arXivDaily arXiv每日学术速递 周一至周五更新

高校专区

Harvard University(哈佛大学)

共收录 1302
2510.01384 2026-05-26 cs.LG

Fine-Tuning Masked Diffusion for Provable Self-Correction

微调掩码扩散以实现可证明的自校正

Jaeyeon Kim, Seunggeun Kim, Taekyun Lee, David Z. Pan, Hyeji Kim, Sham Kakade, Sitan Chen

机构 * Harvard University(哈佛大学) The University of Texas at Austin(德克萨斯大学奥斯汀分校) Kempner Institute(凯姆纳研究所)

AI总结 提出PRISM方法,通过轻量级模型无关的重新掩码策略,在掩码扩散模型中实现可证明的自校正,无需强化学习或验证器,提升低质量令牌检测与修正能力。

Comments Authorship statement: Jaeyeon Kim and Seunggeun Kim contributed equally, and Taekyun Lee is also a co first author

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.07919 2026-05-25 cs.CV

MedVIGIL: Evaluating Trustworthy Medical VLMs Under Broken Visual Evidence

MedVIGIL: 在视觉证据受损下评估可信的医学视觉语言模型

Hanqi Jiang, Junhao Chen, Mingyu Kang, Hyeokjae Kwon, Yi Pan, Lifeng Chen, Weihang You, Haozhen Gong, Ruiyu Yan, Jinglei Lv, Lin Zhao, Hui Ren, Quanzheng Li, Tianming Liu, Xiang Li

机构 * University of Georgia(佐治亚大学) Harvard Medical School(哈佛医学院) Chungbuk National University(Chungbuk国立大学) Chungnam National University Hospital(Chungnam国立大学医院) National University of Singapore(新加坡国立大学) New York University(纽约大学) University of Sydney(悉尼大学) New Jersey Institute of Technology(新泽西理工学院)

AI总结 本文提出MedVIGIL基准,通过300例由放射科医生监督构建的扰动证据测试集,评估医学视觉语言模型在视觉证据失效时的可靠性,并引入MedVIGIL复合评分(MCS)进行审计。

详情

展开后加载摘要…

URL PDF HTML 收藏
2502.20349 2026-05-25 q-bio.NC cs.AI

Naturalistic Computational Cognitive Science: Towards generalizable models and theories that capture the full range of natural behavior

自然主义计算认知科学:迈向能够捕捉自然行为全范围的通用模型与理论

Wilka Carvalho, Andrew Lampinen

机构 * Kempner Institute for the Study of Natural and Artificial Intelligence(Kempner自然与人工智能研究学院) Harvard University(哈佛大学) Google DeepMind(谷歌DeepMind)

AI总结 本文主张通过整合人工智能与认知科学,采用自然主义实验范式与计算模型,以构建能够泛化至自然情境与行为的通用理论。

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.23672 2026-05-25 cs.CV

RiGS: Rigid-aware 4D Gaussian Splatting from a Single Monocular Video

RiGS: 从单目视频中的刚性感知4D高斯泼溅

Chenyu Wu, Wanhua Li, Zhu-Tian Chen, Hanspeter Pfister

机构 * Harvard University(哈佛大学) Zhejiang University(浙江大学) Nanyang Technological University(南洋理工大学) University of Minnesota - Twin Cities(明尼苏达大学-双城分校)

AI总结 提出RiGS,通过静态、刚性和瞬态三种高斯原语分别建模不同时间尺度的运动,并利用目标动态掩码和场景流引导实现单目视频动态场景重建,在新视角合成任务上达到最优性能。

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.23287 2026-05-25 cs.CV

LangFlash: Feed-forward 3D Language Gaussian Splatting from Sparse Unposed Images

LangFlash: 基于前馈的3D语言高斯泼溅从稀疏无位姿图像

Yilong Liu, Wanhua Li, Chen Zhu-Tian, Hanspeter Pfister

机构 * Harvard University(哈佛大学) Nanyang Technological University(南洋理工大学) Tsinghua University(清华大学) University of Minnesota - Twin Cities(明尼苏达大学-双城分校)

AI总结 提出LangFlash,一种前馈框架,从稀疏无位姿多视图图像直接预测3D高斯原语及其语言对齐语义特征,实现低延迟3D重建和语义一致场景理解。

Comments CVPRF 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.23262 2026-05-25 cs.AI

