arXivDaily arXiv每日学术速递 周一至周五更新

期刊&会议

NeurIPS

Conference on Neural Information Processing Systems · 会议 · Machine Learning

共收录 17321
2410.08255 2025-11-25 cs.LG cs.AI

Investigating Representation Universality: Case Study on Genealogical Representations

探讨表示通用性:谱系表示的案例研究

David D. Baek, Yuxiao Li, Max Tegmark

机构 * MIT(麻省理工学院)

AI总结 本研究探讨了大型语言模型在表示谱系信息时的通用性,通过两种实验证据验证图结构表示的通用性,并指出缺乏地面真实表示的挑战。

Comments 14 pages, 7 figures

Journal ref NeurIPS 2025 Workshop on Responsible Foundation Models

详情

展开后加载摘要…

URL PDF HTML 收藏
2111.03201 2025-11-25 cs.LG eess.SP

Compressing Sensor Data for Remote Assistance of Autonomous Vehicles using Deep Generative Models

利用深度生成模型压缩传感器数据以实现自动驾驶车辆的远程协助

Daniel Bogdoll, Johannes Jestram, Jonas Rauch, Christin Scheib, Moritz Wittig, J. Marius Zöllner

AI总结 本文提出利用深度生成模型压缩传感器数据,以提高自动驾驶车辆在需要远程协助时的数据传输效率和重建质量。

Comments Daniel Bogdoll, Johannes Jestram, Jonas Rauch, Christin Scheib and Moritz Wittig contributed equally. Accepted for publication at NeurIPS 2021 ML4AD Workshop

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.18470 2025-11-25 cs.CV

Gaze Beyond the Frame: Forecasting Egocentric 3D Visual Span

超越画面:预测自聚焦三维视觉跨度

Heeseung Yun, Joonil Na, Jaeyeon Kim, Calvin Murdock, Gunhee Kim

机构 * Seoul National University(首尔国立大学) Carnegie Mellon University(卡内基梅隆大学) Reality Labs Research at Meta(Meta现实实验室)

AI总结 本文提出EgoSpanLift方法,通过将自聚焦视觉跨度预测从2D转换到3D,实现对三维环境未来视觉聚焦的高效预测,并构建了大规模基准测试集。

Comments NeurIPS 2025 Spotlight

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.18331 2025-11-25 cs.LG cs.SE

DynamiX: Dynamic Resource eXploration for Personalized Ad-Recommendations

DynamiX: 动态资源探索用于个性化广告推荐

Sohini Roychowdhury, Adam Holeman, Mohammad Amin, Feng Wei, Bhaskar Mehta, Srihari Reddy

机构 * Meta, Ads Data and Representation learning(Meta)

AI总结 Dynamix通过动态资源探索和特征增强提升个性化广告推荐的效率与准确性。

Comments 9 pages, 3 Tables, 5 images. https://openreview.net/pdf?id=oglD54lvcB

Journal ref Neurips 2025 Workshop, Reliable ML from Unreliable Data

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.18199 2025-11-25 stat.ML cs.LG

Improving Forecasts of Suicide Attempts for Patients with Little Data

在数据有限的情况下改进自杀尝试预测

Genesis Hang, Annie Chen, Hope Neveux, Matthew K. Nock, Yaniv Yacoby

机构 * Wellesley College(韦尔斯利学院) Harvard University(哈佛大学)

AI总结 本文提出基于潜在相似性高斯过程的方法,在数据有限的情况下提升对自杀尝试的预测性能。

Comments Accepted at the TS4H Workshop at NeurIPS 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.18162 2025-11-25 cs.CL

Vector Arithmetic in Concept and Token Subspaces

向量算术在概念和令牌子空间中的应用

Sheridan Feucht, Byron Wallace, David Bau

机构 * Northeastern University(东北大学)

AI总结 该研究通过概念和令牌子空间的向量算术揭示语言模型中的语义与表层信息结构。

Comments 9 pages, 6 figures. NeurIPS 2025 Mechanistic Interpretability Workshop

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.17932 2025-11-25 cs.CV cs.GR

Novel View Synthesis from A Few Glimpses via Test-Time Natural Video Completion

通过测试时自然视频补全实现从少量视角合成新视角

Yan Xu, Yixing Wang, Stella X. Yu

AI总结 通过测试时自然视频补全方法,从少量视角生成高质量新视角,无需场景特定训练,有效提升稀疏输入下的场景重建与视角生成质量。

Comments Accepted to NeurIPS 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.17914 2025-11-25 cs.CV cs.AI

