To Distill or Decide? Understanding the Algorithmic Trade-off in Partially Observable Reinforcement Learning
Comments 45 pages, 9 figures, published at NeurIPS 2025
期刊&会议
NeurIPS
Conference on Neural Information Processing Systems · 会议 · Machine Learning
GeoComplete: Geometry-Aware Diffusion for Reference-Driven Image Completion
机构 * National University of Singapore（新加坡国立大学）
Comments Accepted by NeurIPS 2025. Project page: https://bb12346.github.io/GeoComplete/
RAxSS: Retrieval-Augmented Sparse Sampling for Explainable Variable-Length Medical Time Series Classification
机构 * ETH Zurich（苏黎世联邦理工学院） ； University of Freiburg（弗赖堡大学） ； Imperial College London（伦敦帝国学院） ； Technical University of Munich（慕尼黑技术大学）
Comments Accepted at the NeurIPS 2025 Workshop on Learning from Time Series for Health
RoiRL: Efficient, Self-Supervised Reasoning with Offline Iterative Reinforcement Learning
机构 * Criteo AI Lab（Criteo人工智能实验室） ； Ecole Polytechnique Paris（巴黎高等理工学院）
Comments Accepted to the Efficient Reasoning Workshop at NeuRIPS 2025
A Granular Study of Safety Pretraining under Model Abliteration
机构 * Data and Web Science Group, University of Mannheim（曼海姆大学数据与网络科学组） ； Vision and AI Lab, Indian Institute of Science（印度科学院视觉与人工智能实验室） ； Carnegie Mellon University（卡内基梅隆大学） ； Max-Planck-Institute for Informatics, Saarland Informatics Campus（马克斯·普朗克信息研究所，萨尔兰信息校园）
Comments Accepted at NeurIPS 2025 bWorkshop Lock-LLM. *Equal Contribution
Ultra-Efficient Decoding for End-to-End Neural Compression and Reconstruction
机构 * Department of Electrical and Computer Engineering（电气与计算机工程系）
Comments 5 pages, 4 figures, NeurIPS 2025 Workshop MLForSys
MTRec: Learning to Align with User Preferences via Mental Reward Models
机构 * School of Software Engineering, South China University of Technology（华南理工大学软件学院） ； School of Computer Science and Technology, Dalian University of Technology（大连理工大学计算机科学与技术学院） ； Huawei Noah’s Ark Lab（华为诺亚实验室） ； Nanyang Technological University（南洋理工大学）
Journal ref Proceedings of the 39th Conference on Neural Information Processing Systems (NeurIPS 2025)
Towards Provable Emergence of In-Context Reinforcement Learning
Comments NeurIPS 2025, 29 pages
HopaDIFF: Holistic-Partial Aware Fourier Conditioned Diffusion for Referring Human Action Segmentation in Multi-Person Scenarios
机构 * Karlsruhe Institute of Technology（卡尔斯鲁厄理工学院） ； Beijing Institute of Technology（北京理工大学） ； Institute of Automation, Chinese Academy of Sciences（中国科学院自动化研究所） ； Hunan University（湖南大学） ； Shanghai AI Lab（上海人工智能实验室） ； HEBUST
Comments Accepted to NeurIPS 2025. The dataset and code are available at https://github.com/KPeng9510/HopaDIFF
Manipulating 3D Molecules in a Fixed-Dimensional E(3)-Equivariant Latent Space
机构 * Institute for AI Industry Research (AIR), Tsinghua University（人工智能产业研究院（AIR），清华大学） ； Department of Computer Science and Technology, Tsinghua University（计算机科学与技术系，清华大学） ； Qiuzhen College, Tsinghua University（启真学院，清华大学）
Comments This version (v2) includes minor edits. The paper has been accepted to NeurIPS 2025. Code is available at: https://github.com/MuZhao2333/MolFLAE
On the $O(\frac{\sqrt{d}}{K^{1/4}})$ Convergence Rate of AdamW Measured by $\ell_1$ Norm
机构 * Institute of Robotics and Automatic Information Systems, College of Artificial Intelligence, Nankai University, Tianjin, China（机器人与自动信息系统研究所，人工智能学院，南开大学，天津，中国） ； National Key Lab of General AI, School of Intelligence Science and Technology, Peking University, Beijing, China（通用人工智能国家重点实验室，智能科学与技术学院，北京大学，北京，中国）
Comments V2: NeurIPS Camera-Ready. V3: expand upon the conference version by incorporating the analysis of NAdamW
Continuous Thought Machines
机构 * Sakana AI ； University of Tsukuba（东京大学） ； IT University of Copenhagen（哥本哈根IT大学）
Comments Technical report accompanied by online project page: https://pub.sakana.ai/ctm/ Accepted as a spotlight paper at NeurIPS 2025
Toward a Holistic Evaluation of Robustness in CLIP Models
Comments Accepted to IEEE TPAMI, extension of NeurIPS'23 work: A Closer Look at the Robustness of Contrastive Language-Image Pre-Training (CLIP)