Design and Report Benchmarks for Knowledge Work

知识工作的设计与报告基准

Yining Hua, Hongbin Na, Cyrus Ayubcha, Levi Lian

机构 * Harvard University(哈佛大学) University of Technology Sydney(悉尼科技大学) Stanford University(斯坦福大学) Raycaster AI

AI总结 针对当前知识工作评估基准与真实部署脱节的问题,提出三步法(定义工作活动、指定测试设置、评分工作产品)来明确基准任务与工作主张的对应关系,并通过三个案例分析展示设计选择如何影响可支持的工作主张。

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.23134 2026-05-25 cs.LG

Archimedean Copula Inference via Taylor-Mode AD

通过泰勒模式自动微分进行阿基米德Copula推断

Cambridge Yang, Dongdong Li

机构 * Cambridge Yang(剑桥阳) Harvard Medical School(哈佛医学院)

AI总结 提出一个JAX原生框架acopula,利用泰勒模式自动微分实现任意阿基米德生成元下的嵌套Copula精确似然和参数梯度计算,支持高维、任意删失和任意嵌套树,并在多个实际数据集上验证了正确性和效率。

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.23103 2026-05-25 cs.CL cs.AI cs.CY cs.DB

A Fine-Tuned BERT Classifier for Personal-Letter Titles in Late-Ming and Early-Qing Collected Works

用于明清之际文集中个人书信标题的微调BERT分类器

Queenie Luo

机构 * Harvard University(哈佛大学)

AI总结 提出Lepton模型,通过微调BERT-base-chinese对明清文集目录中的标题进行分类,区分个人书信与易混淆的序文,已应用于中国传记数据库识别约五万五千封书信。

详情

展开后加载摘要…

URL PDF HTML 收藏
2604.11679 2026-05-25 cs.CV

Towards Brain MRI Foundation Models for the Clinic: Findings from the FOMO25 Challenge

面向临床的大脑MRI基础模型:来自FOMO25挑战赛的发现

Asbjørn Munk, Stefano Cerri, Vardan Nersesjan, Christian Hedeager Krag, Jakob Ambsdorf, Pablo Rocamora García, Julia Machnio, Peirong Liu, Suhyun Ahn, Nasrin Akbari, Yasmina Al Khalil, Kimberly Amador, Sina Amirrajab, Tal Arbel, Meritxell Bach Cuadra, Ujjwal Baid, Bhakti Baheti, Jaume Banus, Kamil Barbierik, Christoph Brune, Yansong Bu, Baptiste Callard, Yuhan Chen, Cornelius Crijnen, Corentin Dancette, Peter Drotar, Prasad Dutande, Nils D. Forkert, Saurabh Garg, Jakub Gazda, Matej Gazda, Benoît Gérin, Partha Ghosh, Weikang Gong, Pedro M. Gordaliza, Sam Hashemi, Tobias Heimann, Fucang Jia, Jiexin Jiang, Emily Kaczmarek, Chris Kang, Seung Kwan Kang, Mohammad Khazaei, Julien Khlaut, Petros Koutsouvelis, Jae Sung Lee, Yuchong Li, Mengye Lyu, Mingchen Ma, Anant Madabhushi, Klaus H. Maier-Hein, Pierre Manceron, Andrés Martínez Mora, Moona Mazher, Felix Meister, Nataliia Molchanova, Steven A. Niederer, Leonard Nürnberg, Jinah Park, Abdul Qayyum, Jonas Richiardi, Antoine Saporta, Branislav Setlak, Ning Shen, Justin Szeto, Constantin Ulrich, Puru Vaish, Vibujithan Vigneshwaran, Leroy Volmer, Zihao Wang, Siqi Wei, Anthony Winder, Jelmer M. Wolterink, Maxence Wynen, Chang Yang, Si Young Yie, Mostafa Mehdipour Ghazi, Akshay Pai, Espen Jimenez Solem, Sebastian Nørgaard Llambias, Mikael Boesen, Michael Eriksen Benros, Juan Eugenio Iglesias, Mads Nielsen