Rectifying Soft-Label Entangled Bias in Long-Tailed Dataset Distillation

校正长尾数据集蒸馏中的软标签纠缠偏见

Chenyang Jiang, Hang Zhao, Xinyu Zhang, Zhengcen Li, Qiben Shan, Shaocong Wu, Jingyong Su

机构 * Harbin Institute of Technology, Shenzhen(哈尔滨工业大学(深圳)) Pengcheng Laboratory(鹏城实验室)

AI总结 ADSA通过校正长尾数据集蒸馏中的软标签纠缠偏见,显著提升尾部类别准确率和整体准确率。

Comments 10 pages, accepted by NeurIPS 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.17869 2025-11-25 cs.LG

The Horcrux: Mechanistically Interpretable Task Decomposition for Detecting and Mitigating Reward Hacking in Embodied AI Systems

霍克鲁斯:用于检测和缓解具身AI系统中奖励黑客行为的可解释任务分解

Subramanyam Sahoo, Jared Junkin

机构 * Berkeley AI Safety Initiative (BASIS) UC Berkeley(伯克利人工智能安全计划(BASIS)伯克利大学) Department of Electrical and Computer Engineering Johns Hopkins University(电气与计算机工程系约翰霍普金斯大学)

AI总结 本研究提出MITD方法,通过可解释性任务分解有效检测和缓解具身AI系统中的奖励黑客行为,实验表明分解深度可显著降低奖励黑客频率。

Comments Accepted to the NeurIPS (Mexico City) 2025 Workshop on Embodied and Safe-Assured Robotic Systems (E-SARS). Thanks to Aman Chadha

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.17793 2025-11-25 cs.CV cs.LG

Attention Guided Alignment in Efficient Vision-Language Models

注意力引导的高效视觉-语言模型

Shweta Mahajan, Hoang Le, Hyojin Park, Farzad Farhadzadeh, Munawar Hayat, Fatih Porikli

机构 * Qualcomm AI Research(高通人工智能研究)

AI总结 本文提出AGE-VLM,通过交错交叉注意力层和空间知识提取,减少高效视觉-语言模型中的幻觉问题。

Comments 39th Conference on Neural Information Processing Systems (NeurIPS 2025) Workshop on Efficient Reasoning

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.17753 2025-11-25 physics.chem-ph cs.AI

$Δ$-ML Ensembles for Selecting Quantum Chemistry Methods to Compute Intermolecular Interactions

$Δ$-ML集成用于选择量子化学方法计算分子间相互作用

Austin M. Wallace, C. David Sherrill, Giri P. Krishnan

机构 * School of Chemistry and Biochemistry(化学与生物化学系) Georgia Institute of Technology(佐治亚理工学院) Center for Artificial Intelligence in Science and Engineering(科学与工程中的人工智能中心) Center for Computational Molecular Science and Technology(计算分子科学与技术中心) School of Computational Science and Engineering(计算科学与工程系)

AI总结 本文提出基于$Δ$-ML模型集成的方法,用于选择量子化学方法计算分子间相互作用,通过预测误差识别高效方法并验证理论假设。

Comments NeurIPS ML4PS 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.09567 2025-11-25 cs.LG

Let the Experts Speak: Improving Survival Prediction & Calibration via Mixture-of-Experts Heads

让专家发言:通过专家混合头改进生存预测与校准

Todd Morrill, Aahlad Puli, Murad Megjhani, Soojin Park, Richard Zemel

机构 * Department of Computer Science Columbia University USA(哥伦比亚大学计算机科学系) Department of Computer Science New York University USA(纽约大学计算机科学系) Department of Neurology Columbia University Medical Center USA(哥伦比亚大学医学中心神经病学系) Department of Computer Science Barnard College USA(巴纳德学院计算机科学系) Department of Biomedical Informatics Columbia University Medical Center USA(哥伦比亚大学医学中心生物医学信息学系) NewYork-Presbyterian Hospital at Columbia University Medical Center USA(哥伦比亚大学医学中心新英格兰-纽约 Presbyterian 医院)

AI总结 本文提出了一种改进生存预测和校准的混合专家架构,通过更具表现力的专家实现更准确的患者分组和预测。

Comments Accepted as a proceedings paper at the 2025 Machine Learning for Health Symposium and as a workshop paper at the Learning from Time Series for Health workshop at NeurIPS 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.02162 2025-11-25 cs.RO cs.AI cs.HC

Text to Robotic Assembly of Multi Component Objects using 3D Generative AI and Vision Language Models

