NeurIPS

CSI-Bench: A Large-Scale In-the-Wild Dataset for Multi-task WiFi Sensing

CSI-Bench: 一个大规模真实场景下的多任务Wi-Fi传感数据集

Guozhen Zhu, Yuqian Hu, Weihang Gao, Wei-Hsiang Wang, Beibei Wang, K. J. Ray Liu

机构 * Origin Research（Origin研究机构） 纠错

AI总结 CSI-Bench是一个大规模真实场景下的多任务Wi-Fi传感数据集，旨在通过真实环境下的数据收集和标准化评估支持鲁棒且可泛化的模型开发。

Comments 26 pages, 5 figures, accepted by Thirty-Ninth Annual Conference on Neural Information Processing Systems (NeurIPS)

xLSTM-Mixer: Multivariate Time Series Forecasting by Mixing via Scalar Memories

xLSTM-Mixer：通过标量记忆混合实现多变量时间序列预测

Maurice Kraus, Felix Divo, Devendra Singh Dhami, Kristian Kersting

机构 * AI & ML Group, TU Darmstadt（AI与机器学习小组，德累斯顿理工大学） ； TU Eindhoven（埃因霍温理工大学） ； Hessian Center for AI (hessian.AI)（黑森人工智能中心） ； German Research Center for AI (DFKI)（德国人工智能研究中心） ； Centre for Cognitive Science, TU Darmstadt（认知科学中心，德累斯顿理工大学） 纠错

AI总结 xLSTM-Mixer通过融合多种视角和混合架构，实现了多变量时间序列预测的高精度与低内存消耗。

Comments Poster at NeurIPS 2025

The Impact of Quantization on Large Reasoning Model Reinforcement Learning

量化对大推理模型强化学习的影响

Medha Kumar, Zifei Xu, Xin Wang, Tristan Webb

机构 * Pennsylvania State University（宾夕法尼亚州立大学） ； d-Matrix 纠错

AI总结 本文研究了量化对大推理模型强化学习的影响，发现量化感知训练对学习过程有负面影响，而PTQ和QLoRA能提升性能。

Comments Accepted to the NeurIPS 2025 Efficient Reasoning Workshop

Sample-Adaptivity Tradeoff in On-Demand Sampling

按需采样中的样本复杂度与轮次复杂度权衡

Nika Haghtalab, Omar Montasser, Mingda Qiao

机构 * University of California, Berkeley（加州大学伯克利分校） ； Yale University（耶鲁大学） ； University of Massachusetts Amherst（马萨诸塞大学阿默斯特分校） 纠错

AI总结 本文提出了一种新的框架OODS，用于在按需采样中平衡样本和轮次复杂度，实现了近最优的样本复杂度和轮次复杂度界。

Comments 50 pages, to appear at NeurIPS 2025

SNAP: Low-Latency Test-Time Adaptation with Sparse Updates

Hyeongheon Cha, Dong Min Kim, Hye Won Chung, Taesik Gong, Sung-Ju Lee

机构 * School of Electrical Engineering, KAIST, Daejeon, Republic of Korea（韩国科学技术院电子工程学院） ； Department of Computer Science and Engineering, UNIST, Ulsan, Republic of Korea（全南国立大学计算机科学与工程系） 纠错

Journal ref Advances in Neural Information Processing Systems 39 (NeurIPS 2025)

Overfitting in Adaptive Robust Optimization

Karl Zhu, Dimitris Bertsimas

机构 * Operations Research Center（运筹学研究中心） ； Sloan School of Management（斯隆管理学院） 纠错

Comments 4 pages, 1 figure, Accepted to NeurIPS 2025 ML x OR Workshop

MAP Estimation with Denoisers: Convergence Rates and Guarantees

Scott Pesme, Giacomo Meanti, Michael Arbel, Julien Mairal

机构 * Univ. Grenoble Alpes, Inria, CNRS, Grenoble INP, LJK（格勒诺布尔阿尔卑斯大学、法国国家科学研究中心、Inria、Grenoble INP、LJK） 纠错

Comments Uploading the neurips 2025 camera ready version

A Unified Framework for Provably Efficient Algorithms to Estimate Shapley Values

Tyler Chen, Akshay Seshadri, Mattia J. Villani, Pradeep Niroula, Shouvanik Chakrabarti, Archan Ray, Pranav Deshpande, Romina Yalovetzky, Marco Pistoia, Niraj Kumar

机构 * Global Technology Applied Research（全球技术应用研究） 纠错

Comments Accepted at the 39th Conference on Neural Information Processing Systems (NeurIPS 2025); 45 pages, 7 figures, 7 tables

Put CASH on Bandits: A Max K-Armed Problem for Automated Machine Learning

Amir Rezaei Balef, Claire Vernade, Katharina Eggensperger

机构 * Department of Computer Science, University of Tübingen（图宾根大学计算机科学系） 纠错

Comments Accepted at the 39th Conference on Neural Information Processing Systems (NeurIPS 2025)

SOLID: a Framework of Synergizing Optimization and LLMs for Intelligent Decision-Making

