ICML

Bring Reason to Vision: Understanding Perception and Reasoning through Model Merging

Shiqi Chen, Jinghan Zhang, Tongyao Zhu, Wei Liu, Siyang Gao, Miao Xiong, Manling Li, Junxian He

机构 * City University of Hong Kong（香港城市大学） ； Hong Kong University of Science（香港科学大学） ； National University of Singapore（新加坡国立大学） ； Northwestern University（西北大学） 纠错

Comments ICML 2025. Camera-ready version updated. Our code is publicly available at https://github.com/shiqichen17/VLM_Merging

Prediction via Shapley Value Regression

Amr Alkhatib, Roman Bresson, Henrik Boström, Michalis Vazirgiannis

机构 * Örebro University, School of Science and Technology（奥雷布罗大学，科学与技术学院） ； KTH Royal Institute of Technology, School of Electrical Engineering and Computer Science（皇家理工学院，电气工程与计算机科学学院） 纠错

Comments Accepted at ICML 2025

Compositional Flows for 3D Molecule and Synthesis Pathway Co-design

Tony Shen, Seonghwan Seo, Ross Irwin, Kieran Didi, Simon Olsson, Woo Youn Kim, Martin Ester

机构 * School of Computing Science, Simon Fraser University, Canada（计算科学学院，西蒙弗雷泽大学，加拿大） ； Department of Chemistry, KAIST, Republic of Korea（化学系，韩国科学技术院） ； Department of Computer Science（计算机科学系） ； Engineering, Chalmers University of Technology, Sweden（工程学院，查尔姆斯理工大学，瑞典） ； Department of Computer Science, University of Oxford, UK（计算机科学系，牛津大学，英国） ； NVIDIA 纠错

Comments Accepted to ICML 2025, 29 pages, 7 figures, code: https://github.com/tsa87/cgflow

ReVISE: Learning to Refine at Test-Time via Intrinsic Self-Verification

Hyunseok Lee, Seunghyuk Oh, Jaehyung Kim, Jinwoo Shin, Jihoon Tack

机构 * Yonsei University（延世大学） 纠错

Comments Published as conference proceeding for ICML 2025. First two authors contributed equally

Constrained Online Convex Optimization with Polyak Feasibility Steps

Spencer Hutchinson, Mahnoosh Alizadeh

机构 * University of California-Santa Barbara（加州大学圣芭芭拉分校） 纠错

Comments accepted at ICML 2025

Score-of-Mixture Training: Training One-Step Generative Models Made Simple via Score Estimation of Mixture Distributions

Tejas Jayashankar, J. Jon Ryu, Gregory Wornell

机构 * Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology (MIT), Cambridge, Massachusetts, USA（电气工程与计算机科学系，麻省理工学院（MIT），剑桥，马萨诸塞州，美国） 纠错

Comments 29 pages, 12 figures. ICML 2025 (spotlight)

Imitation Learning from a Single Temporally Misaligned Video

William Huey, Huaxiaoyue Wang, Anne Wu, Yoav Artzi, Sanjiban Choudhury

机构 * Cornell University（康奈尔大学） 纠错

Comments ICML 2025

ZebraLogic: On the Scaling Limits of LLMs for Logical Reasoning

Bill Yuchen Lin, Ronan Le Bras, Kyle Richardson, Ashish Sabharwal, Radha Poovendran, Peter Clark, Yejin Choi

机构 * Allen Institute for AI（艾伦人工智能研究所） ； University of Washington（华盛顿大学） ； Stanford University（斯坦福大学） 纠错

Comments Accepted to ICML 2025

BlueGlass: A Framework for Composite AI Safety

Harshal Nandigramwar, Syed Qutub, Kay-Ulrich Scholl

机构 * Intel Labs, Munich, Germany（英特尔慕尼黑实验室） ； University of Stuttgart, Stuttgart, Germany（斯图加特大学） 纠错

Comments Accepted at ICML 2025 [Actionable Interpretability Workshop]

Tiny Reward Models

Sarah Pan

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

Comments 2025 ICML Efficient Systems for Foundation Models Workshop

Fourier Position Embedding: Enhancing Attention's Periodic Extension for Length Generalization

