ICML

Understanding and Mitigating Memorization in Generative Models via Sharpness of Probability Landscapes

Dongjae Jeon, Dueun Kim, Albert No

机构 * Department of Computer Science, Yonsei University, Seoul, Korea（计算机科学系，延世大学，首尔，韩国） ； Department of Artificial Intelligence, Yonsei University, Seoul, Korea（人工智能系，延世大学，首尔，韩国） 纠错

Comments Accepted at ICML 2025 (Spotlight). Code: https://github.com/Dongjae0324/sharpness_memorization_diffusion

QuickMerge++: Fast Token Merging with Autoregressive Prior

Dong Liu, Yanxuan Yu

机构 * Yale University（耶鲁大学） ； Columbia University（哥伦比亚大学） 纠错

Comments The paper has been accepted to ICML Tokshop at https://openreview.net/forum?id=dMdxHd0tRf

Widening the Network Mitigates the Impact of Data Heterogeneity on FedAvg

Like Jian, Dong Liu

机构 * School of Cyber Science and Technology, Beihang University, Beijing, China（网络安全科学与技术学院，北京航空航天大学，北京，中国） 纠错

Comments Accepted by ICML 2025

Event-Aware Sentiment Factors from LLM-Augmented Financial Tweets: A Transparent Framework for Interpretable Quant Trading

Yueyi Wang, Qiyao Wei

机构 * Department of Physics, University of Cambridge, Cambridge, UK（剑桥大学物理系） ； Department of Applied Mathematics and Theoretical Physics, University of Cambridge, Cambridge, UK（剑桥大学应用数学与理论物理系） 纠错

Comments 16 pages, 12 figures, accepted at ICML 2025 New in ML Workshop

Scalable Gaussian Processes with Latent Kronecker Structure

Jihao Andreas Lin, Sebastian Ament, Maximilian Balandat, David Eriksson, José Miguel Hernández-Lobato, Eytan Bakshy

机构 * Meta ； University of Cambridge（剑桥大学） ； Max Planck Institute for Intelligent Systems, Tübingen（图宾根Max Planck智能系统研究所） 纠错

Comments International Conference on Machine Learning 2025

Eigenspectrum Analysis of Neural Networks without Aspect Ratio Bias

Yuanzhe Hu, Kinshuk Goel, Vlad Killiakov, Yaoqing Yang

机构 * dartmouth（达特茅斯学院） ； ucsd（加州大学圣迭戈分校） ； srm（SRM科学与技术学院） ； ucb（加州大学伯克利分校） 纠错

Comments 29 pages, 14 figures, ICML 2025

Beyond Zero Initialization: Investigating the Impact of Non-Zero Initialization on LoRA Fine-Tuning Dynamics

Shiwei Li, Xiandi Luo, Xing Tang, Haozhao Wang, Hao Chen, Weihong Luo, Yuhua Li, Xiuqiang He, Ruixuan Li

机构 * Huazhong University of Science and Technology（华中科技大学） ； Shenzhen Technology University（深圳科技大学） ； FiT, Tencent（FiT，腾讯） 纠错

Comments Accepted by ICML 2025

The Panaceas for Improving Low-Rank Decomposition in Communication-Efficient Federated Learning

Shiwei Li, Xiandi Luo, Haozhao Wang, Xing Tang, Shijie Xu, Weihong Luo, Yuhua Li, Xiuqiang He, Ruixuan Li

机构 * Huazhong University of Science and Technology（华中科技大学） ； Shenzhen Technology University（深圳科技大学） ； FiT, Tencent（FiT，腾讯） 纠错

Comments Accepted by ICML 2025

Advancing AI-Scientist Understanding: Multi-Agent LLMs with Interpretable Physics Reasoning

Yinggan Xu, Hana Kimlee, Yijia Xiao, Di Luo

机构 * NSF Center for Quantum Network（NSF量子网络中心） ； University of California, Los Angeles（加州大学洛杉矶分校） 纠错

Comments ICML 2025 Workshop on MAS

Optimizing Language Models for Inference Time Objectives using Reinforcement Learning