Yinsheng Wang, Tario G You, Léonard Boussioux, Shan Liu

机构 * Department of Industrial & Systems Engineering, University of Washington（工业与系统工程系，华盛顿大学） ； College of Engineering, University of Washington（工程学院，华盛顿大学） ； Michael G. Foster School of Business, University of Washington（Michael G. Foster商学院，华盛顿大学） 纠错

Comments NeurIPS 2025 WORKSHOP ML*OR Workshop: Mathematical Foundations and Operational Integration of Machine Learning for Uncertainty-Aware Decision-Making

Particle Monte Carlo methods for Lattice Field Theory

David Yallup

机构 * Kavli Institute for Cosmology Cambridge, University of Cambridge（卡弗里宇宙学研究所剑桥，剑桥大学） 纠错

Comments To appear in the NeurIPS 2025 workshop, Frontiers in Probabilistic Inference: Sampling Meets Learning

Learning Human-Like RL Agents Through Trajectory Optimization With Action Quantization

Jian-Ting Guo, Yu-Cheng Chen, Ping-Chun Hsieh, Kuo-Hao Ho, Po-Wei Huang, Ti-Rong Wu, I-Chen Wu

机构 * Department of Computer Science, National Yang Ming Chiao Tung University（National Yang Ming Chiao Tung University计算机科学系） ； Institute of Information Science, Academia Sinica（Academia Sinica信息科学研究所） ； Research Center for Information Technology Innovation, Academia Sinica（Academia Sinica信息技术创新研究中心） 纠错

Comments Accepted by the Thirty-Ninth Annual Conference on Neural Information Processing Systems (NeurIPS 2025)

Toward Complete Merger Identification at Cosmic Noon with Deep Learning

Aimee Schechter, Aleksandra Ciprijanovic, Rebecca Nevin, Julie Comerford, Xuejian Shen, Aaron Stemo, Laura Blecha

Comments 5 pages, accepted to the NeurIPS 2025 Machine Learning for the Physical Sciences Workshop

IonCast: A Deep Learning Framework for Forecasting Ionospheric Dynamics

Halil S. Kelebek, Linnea M. Wolniewicz, Michael D. Vergalla, Simone Mestici, Giacomo Acciarini, Bala Poduval, Olga Verkhoglyadova, Madhulika Guhathakurta, Thomas E. Berger, Frank Soboczenski, Atılım Güneş Baydin

机构 * Department of Engineering Science University of Oxford, UK（牛津大学工程科学系） ； Department of Information and Computer Science University of Hawai‘i at Mānoa, USA（夏威夷大学信息与计算机科学系） ； Free Flight Research Lab（自由飞行研究实验室） ； Università degli Studi di Roma Sapienza（罗马大学） ； European Space Agency (ESA)（欧洲航天局） ； University of New Hampshire（新罕布什尔大学） ； NASA Jet Propulsion Laboratory（NASA喷气推进实验室） ； NASA Headquarters（NASA总部） ； Space Weather Technology, Research, and Education Center University of Colorado Boulder（空间天气技术、研究与教育中心，科罗拉多大学博尔德分校） ； Department of Computer Science University of York & King’s College London, UK（约克大学计算机科学系及伦敦国王学院计算机科学系） ； Department of Computer Science University of Oxford, UK（牛津大学计算机科学系） 纠错

Comments 11 pages, 7 figures, 3 tables. Accepted as a poster presentation at the Machine Learning for the Physical Sciences Workshop at NeurIPS 2025

Logit-Based Losses Limit the Effectiveness of Feature Knowledge Distillation

Nicholas Cooper, Lijun Chen, Sailesh Dwivedy, Danna Gurari

机构 * University of Colorado Boulder（科罗拉多大学波得尔） 纠错

Comments NeurIPS Workshop on Symmetry and Geometry in Neural Representations (NeurReps), December 2025

It's LIT! Reliability-Optimized LLMs with Inspectable Tools

Ruixin Zhang, Jon Donnelly, Zhicheng Guo, Ghazal Khalighinejad, Haiyang Huang, Alina Jade Barnett, Cynthia Rudin

机构 * Department of Computer Science（计算机科学系） ； Duke University（杜克大学） 纠错

Comments Accepted to the 39th Conference on Neural Information Processing Systems (NeurIPS 2025) Workshop on Multi-Turn Interactions in Large Language Models

Quantifying the Role of OpenFold Components in Protein Structure Prediction

Tyler L. Hayes, Giri P. Krishnan

机构 * College of Computing（计算机学院） ； Georgia Institute of Technology（佐治亚理工学院） 纠错

Comments Accepted to the NeurIPS 2025 Workshop on Multi-modal Foundation Models and Large Language Models for Life Sciences

Learning Interestingness in Automated Mathematical Theory Formation

George Tsoukalas, Rahul Saha, Amitayush Thakur, Sabrina Reguyal, Swarat Chaudhuri

机构 * UT Austin（得克萨斯大学） ； Princeton University（普林斯顿大学） ； Stanford University（斯坦福大学） 纠错

