ICML

Reproducibility: The New Frontier in AI Governance

Israel Mason-Williams, Gabryel Mason-Williams

机构 * Imperial College London（帝国理工学院） ； King's College London（国王学院） ； Independent Researcher（独立研究者） 纠错

Comments 12 pages,6 figures,Workshop on Technical AI Governance at ICML

Talk Isn't Always Cheap: Understanding Failure Modes in Multi-Agent Debate

Andrea Wynn, Harsh Satija, Gillian Hadfield

机构 * Johns Hopkins University（约翰霍普金斯大学） ； Vector Institute（向量研究所） 纠错

Comments ICML MAS Workshop 2025

Why Is Spatial Reasoning Hard for VLMs? An Attention Mechanism Perspective on Focus Areas

Shiqi Chen, Tongyao Zhu, Ruochen Zhou, Jinghan Zhang, Siyang Gao, Juan Carlos Niebles, Mor Geva, Junxian He, Jiajun Wu, Manling Li

机构 * City University of Hong Kong（香港城市大学） ； National University of Singapore（新加坡国立大学） ； Hong Kong University of Science（香港科学大学） ； Stanford University（斯坦福大学） ； Salesforce Research（Salesforce研究） ； Tel Aviv University（特拉维夫大学） ； Northwestern University（西北大学） 纠错

Comments ICML 2025

ELMO: Efficiency via Low-precision and Peak Memory Optimization in Large Output Spaces

Jinbin Zhang, Nasib Ullah, Erik Schultheis, Rohit Babbar

机构 * Department of Computer Science, Aalto University, Espoo, Finland（艾尔沃斯大学计算机科学系） ； Department of Computer Science, University of Bath, Bath, UK（巴斯大学计算机科学系） 纠错

Comments Accepted to ICML 2025

Isotropy and Geometry of Pretrained Protein LMs

Sheikh Azizul Hakim, Kowshic Roy, M Saifur Rahman

Comments Published in the Proceedings of the ICML 2025 Workshop on Multi-modal Foun- dation Models and Large Language Models for Life Sciences, Vancouver, Canada. 2025

Learning Diffusion Models with Flexible Representation Guidance

Chenyu Wang, Cai Zhou, Sharut Gupta, Zongyu Lin, Stefanie Jegelka, Stephen Bates, Tommi Jaakkola

Comments NeurIPS 2025; Also Oral at ICML 2025 FM4LS workshop

Learning Mixtures of Experts with EM: A Mirror Descent Perspective

Quentin Fruytier, Aryan Mokhtari, Sujay Sanghavi

Comments Accepted to ICML 2025

Journal ref Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:17690-17726, 2025

OmniBal: Towards Fast Instruction-Tuning for Vision-Language Models via Omniverse Computation Balance

Yongqiang Yao, Jingru Tan, Feizhao Zhang, Jiahao Hu, Yazhe Niu, Xin Jin, Bo Li, Pengfei Liu, Ruihao Gong, Dahua Lin, Ningyi Xu

机构 * Shanghai Jiao Tong University（上海交通大学） ； SenseTime Research（商汤科技研究院） ； Central South University（中南大学） ； Tongji University（同济大学） ； Beihang University（北航） ； The Chinese University of Hong Kong（香港中文大学） 纠错

Comments Accepted in ICML 2025

SAM2LoRA: Composite Loss-Guided, Parameter-Efficient Finetuning of SAM2 for Retinal Fundus Segmentation

Sayan Mandal, Divyadarshini Karthikeyan, Manas Paldhe

机构 * Amazon, Toronto ON M5J0A8, Canada（亚马逊公司，多伦多） 纠错

Comments Accepted for publication at the 2025 International Conference on Machine Learning and Applications (ICMLA)

Journal ref 2025 ICMLA, Florida, USA

Learning Representations of Event Time Series with Sparse Autoencoders for Anomaly Detection, Similarity Search, and Unsupervised Classification

Steven Dillmann, Juan Rafael Martínez-Galarza

机构 * Stanford University ； Center for Astrophysics | Harvard \& Smithsonian 纠错

Comments Accepted at the 2025 ICML Workshop on Machine Learning for Astrophysics, Code available at: https://github.com/StevenDillmann/ml-xraytransients-mnras

Invariance Makes LLM Unlearning Resilient Even to Unanticipated Downstream Fine-Tuning

Changsheng Wang, Yihua Zhang, Jinghan Jia, Parikshit Ram, Dennis Wei, Yuguang Yao, Soumyadeep Pal, Nathalie Baracaldo, Sijia Liu

机构 * Michigan State University（密歇根州立大学） ； IBM Research（IBM研究院） 纠错

Comments Accepted by ICML 2025