Ermo Hua, Che Jiang, Xingtai Lv, Kaiyan Zhang, Youbang Sun, Yuchen Fan, Xuekai Zhu, Biqing Qi, Ning Ding, Bowen Zhou

机构 * Tsinghua University（清华大学） ； Northeastern University（东北大学） ； Shanghai Artificial Intelligence Laboratory（上海人工智能实验室） ； Shanghai Jiaotong University（上海交通大学） 纠错

Comments Accepted to ICML 2025

EVOLvE: Evaluating and Optimizing LLMs For In-Context Exploration

Allen Nie, Yi Su, Bo Chang, Jonathan N. Lee, Ed H. Chi, Quoc V. Le, Minmin Chen

机构 * Stanford University（斯坦福大学） ； Google DeepMind（谷歌DeepMind） 纠错

Comments 28 pages. Published at ICML 2025

A Unified View on Learning Unnormalized Distributions via Noise-Contrastive Estimation

J. Jon Ryu, Abhin Shah, Gregory W. Wornell

机构 * Department of EECS, MIT, Cambridge, Massachusetts, USA（麻省理工学院电子工程与计算机科学系） 纠错

Comments 31 pages, 2 figures. ICML 2025

Generative Audio Language Modeling with Continuous-valued Tokens and Masked Next-Token Prediction

Shu-wen Yang, Byeonggeun Kim, Kuan-Po Huang, Qingming Tang, Huy Phan, Bo-Ru Lu, Harsha Sundar, Shalini Ghosh, Hung-yi Lee, Chieh-Chi Kao, Chao Wang

机构 * Graduate Institute of Communication Engineering, National Taiwan University, Taipei, Taiwan（国立台湾大学通信工程研究所） ； Amazon AGI, Bellevue, United States（亚马逊人工智能研究院） 纠错

Comments Accepted by ICML 2025. Project website: https://audiomntp.github.io/

Efficient Molecular Conformer Generation with SO(3)-Averaged Flow Matching and Reflow

Zhonglin Cao, Mario Geiger, Allan dos Santos Costa, Danny Reidenbach, Karsten Kreis, Tomas Geffner, Franco Pellegrini, Guoqing Zhou, Emine Kucukbenli

机构 * NVIDIA ； MIT Center for Bits and Atoms（MIT比特与原子中心） ； MIT Center for Bits（MIT比特与原子中心） 纠错

Comments ICML 2025 poster

QuarterMap: Efficient Post-Training Token Pruning for Visual State Space Models

Tien-Yu Chi, Hung-Yueh Chiang, Diana Marculescu, Kai-Chiang Wu

机构 * National Yang Ming Chiao Tung University（国立阳明交通大学） ； The University of Texas at Austin（德克萨斯大学奥斯汀分校） 纠错

Comments Accepted by Efficient Systems for Foundation Models Workshop at the International Conference on Machine Learning (ICML) 2025

Transformers Don't In-Context Learn Least Squares Regression

Joshua Hill, Benjamin Eyre, Elliot Creager

Comments 21 pages, 16 figures, ICML 2025 Workshop on Reliable and Responsible Foundation Models

A Random Matrix Theory Perspective on the Learning Dynamics of Multi-head Latent Attention

Nandan Kumar Jha, Brandon Reagen

机构 * New York University（纽约大学） 纠错

Comments ICML 2025 Workshop on High-dimensional Learning Dynamics (HiLD)

Geometric Generative Modeling with Noise-Conditioned Graph Networks

Peter Pao-Huang, Mitchell Black, Xiaojie Qiu

机构 * stanford_cs（斯坦福大学计算机科学系） ； ucsd（加州大学圣迭戈分校计算机科学系） ； stanford_genetics（斯坦福大学遗传学系） 纠错

Comments ICML 2025

Continual Reinforcement Learning by Planning with Online World Models

Zichen Liu, Guoji Fu, Chao Du, Wee Sun Lee, Min Lin

机构 * Sea AI Lab（海智人工智能实验室） ； National University of Singapore（新加坡国立大学） 纠错

Comments ICML 2025 Spotlight

ALIGN: Prompt-based Attribute Alignment for Reliable, Responsible, and Personalized LLM-based Decision-Making

