Efficient Diffusion Models for Symmetric Manifolds
机构 * Worcester Polytechnic Institute（沃斯特理工学院） ； Yale University（耶鲁大学）
Comments The conference version of this paper appears in ICML 2025
期刊&会议
ICML
International Conference on Machine Learning · 会议 · Machine Learning
Thickness-aware E(3)-Equivariant 3D Mesh Neural Networks
Comments ICML 2025
Self-Organizing Visual Prototypes for Non-Parametric Representation Learning
机构 * Institute of Computing, University of Campinas, Campinas-SP, Brazil（坎皮纳斯大学计算学院） ； Department of Informatics, University of Oslo, Oslo, Norway（奥斯陆大学信息学院）
Comments Accepted at ICML 2025, code at https://github.com/sthalles/sop
Latent Mamba Operator for Partial Differential Equations
机构 * Department of Electrical Communication Engineering, Indian Institute of Science, Bangalore, India（电子通信工程系，印度科学研究院，班加罗尔，印度） ； Yardi School of AI, Indian Institute of Technology, New Delhi, India（AI学院，印度理工学院，新德里，印度） ； Department of Civil Engineering, Indian Institute of Technology, New Delhi, India（土木工程系，印度理工学院，新德里，印度） ； Department of Computer Science and Automation, Indian Institute of Science, Bangalore, India（计算机科学与自动化系，印度科学研究院，班加罗尔，印度）
Comments Proceedings of the 42 nd International Conference on Machine Learning, Vancouver, Canada. PMLR 267, 2025. Copyright 2025 by the author(s)
Towards Practical Defect-Focused Automated Code Review
Comments Accepted as Spotlight at the 42nd International Conference on Machine Learning (ICML 2025)
Generalization Analysis for Supervised Contrastive Representation Learning under Non-IID Settings
机构 * Nong Minh Hieu（农明辉） ； Antoine Ledent（安托万·莱登）
Comments 40 pages, 3 figures. Published in the 42th Proceedings of International Conference in Machine Learning (ICML 2025)
In-Context Linear Regression Demystified: Training Dynamics and Mechanistic Interpretability of Multi-Head Softmax Attention
Comments Accepted by ICML 2025, related work added
Position: Don't Use the CLT in LLM Evals With Fewer Than a Few Hundred Datapoints
Comments 42 pages, 39 figures. ICML 2025 Spotlight Position Paper
Interpreting CLIP with Hierarchical Sparse Autoencoders
Journal ref Proceedings of the 42st International Conference on Machine Learning (ICML 2025)
MUDDFormer: Breaking Residual Bottlenecks in Transformers via Multiway Dynamic Dense Connections
Comments Accepted to the 42nd International Conference on Machine Learning (ICML'25)
A Physics-Informed Machine Learning Framework for Safe and Optimal Control of Autonomous Systems
Comments 22 Pages, 12 Figures. First two authors have contributed equally. Accepted at ICML 2025
DiffMS: Diffusion Generation of Molecules Conditioned on Mass Spectra
Comments ICML 2025
Outsourced diffusion sampling: Efficient posterior inference in latent spaces of generative models
机构 * Mila -- Qu\'ebec AI Institute ； Jagiellonian University ； CIFAR ； University of Edinburgh
Comments ICML 2025; code: https://github.com/HyperPotatoNeo/Outsourced_Diffusion_Sampling
Devil is in the Details: Density Guidance for Detail-Aware Generation with Flow Models
Comments ICML 2025
From Kernels to Features: A Multi-Scale Adaptive Theory of Feature Learning
Comments 33 pages, 12 figures, accepted at International Conference on Machine Learning 2025
Spurious Correlations in High Dimensional Regression: The Roles of Regularization, Simplicity Bias and Over-Parameterization
机构 * Institute of Science and Technology Austria (ISTA)（奥地利科学与技术研究所）
Comments Revision after ICML 2025 reviews
Distribution-aware Fairness Learning in Medical Image Segmentation From A Control-Theoretic Perspective
Comments ICML 2025 spotlight, see https://openreview.net/forum?id=BUONdewsBa
Mitigating Heterogeneous Token Overfitting in LLM Knowledge Editing
Comments ICML 2025
Complex Wavelet Mutual Information Loss: A Multi-Scale Loss Function for Semantic Segmentation
Comments Accepted at ICML 2025. This version corresponds to the official camera-ready submission
Training Dynamics of In-Context Learning in Linear Attention
Comments ICML 2025 Spotlight
Efficient Logit-based Knowledge Distillation of Deep Spiking Neural Networks for Full-Range Timestep Deployment
Comments Accepted by ICML 2025
How to Synthesize Text Data without Model Collapse?
Comments Accepted at ICML 2025
Preference Adaptive and Sequential Text-to-Image Generation
机构 * International Conference on Machine Learning (ICML 2025)（国际机器学习会议）
Comments Accepted to ICML 2025 Link to PASTA dataset: https://www.kaggle.com/datasets/googleai/pasta-data
Understanding Synthetic Context Extension via Retrieval Heads
机构 * Department of Computer Science, The University of Texas at Austin, Texas, USA（计算机科学系，德克萨斯大学奥斯汀分校）
Comments Published at ICML 2025
SafetyAnalyst: Interpretable, Transparent, and Steerable Safety Moderation for AI Behavior
Comments Accepted to ICML 2025
Graph-constrained Reasoning: Faithful Reasoning on Knowledge Graphs with Large Language Models
Comments Accepted by ICML 2025
Understanding Model Ensemble in Transferable Adversarial Attack
Comments Accepted by ICML 2025
Shielded Diffusion: Generating Novel and Diverse Images using Sparse Repellency
Comments Accepted at ICML 2025
Diss-l-ECT: Dissecting Graph Data with Local Euler Characteristic Transforms
Comments Accepted at ICML 2025
Watermarks in the Sand: Impossibility of Strong Watermarking for Generative Models
机构 * Harvard University（哈佛大学） ； George Mason University（乔治·马歇尔大学） ； Sapienza University of Rome（罗马大学）
Comments ICML 2024. Website: https://hanlin-zhang.com/impossibility-watermarks
