ICML

Align-then-Unlearn: Embedding Alignment for LLM Unlearning

Philipp Spohn, Leander Girrbach, Jessica Bader, Zeynep Akata

机构 * Technical University of Munich（慕尼黑技术大学） ； Munich Center for Machine Learning, MDSI, Helmholtz Munich（慕尼黑机器学习中心，MDSI，海德堡慕尼黑） 纠错

Comments Accepted at ICML 2025 Workshop on Machine Unlearning for Generative AI

SAGDA: Open-Source Synthetic Agriculture Data for Africa

Abdelghani Belgaid, Oumnia Ennaji

机构 * College of Computing, Mohammed VI Polytechnic University（计算机学院，穆莱·易斯二世理工学院） 纠错

Journal ref Proceedings of the ICML 2025 Workshop on Championing Open-source Development in Machine Learning (CODEML '25)

Learning Event Completeness for Weakly Supervised Video Anomaly Detection

Yu Wang, Shiwei Chen

机构 * School of Computer Science ； Technology, Tongji University, Shanghai, China. ； Department of R\&D Data, Microsoft Asia-Pacific Technology CO Ltd, Shanghai, China. 纠错

Comments Accepted by ICML

Enhancing Rating-Based Reinforcement Learning to Effectively Leverage Feedback from Large Vision-Language Models

Tung Minh Luu, Younghwan Lee, Donghoon Lee, Sunho Kim, Min Jun Kim, Chang D. Yoo

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

Comments Accepted to ICML 2025

Lyapunov Learning at the Onset of Chaos

Matteo Benati, Alessandro Londei, Denise Lanzieri, Vittorio Loreto

机构 * Department of Computer, Automatic and Management Engineering. Sapienza University, Via Ariosto 25, Rome, Italy（计算机、自动与管理工程系。萨皮恩扎大学） ； Sony Computer Science Laboratories - Rome. Joint Initiative CREF-SONY, Centro Ricerche Enrico Fermi. Via Panisperna 89/A, 00184, Rome, Italy（索尼计算机科学实验室-罗马。联合倡议 CREF-SONY，恩里科·费米研究中心） ； Sapienza University of Rome, Physics Department.Piazzale A. Moro, 2, 00185, Rome, Italy（罗马萨皮恩扎大学物理系） 纠错

Comments Accepted at ICML 2025, HiLD: High-dimensional Learning Dynamics Workshop

Performance Plateaus in Inference-Time Scaling for Text-to-Image Diffusion Without External Models

Changhyun Choi, Sungha Kim, H. Jin Kim

机构 * Interdisciplinary Program in Artificial Intelligence, Seoul National University（人工智能交叉学科项目，首尔国立大学） ； Aerospace Engineering, Seoul National University（航空航天工程，首尔国立大学） ； ASRI, AIIS, Seoul National University（ASRI、AIIS，首尔国立大学） 纠错

Comments MOSS workshop at ICML 2025 accepted

Is your batch size the problem? Revisiting the Adam-SGD gap in language modeling

Teodora Srećković, Jonas Geiping, Antonio Orvieto

机构 * Max Planck Institute for Intelligent Systems（马克斯·普朗克智能系统研究所） ； ELLIS Institute（ELLIS研究所） ； Tübingen AI Center（图宾根人工智能中心） 纠错

Comments Short version accepted at the 2025 HiLD Workshop at ICML

BSA: Ball Sparse Attention for Large-scale Geometries

Catalin E. Brita, Hieu Nguyen, Lohithsai Yadala Chanchu, Domonkos Nagy, Maksim Zhdanov

机构 * Informatics Institute, University of Amsterdam（阿姆斯特丹大学信息学院） ； AMLab, University of Amsterdam（阿姆斯特丹大学AM实验室） 纠错

Comments Long Context Foundation Models Workshop @ ICML 2025

PROTOCOL: Partial Optimal Transport-enhanced Contrastive Learning for Imbalanced Multi-view Clustering

Xuqian Xue, Yiming Lei, Qi Cai, Hongming Shan, Junping Zhang

机构 * Shanghai Key Laboratory of Intelligent Information Processing, College of Computer Science and Artificial Intelligence, Fudan University, China（上海智能信息处理实验室，计算机科学与人工智能学院，复旦大学） ； College of Computer Science and Technology, Qingdao University, China（计算机科学与技术学院，青岛大学） ； Shanghai Key Laboratory of Navigation and Location-based Services, School of Electronic Information and Electrical Engineering, Shanghai Jiao Tong University, China（上海导航与位置服务实验室，电子信息与电气工程学院，上海交通大学） ； Institute of Science and Technology for Brain-Inspired Intelligence and Key Laboratory of Computational Neuroscience and Brain-Inspired Intelligence, Fudan University, China（脑启发智能科学技术研究院和计算神经科学与脑启发智能重点实验室，复旦大学） 纠错