Yunhao Tang, Kunhao Zheng, Gabriel Synnaeve, Rémi Munos

机构 * Meta GenAI（Meta 生成式人工智能） ； Meta FAIR 纠错

Comments Published as a conference paper at ICML 2025

Linear Bandits with Partially Observable Features

Wonyoung Kim, Sungwoo Park, Garud Iyengar, Assaf Zeevi, Min-hwan Oh

机构 * Seoul National University, Seoul, Korea（首尔国立大学） ； Chung-Ang University, Seoul, Korea（Chung-Ang 大学） ； Columbia University, New York, USA（哥伦比亚大学） 纠错

Comments Accepted in ICML 2025

Adaptive Exploration for Multi-Reward Multi-Policy Evaluation

Alessio Russo, Aldo Pacchiano

机构 * Faculty of Computing and Data Sciences（计算与数据科学系） ； Boston University（波士顿大学） ； Broad Institute of MIT and Harvard（哈佛大学与麻省理工学院Broad研究所） 纠错

Comments Accepted at the International Conference on Machine Learning, 2025

Rethinking Aleatoric and Epistemic Uncertainty

Freddie Bickford Smith, Jannik Kossen, Eleanor Trollope, Mark van der Wilk, Adam Foster, Tom Rainforth

机构 * University of Oxford（牛津大学） 纠错

Comments Published at ICML 2025

Regress, Don't Guess -- A Regression-like Loss on Number Tokens for Language Models

Jonas Zausinger, Lars Pennig, Anamarija Kozina, Sean Sdahl, Julian Sikora, Adrian Dendorfer, Timofey Kuznetsov, Mohamad Hagog, Nina Wiedemann, Kacper Chlodny, Vincent Limbach, Anna Ketteler, Thorben Prein, Vishwa Mohan Singh, Michael Morris Danziger, Jannis Born

机构 * Technical University of Munich（慕尼黑技术大学） ； Munich Center for Machine Learning (MCML)（慕尼黑机器学习中心（MCML）） ； IBM Research Europe（IBM欧洲研究院） ； ETH Zurich（苏黎世联邦理工学院） 纠错

Comments ICML 2025

New Interaction Paradigm for Complex EDA Software Leveraging GPT

Xinyu Wang, Boyu Han, Zhenghan Tai, Jingrui Tian, Yifan Wang, Junyu Yan, Yidong Tian

机构 * McGill University, Canada（麦吉尔大学） ； Stanford University, USA（斯坦福大学） ； University of Toronto, Canada（多伦多大学） ； Tsinghua University, China（清华大学） ； Beihang University, China（北航大学） ； The University of Manchester, UK（曼彻斯特大学） ； University of California, Los Angeles, USA（加州大学洛杉矶分校） ； Beijing Smarton Empower Co., Ltd., China（北京Smarton赋能科技有限公司） 纠错

Comments Accepted to ICML 2025 Workshop on New In Machine Learning (NewInML), 9 pages, 8 figures

Separating Knowledge and Perception with Procedural Data

Adrián Rodríguez-Muñoz, Manel Baradad, Phillip Isola, Antonio Torralba

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

Comments 17 pages, 18 figures, 3 tables, to be published in ICML 2025

OrthoRank: Token Selection via Sink Token Orthogonality for Efficient LLM inference

Seungjun Shin, Jaehoon Oh, Dokwan Oh

机构 * Samsung Advanced Institute of Technology, Korea（三星先进技术研究所，韩国） 纠错

Comments ICML 2025 (final version)

Hierarchical Refinement: Optimal Transport to Infinity and Beyond

Peter Halmos, Julian Gold, Xinhao Liu, Benjamin J. Raphael

机构 * Department of Computer Science, Princeton University（普林斯顿大学计算机科学系） ； Center for Statistics and Machine Learning, Princeton University（普林斯顿大学统计与机器学习中心） 纠错

Comments 33 pages, 9 figures. Accepted to ICML 2025 (Oral). Comments are welcome!

