arXivDaily arXiv每日学术速递 周一至周五更新

高校专区

Massachusetts Institute of Technology(麻省理工学院)

共收录 2206
2605.06505 2026-05-26 cs.LG cs.AI cs.CR

PACZero: PAC-Private Fine-Tuning of Language Models via Sign Quantization

PACZero: 通过符号量化的语言模型PAC隐私微调

Murat Bilgehan Ertan, Xiaochen Zhu, Phuong Ha Nguyen, Marten van Dijk, Srinivas Devadas

机构 * CWI Amsterdam(阿姆斯特丹信息与计算科学研究所) MIT Cambridge(麻省理工学院) Vrije Universiteit Amsterdam(阿姆斯特丹自由大学)

AI总结 提出PACZero系列零阶机制,通过符号量化实现零互信息下的PAC隐私微调,在SST-2和SQuAD上取得竞争性结果。

详情

展开后加载摘要…

URL PDF HTML 收藏
2604.18800 2026-05-26 cs.SI cs.GT cs.LG

Optimal Exploration of New Products under Assortment Decisions

基于分类决策的新产品最优探索

Jackie Baek, Atanas Dinev, Thodoris Lykouris

机构 * Stern School of Business, New York University(纽约大学斯特恩商学院) Massachusetts Institute of Technology(麻省理工学院)

AI总结 研究平台在容量约束下通过分类决策在线学习新产品质量,提出最优探索策略以最小化遗憾,并揭示新产品应与顶级现有产品搭配、同时探索数量由潜力决定等结构洞见。

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.08615 2026-05-26 cs.CV

Inspiration Seeds: Learning Non-Literal Visual Combinations for Generative Exploration

灵感种子:学习用于生成式探索的非字面视觉组合

Kfir Goldberg, Elad Richardson, Yael Vinker

机构 * MIT(麻省理工学院)

AI总结 提出Inspiration Seeds框架,通过CLIP稀疏自编码器提取编辑方向并隔离概念对,实现无需文本提示的两张输入图像的视觉组合生成,支持早期创意阶段的探索性构思。

Comments Project page available at https://kfirgoldberg.github.io/InspirationSeeds/

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.09048 2026-05-26 cs.LG

Guaranteeing Conservation of Integrals with Projection in Physics-Informed Neural Networks

在物理信息神经网络中通过投影保证积分守恒

Anthony Baez, Wang Zhang, Ziwen Ma, Lam Nguyen, Subhro Das, Luca Daniel

机构 * MIT(麻省理工学院) IBM(国际商业机器公司)

AI总结 提出一种投影方法,通过求解约束非线性优化问题,在物理信息神经网络中分别或联合保证线性和二次积分量的守恒,将守恒误差降低三到四个数量级。

详情

展开后加载摘要…

URL PDF HTML 收藏
2509.25339 2026-05-26 cs.CV cs.AI cs.LG eess.IV

VisualOverload: Probing Visual Understanding of VLMs in Really Dense Scenes

VisualOverload: 在真正密集场景中探测VLM的视觉理解

Paul Gavrikov, Wei Lin, M. Jehanzeb Mirza, Soumya Jahagirdar, Muhammad Huzaifa, Sivan Doveh, Serena Yeung-Levy, James Glass, Hilde Kuehne

机构 * Independent Researcher(独立研究者) JKU Linz(林茨JKU) MIT CSAIL Tübingen AI Center(图宾根人工智能中心) Stanford(斯坦福) MIT-IBM Watson AI Lab(MIT-IBM沃森人工智能实验室)

AI总结 提出VisualOverload基准,通过密集场景中的简单视觉任务测试VLM,发现最佳模型仅达69.5%准确率,揭示计数、OCR和逻辑一致性等关键缺陷。

Comments Accepted at CVPR 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.24057 2026-05-26 cs.LG cs.AI

Feature Lottery? A Bifurcation Theory of Concept Emergence

特征彩票?概念涌现的分岔理论

Fuming Yang

机构 * MIT(麻省理工学院)

AI总结 提出一种基于分岔理论的方法,通过损失Hessian驱动的超临界叉形分岔检测表示动力学中的结构涌现,并引入无标签相位坐标β/β_c,在多种设置下验证了四个不同的转变阶段,揭示了特征可解释性的早期可预测性。

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.24045 2026-05-26 cs.LG cs.AI

A Large-Scale Dataset and Benchmark: Do Protein-Ligand Models Learn Binding Sites or Just Binding Likelihood?