机构 * organization= Department of Computer Science, University of Copenhagen , city= Copenhagen , country= Denmark organization= Pioneer Centre for AI , city= Copenhagen , country= Denmark organization= Copenhagen Research Centre for Biological Precision Psychiatry, Mental Health Centre Copenhagen, Copenhagen University Hospital , region= Capital Region of Denmark , city= Copenhagen , country= Denmark organization= Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital Harvard Medical School , city= Boston , state= Massachusetts , country= USA Artificial Intelligence Laboratory, Massachusetts Institute of Technology , city= Boston , state= Massachusetts , country= USA organization= Johns Hopkins University , city= Baltimore , state= Maryland , country= USA organization= Radiological AI Testcenter (RAIT) , region= Capital Region of Denmark , city= Copenhagen , country= Denmark organization= Copenhagen University Hospital, Rigshospitalet , region= Capital Region of Denmark , city= Copenhagen , country= Denmark organization= Copenhagen University Hospital, Bispebjerg \& Frederiksberg Hospital , region= Capital Region of Denmark , city= Copenhagen , country= Denmark organization= Department of Clinical Medicine, Faculty of Health Medical Sciences, University of Copenhagen , city= Copenhagen , country= Denmark organization= Division of Medical Image Computing, German Cancer Research Center (DKFZ) , city= Heidelberg , country= Germany organization= University of British Columbia , city= Vancouver , state= British Columbia , country= Canada organization= Hawkes Institute, Department of Computer Science, University College London , city= London , country= United Kingdom Lung Institute, Faculty of Medicine, Imperial College London , city= London , country= United Kingdom organization= Department of Applied Mathematics, Technical Medical Centre, University of Twente , city= Enschede , country= Netherlands organization= IISLAB, Technical University of Košice , city= Košice , country= Slovakia organization= 2nd Department of Internal Medicine, Pavol Jozef Safarik University L Pasteur University Hospital , city= Košice , country= Slovakia organization= Fudan University , city= Shanghai , country= China organization= Shenzhen Technology University , city= Shenzhen , country= China organization= Department of Radiology, Lausanne University Hospital University of Lausanne , city= Lausanne , country= Switzerland organization= Louvain Neuroinflammation Imaging Lab (NIL), Université Catholique de Louvain , city= Brussels , country= Belgium organization= University of Applied Sciences organization= CIBM Center for Biomedical Imaging , city= Lausanne , country= Switzerland organization= Department of Radiation Oncology (Maastro), GROW Research Institute for Oncology Reproduction, Maastricht University Medical Centre+ , city= Maastricht , country= The Netherlands organization= Department of Biomedical Engineering, Medical Image Analysis, Eindhoven University of Technology , city= Eindhoven , country= The Netherlands organization= Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences , city= Shenzhen , country= China organization= McGill University Mila - Quebec AI Institute , city= Montreal , country= Canada organization= Hotchkiss Brain Institute Department of Radiology, University of Calgary , city= Calgary , state= Alberta , country= Canada organization= Department of Radiology, University of Calgary , city= Calgary , state= Alberta , country= Canada organization= Alberta Children's Hospital Research Institute, Department of Clinical Neuroscience, University of Calgary , city= Calgary , state= Alberta , country= Canada organization= The Wallace H. Coulter Department of Biomedical Engineering, Georgia Tech Emory University , city= Atlanta , state= Georgia , country= USA organization= SGGS College of Engineering organization= Seoul National University , city= Seoul , country= South Korea organization= The D-Lab, Department of Precision Medicine, GROW Research Institute for Oncology Reproduction, Maastricht University , city= Maastricht , country= The Netherlands organization= Artificial Intelligence in Medicine (AIM) Program, Mass General Brigham, Harvard Medical School , city= Boston , state= Massachusetts , country= USA Nuclear Medicine, CARIM \& GROW, Maastricht University , city= Maastricht , country= The Netherlands organization= Department of Radiation Oncology, Dana-Farber Cancer Institute, Brigham Women’s Hospital, Harvard Medical School , city= Boston , state= Massachusetts , country= USA Learning Group, Heidelberg University Hospital , city= Heidelberg , country= Germany

AI总结 针对临床脑MRI数据异质且标注成本高的问题,FOMO25挑战赛通过自监督预训练(FOMO60K数据集)评估了16个团队的基础模型,发现自监督预训练能提升域迁移泛化性,但不同任务需不同预训练目标,且模型规模扩展收益有限。

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.07235 2026-05-25 cs.LG cs.AI cs.IT math.IT

ArcMark: Distortion-Free Multi-Byte LLM Watermark via Optimal Transport

ArcMark: 通过最优传输实现无失真的多字节大语言模型水印

Atefeh Gilani, Sajani Vithana, Carol Xuan Long, Oliver Kosut, Lalitha Sankar, Flavio P. Calmon

机构 * Arizona State University(亚利桑那州立大学) Harvard University(哈佛大学)