ActionPiece: Contextually Tokenizing Action Sequences for Generative Recommendation

Yupeng Hou, Jianmo Ni, Zhankui He, Noveen Sachdeva, Wang-Cheng Kang, Ed H. Chi, Julian McAuley, Derek Zhiyuan Cheng

机构 * University of California, San Diego（加州大学圣地亚哥分校） ； Google DeepMind（谷歌DeepMind） 纠错

Comments ICML 2025 (Spotlight)

Sub-Sequential Physics-Informed Learning with State Space Model

Chenhui Xu, Dancheng Liu, Yuting Hu, Jiajie Li, Ruiyang Qin, Qingxiao Zheng, Jinjun Xiong

机构 * University at Buffalo, SUNY（布法罗大学，纽约州立大学） ； University of Notre Dame（诺丁汉大学） 纠错

Comments ICML 2025

LieRE: Lie Rotational Positional Encodings

Sophie Ostmeier, Brian Axelrod, Maya Varma, Michael E. Moseley, Akshay Chaudhari, Curtis Langlotz

机构 * Computer Science Department, Stanford University, USA（斯坦福大学计算机科学系） ； Radiology Department, Stanford University, USA（斯坦福大学放射学系） 纠错

Comments Final proceedings version at ICML

Central Path Proximal Policy Optimization

Nikola Milosevic, Johannes Müller, Nico Scherf

机构 * Max Planck Institute for Human Cognitive and Brain Sciences（马克斯·普朗克人类认知与脑科学研究所） ； Center for Scalable Data Analytics and Artificial Intelligence (ScaDS.AI)（可扩展数据与人工智能中心） ； Institut für Mathematik, Technische Universität Berlin（柏林技术大学数学研究所） 纠错

Journal ref The Exploration in AI Today Workshop at ICML 2025

A Closer Look at Multimodal Representation Collapse

Abhra Chaudhuri, Anjan Dutta, Tu Bui, Serban Georgescu

机构 * Fujitsu Research of Europe（富士通欧洲研究） 纠错

Comments International Conference on Machine Learning (ICML) 2025 (Spotlight)

Neighbour-Driven Gaussian Process Variational Autoencoders for Scalable Structured Latent Modelling

Xinxing Shi, Xiaoyu Jiang, Mauricio A. Álvarez

机构 * Department of Computer Science, University of Manchester, Manchester, UK（计算机科学系，曼彻斯特大学，曼彻斯特，英国） 纠错

Comments ICML 2025

Sparse Training from Random Initialization: Aligning Lottery Ticket Masks using Weight Symmetry

Mohammed Adnan, Rohan Jain, Ekansh Sharma, Rahul G. Krishnan, Yani Ioannou

机构 * Schulich School of Engineering, University of Calgary ； Vector Institute for AI ； Dept.\ of Computer Science, University of Toronto 纠错

Comments Accepted at ICML 2025

SAND: One-Shot Feature Selection with Additive Noise Distortion

Pedram Pad, Hadi Hammoud, Mohamad Dia, Nadim Maamari, L. Andrea Dunbar

机构 * EPFL（瑞士洛桑联邦理工学院） 纠错

Comments Proceedings of the 42nd International Conference on Machine Learning (ICML), Vancouver, Canada. PMLR 267, 2025

Best Subset Selection: Optimal Pursuit for Feature Selection and Elimination

Zhihan Zhu, Yanhao Zhang, Yong Xia

Comments Accepted to ICML 2025

Embedding Safety into RL: A New Take on Trust Region Methods

Nikola Milosevic, Johannes Müller, Nico Scherf

机构 * Max Planck Institute for Human Cognitive and Brain Sciences（马克斯·普朗克人类认知与脑科学研究所） ； Center for Scalable Data Analytics and Artificial Intelligence（可扩展数据分析与人工智能中心） 纠错

Comments Accepted at ICML 2025

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

RULEBREAKERS: Challenging LLMs at the Crossroads between Formal Logic and Human-like Reasoning

Jason Chan, Robert Gaizauskas, Zhixue Zhao

机构 * University of Sheffield, Sheffield, UK（谢菲尔德大学） 纠错

Comments Accepted by ICML 2025

SLiM: One-shot Quantization and Sparsity with Low-rank Approximation for LLM Weight Compression

Mohammad Mozaffari, Amir Yazdanbakhsh, Maryam Mehri Dehnavi

机构 * University of Toronto（多伦多大学） ； Google DeepMind（谷歌DeepMind） ； NVIDIA Research（NVIDIA研究） 纠错

Comments Published at Proceedings of the 42 nd International Conference on Machine Learning (ICML 2025)