A Near-Optimal Single-Loop Stochastic Algorithm for Convex Finite-Sum Coupled Compositional Optimization
机构 * Department of Computer Science ； Engineering, Texas A\&M University, College Station, TX, USA
Comments In ICML 2025
期刊&会议
ICML
International Conference on Machine Learning · 会议 · Machine Learning
Insights into a radiology-specialised multimodal large language model with sparse autoencoders
机构 * Microsoft Research, Health Futures, Cambridge, United Kingdom（微软研究院，健康未来，剑桥，英国）
Comments Actionable Interpretability Workshop at ICML 2025. 24 pages, 7 figures, 5 tables
Statistical and Computational Guarantees of Kernel Max-Sliced Wasserstein Distances
机构 * School of Data Science, The Chinese University of Hong Kong, Shenzhen, Shenzhen, China（数据科学学院，香港中文大学（深圳）） ； School of Artificial Intelligence, The Chinese University of Hong Kong, Shenzhen, Shenzhen, China（人工智能学院，香港中文大学（深圳）） ； Department of Mathematics, Michigan State University, East Lansing, USA（数学系，密歇根州立大学） ； School of Industrial and Systems Engineering, Georgia Institute of Technology, Atlanta, USA（工业与系统工程学院，佐治亚理工学院）
Comments Accepted by ICML-2025
Using Multiple Input Modalities Can Improve Data-Efficiency and O.O.D. Generalization for ML with Satellite Imagery
机构 * Department of Computer Science, University of Colorado Boulder（计算机科学系，科罗拉多大学博尔德分校）
Comments 17 pages, 9 figures, 7 tables. Accepted to TerraBytes@ICML 2025
Unsupervised Concept Discovery Mitigates Spurious Correlations
机构 * Mila-Quebec AI Institute, Canada（魁北克AI研究所） ； Samsung - SAIT AI Lab, Montreal, Canada（三星SAIT人工智能实验室） ； University of Montreal, Canada（蒙特利尔大学） ； Institute of Science and Technology Austria（奥地利科学与技术研究所） ； National University of Singapore（新加坡国立大学）
Journal ref ICML 2024
Reasoning-Finetuning Repurposes Latent Representations in Base Models
机构 * Rudolf Peierls Centre for Theoretical Physics, Oxford（鲁道夫·佩尔尔斯理论物理中心，牛津） ； University of Oxford（牛津大学）
Comments 6 pages, 6 figures. ICML 2025 Workshop on Actionable Interpretability
FOUNDER: Grounding Foundation Models in World Models for Open-Ended Embodied Decision Making
机构 * School of Artificial Intelligence, Nanjing University, China（人工智能学院，南京大学） ； National Key Laboratory for Novel Software Technology, Nanjing University, China（新型软件技术国家重点实验室，南京大学）
Comments Accepted by Forty-Second International Conference on Machine Learning (ICML 2025)
DSSD: Efficient Edge-Device LLM Deployment and Collaborative Inference via Distributed Split Speculative Decoding
Comments ICML 2025
Improving Transformer World Models for Data-Efficient RL
机构 * Google DeepMind（谷歌DeepMind）
Comments ICML 2025
Formalising causal inference as prediction on a target population
机构 * University of Tübingen and Tübingen AI Center（图宾根大学和图宾根人工智能中心）
Comments Presented at the Humans, Algorithmic Decision-Making and Society Workshop at ICML 2024
QuRe: Query-Relevant Retrieval through Hard Negative Sampling in Composed Image Retrieval
Comments Accepted to ICML 2025
The Challenge of Teaching Reasoning to LLMs Without RL or Distillation