AI总结 本文提出ArcMark,一种基于编码和信息论原理的无失真水印方法,能够在数百个token中可靠嵌入多字节信息,且不改变LLM的下一token分布。

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.22581 2026-05-22 cs.CV cs.AI cs.LG

SceneAligner: 3D-Grounded Floorplan Localization in the Wild

SceneAligner: 在真实场景中实现基于3D的平面定位

Junhyeong Cho, Ruojin Cai, Hadar Averbuch-Elor

机构 * Cornell University(康奈尔大学) Kempner Institute, Harvard University(哈佛大学 Kempner 院)

AI总结 本文提出了一种在真实场景中实现基于3D重建的平面定位方法,通过将任务 grounding 在场景的重建3D表示中,解决了现有方法在大规模建筑和栅格化平面图中应用受限的问题。

Comments Project Page: https://Cornell-VAILab.github.io/SceneAligner

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.22471 2026-05-22 cs.LG

Lost in Tokenization: Fundamental Trade-offs in Graph Tokenization for Transformers

迷失在标记化中:图标记化在Transformer中的基本权衡

Maya Bechler-Speicher, Gilad Yehudai, Gil Harari, Clayton Sanford, Amir Globerson, Joan Bruna

机构 * Courant Institute of Mathematical Sciences, New York University(纽约大学数学科学学院) John A. Paulson School of Engineering and Applied Sciences, Harvard University(哈佛大学工程与应用科学学院) Google Research(谷歌研究) Tel-Aviv University(特拉维夫大学)

AI总结 本文研究了图标记化在Transformer中的基本权衡,探讨了不同标记化方法对模型表达能力的影响,并通过实验验证了不同任务对不同结构视图的偏好。

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.07870 2026-05-22 cond-mat.dis-nn cs.AI stat.ML

Spectral Dynamics in Deep Networks: Feature Learning, Outlier Escape, and Learning Rate Transfer

深度网络中的谱动力学:特征学习、异常值逃逸和学习率转移

Clarissa Lauditi, Cengiz Pehlevan, Blake Bordelon

机构 * John A. Paulson School of Engineering and Applied Sciences, Harvard University(哈佛大学约翰A·保罗森工程与应用科学学院) Kempner Institute for the Study of Natural and Artificial Intelligence, Harvard University(哈佛大学自然与人工智能研究学院) Center for Mathematical Sciences and Applications, Harvard University(哈佛大学数学科学中心) Oden Institute for Computational Engineering and Sciences & Dept. of Neuroscience, UT Austin(得克萨斯大学奥斯汀分校奥登计算工程与科学学院及神经科学系)

AI总结 本文研究了在宽神经网络中通过(随机)梯度下降训练时隐藏权重谱的演变,提出了一种双层动态平均场理论(DMFT)来联合跟踪具有尖峰集合的隐藏权重谱动态,其中尖峰方向在随机体上保持统计依赖性。该框架应用于两种设置:(1)无限宽度非线性网络在均值场/μP缩放下,以及(2)深度线性网络在比例高维极限下。理论预测了异常值如何随训练时间、宽度、输出尺度和初始化方差演变。在深度线性网络中,μP产生与宽度一致的异常值动态和超参数转移,包括主导NTK模式向稳定性边缘(EoS)的宽度稳定增长。相比之下,NTK参数化表现出强烈依赖宽度的异常值动态,尽管收敛到一个稳定的宽网络极限。我们展示了这种体+异常值图像是描述简单任务的,但涉及大量输出的任务(如ImageNet分类或GPT语言建模)则更适合通过重构谱体来描述。我们开发了一个具有大量输出通道的玩具模型,重现了这一现象,并展示了足够宽的网络下谱边缘仍会收敛。

Comments Updating related works + discussion

详情

展开后加载摘要…

URL PDF HTML 收藏
2601.20205 2026-05-22 cs.LG

Hyperparameter Transfer with Mixture-of-Expert Layers

通过专家混合层进行超参数迁移

Tianze Jiang, Blake Bordelon, Cengiz Pehlevan, Boris Hanin

机构 * Operations Research Financial Engineering, Princeton University, Princeton, NJ, USA Center of Mathematical Sciences Applications, Harvard University, Cambridge, MA, USA John A. Paulson School of Engineering Applied Sciences, Center for Brain Science, Kempner Institute for the Study of Natural Artificial Intelligence, Harvard University, Cambridge, MA, USA

