Teaching LLMs to Speak Spectroscopy
Comments 6 pages, 1 figure, Accepted to the Machine Learning for Astrophysics (ML4Astro) Colocated Workshop at ICML 2025
期刊&会议
ICML
International Conference on Machine Learning · 会议 · Machine Learning
Curse of High Dimensionality Issue in Transformer for Long-context Modeling
机构 * South China University of Technology（华南理工大学） ； Peng Cheng Laboratory（鹏城实验室） ； Pazhou Laboratory（Pazhou实验室） ； Key Laboratory of Big Data（大数据重点实验室）
Comments Accepted at ICML 2025
Identifying Causal Direction via Variational Bayesian Compression
机构 * Deakin Applied Artificial Intelligence Initiative, Deakin University, Geelong, Australia（德金应用人工智能倡议，德金大学，澳大利亚Geelong）
Comments Accepted at the 42nd International Conference on Machine Learning (ICML2025)
Measuring Diversity in Synthetic Datasets
机构 * School of Computer Science（计算机科学学院） ； Engineering, Sun Yat-sen University（工程，中山大学） ； School of Software Engineering（软件工程学院） ； Sun Yat-sen University（中山大学） ； Zhuhai Key Laboratory of Trusted Large Language Models（珠海可信大语言模型实验室） ； Tencent AI Lab（腾讯AI实验室） ； National University of Singapore（新加坡国立大学） ； Shenzhen University（深圳大学） ； Shanghai Jiao Tong University（上海交通大学）
Comments Accepted by ICML 2025
Efficient Distributed Optimization under Heavy-Tailed Noise
机构 * Department of Statistics, University of Chicago（芝加哥大学统计系） ； Meta ； Department of Computer Science, University of Chicago（芝加哥大学计算机科学系）
Comments Accepted to ICML 2025
Diversifying Policy Behaviors with Extrinsic Behavioral Curiosity
机构 * School of Data Science, The Chinese University of Hong Kong, Shenzhen, China（数据科学学院，香港中文大学（深圳）） ； CFAR, Agency for Science, Technology and Research, Singapore（CFAR，新加坡科技研究局） ； College of Computing and Data Science, Nanyang Technological University (NTU), Singapore（计算与数据科学学院，南洋理工大学（NTU））
Comments 20 pages, conference paper
Journal ref International Conference on Machine Learning (ICML 2025)
Collaborative Mean Estimation Among Heterogeneous Strategic Agents: Individual Rationality, Fairness, and Truthful Contribution
机构 * University of Wisconsin-Madison（威斯康星大学麦迪逊分校） ； Cornell University（康奈尔大学）
Comments ICML 2025
Clipping Improves Adam-Norm and AdaGrad-Norm when the Noise Is Heavy-Tailed
机构 * Moscow Institute of Physics and Technology, Russia（莫斯科物理技术学院） ； Ivannikov Institute for System Programming RAS, Russia（伊万诺夫系统编程研究所） ； Sber AI Lab, Russia（俄罗斯Sber人工智能实验室） ； Innopolis University, Russia（因诺波利斯大学） ； The Russian Presidential Academy of National Economy and Public Administration, Russia（俄罗斯总统经济与公共行政学院） ； Skolkovo Institute of Science and Technology, Russia（斯克沃尔佐夫科学与技术研究所） ； Mohamed bin Zayed University of Artificial Intelligence, UAE（穆罕默德·本·扎耶德人工智能大学） ； Machine Learning and Optimization Laboratory (MLO), EPFL, Lausanne, Switzerland（机器学习与优化实验室（MLO），EPFL，拉瓦尔普勒斯，瑞士）
Comments ICML 2025. 65 pages, 12 figures. Changes in V3: extended results for the methods with coordinate-wise stepsizes and new experiments
Provable In-Context Vector Arithmetic via Retrieving Task Concepts