大规模数据集与基准:蛋白质-配体模型学习的是结合位点还是仅仅结合可能性?

Zhaohan Meng, Zhen Bai, Ke Yuan, Iadh Ounis, Zaiqiao Meng, Hao Xu, Joseph Loscalzo

机构 * School of Computing Science(计算科学学院) School of Cancer Sciences(癌症科学学院) School of Life Science and Technology(生命科学与技术学院) Institute of Science Tokyo(东京科学研究院) Cancer Research UK Scotland Institute(英国癌症研究会苏格兰研究所) Language Technology Lab(语言技术实验室) Department of Medicine, Brigham and Women’s Hospital, Harvard Medical School(哈佛医学院内科部,布里格斯妇女医院) The Broad Institute of MIT and Harvard(MIT和哈佛大学Broad研究所)

AI总结 针对现有基准无法评估模型是否定位结合位点的问题,提出包含约10万对蛋白质-配体的InteractBind数据集和细粒度基准,通过结合位点定位任务揭示模型在强二元预测下定位能力有限。

Comments Under Review for the NeurIPS 2026 Conference, Track on Evaluations and Datasets

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.24002 2026-05-26 physics.chem-ph cond-mat.mtrl-sci cs.AI physics.comp-ph

Harnessing AtomisticSkills for Agentic Atomistic Research

利用原子技能实现代理原子研究

Bowen Deng, Bohan Li, Matthew Cox, Hoje Chun, Juno Nam, Artur Lyssenko, Sathya Edamadaka, Jurgis Ruza, Xiaochen Du, Nofit Segal, Jesus Diaz Sanchez, Mingrou Xie, Ty Perez, Yu Yao, Miguel Steiner, Sauradeep Majumdar, Charles B. Musgrave, Anirban Chandra, Abhirup Patra, Detlef Hohl, Connor W. Coley, Ju Li, Rafael Gómez-Bombarelli

机构 * Department of Materials Science Engineering, Massachusetts Institute of Technology, Cambridge, MA 02139, USA Department of Chemical Engineering, Massachusetts Institute of Technology, Cambridge, MA 02139, USA Department of Chemistry, Kookmin University, Seoul 02707, Republic of Korea Harvard University, Department of Chemistry Department of Chemistry, Massachusetts Institute of Technology, Cambridge, MA 02139, USA Department of Nuclear Science Shell Information Technology International Inc., Texas 77082, United States Shell International Exploration \& Production Inc., Texas 77079, United States

AI总结 提出AtomisticSkills框架,通过分层分解科学工作流为技能和工具,使通用AI编码代理能够进行原子级研究,并在多个科学任务中验证其能力。

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.23917 2026-05-26 cs.CL

Multi-Persona Debate System for Automated Scientific Hypothesis Generation

用于自动科学假设生成的多角色辩论系统

Jaeha Oh, Byungchan Kim, Ju Li, Yang Jeong Park, Jin-Sung Park

机构 * Department of Materials Science & Engineering, Ajou University(材料科学与工程系,阿乔大学) Department of Energy Systems Research, Ajou University(能源系统研究系,阿乔大学) Department of Nuclear Science and Engineering, Massachusetts Institute of Technology(核科学与工程系,麻省理工学院) Department of Materials Science and Engineering, Massachusetts Institute of Technology(材料科学与工程系,麻省理工学院) Department of Materials Science and Engineering, Ulsan National Institute of Science and Technology(材料科学与工程系,乌山国家科学与技术研究院) Graduate School of Artificial Intelligence, Ulsan National Institute of Science and Technology(人工智能研究生院,乌山国家科学与技术研究院)

AI总结 提出多角色辩论系统(MPDS),结合文献检索、长上下文大语言模型推理、语料驱动角色归纳和结构化多智能体辩论,自动生成科学假设,在电池材料研究中验证其有效性。