AI总结 本文提出了一种新的参数化方法,用于在扩展模型宽度、深度、专家数量和专家(隐藏)大小时,通过专家混合层的变压器模型进行超参数迁移,该方法基于动态平均场理论分析,实验证明其在不同规模模型间可靠地迁移超参数。

Comments ICML 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2512.04111 2026-05-22 cs.SE cs.AI cs.HC

CentaurEval: Benchmarking Human-in-the-Loop Value in Agentic Coding

CentaurEval: 评估人机协同在编程中的价值

Hanjun Luo, Chiming Ni, Jiaheng Wen, Zhimu Huang, Yiran Wang, Bingduo Liao, Sylvia Chung, Yingbin Jin, Xinfeng Li, Wenyuan Xu, XiaoFeng Wang, Hanan Salam

机构 * New York University Abu Dhabi(纽约大学阿布扎克校区) Nanyang Technological University(南洋理工大学) University of Illinois Urbana-Champaign(伊利诺伊大学厄巴纳-香槟分校) Harvard University(哈佛大学) Zhejiang University(浙江大学) University of Electronic Science and Technology of China(电子科技大学) Beijing University of Technology(北京理工大学) The Hong Kong Polytechnic University(香港理工大学)

AI总结 本文提出CentaurEval基准测试,用于评估人机协同在编程中的价值。该基准测试通过协作必要问题模板,结合人类推理和AI效率,展示了人机协作在编程任务中的显著优势。

Comments Accepted by ICML 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.21859 2026-05-22 q-bio.PE cs.LG q-bio.QM

PhylaFlow: Hybrid Flow Matching in Billera-Holmes-Vogtmann Tree Space for Phylogenetic Inference

PhylaFlow:在Billera-Holmes-Vogtmann树空间中进行混合流匹配用于系统发育推断

Yasha Ektefaie, Leo Cui, Shrey Jain, Marinka Zitnik, Pardis Sabeti

机构 * Eric and Wendy Schmidt Center, Broad Institute of MIT and Harvard(埃里克和wendy Schmidt中心,MIT和哈佛大学Broad研究所) Department of Biomedical Informatics, Harvard Medical School(哈佛医学院生物医学信息学系) Centennial High School(Centennial高中) Infectious Disease and Microbiome Program, Broad Institute of MIT and Harvard(传染病与微生物组计划,MIT和哈佛大学Broad研究所)

AI总结 该研究提出PhylaFlow模型,通过在Billera-Holmes-Vogtmann树空间中学习后验盆地运输,实现混合流匹配,从而提高系统发育推断的效率和准确性。

Comments 9 pages, 3 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.21849 2026-05-22 cs.LG cs.CL

Geometry-Adaptive Explainer for Faithful Dictionary-Based Interpretability under Distribution Shift

基于几何适应的解释器:在分布偏移下字典基础可解释性的忠实性

Sungjun Lim, Heedong Kim, Andrew Lee, Kyungwoo Song

机构 * Yonsei University(延世大学) Harvard University(哈佛大学)

AI总结 本文提出了一种几何适应解释器(GAE),用于在分布偏移下提高基于字典的可解释性。通过重新对齐解释器的字典与偏移活跃子空间,同时保持原始特征结构,GAE在无监督的情况下减少了分布偏移下的忠实性差距。

详情

展开后加载摘要…

URL PDF HTML 收藏
2603.11679 2026-05-22 cs.AI

LLMs can construct powerful representations and streamline sample-efficient supervised learning

LLMs can construct powerful representations and streamline sample-efficient supervised learning

Ilker Demirel, Lawrence Shi, Zeshan Hussain, David Sontag

机构 * MIT(麻省理工学院) Harvard Medical School(哈佛医学院)

AI总结 本文提出了一种基于LLM的代理流程,通过生成全局 rubric 来提升多模态数据的表示能力,并在15个临床任务中显著优于传统方法。

详情

展开后加载摘要…

URL PDF HTML 收藏
2510.08759 2026-05-22 cs.CV cs.RO

Dissecting Embodied Abilities in Multimodal Language Models through Skill-level Evaluation and Diagnosis

通过技能级评估与诊断解构多模态语言模型的具身能力

Yu Qi, Haibo Zhao, Ziyu Guo, Siyuan Ma, Ziyan Chen, Yaokun Han, Renrui Zhang, Zitiantao Lin, Yizhe Zhu, Shiji Xin, Yijian Huang, Boce Hu, Kai Cheng, Peiheng Wang, Jiazheng Liu, Jiayi Zhang, Yizhe Zhu, Wenqing Wang, Yiran Qin, Haojie Huang, Lawson L. S. Wong