Comments 15 pages, 7 figures, accepted by the Forty-Second International Conference on Machine Learning

SAFE: Finding Sparse and Flat Minima to Improve Pruning

Dongyeop Lee, Kwanhee Lee, Jinseok Chung, Namhoon Lee

Comments ICML 2025

COGNATE: Acceleration of Sparse Tensor Programs on Emerging Hardware using Transfer Learning

Chamika Sudusinghe, Gerasimos Gerogiannis, Damitha Lenadora, Charles Block, Josep Torrellas, Charith Mendis

机构 * University of Illinois Urbana-Champaign, USA（伊利诺伊大学厄巴纳-香槟分校） 纠错

Comments Accepted at the 42nd International Conference on Machine Learning

Bayesian Neural Scaling Law Extrapolation with Prior-Data Fitted Networks

Dongwoo Lee, Dong Bok Lee, Steven Adriaensen, Juho Lee, Sung Ju Hwang, Frank Hutter, Seon Joo Kim, Hae Beom Lee

机构 * Yonsei University（延世大学） ； University of Freiburg（弗莱堡大学） ； Korea University（韩国大学） 纠错

Comments Accepted to ICML 2025

Watch Out Your Album! On the Inadvertent Privacy Memorization in Multi-Modal Large Language Models

Tianjie Ju, Yi Hua, Hao Fei, Zhenyu Shao, Yubin Zheng, Haodong Zhao, Mong-Li Lee, Wynne Hsu, Zhuosheng Zhang, Gongshen Liu

机构 * Shanghai Jiao Tong University（上海交通大学） ； National University of Singapore（新加坡国立大学） 纠错

Comments Accepted at ICML 2025

Riemann Tensor Neural Networks: Learning Conservative Systems with Physics-Constrained Networks

Anas Jnini, Lorenzo Breschi, Flavio Vella

机构 * Department of Information Engineering and Computer Science, University of Trento, Trento, Italy（信息工程与计算机科学系，特伦托大学，意大利特伦托） 纠错

Comments To be published in the Proceedings of the Forty-Second International Conference on Machine Learning

Alpha-SQL: Zero-Shot Text-to-SQL using Monte Carlo Tree Search

Boyan Li, Jiayi Zhang, Ju Fan, Yanwei Xu, Chong Chen, Nan Tang, Yuyu Luo

Comments ICML 2025

Idiosyncrasies in Large Language Models

Mingjie Sun, Yida Yin, Zhiqiu Xu, J. Zico Kolter, Zhuang Liu

机构 * Carnegie Mellon University（卡内基梅隆大学） ； Princeton University（普林斯顿大学） ； University of Pennsylvania（宾夕法尼亚大学） 纠错

Comments Published in ICML 2025. Website at https://eric-mingjie.github.io/llm-idiosyncrasies/index.html

Improved Online Confidence Bounds for Multinomial Logistic Bandits

Joongkyu Lee, Min-hwan Oh

机构 * Seoul National University, Seoul, Korea（首尔国立大学） 纠错

Comments Accepted at ICML 2025

SelfCite: Self-Supervised Alignment for Context Attribution in Large Language Models

Yung-Sung Chuang, Benjamin Cohen-Wang, Shannon Zejiang Shen, Zhaofeng Wu, Hu Xu, Xi Victoria Lin, James Glass, Shang-Wen Li, Wen-tau Yih

机构 * Massachusetts Institute of Technology（麻省理工学院） ； Meta FAIR 纠错

Comments ICML 2025 main conference paper. The source code is available at https://github.com/facebookresearch/SelfCite

LOB-Bench: Benchmarking Generative AI for Finance -- an Application to Limit Order Book Data

Peer Nagy, Sascha Frey, Kang Li, Bidipta Sarkar, Svitlana Vyetrenko, Stefan Zohren, Ani Calinescu, Jakob Foerster

机构 * Oxford-Man Institute of Quantitative Finance, University of Oxford（牛津大学量化金融研究所） ； Department of Computer Science, University of Oxford（牛津大学计算机科学系） ； Department of Statistics, University of Oxford（牛津大学统计学系） ； Foerster Lab for AI Research, University of Oxford（福尔斯特人工智能研究实验室） ； J.P. Morgan AI Research（摩根大通人工智能研究） 纠错

Journal ref Proceedings of the 42nd International Conference on Machine Learning, Vancouver, Canada. PMLR 267, 2025