Comments 31 pages with 7 main figures, 4 supplementary figures and 1 supplementary table

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.23916 2026-05-26 cs.IR cs.AI econ.GN q-fin.EC

Agent-Facing Information Design in LLM Tool Registries

面向智能体的LLM工具注册表信息设计

Haochuan Kevin Wang

机构 * Massachusetts Institute of Technology(麻省理工学院)

AI总结 本研究首次系统性地分析了LLM工具注册表中广告式描述对智能体选择的影响,发现法律上允许的夸大宣传(如主观最高级表述)完全主导优化效果,而虚假声明无额外影响,并提出了分离选择导向与营销导向描述及智能体注意力质量分数等注册表设计建议。

详情

展开后加载摘要…

URL PDF HTML 收藏
2509.08150 2026-05-26 cs.CL

Verbalized Algorithms: Classical Algorithms are All You Need (Mostly)

言语化算法:经典算法就是你所需要的(大部分)

Supriya Lall, Christian Farrell, Hari Pathanjaly, Marko Pavic, Sarvesh Chezhian, Masataro Asai

机构 * MIT CSAIL(MIT 计算与人工智能实验室) MIT-IBM Watson AI Lab(MIT-IBM 沃森人工智能实验室) IBM Infrastructure(IBM 基础设施) Marist University(马里斯特大学) UC Irvine(加州大学尔湾分校) IBM Research Cambridge, USA(IBM 英国剑桥研究中心)

AI总结 提出言语化算法(VA)范式,将LLM作为可靠的基本操作(如字符串比较)集成到经典算法中,以提升推理的准确性和效率。

Comments Accepted in NeurIPS 2025 Workshop on Efficient Reasoning; Submitted to Position Paper Track at Neurips 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2604.11759 2026-05-25 cs.AI

Retrieval Is Not Enough: Why Organizational AI Needs Epistemic Infrastructure

检索是不够的:为什么组织AI需要认知基础设施

Federico Bottino, Carlo Ferrero, Nicholas Dosio, Pierfrancesco Beneventano

机构 * Kakashi Ventures Accelerator (KVA)(Kakashi Ventures加速器) Massachusetts Institute of Technology(麻省理工学院)

AI总结 本文提出OIDA框架,通过类型化知识对象、知识重力引擎和问题原语,解决组织AI中检索系统缺乏认知结构的问题,并引入认知质量评分(EQS)进行评估。

Comments 10 pages, 2 figures, 8 tables, 6 appendices

详情

展开后加载摘要…

URL PDF HTML 收藏
2603.23565 2026-05-25 cs.LG cs.AI

Safe Reinforcement Learning with Preference-based Constraint Inference

基于偏好的约束推断的安全强化学习

Chenglin Li, Grant Ruan, Hua Geng

机构 * Department of Automation, Tsinghua University, Beijing, China Laboratory for Information \& Decision Systems, Massachusetts Institute of Technology, Cambridge, MA, USA

AI总结 提出偏好约束强化学习(PbCRL),通过引入死区机制和信噪比损失,从人类偏好中推断安全约束,实现更好的约束对齐和策略学习。

Comments Accepted by the 43rd International Conference on Machine Learning (ICML 2026)

详情

展开后加载摘要…

URL PDF HTML 收藏
2501.08222 2026-05-25 cs.RO

Data-driven Spatial Classification using Multi-Arm Bandits for Monitoring with Energy-Constrained Mobile Robots

基于多臂老虎机的数据驱动空间分类用于能量受限移动机器人监测

Xiaoshan Lin, Siddharth Nayak, Stefano Di Cairano, Abraham P. Vinod

机构 * Aerospace Engineering and Mechanics department, University of Minnesota(明尼苏达大学航空航天工程与力学系) Aeronautics and Astronautics department, Massachusetts Institute of Technology(麻省理工学院航空与航天系) Mitsubishi Electric Research Laboratories (MERL)(三菱电机研究实验室)

AI总结 提出一种结合多臂老虎机高维规划与整数规划低维路径规划的层级方法,解决移动机器人团队在能量约束下对监测区域进行快速空间分类的问题。

Comments 8 pages, 6 figures. See https://www.youtube.com/watch?v=gzulpOcVYzg for an overview of the approach along with videos of the hardware experiments