机构 * Northeastern University, Boston, MA, USA The Chinese University of Hong Kong, Hong Kong, China Peking University, Beijing, China Westlake University, Hangzhou, China Harvard University, Cambridge, MA, USA Purdue University, West Lafayette, IN, USA University of Oxford, Oxford, United Kingdom

AI总结 本文提出BEAR基准,通过分解具身任务为14个原子技能进行细粒度评估,发现感知能力是推理失败的主要瓶颈,并提出BEAR-Agent多模态对话代理,显著提升具身技能性能。

Comments Accepted to ICML 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2509.06503 2026-05-22 cs.AI q-bio.QM

An AI system to help scientists write expert-level empirical software

一种帮助科学家编写专家级经验软件的AI系统

Eser Aygün, Anastasiya Belyaeva, Gheorghe Comanici, Marc Coram, Hao Cui, Jake Garrison, Renee Johnston Anton Kast, Cory Y. McLean, Peter Norgaard, Zahra Shamsi, David Smalling, James Thompson, Subhashini Venugopalan, Brian P. Williams, Chujun He, Sarah Martinson, Martyna Plomecka, Lai Wei, Yuchen Zhou, Qian-Ze Zhu, Matthew Abraham, Erica Brand, Anna Bulanova, Jeffrey A. Cardille, Chris Co, Scott Ellsworth, Grace Joseph, Malcolm Kane, Ryan Krueger, Johan Kartiwa, Dan Liebling, Jan-Matthis Lueckmann, Paul Raccuglia, Xuefei, Wang, Katherine Chou, James Manyika, Yossi Matias, John C. Platt, Lizzie Dorfman, Shibl Mourad, Michael P. Brenner

机构 * Google DeepMind(谷歌DeepMind) Google Research(谷歌研究) Google Platforms and Devices(谷歌平台与设备) Massachusetts Institute of Technology(麻省理工学院) School of Engineering and Applied Sciences, Harvard University(哈佛大学工程与应用科学学院)

AI总结 本文提出Empirical Research Assistance (ERA)系统,利用大型语言模型和树搜索技术,自动创建高质量的科学软件,以加速计算实验的开发,从而提高科研效率。

Comments 78 pages, 31 figures, 22 tables

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.21455 2026-05-21 cs.LG

Mitigating Label Bias with Interpretable Rubric Embeddings

通过可解释的评分标准嵌入缓解标签偏差

Calvin Isley, Johann D. Gaebler, Sharad Goel

机构 * Harvard Kennedy School(哈佛肯尼迪学校) Harvard University(哈佛大学)

AI总结 本文提出通过可解释的评分标准嵌入来缓解标签偏差问题,通过理论和实验证明该方法在合理条件下能减少标签偏差并提升群体质量评估。

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.21427 2026-05-21 cs.AI cs.DC

PALS: Power-Aware LLM Serving for Mixture-of-Experts Models

PALS: 为混合专家模型的功率感知LLM服务

Can Hankendi, Rana Shahout, Minlan Yu, Ayse K. Coskun

机构 * Boston University(波士顿大学) Harvard University School of Engineering(哈佛大学工程与应用科学学院) Harvard University(哈佛大学)

AI总结 本文提出PALS,一种功率感知的LLM服务运行时,通过将GPU功率上限作为可控制的参数与软件参数如批大小联合优化,提升能效并减少在功率限制下的服务质量违规。

Comments 13 pages, 10 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.21324 2026-05-21 q-bio.NC cs.LG

Stimulus symmetries can confound representational similarity analyses

刺激对称性可能混淆表征相似性分析

Farhad Pashakhanloo, Jacob A. Zavatone-Veth

机构 * Center for Brain Science(脑科学中心) Society of Fellows(fellows 社会) Harvard University(哈佛大学)

AI总结 研究探讨了网络输入对称性如何影响表征相似性矩阵(RSMs)的分析,指出不同配置可能导致不同的RSMs,并展示了随机梯度下降或能量正则化如何生成稀疏漂移代码,从而导致漂移RSMs。

Comments 40 pages

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.20782 2026-05-21 cs.LG

Causal Machine Learning Is Not a Panacea: A Roadmap for Observational Causal Inference in Health