通过3D生成AI和视觉语言模型实现多组件物体的文本到机器人组装

Alexander Htet Kyaw, Richa Gupta, Dhruv Shah, Anoop Sinha, Kory Mathewson, Stefanie Pender, Sachin Chitta, Yotto Koga, Faez Ahmed, Lawrence Sass, Randall Davis

机构 * Massachusetts Institute of Technology (MIT)(麻省理工学院) MIT(麻省理工学院) Google DeepMind(谷歌DeepMind) Google, Paradigms of Intelligence(谷歌、范式智能) Autodesk Research(Autodesk研究) MIT Mechanical Engineering(麻省理工学院机械工程系) MIT Architecture(麻省理工学院建筑系) MIT CSAIL(麻省理工学院计算机科学与人工智能实验室)

AI总结 本文提出利用3D生成AI和视觉语言模型实现多组件物体的文本到机器人组装,通过多模态推理分解生成网格并优化组件分配。

Comments Accepted to NeurIPS 2025, Conference on Neural Information Processing Systems, Creative AI Track

详情

展开后加载摘要…

URL PDF HTML 收藏
2510.26451 2025-11-25 cs.LG cs.AI

Robust Graph Condensation via Classification Complexity Mitigation

通过分类复杂性缓解实现鲁棒图压缩

Jiayi Luo, Qingyun Sun, Beining Yang, Haonan Yuan, Xingcheng Fu, Yanbiao Ma, Jianxin Li, Philip S. Yu

机构 * SKLCCSE, School of Computer Science and Engineering, Beihang University(北京航空航天大学信息与电子技术学院) Laboratory for Foundations of Computer Science, University of Edinburgh(爱丁堡大学计算机科学基础实验室) Key Lab of Education Blockchain and Intelligent Technology, Guangxi Normal University(广西师范大学教育区块链与智能技术重点实验室) Gaoling School of Artificial Intelligence, Renmin University of China(中国人民大学光荣人工智能学院) Department of Computer Science, University of Illinois, Chicago(伊利诺伊大学芝加哥分校计算机科学系)

AI总结 本文提出MRGC框架,通过引入流形约束模块,提升图压缩在对抗攻击下的鲁棒性。

Comments Accepted by Neurips 2025 (Spotlight)

详情

展开后加载摘要…

URL PDF HTML 收藏
2509.19659 2025-11-25 cs.CV

Bias in the Picture: Benchmarking VLMs with Social-Cue News Images and LLM-as-Judge Assessment

图像中的偏见:使用社会线索新闻图像和LLM作为裁判的基准测试

Aravind Narayanan, Vahid Reza Khazaie, Shaina Raza

机构 * Vector Institute for AI(向量人工智能研究院)

AI总结 本文通过社会线索新闻图像基准测试,揭示了VLMs在视觉上下文中存在系统性偏见,特别是性别和职业属性,指出高忠实度并不必然降低偏见,提出公平的多模态评估方法。

Comments Accepted to NeurIPS 2025 Workshop (Evaluating the Evolving LLM Lifecycle)

详情

展开后加载摘要…

URL PDF HTML 收藏
2509.17134 2025-11-25 cs.GT

Tight Bounds On the Distortion of Randomized and Deterministic Distributed Voting

随机化和确定性分布式投票的扭曲紧界

Mohammad Ali Abam, Davoud Kareshki, Marzieh Nilipour, Mohammad Hossein Paydar, Masoud Seddighin

AI总结 研究随机化和确定性分布式投票机制的扭曲界,改进了avgmax、maxavg和maxmax的上界,提出了不同随机化设置下的紧界和近紧界。

Comments 36 pages, 12 figures, Accepted at NeurIPS 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2509.13626 2025-11-25 cs.IR cs.AI

Mind the Gap: Aligning Knowledge Bases with User Needs to Enhance Mental Health Retrieval

注意差距:通过用户需求对齐知识库以增强心理健康检索

Amanda Chan, James Jiayu Liu, He Kai, Onno P. Kampman

机构 * Princeton University(普林斯顿大学) National University of Singapore(国立新加坡大学) MOH Office for Healthcare Transformation(卫生部医疗转型办公室)

AI总结 通过用户需求对齐知识库,提升心理健康检索性能,减少内容创建需求,实现高质量信息检索。

Comments 25 pages, 3 figures, submitted to NeurIPS 2025 GenAI4Health

详情

展开后加载摘要…

URL PDF HTML 收藏
2509.01736 2025-11-25 hep-ph cs.LG

Multimodal Generative Flows for LHC Jets

多模态生成流用于LHC喷注

Darius A. Faroughy, Manfred Opper, Cesar Ojeda

机构 * NHETC, Rutgers University(罗格斯大学) TU Berlin(柏林技术大学) University of Potsdam(波茨坦大学)

