When and How Does CLIP Enable Domain and Compositional Generalization?
机构 * University of Freiburg（弗赖堡大学）
Comments ICML 2025 (Spotlight)
期刊&会议
ICML
International Conference on Machine Learning · 会议 · Machine Learning
Revisiting Non-Acyclic GFlowNets in Discrete Environments
机构 * HSE University, Moscow, Russia ； CMAP – CNRS – \'E cole polytechnique – Institut Polytechnique de Paris, 91128, Palaiseau, France ； Université Paris-Saclay, CNRS, LMO, 91405, Orsay, France
Comments ICML 2025; minor corrections in proofs of Proposition 3.6 and 3.8 in v3, all results remain unchanged
Temporal Query Network for Efficient Multivariate Time Series Forecasting
机构 * South China University of Technology（华南理工大学） ； Pengcheng Laboratory（鹏城实验室）
Comments ICML 2025
GAPrompt: Geometry-Aware Point Cloud Prompt for 3D Vision Model
机构 * Wangxuan Institute of Computer Technology, Peking University, Beijing, China（王轩计算机技术研究所，北京大学，北京，中国） ； State Key Laboratory of CAD\&CG, Zhejiang University, Hangzhou, China（CAD\&CG国家重点实验室，浙江大学，杭州，中国） ； School of Artificial Intelligence and Automation, Huazhong University of Science and Technology, Wuhan, China（人工智能与自动化学院，华中科技大学，武汉，中国）
Comments Accepted by ICML 2025
Scaling Video-Language Models to 10K Frames via Hierarchical Differential Distillation
机构 * Gaoling School of Artificial Intelligence, Renmin University of China（中国人民大学北京校区人工智能学院） ； School of Artificial Intelligence, Wuhan University（武汉大学人工智能学院）
Comments Accepted by ICML 2025
Automatic Reward Shaping from Confounded Offline Data
机构 * Causal AI Lab, Columbia University, New York, USA（因果AI实验室，哥伦比亚大学，纽约，美国） ； Department of Electrical Engineering and Computer Science, Syracuse University, New York, USA（电气工程与计算机科学系， Syracuse大学，纽约，美国）
Comments ICML 2025
When Do Neural Networks Learn World Models?
机构 * Department of Automation, Tsinghua University, Beijing, China（自动化系，清华大学，北京，中国）
Comments ICML 2025; ICLR 2025 World Models Workshop (oral, outstanding paper award)
Triplet Interaction Improves Graph Transformers: Accurate Molecular Graph Learning with Triplet Graph Transformers
机构 * Department of Computer Science, Rensselaer Polytechnic Institute, Troy, NY, USA（理海理工学院计算机科学系） ； IBM Thomas J. Watson Research Center, Yorktown Heights, NY, USA（IBM 玛丽·沃森研究中心）
Comments ICML'24 Accepted Version, 25 pages, 10 figures, 18 tables
Journal ref PMLR 235:20768- 20792, 2024
A Theoretical Justification for Asymmetric Actor-Critic Algorithms
机构 * Montefiore Institute, University of Liège（利根大学蒙特菲尔研究所） ； Department of Electrical and Computer Engineering, McGill University（麦吉尔大学电气与计算机工程系）
Comments 8 pages, 31 pages total
Journal ref International Conference on Machine Learning, 2025
Towards a General Time Series Forecasting Model with Unified Representation and Adaptive Transfer
机构 * East China Normal University, Shanghai, China（东华大学） ； Aalborg University, Aalborg, Denmark（奥尔堡大学） ； Huawei Noah's Ark Lab, Shenzhen, China（华为诺亚实验室）
Comments Accepted by the Forty-second International Conference on Machine Learning (ICML2025)
Toward Efficient and Scalable Design of In-Memory Graph-Based Vector Search
Comments Presented at ICML 2025 VecDB Workshop; an extended version appeared in ACM SIGMOD 2025 ('Graph-Based Vector Search: An Experimental Evaluation of the State-of-the-Art')
KABB: Knowledge-Aware Bayesian Bandits for Dynamic Expert Coordination in Multi-Agent Systems
机构 * Department of XXX, University of YYY, Location, Country（YYY大学XXX系） ； School of ZZZ, Institute of WWW, Location, Country（WWW研究所ZZZ学院）
Comments Accepted by the Main Conference of ICML 2025. Code: https://github.com/HCP-AI-Research-Lab/KABB
BRIDGE: Bootstrapping Text to Control Time-Series Generation via Multi-Agent Iterative Optimization and Diffusion Modeling
机构 * Microsoft Research（微软研究院） ； The University of Manchester, UK（英国曼彻斯特大学） ； Nanjing University, China（南京大学） ； Imperial College London, Imperial Global Singapore, Singapore（伦敦帝国学院，新加坡帝国全球分校，新加坡） ； The University of Tokyo, Japan（日本东京大学）
Comments ICML 2025 Main Conference
Do Sparse Autoencoders Generalize? A Case Study of Answerability
机构 * University of Oxford（牛津大学） ； University of Oxford, Oxford, United Kingdom（牛津大学） ； MIT-IBM Watson AI Lab（麻省理工-IBM Watson人工智能实验室）