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.23551 2026-05-25 cs.LG cs.AI

Goal-Conditioned Agents that Learn Everything All at Once

目标条件智能体一次性学习所有内容

Michael Matthews, Matthew Jackson, Michael Beukman, Thomas Foster, Alistair Letcher, Scott Fujimoto, Cédric Colas, Jakob Foerster

机构 * University of Oxford(牛津大学) McGill University(麦吉尔大学) MIT(麻省理工学院) Inria(法国国家信息与自动化研究所)

AI总结 提出LEO方法,通过联合输出所有目标的值和动作,实现高效并行全目标学习,在Craftax上显著优于其他方法,在连续控制任务上与基线相当,速度提升超过250倍。

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.23275 2026-05-25 cs.LG

Diffusion Domain Expansion: Learning to Coordinate Pre-trained Diffusion Models

扩散域扩展:学习协调预训练扩散模型

Egor Lifar, Semyon Savkin, Timur Garipov, Shangyuan Tong, Tommi Jaakkola

机构 * MIT CSAIL(麻省理工学院计算机科学与人工智能实验室)

AI总结 提出扩散域扩展(DDE)方法,通过紧凑可训练网络协调预训练扩散模型的去噪输出,以生成更大对象并处理更复杂条件,在长音频生成和条件图像生成中优于其他方法。

Comments Accepted as poster at ICML 2024 Workshop on Structured Probabilistic Inference and Generative Modeling (SPIGM)

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.23193 2026-05-25 cs.HC cs.CL cs.CY cs.MA

CultivAgents: Cultivating Relationship-Centered Multi-Agent Systems for Personalized Gardening

CultivAgents:培育以关系为中心的多智能体系统以实现个性化园艺

Yiyang Wang, Moeiini Reilly, Britney Johnson, Kefei Yan, Alex Cabral, Josiah Hester

机构 * Georgia Institute of Technology(佐治亚理工学院) Massachusetts Institute of Technology(麻省理工学院)

AI总结 提出 CultivAgents,一个基于关怀伦理、协调经验、环境和民族植物学三个专业智能体的多智能体系统,为社区园丁提供个性化、社会文化背景化的园艺支持,并通过混合方法研究验证其有效性。

Comments Preprint, 9 pages. Website: https://hello-diana.github.io/CultivAgents/

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.23098 2026-05-25 cs.RO

UfM*: Uncertainty from Motion* for DNN Depth Estimation Using Gaussians

UfM*:基于高斯分布的运动不确定性用于DNN深度估计

Soumya Sudhakar, Sertac Karaman, Vivienne Sze

机构 * Massachusetts Institute of Technology(麻省理工学院)

AI总结 提出UfM*算法,利用紧凑高斯混合模型高效计算多视图不一致性,实现单次推理的不确定性估计,在资源受限机器人上达到实时性能。

Comments 18 pages, 15 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.20087 2026-05-25 cs.CL cs.AI

ThoughtTrace: Understanding User Thoughts in Real-World LLM Interactions

ThoughtTrace: 理解真实世界LLM交互中的用户想法

Chuanyang Jin, Binze Li, Haopeng Xie, Cathy Mengying Fang, Tianjian Li, Shayne Longpre, Hongxiang Gu, Maximillian Chen, Tianmin Shu

机构 * Johns Hopkins University(约翰霍普金斯大学) Massachusetts Institute of Technology(麻省理工学院) Google Research(谷歌研究)

AI总结 本文提出ThoughtTrace,首个大规模数据集,通过配对真实多轮人机对话与用户自述想法(提示原因和对助手回复的反应),揭示用户认知动态,并展示其在行为预测和个性化对齐中的价值。

Comments 53 pages, 23 figures, 4 tables. Project website: https://thoughttrace-project.github.io/

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.10219 2026-05-25 math.OC cs.CC cs.LG

Parameterized Complexity of Stationarity Testing for Piecewise-Affine Functions and Shallow CNN Losses

分段仿射函数与浅层CNN损失的平稳性检验的参数化复杂性

Yuhan Ye

机构 * MIT(麻省理工学院)

