Result of Architectural Constraints

为何线性可解释性有效:架构约束下的不变子空间

Andres Saurez, Yousung Lee, Dongsoo Har

机构 * Korea Advanced Institute of Science and Technology(韩国科学技术院)

AI总结 本文提出不变子空间必要性定理,揭示Transformer架构约束下线性可解释性方法有效的原因,统一了线性探针和稀疏自编码器。

Comments Submitted to ICML 2026. 19 pages, 13 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.09158 2026-02-11 cs.LG cs.AI

What do Geometric Hallucination Detection Metrics Actually Measure?

几何幻觉检测度量实际上测量什么?

Eric Yeats, John Buckheit, Sarah Scullen, Brendan Kennedy, Loc Truong, Davis Brown, Bill Kay, Cliff Joslyn, Tegan Emerson, Michael J. Henry, John Emanuello, Henry Kvinge

机构 * Pacific Northwest National Laboratory(太平洋西北国家实验室) University of Washington(华盛顿大学) University of Pennsylvania(宾夕法尼亚大学) Colorado State University(科罗拉多州立大学) University of Texas, El Paso(德克萨斯大学埃尔帕索分校) Laboratory for Advanced Cybersecurity Research, National Security Agency(国家安全局高级网络安全研究实验室)

AI总结 本文研究几何统计在检测幻觉中的作用,通过合成数据集分析不同属性对幻觉检测的影响,并提出归一化方法提升多领域检测性能。

Comments Published at the 2025 ICML Workshop on Reliable and Responsible Foundation Models

详情

展开后加载摘要…

URL PDF HTML 收藏
2212.00133 2026-02-11 cs.LG math.OC stat.ML

Universal Neural Optimal Transport

通用神经最优传输

Jonathan Geuter, Gregor Kornhardt, Ingimar Tomasson, Vaios Laschos

机构 * Harvard John A. Paulson School of Engineering and Applied Sciences(哈佛大学约翰·A·保罗森工程与应用科学学院) Kempner Institute at Harvard University(哈佛大学凯门研究所) Weierstrass Institute, Berlin, Germany(魏尔斯特拉斯研究所)

AI总结 UNOT通过傅里叶神经算子和对抗训练,实现高效准确的最优传输距离和计划预测,并在Wasserstein空间几何和Sinkhorn算法初始化中表现出色。

Comments 37 pages, 19 figures, accepted to ICML 2025

Journal ref Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:19196-19232, 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.08858 2026-02-10 cs.CV cs.AI

FlattenGPT: Depth Compression for Transformer with Layer Flattening

FlattenGPT: Transformer中基于层扁平化的深度压缩

Ruihan Xu, Qingpei Guo, Yao Zhu, Xiangyang Ji, Ming Yang, Shiliang Zhang

机构 * State Key Laboratory of Multimedia Information Processing, School of Computer Science, Peking University(多媒体信息处理国家重点实验室,计算机科学学院,北京大学) Tsinghua University(清华大学)

AI总结 FlattenGPT通过层扁平化技术实现Transformer模型的深度压缩,有效提升效率并保持性能,适用于多种模型类型和参数规模。

Comments Submitted to ICML 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.04915 2026-02-10 cs.LG cs.AI

SLAY: Geometry-Aware Spherical Linearized Attention with Yat-Kernel

SLAY:基于Yat-kernel的几何感知球面线性化注意力

Jose Miguel Luna, Taha Bouhsine, Krzysztof Choromanski

机构 * Columbia University(哥伦比亚大学) Google DeepMind(谷歌DeepMind)

AI总结 SLAY通过几何感知的球面线性化注意力机制,实现了接近softmax注意力的性能,同时保持线性时间复杂度。

Comments ICML 2026, 8 pages main body, 27 pages total

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.06229 2026-02-09 cs.LG cs.AI

SR4-Fit: An Interpretable and Informative Classification Algorithm Applied to Prediction of U.S. House of Representatives Elections

SR4-Fit:一种可解释且信息丰富的分类算法应用于美国众议院选举预测

Shyam Sundar Murali Krishnan, Dean Frederick Hougen

机构 * School of Computer Science(计算机科学学院) Gallogly College of Engineering(加洛格利工程学院) University of Oklahoma(俄克拉荷马大学)