Bharadwaj Ravichandran, David Joy, Paul Elliott, Brian Hu, Jadie Adams, Christopher Funk, Emily Veenhuis, Anthony Hoogs, Arslan Basharat

机构 * kitware（Kitware公司） 纠错

Comments 10 pages total (including appendix), ICML 2025 Workshop on Reliable and Responsible Foundation Models

Fixed-Confidence Multiple Change Point Identification under Bandit Feedback

Joseph Lazzaro, Ciara Pike-Burke

机构 * Department of Mathematics, Imperial College London, London, England（伦敦帝国学院数学系） 纠错

Comments ICML 2025

ToxBench: A Binding Affinity Prediction Benchmark with AB-FEP-Calculated Labels for Human Estrogen Receptor Alpha

Meng Liu, Karl Leswing, Simon K. S. Chu, Farhad Ramezanghorbani, Griffin Young, Gabriel Marques, Prerna Das, Anjali Panikar, Esther Jamir, Mohammed Sulaiman Shamsudeen, K. Shawn Watts, Ananya Sen, Hari Priya Devannagari, Edward B. Miller, Muyun Lihan, Howook Hwang, Janet Paulsen, Xin Yu, Kyle Gion, Timur Rvachov, Emine Kucukbenli, Saee Gopal Paliwal

机构 * NVIDIA 纠错

Comments Workshop on Generative AI for Biology at ICML 2025

Optimizing Sequential Multi-Step Tasks with Parallel LLM Agents

Enhao Zhang, Erkang Zhu, Gagan Bansal, Adam Fourney, Hussein Mozannar, Jack Gerrits

机构 * University of Washington, Seattle, USA（华盛顿大学） ； Microsoft Research, Redmond, USA（微软研究院） 纠错

Comments ICML 2025 Workshop on MAS

DeltaSHAP: Explaining Prediction Evolutions in Online Patient Monitoring with Shapley Values

Changhun Kim, Yechan Mun, Sangchul Hahn, Eunho Yang

机构 * KAIST（韩国科学技术院） 纠错

Comments Accepted to ICML 2025 Workshop on Actionable Interpretability. Code is available at https://github.com/AITRICS/DeltaSHAP

Teaching Models to Verbalize Reward Hacking in Chain-of-Thought Reasoning

Miles Turpin, Andy Arditi, Marvin Li, Joe Benton, Julian Michael

机构 * Scale AI ； Harvard College（哈佛学院） ； Independent（独立） 纠错

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

Consistency in Language Models: Current Landscape, Challenges, and Future Directions

Jekaterina Novikova, Carol Anderson, Borhane Blili-Hamelin, Domenic Rosati, Subhabrata Majumdar

机构 * Dalhousie University（达尔豪斯大学） ； Vanguard, Enteprise AI Research（先锋企业人工智能研究所） 纠错

Comments Accepted in ICML 2025 Workshop on Reliable and Responsible Foundation Models

DataDecide: How to Predict Best Pretraining Data with Small Experiments

Ian Magnusson, Nguyen Tai, Ben Bogin, David Heineman, Jena D. Hwang, Luca Soldaini, Akshita Bhagia, Jiacheng Liu, Dirk Groeneveld, Oyvind Tafjord, Noah A. Smith, Pang Wei Koh, Jesse Dodge

机构 * Allen Institute for AI ； Paul G. Allen School of Computer Science \& Engineering, University of Washington ； University of Pennsylvania 纠错

Comments ICML 2025

Auditing Prompt Caching in Language Model APIs

Chenchen Gu, Xiang Lisa Li, Rohith Kuditipudi, Percy Liang, Tatsunori Hashimoto

机构 * stanford（斯坦福大学） 纠错

Comments Accepted at ICML 2025

Lower Bounds for Chain-of-Thought Reasoning in Hard-Attention Transformers

Alireza Amiri, Xinting Huang, Mark Rofin, Michael Hahn

机构 * Sharif University of Technology（沙里夫技术大学） ； Saarland University（萨尔兰大学） 纠错

Comments ICML 2025