Comments Accepted workshop paper at the ICML 2025 Workshop on Reliable and Responsible Foundation Models
CRANE: Reasoning with constrained LLM generation
机构 * Department of Computer Science, University of Illinois Urbana-Champaign, USA（计算机科学系，伊利诺伊大学厄巴纳-香槟分校）
Comments Accepted at ICML 2025, Code at: https://github.com/uiuc-focal-lab/CRANE
Towards Generative Ray Path Sampling for Faster Point-to-Point Ray Tracing
Comments 6 pages, 6 figures, accepted at IEEE ICMLCN 2025
Journal ref 2025 IEEE International Conference on Machine Learning for Communication and Networking (ICMLCN)
World Model Implanting for Test-time Adaptation of Embodied Agents
机构 * Department of Computer Science and Engineering, Sungkyunkwan University（计算机科学与工程系，顺天大学）
Journal ref Forty-second International Conference on Machine Learning, 2025
LotteryCodec: Searching the Implicit Representation in a Random Network for Low-Complexity Image Compression
Journal ref International Conference on Machine Learning (2025)
Integrating Intermediate Layer Optimization and Projected Gradient Descent for Solving Inverse Problems with Diffusion Models
机构 * uestc（电子科技大学）
Comments ICML 2025
Defending LVLMs Against Vision Attacks through Partial-Perception Supervision
机构 * College of Control Science and Engineering, Zhejiang University, China（控制科学与工程学院，浙江大学，中国） ； Huzhou Institute of Industrial Control Technology, China（湖州工业控制技术研究所，中国） ； School of Computer Science and Engineering, Nanyang Technological University, Singapore（计算机科学与工程学院，南洋理工大学，新加坡） ； School of Computer Science, Shanghai Jiao Tong University, China（计算机科学学院，上海交通大学，中国） ； School of Computing, National University of Singapore, Singapore（计算学院，新加坡国立大学，新加坡） ； Research, Singapore（研究，新加坡）
Comments Accepted to ICML 2025
PAK-UCB Contextual Bandit: An Online Learning Approach to Prompt-Aware Selection of Generative Models and LLMs
机构 * Department of Computer Science and Engineering, The Chinese University of Hong Kong（香港中文大学计算机科学与工程系） ； Independent Researcher（独立研究者）
Comments accepted to ICML 2025
LapSum -- One Method to Differentiate Them All: Ranking, Sorting and Top-k Selection
机构 * Faculty of Mathematics ； Computer Science, Jagiellonian University, Krak\'ow, Poland ； University of Illinois, Urbana-Champaign, USA
Comments Accepted at ICML 2025
Temporal Representation Learning for Real-Time Ultrasound Analysis
机构 * Department of Computer Science, ETH Zurich, Switzerland（苏黎世联邦理工学院计算机科学系） ； Department of Biomedical Engineering, University of Basel, Switzerland（巴塞尔大学生物医学工程系）
Comments ICMl 2025 Workshop
Localizing and Mitigating Memorization in Image Autoregressive Models
机构 * CISPA Helmholtz Center for Information Security（CISPA海德堡信息安全中心）
Comments Accepted at ICML 2025 Workshop on the Impact of Memorization on Trustworthy Foundation Models
Data Cartography for Detecting Memorization Hotspots and Guiding Data Interventions in Generative Models
机构 * Illinois Mathematics and Science Academy（伊利诺伊数学与科学学院）
Comments 6 pages, 2 figures, 1 table; Presented at the 42nd International Conference on Machine Learning (ICML), winning the "Best Poster" award at ICML's workshop for data in generative models (DIG-BUGS)
A Generalization Theory for Zero-Shot Prediction
机构 * University of Washington（华盛顿大学）
Comments Published at ICML '25 (Oral)
Gaussian Mixture Flow Matching Models
机构 * Stanford University（斯坦福大学） ； Adobe Research（Adobe研究）
Comments ICML 2025. Code: https://github.com/Lakonik/GMFlow
Roll the dice & look before you leap: Going beyond the creative limits of next-token prediction
机构 * Google Research, US（谷歌研究） ； Carnegie Mellon University, Pittsburgh, US（卡内基梅隆大学）
Comments ICML 2025 (oral)
Stochastic Control for Fine-tuning Diffusion Models: Optimality, Regularity, and Convergence
Comments 31 pages; first version: Dec 2024; this version: Aug 2025
Journal ref Forty-Second International Conference on Machine Learning 2025
Revisiting Instance-Optimal Cluster Recovery in the Labeled Stochastic Block Model
Comments Accepted for presentation at the Forty-Second International Conference on Machine Learning (ICML 2025)