AI总结 SR4-Fit是一种新型可解释分类算法,通过结合稀疏正则化和规则拟合技术,实现了在众议院选举预测中的高准确性和可解释性,同时优于现有黑箱和规则算法。

Comments 8 pages, 2 figures, 7 tables, to appear in the 24th IEEE AMLA International Conference on Machine Learning and Applications (ICMLA'25)

详情

展开后加载摘要…

URL PDF HTML 收藏
2403.01673 2026-02-06 stat.ML cs.AI cs.LG

CATS: Enhancing Multivariate Time Series Forecasting by Constructing Auxiliary Time Series as Exogenous Variables

CATS: 通过构建辅助时间序列作为外生变量增强多变量时间序列预测

Jiecheng Lu, Xu Han, Yan Sun, Shihao Yang

机构 * Georgia Institute of Technology(佐治亚理工学院) Amazon Web Services(亚马逊网络服务)

AI总结 CATS通过构建辅助时间序列作为外生变量,有效提升多变量时间序列预测的性能,实现高效且可转移的预测解决方案。

Comments Camera-ready version. Accepted at ICML 2024

Journal ref Proceedings of the Forty-first International Conference on Machine Learning (ICML 2024)

详情

展开后加载摘要…

URL PDF HTML 收藏
2309.16858 2026-02-06 stat.ML cs.LG

Improved Generalization Bounds for Transductive Learning by Transductive Local Complexity and Its Applications

通过传递学习局部复杂度改进传递学习的泛化界及其应用

Yingzhen Yang

机构 * School of Computing and Augmented Intelligence(计算与增强智能学院) Arizona State University(亚利桑那州立大学)

AI总结 本文通过引入传递局部复杂度,改进了传递学习的泛化界,并在二值类和核学习中取得了新的理论突破。

Comments The ICML 2025 conference version (https://openreview.net/pdf?id=NRVdvg7VMn) is a special case of this paper where the chain length is fixed at 2 (i.e.,$Q=2$, see Def. 5.1), and its main results follow directly from the results here. This paper further provides a nearly optimal excess risk bound for realizable transductive learning and a stronger bound for transductive kernel learning

详情

展开后加载摘要…

URL PDF HTML 收藏
2110.04598 2026-02-06 cs.LG

Self-explaining Neural Network with Concept-based Explanations for ICU Mortality Prediction

具有基于概念的解释的自解释神经网络用于ICU死亡率预测

Sayantan Kumar, Sean C. Yu, Thomas Kannampallil, Zachary Abrams, Andrew Michelson, Philip R. O. Payne

机构 * Department of Computer Science and Engineering, Washington University in St. Louis, St. Louis, MO, USA(计算机科学与工程系,华盛顿大学圣路易斯分校) Institute for Informatics, Washington University School of Medicine, St. Louis, MO, USA(信息学院,华盛顿大学医学学院) Department of Anaesthesiology, Washington University School of Medicine, St. Louis, MO, USA(麻醉学系,华盛顿大学医学学院) Department of Pulmonary Critical Care and Medicine, Washington University School of Medicine, St. Louis, MO, USA(呼吸科重症医学系,华盛顿大学医学学院)

AI总结 本文提出了一种基于概念的自解释神经网络,用于ICU患者死亡率预测,通过联合训练生成解释和预测,提升模型的可解释性与预测性能。

Comments Workshop on Interpretable ML in Healthcare at International Conference on Machine Learning (ICML 2022)

Journal ref BCB '22: Proceedings of the 13th ACM International Conference on Bioinformatics, Computational Biology and Health Informatics Article No.: 8, Pages 1 - 9, 2022

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.04904 2026-02-06 cs.LG cs.AI cs.MM eess.IV

DCER: Dual-Stage Compression and Energy-Based Reconstruction

DCER:双阶段压缩与基于能量的重建

Yiwen Wang, Jiahao Qin

机构 * Yiwen Wang(无) Jiahao Qin(无)

AI总结 DCER通过双阶段压缩和基于能量的重建解决多模态融合中的噪声和缺失模态问题,实现鲁棒性提升。

Comments 13 pages, 2 figures, 8 tables. Submitted to ICML 2026. Code will be available on GitHub

详情

展开后加载摘要…

URL PDF HTML 收藏
2502.07244 2026-02-06 cs.LG cs.AI stat.ML

Linear Transformers as VAR Models: Aligning Autoregressive Attention Mechanisms with Autoregressive Forecasting

线性变换器作为VAR模型:将自回归注意力机制与自回归预测对齐

Jiecheng Lu, Shihao Yang

机构 * Georgia Institute of Technology(佐治亚理工学院)