机构 * Department of Computer Science, City University of Hong Kong, Hong Kong SAR ； Center for Advanced Intelligence Project, RIKEN, Japan ； Institute of High Performance Computing (IHPC), Agency for Science, Technology ； Research (A STAR), Singapore ； Centre for Frontier AI Research (CFAR), Agency for Science, Technology ； College of Computing ； Data Science, Nanyang Technological University, Singapore ； Department of Mathematical Informatics, The University of Tokyo, Japan ； School of Mathematics ； Statistics, The University of Sydney, Australia
Comments Accepted by the 42nd International Conference on Machine Learning (ICML 2025)
Improving Diversity in Language Models: When Temperature Fails, Change the Loss
Comments Forty-Second International Conference on Machine Learning, ICML2025
The Importance of Being Lazy: Scaling Limits of Continual Learning
机构 * Electrical Engineering, ETH Zurich, Switzerland（电子工程系，苏黎世联邦理工学院） ； Dept. of Computer Science, ETH Zurich（计算机科学系，苏黎世联邦理工学院） ； ETH AI Center（苏黎世联邦理工学院人工智能中心） ； Astronomy, University of Padua, Italy（天文学系，帕多瓦大学）
Comments Proceedings of the 42nd International Conference on Machine Learning (2025). JG and AB contributed equally to this work
Gradual Transition from Bellman Optimality Operator to Bellman Operator in Online Reinforcement Learning
机构 * The University of Tokyo（东京大学）
Comments Accepted at ICML 2025. Source code: https://github.com/motokiomura/annealed-q-learning
RocketKV: Accelerating Long-Context LLM Inference via Two-Stage KV Cache Compression
机构 * NVIDIA, Santa Clara, USA（NVIDIA公司） ； Georgia Institute of Technology, Atlanta, USA（佐治亚理工学院）
Comments ICML 2025
Accelerating Linear Recurrent Neural Networks for the Edge with Unstructured Sparsity
机构 * Neuromorphic Computing Lab, Intel Corporation, USA ； Institute of Informatics, LMU Munich, Germany ； Bernoulli Institute \& CogniGron, University of Groningen, Netherlands
Comments ICML 2025
Towards flexible perception with visual memory
机构 * Google DeepMind（谷歌DeepMind）
Comments ICML 2025 camera ready version
Importance Corrected Neural JKO Sampling
机构 * University College London（伦敦大学学院）
Comments Accepted at ICML 2025
Discrete Neural Algorithmic Reasoning
机构 * Yandex Research（Yandex研究院）
Comments Forty-Second International Conference on Machine Learning (ICML 2025)
Pivoting Factorization: A Compact Meta Low-Rank Representation of Sparsity for Efficient Inference in Large Language Models
机构 * Center for Complex Network Intelligence (CCNI), Tsinghua Laboratory of Brain and Intelligence (THBI), Department of Psychological and Cognitive Sciences（复杂网络智能中心（CCNI）、清华脑智能实验室（THBI）、心理与认知科学系） ； Department of Computer Science（计算机科学系） ； Department of Biomedical Engineering, Tsinghua University, China（生物医学工程系，清华大学，中国）
Comments ICML 2025
Sparse Spectral Training and Inference on Euclidean and Hyperbolic Neural Networks
机构 * Center for Complex Network Intelligence (CCNI), Tsinghua Laboratory of Brain and Intelligence (THBI), Department of Psychological and Cognitive Sciences（复杂网络智能中心（CCNI）、清华脑智能实验室（THBI）、心理与认知科学系） ； Department of Computer Science（计算机科学系） ； Department of Biomedical Engineering, Tsinghua University, China（生物医学工程系，清华大学，中国）
Comments ICML 2025
Position: Causal Machine Learning Requires Rigorous Synthetic Experiments for Broader Adoption