机构 * Nvidia Corporation（Nvidia公司） ； Institute for Artificial Intelligence Research（人工智能研究机构） ； Faculty of Technical Sciences, University of Novi Sad（诺维萨德大学技术科学学院） ； Department of Mathematics and Informatics, Faculty of Sciences, University of Novi Sad（诺维萨德大学科学学院数学与信息学系） ； University of Illinois Urbana-Champaign（伊利诺伊大学厄巴纳-香槟分校） ； University of Chicago（芝加哥大学） ； Georgia Institute of Technology（佐治亚理工学院） ； Department of Computer and Information Science, School of Engineering and Applied Science, University of Pennsylvania（宾夕法尼亚大学计算机与信息科学系） ； Massachusetts Institute of Technology（麻省理工学院）
Comments Accepted at the Second AI for Math Workshop at the 42nd International Conference on Machine Learning (ICML 2025)
Simple Mechanistic Explanations for Out-Of-Context Reasoning
机构 * MIT（麻省理工学院）
Comments ICML 2025 Workshop R2-FM
BuildEvo: Designing Building Energy Consumption Forecasting Heuristics via LLM-driven Evolution
机构 * Department of the Built Environment, National University of Singapore（新加坡国立大学建筑环境系） ； Department of Industrial and Systems Engineering, Korea Advanced Institute of Science and Technology（韩国科学技术院工业与系统工程系）
Comments ICML 2025 CO-Build Workshop Poster
Dual form Complementary Masking for Domain-Adaptive Image Segmentation
机构 * University of Science and Technology of China（中国科学技术大学） ； Institute of Artificial Intelligence, Hefei Comprehensive National Science Center（合肥综合性国家科学中心人工智能研究院） ； Data Science Institute, Imperial College London（伦敦帝国理工学院数据科学研究院） ； iFLYTEK CO., LTD（iFLYTEK公司）
Comments Accepted by ICML 2025
BlockBPE: Parallel BPE Tokenization
机构 * Amos You（独立研究者）
Comments ES-FoMo III: 3rd Workshop on Efficient Systems for Foundation Models (ICML 2025)
Image-Based Multi-Survey Classification of Light Curves with a Pre-Trained Vision Transformer
机构 * Department of Computer Science, Universidad de Concepción（康塞普西翁大学计算机科学系） ； Center for Data and Artificial Intelligence, Universidad de Concepción（康塞普西翁大学数据与人工智能中心） ； Millennium Institute of Astrophysics (MAS)（天文物理千年研究所） ； Heidelberg Institute for Theoretical Studies（海德堡理论研究 institute） ； Center for Mathematical Modeling (CMM), Universidad de Chile（智利大学数学建模中心） ； John A. Paulson School of Engineering and Applied Sciences, Harvard University（哈佛大学工程与应用科学学院） ； Data and Artificial Intelligence Initiative (IDIA), Faculty of Physical and Mathematical Sciences, Universidad de Chile（智利大学数据与人工智能倡议，物理与数学科学学院） ； Instituto de Astrofísica, Pontificia Universidad Católica de Chile（智利天主大学天文研究所） ； Centro de Astro-Ingeniería, Pontificia Universidad Católica de Chile（智利天主大学天文工程中心） ； European Southern Observatory (ESO)（欧洲南 Observatory） ； Dept. of Electrical Engineering, University of Chile（智利大学电气工程系） ； Instituto de Alta Investigación, Universidad de Tarapacá（塔拉帕卡大学高级研究 institute）
Comments Accepted at the 2025 Workshop on Machine Learning for Astrophysics at the International Conference on Machine Learning (ICML)
General Modular Harness for LLM Agents in Multi-Turn Gaming Environments
机构 * Halıcıoğlu Data Science Institute (HDSI), University of California San Diego, La Jolla, CA, USA（Halıcıoğlu数据科学研究所（HDSI），加州大学圣地亚哥分校，拉贾拉，加州，美国）
Comments 8 pages, ICML MAS workshop
Implicit Bias of Gradient Descent for Non-Homogeneous Deep Networks
机构 * University of California, Berkeley（加州大学伯克利分校） ； Columbia University（哥伦比亚大学） ； Lawrence Berkeley National Laboratory（伯克利国家实验室） ； Google DeepMind（谷歌DeepMind）
Comments 79 pages, appeared in Proceedings of the 42nd International Conference on Machine Learning, Vancouver, Canada
Journal ref ICML 2025
Organize the Web: Constructing Domains Enhances Pre-Training Data Curation
