ICML

Stealing That Free Lunch: Exposing the Limits of Dyna-Style Reinforcement Learning

Brett Barkley, David Fridovich-Keil

机构 * Department of Computer Science, University of Texas at Austin, Austin, TX, USA（德克萨斯大学奥斯汀分校计算机科学系） ； Department of Aerospace Engineering and Engineering Mechanics, University of Texas at Austin, Austin, TX, USA（德克萨斯大学奥斯汀分校航空航天工程与工程力学系） 纠错

Comments Accepted to ICML 2025

Network Sparsity Unlocks the Scaling Potential of Deep Reinforcement Learning

Guozheng Ma, Lu Li, Zilin Wang, Li Shen, Pierre-Luc Bacon, Dacheng Tao

机构 * Nanyang Technical University（南洋理工大学） ； Mila - Quebec AI Institute（魁北克AI研究院） ； University of Oxford（牛津大学） 纠错

Comments Accepted to ICML 2025

LSCD: Lomb-Scargle Conditioned Diffusion for Time series Imputation

Elizabeth Fons, Alejandro Sztrajman, Yousef El-Laham, Luciana Ferrer, Svitlana Vyetrenko, Manuela Veloso

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

Comments In ICML 2025

On Training-Test (Mis)alignment in Unsupervised Combinatorial Optimization: Observation, Empirical Exploration, and Analysis

Fanchen Bu, Kijung Shin

机构 * School of Electrical Engineering, KAIST（韩国庆熙大学电气工程学院） ； AI4CO Open-Source Community（AI4CO开源社区） ； Kim Jaechul Graduate School of Artificial Intelligence, KAIST（韩国庆熙大学人工智能研究生院） 纠错

Comments 2nd Workshop on Test-Time Adaptation: Putting Updates to the Test @ ICML 2025

SparseLoRA: Accelerating LLM Fine-Tuning with Contextual Sparsity

Samir Khaki, Xiuyu Li, Junxian Guo, Ligeng Zhu, Chenfeng Xu, Konstantinos N. Plataniotis, Amir Yazdanbakhsh, Kurt Keutzer, Song Han, Zhijian Liu

机构 * University of Toronto（多伦多大学） ； MIT（麻省理工学院） ； Google DeepMind（谷歌DeepMind） 纠错

Comments ICML 2025. The first three authors contributed equally to this work. Project page: https://z-lab.ai/projects/sparselora

Signatures to help interpretability of anomalies

Emmanuel Gangler, Emille E. O. Ishida, Matwey V. Kornilov, Vladimir Korolev, Anastasia Lavrukhina, Konstantin Malanchev, Maria V. Pruzhinskaya, Etienne Russeil, Timofey Semenikhin, Sreevarsha Sreejith, Alina A. Volnova

机构 * Université Clermont Auvergne, CNRS, LPCA（克莱蒙特奥弗涅大学、法国国家科学研究中心、LPCA研究所） ； National Research University Higher School of Economics（俄罗斯国民研究大学高等经济学院） ； The Oskar Klein Centre, Department of Astronomy, Stockholm University（斯德哥尔摩大学奥斯特基尔克中心、天文学系） ； Faculty of Physics, Lomonosov Moscow State University（莫斯科罗蒙诺索夫莫斯科国立大学物理系） ； McWilliams Center for Cosmology and Astrophysics, Department of Physics, Carnegie Mellon University（卡内基梅隆大学天体物理学与宇宙学中心、物理系） ； Sternberg Astronomical Institute, Lomonosov Moscow State University（莫斯科罗蒙诺索夫莫斯科国立大学斯特尔贝格天文研究所） ； Physics department, University of Surrey（萨里大学物理系） ； Space Research Institute of the Russian Academy of Sciences（俄罗斯科学院空间研究 institutes） 纠错

Comments 7 pages, 3 figure, proceedings of the International Conference on Machine Learning for Astrophysics (ML4ASTRO2)

Perceptual-GS: Scene-adaptive Perceptual Densification for Gaussian Splatting

Hongbi Zhou, Zhangkai Ni

机构 * School of Computer Science and Technology, Tongji University, Shanghai, China（计算机科学与技术学院，同济大学，上海，中国） 纠错

Comments Accepted to International Conference on Machine Learning (ICML) 2025

Info-Coevolution: An Efficient Framework for Data Model Coevolution

Ziheng Qin, Hailun Xu, Wei Chee Yew, Qi Jia, Yang Luo, Kanchan Sarkar, Danhui Guan, Kai Wang, Yang You

机构 * National University of Singapore, School of Computing, 11 Research Link, Singapore（新加坡国立大学计算机学院） ； ByteDance Ltd, Singapore（字节跳动有限公司） 纠错