因果机器学习并非万能:健康领域观察性因果推断的路线图

Donna Tjandra, Trenton Chang, Sonali Parbhoo, Rajesh Ranganath, Andre Kurepa Waschka, William Mitchell, Maggie Makar, Shalmali Joshi, Finale Doshi-Velez, Leo Anthony Celi, Jenna Wiens

机构 * Division of Computer Science and Engineering, University of Michigan(密歇根大学计算机科学与工程系) Department of Electrical and Electronic Engineering, Imperial College London(伦敦帝国理工学院电子与电气工程系) Courant Institute of Mathematical Sciences, New York University(纽约大学Courant数学科学研究所) Center for Data Science, New York University(纽约大学数据科学中心) Department of Mathematics & Statistics, Elon University(埃洛伊大学数学与统计学系) Department of Ophthalmology, Cambridge University Hospitals(剑桥大学医院眼科部) Department of Biomedical Informatics, Columbia University(哥伦比亚大学生物医学信息学系) School of Engineering and Applied Science, Harvard University(哈佛大学工程与应用科学学院) Laboratory for Computational Physiology, Institute for Medical Engineering and Science, Massachusetts Institute of Technology(麻省理工学院医学工程与科学研究所计算生理学实验室) Department of Medicine, Beth Israel Deaconess Medical Center(贝斯以色列德aconess医疗中心医学部) Department of Biostatistics, Harvard T.H. Chan School of Public Health(哈佛T.H. Chan公共卫生学院生物统计学系)

AI总结 本文探讨了因果机器学习在观察性数据中的应用,强调了验证有效性假设和合理使用因果机器学习的重要性,提出了加强因果分析严谨性和可解释性的模板。

详情

展开后加载摘要…

URL PDF HTML 收藏
2512.14896 2026-05-21 cs.CL cs.AI

DrugRAG: Enhancing Pharmacy LLM Performance Through A Novel Retrieval-Augmented Generation Pipeline

DrugRAG: 通过一种新颖的检索增强生成流水线提升药学LLM性能

Houman Kazemzadeh, Kiarash Mokhtari Dizaji, Seyed Reza Tavakoli, Farbod Davoodi, MohammadReza KarimiNejad, Parham Abed Azad, Fatemeh Latifi, Ali Sabzi, Armin Khosravi, Siavash Ahmadi, Babak Khalaj, Mohammad Hossein Rohban, Glolamali Aminian, Zohreh Amoozgar, Tahereh Javaheri

机构 * Department of Medicinal Chemistry, Faculty of Pharmacy, Tehran University of Medical Sciences(药学系,泰赫兰医科大学) Department of Computer Sciences, Faculty of Mathematics and Computer Sciences, Amir Kabir University of Technology(计算机科学系,阿米尔·卡比尔技术大学) Department of Mathematical Sciences, Sharif University of Technology(数学科学系,沙菲克技术大学) Department of Computer Sciences, Missouri University of Science and Technology(计算机科学系,密苏里科学与技术大学) Department of Computer Engineering, Sharif University of Technology(计算机工程系,沙菲克技术大学) Department of Faculty of Interdisciplinary Science and Technology, Tarbiat Modares University(跨学科科学与技术学院,塔里亚特莫达res大学) Electronics Research Institute, Sharif University of Technology(电子研究所,沙菲克技术大学) Department of Electrical Engineering, Sharif University of Technology(电气工程系,沙菲克技术大学) The Alan Turing Institute, London, United Kingdom(艾伦·图灵研究所,伦敦,英国) Department of Radiation Oncology, Massachusetts General Hospital & Harvard Medical School(放射肿瘤科,麻省总医院及哈佛医学院) Health Informatics Lab, Metropolitan College, Boston University(健康信息学实验室,波士顿大学)

AI总结 本研究评估了大型语言模型在药学执业资格问答任务中的性能,并开发了一种外部知识整合方法以提高准确性,通过DrugRAG流水线整合结构化药物知识,从而提升药学相关问答任务的LLM性能。

Comments 14 pages, 2 figures, 2 tables. The revised version includes McNemar's paired statistical analysis, Wilson confidence intervals, expanded methodological clarifications, a revised discussion of evidence retrieval, improved reproducibility details, and updated limitations

详情

展开后加载摘要…

URL PDF HTML 收藏
2509.22963 2026-05-21 cs.LG

Reinforcement Learning with Discrete Diffusion Policies for Combinatorial Action Spaces