AI总结 本文提出SAMoVAR,一种将Transformer架构与自回归目标对齐的线性变换器变体,通过整合可解释的动态VAR权重,提升时间序列预测的性能和可解释性。

Comments Camera-ready version. Accepted at ICML 2025

Journal ref Proceedings of the Forty-second International Conference on Machine Learning (ICML 2025)

详情

展开后加载摘要…

URL PDF HTML 收藏
2502.01411 2026-02-06 cs.CV

Human Body Restoration with One-Step Diffusion Model and A New Benchmark

一步扩散模型与新基准的人体恢复

Jue Gong, Jingkai Wang, Zheng Chen, Xing Liu, Hong Gu, Yulun Zhang, Xiaokang Yang

机构 * Shanghai Jiao Tong University, China(上海交通大学) vivo Mobile Communication Co., Ltd, China(vivo移动通信有限公司)

AI总结 本文提出了一种一步扩散模型OSDHuman和新基准数据集PERSONA,用于提升人体恢复的视觉质量和定量指标。

Comments 8 pages, 9 figures. Accepted at ICML 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2410.03159 2026-02-06 cs.LG cs.AI stat.ML

WAVE: Weighted Autoregressive Varying Gate for Time Series Forecasting

WAVE:带有自回归和移动平均组件的加权自回归变门机制用于时间序列预测

Jiecheng Lu, Xu Han, Yan Sun, Shihao Yang

机构 * Georgia Institute of Technology(佐治亚理工学院)

AI总结 WAVE通过整合ARMA结构提升时间序列预测性能,结合自回归和移动平均组件,实现更高效的长程和局部时间模式捕捉。

Comments Camera-ready version. Accepted at ICML 2025

Journal ref Proceedings of the Forty-second International Conference on Machine Learning (ICML 2025)

详情

展开后加载摘要…

URL PDF HTML 收藏
2506.15732 2026-02-05 cs.AI cs.LG

Can LLMs Reconcile Knowledge Conflicts in Counterfactual Reasoning

大语言模型能否在反事实推理中调和知识冲突

Khurram Yamin, Gaurav Ghosal, Bryan Wilder

机构 * Department of Machine Learning(机器学习系) Carnegie Mellon University(卡内基梅隆大学)

AI总结 本研究探讨了大语言模型在反事实推理中整合参数知识的能力,发现其普遍挣扎并存在知识退化问题。

Comments ICML 2025 Workshop on Scaling up Intervention Models

详情

展开后加载摘要…

URL PDF HTML 收藏
2502.19758 2026-02-05 cs.LG cs.AI

Learning with Exact Invariances in Polynomial Time

在多项式时间内学习精确不变性

Ashkan Soleymani, Behrooz Tahmasebi, Stefanie Jegelka, Patrick Jaillet

机构 * School of CIT, MCML(信息科技学院、微系统实验室) MDSI, Technical University of Munich (TUM)(慕尼黑工业大学微系统研究所) MIT EECS(麻省理工学院电子工程与计算机科学系) MIT CSAIL(麻省理工学院计算机科学与人工智能实验室) MIT LIDS(麻省理工学院媒体实验室)

AI总结 本文提出了一种在多项式时间内实现精确不变性的算法,通过利用输入空间的几何属性,达到与核回归相同的泛化误差,是该领域首个实现精确不变性的多项式时间方法。

Journal ref International Conference on Machine Learning (ICML) 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2406.03682 2026-02-05 cs.LG

A Universal Class of Sharpness-Aware Minimization Algorithms

一种通用的锐度感知最小化算法类

Behrooz Tahmasebi, Ashkan Soleymani, Dara Bahri, Stefanie Jegelka, Patrick Jaillet

机构 * MIT CSAIL(麻省理工学院计算机科学与人工智能实验室) MIT LIDS(麻省理工学院领导力与决策科学研究所) MIT(麻省理工学院) TU Munich(慕尼黑技术大学) Google DeepMind(谷歌DeepMind)

AI总结 本文提出了一种通用的锐度感知最小化算法类,通过引入新的锐度度量,解决了神经网络中参数不变性问题,并展示了Frob-SAM和Det-SAM等实例。

Comments ICML 2024. Code is available at http://github.com/dbahri/universal_sam

Journal ref International Conference on Machine Learning (ICML) 2024

详情

展开后加载摘要…

URL PDF HTML 收藏