Comments V1

Journal ref ICML 2025

Build Agent Advocates, Not Platform Agents

Sayash Kapoor, Noam Kolt, Seth Lazar

Comments Accepted to ICML 2025 position paper track

From RAG to Memory: Non-Parametric Continual Learning for Large Language Models

Bernal Jiménez Gutiérrez, Yiheng Shu, Weijian Qi, Sizhe Zhou, Yu Su

机构 * The Ohio State University, Columbus, OH, USA（俄亥俄州立大学） ； University of Illinois Urbana-Champaign, IL, USA（伊利诺伊大学厄巴纳-香槟分校） 纠错

Comments ICML 2025. Code and data are available at: https://github.com/OSU-NLP-Group/HippoRAG

Layer-wise Alignment: Examining Safety Alignment Across Image Encoder Layers in Vision Language Models

Saketh Bachu, Erfan Shayegani, Rohit Lal, Trishna Chakraborty, Arindam Dutta, Chengyu Song, Yue Dong, Nael Abu-Ghazaleh, Amit K. Roy-Chowdhury

Comments Accepted by ICML 2025 as a spotlight poster

Adapting While Learning: Grounding LLMs for Scientific Problems with Intelligent Tool Usage Adaptation

Bohan Lyu, Yadi Cao, Duncan Watson-Parris, Leon Bergen, Taylor Berg-Kirkpatrick, Rose Yu

机构 * Tsinghua University（清华大学） ； University of California, San Diego（加州大学圣地亚哥分校） 纠错

Comments 37 pages, 16 figures

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

Disentangling and Integrating Relational and Sensory Information in Transformer Architectures

Awni Altabaa, John Lafferty

机构 * Yale University（耶鲁大学） ； Wu Tsai Institute, Yale University（耶鲁大学吴莎研究所） 纠错

Comments ICML 2025

Radio Galaxy Zoo: EMU -- paving the way for EMU cataloging using AI and citizen science

Hongming Tang, Eleni Vardoulaki, RGZ EMU collaboration

Comments 6 pages, 2 figures, The 2nd edition of the International Conference on Machine Learning for Astrophysics (ML4ASTRO2), conference paper accepted

Solving Zero-Sum Convex Markov Games

Fivos Kalogiannis, Emmanouil-Vasileios Vlatakis-Gkaragkounis, Ian Gemp, Georgios Piliouras

机构 * Department of Computer Science & Engineering, University of California San Diego（计算机科学与工程系，加州大学圣地亚哥分校） ； Department of Computer Sciences, University of Wisconsin-Madison（计算机科学系，威斯康星大学麦迪逊分校） ； Archimedes, Athena Research Center, Greece（阿基米德，亚提卡研究中心，希腊） ； Google DeepMind, London, UK（谷歌深Mind，伦敦，英国） 纠错

Comments To appear in the Proceedings of the 2025 International Conference on Machine Learning (ICML 2025)

Mitigating Over-Squashing in Graph Neural Networks by Spectrum-Preserving Sparsification

Langzhang Liang, Fanchen Bu, Zixing Song, Zenglin Xu, Shirui Pan, Kijung Shin

机构 * Kim Jaechul Graduate School of Artificial Intelligence, KAIST, Seoul, Republic of Korea（韩国釜山科学技术院人工智能研究生院） ； School of Electrical Engineering, KAIST, Daejeon, Republic of Korea（韩国釜山科学技术院电子工程学院） ； Department of Engineering, University of Cambridge, Cambridge, United Kingdom（剑桥大学工程系） ； Incubation Institute, Fudan University, Shanghai, China（复旦大学孵化院） ； Shanghai Academy of Artificial Intelligence for Science, Shanghai, China（上海人工智能科学研究院） ； School of Information and Communication Technology, Griffith University, Gold Coast, Australia（格里菲斯大学信息与通信技术学院） 纠错

Comments Published as a conference paper at ICML 2025

Contour Integration Underlies Human-Like Vision

Ben Lonnqvist, Elsa Scialom, Abdulkadir Gokce, Zehra Merchant, Michael H. Herzog, Martin Schrimpf

机构 * École Polytechnique Fédérale de Lausanne (EPFL), Switzerland（瑞士联邦理工学院（EPFL）） 纠错

Comments ICML Camera Ready

Rethinking External Slow-Thinking: From Snowball Errors to Probability of Correct Reasoning

Zeyu Gan, Yun Liao, Yong Liu

机构 * Gaoling School of Artificial Intelligence, Renmin University of China, Beijing, China（中国人民大学人工智能学院） ； Beijing Key Laboratory of Research on Large Models（北京大型模型研究关键实验室） ； Engineering Research Center of Next-Generation Intelligent Search（下一代智能搜索工程研究中心） ； College of Artificial Intelligence, Tianjin University of Science（天津科技大学人工智能学院） 纠错

Comments Published as a conference paper in ICML 2025

Ladder-residual: parallelism-aware architecture for accelerating large model inference with communication overlapping

Muru Zhang, Mayank Mishra, Zhongzhu Zhou, William Brandon, Jue Wang, Yoon Kim, Jonathan Ragan-Kelley, Shuaiwen Leon Song, Ben Athiwaratkun, Tri Dao