基于离散扩散策略的强化学习

Haitong Ma, Ofir Nabati, Aviv Rosenberg, Bo Dai, Oran Lang, Craig Boutilier, Na Li, Shie Mannor, Lior Shani, Guy Tenneholtz

机构 * Google Research(谷歌研究) Harvard University(哈佛大学) Google DeepMind(谷歌DeepMind) Nvidia Research(Nvidia研究)

AI总结 本文提出了一种新的框架,用于在复杂的组合动作空间中训练高效的离散扩散模型策略,通过高效的在线训练过程和策略镜像下降方法,实现了稳定的策略改进,并在多个挑战性组合基准上取得了最先进的性能。

Comments 22 pages, 10 figures. Haitong Ma and Ofir Nabati contributed equally to this paper

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.20314 2026-05-21 cs.LG cs.AI

Less Data, Faster Training: repeating smaller datasets speeds up learning via sampling biases

数据更少,训练更快:重复较小的数据集通过采样偏差加速学习

Jingwen Liu, Ezra Edelman, Surbhi Goel, Bingbin Liu

机构 * Columbia University(哥伦比亚大学) University of Pennsylvania(宾夕法尼亚大学) Harvard University(哈佛大学)

AI总结 研究探讨了'小数据与大数据差距'现象,即使用更少样本重复训练比使用更大数据集更节省计算资源,通过层间增长和采样偏差机制实现加速,为优化提供了新的归纳偏差。

Comments ICML 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.20299 2026-05-21 cs.LG cs.AI cs.RO

Mechanisms of Misgeneralization in Physical Sequence Modeling

物理序列建模中泛化错误的机制

Kento Nishi, Raphael Tang, Karun Kumar, Core Francisco Park, Hidenori Tanaka

机构 * Harvard College(哈佛大学) Harvard John A. Paulson School of Engineering and Applied Sciences(哈佛大学约翰·A·保罗森工程与应用科学学院) Comcast AI CBS-NTT Program in Physics of Intelligence, Harvard University(哈佛大学物理智能计划) Physics of Artificial Intelligence Group, NTT Research, Inc., Sunnyvale, CA, USA(人工智能物理研究组,NTT研究公司,美国加利福尼亚州山景城) Microsoft(微软)

AI总结 本文研究了物理序列建模中由于局部误差传播导致的物理泛化错误,提出了一种数据偏差核来预测物理量的质量变化,并提出了基于核的干预策略。

Comments Preprint. kentonishi.com/physical-misgeneralization

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.20218 2026-05-21 physics.soc-ph cs.AI cs.SI

Network-Based Interventions for HIV Prevention via Cascade-Aware Suppression of Transmission

基于网络的HIV预防干预:通过 cascade 意识的传播抑制

Akseli Kangaslahti, Davin Choo, Milind Tambe, Alastair van Heerden, Cheryl Johnson

机构 * Harvard University(哈佛大学) University of Witwatersrand(沃尔特·斯通大学) Wits Health Consortium(沃茨健康联盟) World Health Organization(世界卫生组织)

AI总结 本文提出了一种基于网络的HIV预防干预方法,通过考虑传播链的抑制来减少新的感染传播。核心方法是将问题建模为一个约束优化问题,并提出了一种多项式时间的近似算法CAST,该算法在多项式时间内达到近似比。主要贡献是证明了该算法在真实世界HIV网络上的有效性。

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.06395 2026-05-21 cs.LG cs.AI eess.SP

Consistent Geometric Deep Learning via Hilbert Bundles and Cellular Sheaves

通过希尔伯特丛和细胞sheaf实现一致的几何深度学习

Kartik Tandon, Julian Gould, Tanishq Bhatia, Francesca Dominici, Alejandro Ribeiro, Claudio Battiloro

机构 * University of Pennsylvania(宾夕法尼亚大学) Sakana AI Northeastern University(东北大学) Harvard University(哈佛大学)

AI总结 本文提出了一种新的卷积学习框架,用于在流形上支持的可能无限维信号,通过希尔伯特丛关联的连接拉普拉斯算子作为卷积算子,引入了称为HilbNets的滤波器和神经网络,并通过两阶段采样过程实现,证明了采样诱导的希尔伯特细胞sheaf的sheaf拉普拉斯收敛于底层连接拉普拉斯,从而在无限维丛设置中推广了Belkin和Niyogi的收敛结果,最终在合成和现实任务中验证了该框架。

Comments 51 pages, 3 figures, 5 tables

详情

展开后加载摘要…

URL PDF HTML 收藏