AI总结 研究连续分段仿射函数近似一阶平稳性检验的参数化复杂性,以维度为参数给出XP算法和W[1]-硬度,并推广到浅层ReLU CNN训练损失。

Comments 32 pages, 1 figure, 1 table

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
2603.18551 2026-05-25 math.OC cs.CC cs.LG

Learning Decision-Sufficient Representations for Linear Optimization

学习线性优化的决策充分表示

Yuhan Ye, Saurabh Amin, Asuman Ozdaglar

机构 * MIT(麻省理工学院)

AI总结 针对未知成本向量位于先验集中的线性规划,提出点态充分性概念并设计多项式时间切割平面算法构建压缩数据集,实现与决策相关维度d*相关的PAC保证和泛化界。

Comments 45 pages plus appendix, 2 figures. Accepted at COLT 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.18176 2026-05-25 cs.CL

Improving Sampling for Masked Diffusion Models via Information Gain

通过信息增益改进掩码扩散模型的采样

Kaisen Yang, Jayden Teoh, Kaicheng Yang, Yitong Zhang, Alex Lamb

机构 * Department of Computer Science and Technology, Tsinghua University(清华大学计算机科学与技术系) Massachusetts Institute of Technology(麻省理工学院) Shanghai Jiao Tong University(上海交通大学) Beihang University(北航) College of AI, Tsinghua University(清华大学人工智能学院)

AI总结 提出一种无需训练的Info-Gain采样器,利用掩码扩散模型的双向结构平衡即时不确定性与剩余掩码位置的信息增益,在推理、编码、创意写作和图像生成任务中优于现有采样器。

Comments https://github.com/yks23/Information-Gain-Sampler Accepted by ICML2026 Accepted by ICML2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2601.07545 2026-05-25 cs.LG stat.ML

Near-Optimal Private Linear Regression via Iterative Hessian Mixing

通过迭代Hessian混合实现近最优私有线性回归

Omri Lev, Moshe Shenfeld, Vishwak Srinivasan, Katrina Ligett, Ashia C. Wilson

机构 * Department of EECS, Massachusetts Institute of Technology, US(麻省理工学院电子工程与计算机科学系) School of Computer Science and Engineering, The Hebrew University of Jerusalem, IL(耶路撒冷希伯来大学计算机科学与工程学院)

AI总结 提出迭代Hessian混合(IHM)算法,基于高斯草图方法实现差分隐私线性回归,在理论上消除了AdaSSP方法中的维度相关乘法因子,并在实证中持续优于现有基线。

详情

展开后加载摘要…

URL PDF HTML 收藏
2510.12787 2026-05-25 cs.AI cs.MA

Ax-Prover: A Deep Reasoning Agentic Framework for Theorem Proving in Mathematics and Quantum Physics

Ax-Prover:用于数学和量子物理定理证明的深度推理智能体框架

Benjamin Breen, Marco Del Tredici, Jacob McCarran, Javier Aspuru Mijares, Weichen Winston Yin, Kfir Sulimany, Jacob M. Taylor, Frank H. L. Koppens, Dirk Englund

机构 * Axiomatic_AI(公理人工智能) Massachusetts Institute of Technology (MIT)(麻省理工学院) Institut de Ciències Fotòniques (ICFO)(光子科学研究所) Institució Catalana de Recerca i Estudis Avançats (ICREA)(加泰罗尼亚高级研究与高等学院)

AI总结 提出多智能体系统Ax-Prover,通过将大语言模型与Lean工具结合,实现跨科学领域的自动化定理证明,并在抽象代数和量子理论新基准上显著超越现有方法。

详情

展开后加载摘要…

URL PDF HTML 收藏
2510.04406 2026-05-25 stat.ML cs.LG

Decomposition-Based Modular Conformal Prediction for Two-Stage Modeling

基于分解的模块化共形预测用于两阶段建模

William Zhang, Saurabh Amin, Georgia Perakis

机构 * Operations Research Center, Massachusetts Institute of Technology, Cambridge, MA, USA(麻省理工学院运筹学研究中心) Laboratory for Information and Decision Systems, Massachusetts Institute of Technology, Cambridge, MA, US(麻省理工学院信息与决策系统实验室) Sloan School of Management, Massachusetts Institute of Technology, Cambridge, MA, USA(麻省理工学院斯隆管理学院)