Comments NeurIPS 2025 Spotlight

Conflict Adaptation in Vision-Language Models

Xiaoyang Hu

机构 * Brown University（布朗大学） 纠错

Comments Workshop on Interpreting Cognition in Deep Learning Models at NeurIPS 2025

Automated Detection of Visual Attribute Reliance with a Self-Reflective Agent

Christy Li, Josep Lopez Camuñas, Jake Thomas Touchet, Jacob Andreas, Agata Lapedriza, Antonio Torralba, Tamar Rott Shaham

机构 * MIT CSAIL（麻省理工学院计算机科学与人工智能实验室） ； Universitat Oberta de Catalunya（巴塞罗那开放大学） ； Louisiana Tech（路易斯安那理工大学） ； Northeastern University（东北大学） 纠错

Comments 32 pages, 10 figures, Neurips 2025

Sync or Sink: Bounds on Algorithmic Collective Action with Noise and Multiple Groups

Aditya Karan, Prabhat Kalle, Nicholas Vincent, Hari Sundaram

Comments Updated with feedback for NeurIPS workshop

MMG: Mutual Information Estimation via the MMSE Gap in Diffusion

Longxuan Yu, Xing Shi, Xianghao Kong, Tong Jia, Greg Ver Steeg

机构 * University of California, Riverside（加州大学河滨分校） ； University of California, San Diego（加州大学圣地亚哥分校） 纠错

Comments Accepted to the SPIGM Workshop at NeurIPS 2025

RAT: Bridging RNN Efficiency and Attention Accuracy via Chunk-based Sequence Modeling

Xiuying Wei, Anunay Yadav, Razvan Pascanu, Caglar Gulcehre

机构 * CLAIRE, EPFL（EPFL的CLAIRE） ； Google DeepMind（谷歌DeepMind） 纠错

Comments Accepted by NeurIPS 2025

Learning in Compact Spaces with Approximately Normalized Transformer

Jörg K. H. Franke, Urs Spiegelhalter, Marianna Nezhurina, Jenia Jitsev, Frank Hutter, Michael Hefenbrock

机构 * University of Freiburg（弗赖堡大学） ； ELLIS Institute Tübingen（图宾根ELLIS研究所） ； Open-Sci Collective（开放科学集体） ； LAION ； Jülich Supercomputing Centre (JSC)（尤利奇超级计算中心） ； Prior Labs（Prior实验室） 纠错

Comments In Proceedings of the 39th Conference on Neural Information Processing Systems (NeurIPS 2025)

AdCare-VLM: Towards a Unified and Pre-aligned Latent Representation for Healthcare Video Understanding

Md Asaduzzaman Jabin, Hanqi Jiang, Yiwei Li, Patrick Kaggwa, Eugene Douglass, Juliet N. Sekandi, Tianming Liu

机构 * University of Georgia（佐治亚大学） 纠错

Comments 39th Conference on Neural Information Processing Systems (NeurIPS 2025) Workshop: 7th International Workshop on Large Scale Holistic Video Understanding: Toward Video Foundation Models

Journal ref Neural Information Processing Systems (NeurIPS 2025)

The Effect of Optimal Self-Distillation in Noisy Gaussian Mixture Model

Kaito Takanami, Takashi Takahashi, Ayaka Sakata

Comments Accepted at NeurIPS 2025

AdamHD: Decoupled Huber Decay Regularization for Language Model Pre-Training

Fu-Ming Guo, Yingfang Fan

机构 * Stanford University（斯坦福大学） ； Harvard Medical School（哈佛医学院） 纠错

Comments 39th Conference on Neural Information Processing Systems (NeurIPS 2025) Workshop: GPU-Accelerated and Scalable Optimization (ScaleOpt)

Journal ref 39th Conference on Neural Information Processing Systems (NeurIPS 2025) Workshop: GPU-Accelerated and Scalable Optimization (ScaleOpt)

Encoding and Understanding Astrophysical Information in Large Language Model-Generated Summaries

Kiera McCormick, Rafael Martínez-Galarza

机构 * Department of Computer Science（计算机科学系） ； Johns Hopkins University（约翰霍普金斯大学） ； AstroAI Center for Astrophysics（天体物理中心） 纠错

Comments Accepted to the Machine Learning and the Physical Sciences Workshop at NeurIPS 2025, 11 pages, 4 figures

High-resolution weak lensing mass mapping from DES-Y3 data using diffusion-based prior

Supranta S. Boruah, Michael Jacob, Bhuvnesh Jain, Riya Maiya, Raghav Venkataramanan

Comments 6 pages, 3 figures, Accepted at Neurips Machine Learning and the Physical Sciences workshop

BEDLAM2.0: Synthetic Humans and Cameras in Motion

Joachim Tesch, Giorgio Becherini, Prerana Achar, Anastasios Yiannakidis, Muhammed Kocabas, Priyanka Patel, Michael J. Black

机构 * Max Planck Institute for Intelligent Systems（马克斯·普朗克智能系统研究所） ； Meshcapade GmbH（Meshcapade公司） 纠错

Comments NeurIPS 2025 (Datasets and Benchmarks track, oral). Project website: https://bedlam2.is.tue.mpg.de