AI总结 本文提出一种基于变换器的多模态生成流,用于联合建模LHC喷注的连续动力学特征和离散量子数,实现了高保真的喷注生成。

Comments Accepted at NeurIPS 2025 ML4PS workshop

详情

展开后加载摘要…

URL PDF HTML 收藏
2506.11088 2025-11-25 cs.CL cs.AI

One SPACE to Rule Them All: Jointly Mitigating Factuality and Faithfulness Hallucinations in LLMs

一个空间统治它们全部:联合缓解大语言模型中的事实性和忠实性幻觉

Pengbo Wang, Chaozhuo Li, Chenxu Wang, Liwen Zheng, Litian Zhang, Xi Zhang

机构 * Beijing University of Posts and Telecommunications(北京邮电大学) Shihezi University(石河子大学)

AI总结 SPACE通过联合编辑共享激活子空间,有效缓解大语言模型中的事实性和忠实性幻觉问题。

Comments Accepted as NIPS 2025 poster

Journal ref NeurIPS 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2505.19536 2025-11-25 cs.CV cs.AI cs.CL

FlowCut: Rethinking Redundancy via Information Flow for Efficient Vision-Language Models

FlowCut: 通过信息流重新思考冗余性以提高视觉-语言模型的效率

Jintao Tong, Wenwei Jin, Pengda Qin, Anqi Li, Yixiong Zou, Yuhong Li, Yuhua Li, Ruixuan Li

机构 * School of Computer Science and Technology, Huazhong University of Science and Technology(华中科技大学计算机科学与技术学院) Xiaohongshu Inc.(小红书公司) Shanghai Jiao Tong University(上海交通大学)

AI总结 FlowCut通过信息流视角改进视觉-语言模型的冗余识别,实现更高效的剪枝效果。

Comments Accepted by NeurIPS 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2505.18455 2025-11-25 stat.ML cs.LG

On Minimax Estimation of Parameters in Softmax-Contaminated Mixture of Experts

在软最大化污染混合专家模型中参数的最小最大估计

Fanqi Yan, Huy Nguyen, Dung Le, Pedram Akbarian, Nhat Ho, Alessandro Rinaldo

机构 * Department of Computer Science(计算机科学系) Department of Statistics and Data Sciences(统计学与数据科学系) Department of Electrical and Computer Engineering(电气与计算机工程系)

AI总结 本文研究了软最大化污染混合专家模型中参数的最小最大估计问题,探讨了新提示对参数估计的影响,并推导了在可区分性条件下的最优估计速率。

Comments Accepted to NeurIPS 2025. Fanqi Yan, Huy Nguyen, and Dung Le contributed equally to this work

详情

展开后加载摘要…

URL PDF HTML 收藏
2505.16864 2025-11-25 cs.CV

Training-Free Efficient Video Generation via Dynamic Token Carving

无需训练的高效视频生成 via 动态令牌雕刻

Yuechen Zhang, Jinbo Xing, Bin Xia, Shaoteng Liu, Bohao Peng, Xin Tao, Pengfei Wan, Eric Lo, Jiaya Jia

AI总结 Jenga通过动态注意力雕刻和逐步分辨率生成,实现无需训练的高效视频生成,显著提升生成速度并保持生成质量。

Comments NeurIPS 2025, Project Page: https://julianjuaner.github.io/projects/jenga/

详情

展开后加载摘要…

URL PDF HTML 收藏
2502.00313 2025-11-25 cs.GT cs.AI cs.CL cs.MA

Distributive Fairness in Large Language Models: Evaluating Alignment with Human Values

大语言模型中的分配公平性:评估与人类价值观的对齐

Hadi Hosseini, Samarth Khanna

机构 * Penn State University(宾夕法尼亚州立大学)

AI总结 本文研究了大语言模型在分配公平性方面的表现,发现其与人类价值观存在偏差,并探讨了提升对齐性的策略。

Comments Accepted at NeurIPS 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2501.07502 2025-11-25 cs.LG cs.AI

Human-Inspired Multi-Level Reinforcement Learning

受人类启发的多级强化学习

Mingkang Wu, Devin White, Vernon Lawhern, Nicholas R. Waytowich, Yongcan Cao

机构 * The University of Texas at San Antonio(德克萨斯大学圣安东尼奥分校) Army Educational Outreach Program(陆军教育推广计划) DEVCOM Army Research Lab(DEVCOM陆军研究实验室)