机构 * Department of Computer Science, University of Bath, Bath, UK（英国巴斯大学计算机科学系） ； TAU, LISN, INRIA Saclay, France（法国萨克利实验室） ； INSA Rouen Normandy, Normandy University, LITIS, Rouen, France（法国诺曼底大学）
Comments Accepted at ICML 2025
Vulnerability-Aware Alignment: Mitigating Uneven Forgetting in Harmful Fine-Tuning
机构 * The Chinese University of Hong Kong（香港中文大学） ； Microsoft Research Asia（微软亚洲研究院） ； Shenzhen Campus of Sun Yat-sen University（中山大学深圳校区）
Comments ICML 2025
Fast Tensor Completion via Approximate Richardson Iteration
机构 * MIT（麻省理工学院） ； Google Research（谷歌研究）
Comments 18 pages, 4 figures
Journal ref Proceedings of the 42nd International Conference on Machine Learning (ICML 2025)
Zero-Shot Generalization of Vision-Based RL Without Data Augmentation
机构 * Department of Computer Science, University of Southern California, Los Angeles, USA（计算机科学系，南加州大学，洛杉矶，美国）
Comments Published at ICML 2025
Multidimensional Adaptive Coefficient for Inference Trajectory Optimization in Flow and Diffusion
机构 * Music and Audio Research Group, Department of Intelligence and Information, Seoul National University, Seoul, Korea（智能信息系音频研究组，首尔国立大学，韩国首尔） ； Interdisciplinary Program in Artificial Intelligence, Seoul National University, Seoul, Korea（人工智能跨学科项目，首尔国立大学，韩国首尔） ； Artificial Intelligence Institute, Seoul National University, Seoul, Korea（人工智能研究所，首尔国立大学，韩国首尔） ； School of Computing, KAIST, Daejeon, Korea（计算学院，韩国科学技术院，韩国大田）
Comments ICML 2025 Paper
ReferSplat: Referring Segmentation in 3D Gaussian Splatting
机构 * MoE Key Laboratory of Interdisciplinary Research of Computation and Economics（跨学科计算与经济学研究实验室） ； Shanghai University of Finance（上海财经大学） ； Institute of Big Data, College of Computer Science and Artificial Intelligence（大数据研究所，计算机科学与人工智能学院） ； Fudan University, Shanghai, China（复旦大学，上海，中国） ； Nanyang Technological University, Singapore（南洋理工大学，新加坡） ； Sun Yat-sen University, Guangzhou, China（中山大学，广州，中国）
Comments ICML 2025 Oral, Code: https://github.com/heshuting555/ReferSplat
Parity Requires Unified Input Dependence and Negative Eigenvalues in SSMs
机构 * Chandar Research Lab（昌达尔研究实验室） ； Mila - Quebec AI Institute（蒙特利尔AI研究所） ； Queens University（皇后大学） ； CIFAR AI Chair（CIFAR人工智能主席） ； McGill University（麦吉尔大学） ； Royal Military College of Canada（加拿大皇家军事学院）
Comments 5 pages. Accepted at ICML 2025 Workshop on Methods and Opportunities at Small Scale
Optimal and Practical Batched Linear Bandit Algorithm
机构 * snu（首尔国立大学）
Comments Accepted at ICML 2025
Position: Certified Robustness Does Not (Yet) Imply Model Security
机构 * School of Computing and Information Systems, University of Melbourne, Australia（计算与信息系统学院，墨尔本大学，澳大利亚）
Comments 9 pages, ICML, 2025
Winner-takes-all for Multivariate Probabilistic Time Series Forecasting
Comments ICML 2025
Constrained Hamiltonian Systems on Observation-Induced Fiber Bundles: Theory of Symmetry and Integrability
Comments This paper establishes the complete mathematical theory underlying the practical framework presented in "Learning Dynamics under Environmental Constraints via Measurement-Induced Bundle Structures" (ICML 2025 (Forty-Second International Conference on Machine Learning) Spotlight, top 2.5% of ~12,000 submissions)