机构 * Together AI ； University of Southern California ； MIT-IBM Watson Lab ； University of Sydney ； Massachusetts Institute of Technology ； Princeton University 纠错

Comments ICML 2025

Core Knowledge Deficits in Multi-Modal Language Models

Yijiang Li, Qingying Gao, Tianwei Zhao, Bingyang Wang, Haoran Sun, Haiyun Lyu, Robert D. Hawkins, Nuno Vasconcelos, Tal Golan, Dezhi Luo, Hokin Deng

Comments Accepted by ICML 2025. Project page at https://williamium3000.github.io/core-knowledge and code is available at https://github.com/williamium3000/core-knowledge

Towards Robust Causal Effect Identification Beyond Markov Equivalence

Kai Z. Teh, Kayvan Sadeghi, Terry Soo

Comments ICML 2025 workshop - Scaling Up Intervention Models

Pixel-level Certified Explanations via Randomized Smoothing

Alaa Anani, Tobias Lorenz, Mario Fritz, Bernt Schiele

机构 * Max Planck Institute for Informatics（马克斯·普朗克研究所信息学研究所） ； CISPA Helmhotz Center for Information Security（CISPA海德堡信息安全中心） 纠错

Journal ref International Conference on Machine Learning (ICML), 2025

RE-IMAGINE: Symbolic Benchmark Synthesis for Reasoning Evaluation

Xinnuo Xu, Rachel Lawrence, Kshitij Dubey, Atharva Pandey, Risa Ueno, Fabian Falck, Aditya V. Nori, Rahul Sharma, Amit Sharma, Javier Gonzalez

机构 * Microsoft Research Cambridge, UK（微软研究院剑桥分部） ； Microsoft Research India, India（微软研究院印度分部） 纠错

Comments ICML 2025

Provable Maximum Entropy Manifold Exploration via Diffusion Models

Riccardo De Santi, Marin Vlastelica, Ya-Ping Hsieh, Zebang Shen, Niao He, Andreas Krause

机构 * ETH Zurich, 8092 Zurich, Switzerland（苏黎世联邦理工学院） ； ETH AI Center, Zurich, Switzerland（苏黎世联邦理工学院人工智能中心） 纠错

Comments ICML 2025

When Model Knowledge meets Diffusion Model: Diffusion-assisted Data-free Image Synthesis with Alignment of Domain and Class

Yujin Kim, Hyunsoo Kim, Hyunwoo J. Kim, Suhyun Kim

机构 * Korea University（韩国大学） ； Korea Advanced Institute of Science and Technology（韩国科学技术院） ； Kyung Hee University（庆熙大学） ； Korea Institute of Science（韩国科学研究院） 纠错

Comments Published at ICML 2025

Unlocking Post-hoc Dataset Inference with Synthetic Data

Bihe Zhao, Pratyush Maini, Franziska Boenisch, Adam Dziedzic

机构 * cispa（CISPA 河岸信息安全中心） ； cmu（卡内基梅隆大学） 纠错

Comments Accepted at ICML 2025

Multi-Task Reward Learning from Human Ratings

Mingkang Wu, Devin White, Evelyn Rose, Vernon Lawhern, Nicholas R Waytowich, Yongcan Cao

机构 * Department of Electrical and Computer Engineering, University of Texas at San Antonio, Texas, USA（电子与计算机工程系，德克萨斯州立大学圣安东尼奥分校） ； DEVCOM Army Research Lab, USA（陆军研究实验室） 纠错

Comments Accepted to the workshop on Models of Human Feedback for AI Alignment at the 42nd International Conference on Machine Learning

Too Big to Think: Capacity, Memorization, and Generalization in Pre-Trained Transformers

Joshua Barron, Devin White

机构 * Amazon（亚马逊） 纠错

Comments Accepted for oral presentation to Tiny Titans: The next wave of On-Device Learning for Foundational Models Workshop at the 42nd International Conference on Machine Learning

MAD-MAX: Modular And Diverse Malicious Attack MiXtures for Automated LLM Red Teaming

Stefan Schoepf, Muhammad Zaid Hameed, Ambrish Rawat, Kieran Fraser, Giulio Zizzo, Giandomenico Cornacchia, Mark Purcell

机构 * Univ. of Cambridge, Cambridge, UK.（剑桥大学） ； IBM Research Europe, Dublin, Ireland（IBM欧洲研究院） 纠错

Comments Data in Generative Models Workshop: The Bad, the Ugly, and the Greats (DIG-BUGS) at ICML 2025

Learning Strategic Language Agents in the Werewolf Game with Iterative Latent Space Policy Optimization

Zelai Xu, Wanjun Gu, Chao Yu, Yi Wu, Yu Wang

机构 * Tsinghua University, Beijing, China（清华大学） ； Beijing Zhongguancun Academy, Beijing, China（北京中关村学院） ； Shanghai Qi Zhi Institute, Shanghai, China（上海启智研究所） 纠错

Comments Published in ICML 2025