AI总结 提出一种针对两阶段顺序模型的共形预测框架,通过将预测残差分解为阶段特定成分,利用族系错误率控制校准缩放参数,实现不确定性归因和鲁棒覆盖。

Comments 11 pages, (37 with appendix), 15 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
2509.26383 2026-05-25 cs.CL cs.AI

Efficient and Transferable Agentic Knowledge Graph RAG via Reinforcement Learning

基于强化学习的高效可迁移智能知识图谱检索增强生成

Junhong Lin, Shicheng Liu, Jinyeop Song, Song Wang, Julian Shun, Yada Zhu

机构 * MIT CSAIL(麻省理工学院CSAIL) University of Virginia(弗吉尼亚大学) IBM Research(IBM研究院)

AI总结 提出KG-R1框架,通过强化学习将单个智能体与知识图谱交互,统一检索、推理和生成过程,在KGQA基准上以更少生成token提升准确率,并展现跨图谱的零样本迁移能力。

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.22884 2026-05-25 cs.LG cs.AI

Tensor Cache: Eviction-conditioned Associative Memory for Transformers

Tensor Cache: 基于驱逐条件的Transformer联想记忆

Kabir Swain, Sijie Han, Daniel Karl I. Weidele, Mauro Martino, Antonio Torralba

机构 * Massachusetts Institute of Technology, Cambridge, MA, USA(麻省理工学院) IBM Research, Cambridge, MA, USA(IBM研究院) University of Toronto, Toronto, Canada(多伦多大学)

AI总结 提出Tensor Cache,一种双层缓存结构,将滑动窗口注意力作为L1缓存,将窗口驱逐的KV对压缩为外积快速权重记忆作为L2缓存,通过门控融合和端到端训练,在有限状态基线中提升记忆-质量边界。

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.20102 2026-05-25 cs.LG cs.AI

BarrierSteer: LLM Safety via Learning Barrier Steering

BarrierSteer: 通过学习障碍引导实现大语言模型安全

Thanh Q. Tran, Arun Verma, Kiwan Wong, Bryan Kian Hsiang Low, Daniela Rus, Wei Xiao

机构 * Department of Computer Science, National University of Singapore(新加坡国立大学计算机科学系) Singapore-MIT Alliance for Research and Technology Centre(新加坡-麻省理工联合研究中心) CSAIL, Massachusetts Institute of Technology(麻省理工学院计算机科学与人工智能实验室) Worcester Polytechnic Institute(沃斯堡理工学院)

AI总结 提出 BarrierSteer 框架,通过将学习到的非线性安全约束作为控制障碍函数嵌入潜在表示空间,在推理时引导模型生成安全内容,理论保证且不修改模型参数。

Comments This paper introduces SafeBarrier, a framework that enforces safety in large language models by steering their latent representations with control barrier functions during inference, reducing adversarial and unsafe outputs

详情

展开后加载摘要…

URL PDF HTML 收藏
2601.14652 2026-05-25 cs.AI cs.CL cs.MA

MAS-Orchestra: Understanding and Improving Multi-Agent Reasoning Through Holistic Orchestration and Controlled Benchmarks

MAS-Orchestra:通过整体编排和受控基准理解与改进多智能体推理

Zixuan Ke, Yifei Ming, Austin Xu, Ryan Chin, Xuan-Phi Nguyen, Prathyusha Jwalapuram, Jiayu Wang, Semih Yavuz, Caiming Xiong, Shafiq Joty

机构 * Salesforce Research(Salesforce研究院) University of Wisconsin-Madison(威斯康星大学麦迪逊分校) Massachusetts Institute of Technology(麻省理工学院)

AI总结 提出MAS-Orchestra框架,将多智能体系统编排形式化为函数调用的强化学习问题,并引入受控基准MASBENCH,揭示MAS收益依赖于任务结构等因素,在数学推理等任务上取得一致改进且效率提升10倍以上。

Comments ICML 2026

详情

展开后加载摘要…

URL PDF HTML 收藏