Scaling Laws for Upcycling Mixture-of-Experts Language Models

Seng Pei Liew, Takuya Kato, Sho Takase

Comments ICML 2025. 16 figures, 8 tables. Code available at https://github.com/sbintuitions/sparse-upcycling-scaling-laws

Deep Linear Network Training Dynamics from Random Initialization: Data, Width, Depth, and Hyperparameter Transfer

Blake Bordelon, Cengiz Pehlevan

机构 * John Paulson School of Engineering and Applied Sciences, Harvard University（哈佛大学约翰·保罗森工程与应用科学学校） ； Center for Brain Sciences（脑科学研究中心） ； Kempner Institute（凯普纳研究所） 纠错

Comments ICML Camera Ready

Unifying Specialized Visual Encoders for Video Language Models

Jihoon Chung, Tyler Zhu, Max Gonzalez Saez-Diez, Juan Carlos Niebles, Honglu Zhou, Olga Russakovsky

机构 * Department of Computer Science, Princeton University, Princeton, NJ, United States（普林斯顿大学计算机科学系） ； Salesforce Research, Palo Alto, CA, United States（Salesforce研究） 纠错

Comments Accepted to ICML 2025 as a Poster. Project page: https://tylerzhu.com/merv/

Fixing the Loose Brake: Exponential-Tailed Stopping Time in Best Arm Identification

Kapilan Balagopalan, Tuan Ngo Nguyen, Yao Zhao, Kwang-Sung Jun

机构 * University of Arizona（亚利桑那大学） 纠错

Comments Accepted by ICML 2025. This version has fixed a minor typo in Lemma C.3. upon the camera-ready version

Failure Modes of LLMs for Causal Reasoning on Narratives

Khurram Yamin, Shantanu Gupta, Gaurav R. Ghosal, Zachary C. Lipton, Bryan Wilder

机构 * Carnegie Mellon（卡内基梅隆大学） 纠错

Comments ICML 2025 Workshop on Scaling up Intervention Models

Convex Markov Games: A New Frontier for Multi-Agent Reinforcement Learning

Ian Gemp, Andreas Haupt, Luke Marris, Siqi Liu, Georgios Piliouras

机构 * College of Computing, MIT, Cambridge, MA, USA（麻省理工学院计算机学院） ； Google DeepMind, London, UK（谷歌深Mind） 纠错

Comments Published at ICML 2025

EffiCoder: Enhancing Code Generation in Large Language Models through Efficiency-Aware Fine-tuning

Dong Huang, Guangtao Zeng, Jianbo Dai, Meng Luo, Han Weng, Yuhao Qing, Heming Cui, Zhijiang Guo, Jie M. Zhang

机构 * University of Hong Kong（香港大学） ； Singapore University of Technology（新加坡科技设计大学） ； University of Edinburgh（爱丁堡大学） ； National University of Singapore（新加坡国立大学） ； Beijing University of Posts（北京邮电大学） ； University of Cambridge（剑桥大学） ； King’s College London（伦敦国王学院） 纠错

Comments Accepted by ICML 2025

Mind the Gap: a Spectral Analysis of Rank Collapse and Signal Propagation in Attention Layers

Thiziri Nait Saada, Alireza Naderi, Jared Tanner

机构 * Mathematical Institute, University of Oxford（牛津大学数学研究所） 纠错

Comments International Conference on Machine Learning

DeFoG: Discrete Flow Matching for Graph Generation

Yiming Qin, Manuel Madeira, Dorina Thanou, Pascal Frossard

机构 * EPFL, Lausanne, Switzerland（苏黎世联邦理工学院） 纠错

Comments The first two authors contributed equally to this work. Accepted at International Conference on Machine Learning (ICML) 2025

Journal ref International Conference on Machine Learning (ICML) 2025

How Much Can We Forget about Data Contamination?

Sebastian Bordt, Suraj Srinivas, Valentyn Boreiko, Ulrike von Luxburg

机构 * University of Tübingen, Tübingen AI Center, Germany ； Bosch Research North America \& Bosch Center for Artificial Intelligence (BCAI), Sunnyvale, USA 纠错

Comments ICML 2025 camera ready

Large Language Model (LLM)-enabled In-context Learning for Wireless Network Optimization: A Case Study of Power Control

Hao Zhou, Chengming Hu, Dun Yuan, Ye Yuan, Di Wu, Xue Liu, Charlie Zhang

Comments The latest version of this work has been accepted by ICML 2025 Workshop on ML4Wireless, and the revised title is "Prompting Wireless Networks: Reinforced In-Context Learning for Power Control"