AI总结 本文提出了一种受人类启发的多级强化学习方法,通过提取多级信息实现有效学习,结合低级奖励信号和高级方向信息,提升策略优化效果。

Comments Accepted to the Aligning Reinforcement Learning Experimentalists and Theorists Workshop at NeurIPS 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2412.03906 2025-11-25 cs.LG stat.ML

Final-Model-Only Data Attribution with a Unifying View of Gradient-Based Methods

仅最终模型的数据归因与基于梯度方法的统一视角

Dennis Wei, Inkit Padhi, Soumya Ghosh, Amit Dhurandhar, Karthikeyan Natesan Ramamurthy, Maria Chang

机构 * IBM Research(IBM研究院) Merck Research Labs(默克研究实验室)

AI总结 本文提出在仅拥有最终模型的情况下,通过统一基于梯度的方法来评估模型对训练实例的敏感性,并实证分析了不同方法的近似质量。

Comments Published at the Thirty-Ninth Annual Conference on Neural Information Processing Systems (NeurIPS 2025). 28 pages, 11 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
2411.13093 2025-11-25 cs.CV cs.AI

Video-RAG: Visually-aligned Retrieval-Augmented Long Video Comprehension

Video-RAG: 基于视觉对齐的检索增强长视频理解

Yongdong Luo, Xiawu Zheng, Guilin Li, Shukang Yin, Haojia Lin, Chaoyou Fu, Jinfa Huang, Jiayi Ji, Fei Chao, Jiebo Luo, Rongrong Ji

机构 * Key Laboratory of Multimedia Trusted Perception and Efficient Computing, Ministry of Education of China, Xiamen University(中国教育部多媒体可信感知与高效计算重点实验室,厦门大学) Nanjing University(南京大学) University of Rochester(罗切斯特大学)

AI总结 Video-RAG通过视觉对齐的辅助文本提升长视频理解性能,无需训练且兼容性强,显著优于专有模型。

Comments Accepted at NeurIPS 2025. Camera-ready version

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.17400 2025-11-24 cs.CV cs.AI

Sparse Mixture-of-Experts for Multi-Channel Imaging: Are All Channel Interactions Required?

稀疏专家混合模型用于多通道成像:所有通道交互都必要吗?

Sukwon Yun, Heming Yao, Burkhard Hoeckendorf, David Richmond, Aviv Regev, Russell Littman

机构 * University of North Carolina at Chapel Hill(北卡罗来纳大学教堂山分校) Research and Early Development (gRED), Genentech(基因泰克研发与早期开发部) Biology Research — AI Development (BRAID), Genentech(基因泰克生物学研究——人工智能开发部)

AI总结 本文提出MoE-ViT,通过稀疏专家混合模型优化多通道图像处理,提升效率而不牺牲性能。

Comments This has been accepted at the NeurIPS AI4Science Workshop 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.17399 2025-11-24 cs.LG

Stable Coresets via Posterior Sampling: Aligning Induced and Full Loss Landscapes

通过后验采样实现稳定的聚类:对诱导和完整损失景观的对齐

Wei-Kai Chang, Rajiv Khanna

机构 * Purdue University(普渡大学)

AI总结 本文提出了一种基于后验采样的稳定聚类方法,通过优化损失景观对齐,提升模型稳定性与泛化能力。

Comments neurips 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2506.02813 2025-11-24 q-bio.NC cs.NE

Brain-Like Processing Pathways Form in Models With Heterogeneous Experts

具有异质专家的模型中形成类脑处理路径

Jack Cook, Danyal Akarca, Rui Ponte Costa, Jascha Achterberg

AI总结 本文提出混合路径模型,通过诱导偏差研究大脑如何形成任务特定路径,并验证其在不同难度任务中的有效性。

Comments Accepted at 39th Conference on Neural Information Processing Systems (NeurIPS 2025); 31 pages, 16 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.17258 2025-11-24 cs.LG

Enforcing governing equation constraints in neural PDE solvers via training-free projections

通过无训练投影强制神经PDE求解器中的约束方程

Omer Rochman, Gilles Louppe

机构 * University of Liège(利根大学)

AI总结 本文提出两种无训练投影方法,用于减少神经PDE求解器中约束方程的违反情况,提升求解精度。

Comments Machine Learning and the Physical Sciences, Neurips 2025, San Diego

详情

展开后加载摘要…

URL PDF HTML 收藏