ICML

Random Initialization of Gated Sparse Adapters

Vi Retault, Yohaï-Eliel Berreby

机构 * GIGL Department, Polytechnique Montréal, Montréal, QC, Canada（波士顿理工大学蒙特利尔分校GIGL部门） ； Department of Physiology, McGill University, Montréal, QC, Canada（麦吉尔大学生理学系） ； Mila - Quebec AI Institute（魁北克AI研究所） 纠错

Comments 13 pages (8 main), 6 figures (4 main). Accepted by NewInML workshop @ ICML 2025 on June 27, 2025

UI-Evol: Automatic Knowledge Evolving for Computer Use Agents

Ziyun Zhang, Xinyi Liu, Xiaoyi Zhang, Jun Wang, Gang Chen, Yan Lu

机构 * School of Software and Microelectronics, Peking University（北京大学软件与微电子学院） ； Microsoft Research Asia（微软亚洲研究院） 纠错

Comments Accepted to ICML 2025 Workshop on Computer Use Agents

Transfer Learning for Onboard Cloud Segmentation in Thermal Earth Observation: From Landsat to a CubeSat Constellation

Niklas Wölki, Lukas Kondmann, Christian Mollière, Martin Langer, Julia Gottfriedsen, Martin Werner

机构 * OroraTech GmbH（OroraTech公司） ； Technical University of Munich (TUM)（慕尼黑技术大学） 纠错

Comments This work was presented at the TerraBytes Workshop at the 42nd International Conference on Machine Learning. This version is not part of the official ICML proceedings

Recommendations and Reporting Checklist for Rigorous & Transparent Human Baselines in Model Evaluations

Kevin L. Wei, Patricia Paskov, Sunishchal Dev, Michael J. Byun, Anka Reuel, Xavier Roberts-Gaal, Rachel Calcott, Evie Coxon, Chinmay Deshpande

机构 * Harvard University, Cambridge, MA, USA ； Stanford University, Stanford, CA, USA ； Center for Democracy \& Technology, Washington, D.C., USA ； Max Planck School of Cognition, Leipzig, Germany ； not those of RAND or its research sponsors, clients, or grantors. 纠错

Comments A version of this paper has been accepted to ICML 2025 as a position paper (spotlight), with the title: "Position: Human Baselines in Model Evaluations Need Rigor and Transparency (With Recommendations & Reporting Checklist)."

Journal ref Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:82265-82325, 2025

Dequantified Diffusion-Schr{ö}dinger Bridge for Density Ratio Estimation

Wei Chen, Shigui Li, Jiacheng Li, Junmei Yang, John Paisley, Delu Zeng

机构 * The School of Mathematics, South China University of Technology, Guangzhou 510006, China（华南理工大学数学学院） ； The School of Electronic and Information Engineering, South China University of Technology, Guangzhou 510006, China（华南理工大学电子与信息工程学院） ； The Department of Electrical Engineering, Columbia University, New York, NY 10027, USA（哥伦比亚大学电气工程系） 纠错

Journal ref ICML 2025: Proceedings of the 42nd International Conference on Machine Learning, 2025

Hyper-Transforming Latent Diffusion Models

Ignacio Peis, Batuhan Koyuncu, Isabel Valera, Jes Frellsen

机构 * Department of Applied Mathematics and Computer Science, Technical University of Denmark, Copenhagen, Denmark（丹麦技术大学应用数学与计算机科学系） ； Pioneer Centre for Artificial Intelligence, Copenhagen, Denmark（哥本哈根人工智能先锋中心） ； Saarland University, Saarbrücken, Germany（萨尔兰大学） ； Zuse School ELIZA（祖斯学校ELIZA） 纠错

Journal ref Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:48714-48733, 2025

Differential privacy guarantees of Markov chain Monte Carlo algorithms

Andrea Bertazzi, Tim Johnston, Gareth O. Roberts, Alain Durmus

机构 * University of Warwick, UK（沃里克大学） 纠错

Journal ref ICML 2025

Can Classic GNNs Be Strong Baselines for Graph-level Tasks? Simple Architectures Meet Excellence

Yuankai Luo, Lei Shi, Xiao-Ming Wu

机构 * Beihang University（北洋大学） ； The Hong Kong Polytechnic University（香港理工大学） 纠错

Comments ICML 2025

Wait-Less Offline Tuning and Re-solving for Online Decision Making

Jingruo Sun, Wenzhi Gao, Ellen Vitercik, Yinyu Ye

机构 * Department of Management Science \& Engineering, Stanford University ； Institute for Computational ； Mathematical Engineering, Stanford University ； Department of Computer Science, Stanford University ； Chinese University of Hong Kong (Shen Zhen) ； Hong Kong University of Science 纠错

Comments In this version, we achieve a tighter regret bound with the warm start for the first batch. We also make the proof more elegant by manually accepting all subsequent orders once the constraint is violated. In this way, we do not need to introduce the concept of stopping time for the analysis of the LP-based method

Journal ref Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:57419-57449, 2025

Dataset Distillation for Offline Reinforcement Learning

Jonathan Light, Yuanzhe Liu, Ziniu Hu

机构 * Department of Computer Science（计算机科学系） ； Rensselaer Polytechnic Institute（伦塞拉尔理工学院） ； Computing + Mathematical Sciences Department（计算+数学科学系） 纠错

Comments ICML 2024 DMLR Workshop Our project site is available at https://datasetdistillation4rl.github.io We also provide our implementation at https://github.com/ggflow123/DDRL

Point-Level Topological Representation Learning on Point Clouds

Vincent P. Grande, Michael T. Schaub

机构 * rwth Aachen University（亚琛工业大学） 纠错

Comments 47 pages, 18 figures, comments welcome

Journal ref International Conference on Machine Learning (ICML 2025)

Outlier Gradient Analysis: Efficiently Identifying Detrimental Training Samples for Deep Learning Models

Anshuman Chhabra, Bo Li, Jian Chen, Prasant Mohapatra, Hongfu Liu

机构 * University of South Florida, Tampa, FL, USA（佛罗里达州立大学） ； Brandeis University, Waltham, MA, USA（布兰迪大学） ； Tsinghua University（清华大学） 纠错

Comments Accepted to ICML 2025 (Oral)