Comments Accepted at ICML 2025. Project page: https://weborganizer.allen.ai
ViTally Consistent: Scaling Biological Representation Learning for Cell Microscopy
机构 * Valence Labs（Valence实验室） ； University of Manchester（曼彻斯特大学）
Comments ICML 2025 main-track paper (42nd International Conference on Machine Learning). Formerly appeared as best paper runner-up at NeurIPS 2024 Foundation Models for Science Workshop (38th Conference on Neural Information Processing Systems). 18 pages, 7 figures
RAGGED: Towards Informed Design of Scalable and Stable RAG Systems
机构 * Machine Learning Department, Carnegie Mellon University, Pittsburgh, USA（卡内基梅隆大学机器学习系） ； The Language Technologies Institute, Carnegie Mellon University, Pittsburgh, USA（卡内基梅隆大学语言技术研究所）
Comments Project page: https://github.com/neulab/ragged
Journal ref ICML 2025
FMC: Formalization of Natural Language Mathematical Competition Problems
机构 * State Key Laboratory for Multimedia Information Processing, School of Computer Science, PKU-Anker LLM Lab, Peking University, Beijing, China（多媒体信息处理国家重点实验室，计算机科学学院，PKU-Anker LLM实验室，北京大学，北京，中国） ； Paul G. Allen School of Computer Science（保罗·G·阿伦计算机科学学校） ； Engineering, University of Washington, U.S.A.（工程，华盛顿大学，美国）
Comments Accepted in ICML 2025 AI4MATH Workshop
WhisperKit: On-device Real-time ASR with Billion-Scale Transformers
机构 * Argmax Inc.（Argmax公司） ； University of California, Los Angeles, USA（加州大学洛杉矶分校）
Comments ICML 2025 - On-Device Learning for Foundational Models Workshop
Hi Robot: Open-Ended Instruction Following with Hierarchical Vision-Language-Action Models
机构 * Stanford University（斯坦福大学） ； University of California, Berkeley（加州大学伯克利分校）
Comments ICML 2025
X Hacking: The Threat of Misguided AutoML
机构 * Deutsches Forschungszentrum für Künstliche Intelligenz GmbH.（德国人工智能研究中心有限公司） ； Institute of Informatics, Ludwig-Maximilians-Universität München（慕尼黑大学信息学院）
Comments Accepted to ICML 2025
Function-to-Style Guidance of LLMs for Code Translation
机构 * Harbin Institute of Technology, Shenzhen, China.（哈尔滨工业大学深圳研究院） ； Huawei Translation Services Center, Beijing, China.（华为翻译服务中心）
Comments This paper has been accepted by ICML 2025. Models and benchmarks can be found at https://www.modelscope.cn/collections/F2STrans-42526ff95dd843
Spatial Reasoners for Continuous Variables in Any Domain
机构 * Max Planck Institute for Informatics, Saarland Informatics Campus, Germany（马克斯·普朗克研究所信息学院，萨尔兰信息校区，德国）
Comments For the project documentation see https://spatialreasoners.github.io/ . The SRM project website is available at https://geometric-rl.mpi-inf.mpg.de/srm/ . The work was published on ICML 2025 CODEML workshop
The Price of Freedom: Exploring Expressivity and Runtime Tradeoffs in Equivariant Tensor Products
机构 * Department of EECS, Massachusetts Institute of Technology, Cambridge MA 02139, USA（麻省理工学院电子工程与计算机科学系） ； Department of CSE, Massachusetts Institute of Technology, Cambridge MA 02139, USA（麻省理工学院计算机科学与工程系）
Comments Published at ICML 2025. 27 pages, 10 figures
On the Similarities of Embeddings in Contrastive Learning
机构 * Department of Statistics and Data Science, Yonsei University, Seoul, Korea（统计与数据科学系，延世大学，首尔，韩国）
Comments contrastive learning, representation learning, embedding, similarity, negative pair, positive pair
Journal ref International Conference on Machine Learning (ICML) 2025
