arXivDaily arXiv每日学术速递 周一至周五更新

高校专区

Princeton University(普林斯顿大学)

共收录 78
2607.02050 2026-07-07 cs.LG cs.SY eess.SY 新提交

A Memory Efficient Unified Algorithm for Online Learning of Linear Dynamical Systems

一种内存高效的线性动力系统在线学习统一算法

Yuval Ran-Milo, Angelos Assos, Elad Hazan

机构 * Tel Aviv University(特拉维夫大学) Princeton University(普林斯顿大学)

AI总结 针对低不稳定性复杂度的线性动力系统,提出一种在线预测算法,参数复杂度为Õ(k),实现亚线性遗憾,内存适应内在复杂度而非全状态维度。

Comments 34 pages, 1 figure

详情

展开后加载摘要…

URL PDF HTML 收藏
2607.02329 2026-07-03 cs.AI cond-mat.mtrl-sci physics.comp-ph 新提交

Grounded autonomous research: a fault-tolerant LLM pipeline from corpus to manuscript in frontier computational physics

基于语料库的前沿计算物理容错LLM流水线:从语料到论文的自主研究

Haonan Huang

机构 * Princeton University(普林斯顿大学)

AI总结 提出一个端到端LLM流水线,从11,083篇arXiv论文语料库出发,自主完成前沿计算物理研究,包括构思方向、复现文献、第一性原理计算和撰写论文,通过冗余设计实现容错。

Comments 39 pages, 5 figures. Accepted at the ICML 2026 AI for Science Workshop (https://openreview.net/forum?id=R5YXaPgUAx). Includes the pipeline-generated companion physics manuscript as an appendix. Data and scaffolding archive: https://doi.org/10.5281/zenodo.21126996

详情

展开后加载摘要…

URL PDF HTML 收藏
2607.00918 2026-07-02 cs.CL cs.AI cs.MA 新提交

From Personas to Plot: Character-Grounded Multi-Agent Story Generation for Long-Form Narratives

从角色到情节:基于角色的多智能体长篇小说生成

Aayush Aluru, Chloe Ho, Muhammad Hammouri, Kerry Luo, Myra Malik, Ryan Lagasse, Arjun Bahuguna, Vasu Sharma

机构 * Pocket FM Princeton University(普林斯顿大学) University of Michigan(密歇根大学) University of Maryland(马里兰大学) Universitat Pompeu Fabra(庞培法布拉大学)

AI总结 提出MAGNET多智能体叙事引擎和ATLAS幻觉检测管线,通过角色代理和世界状态追踪,在100页故事中减少41%注释和50%幻觉,实现连贯长篇小说生成。

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.32012 2026-07-01 cs.LG cs.CV 新提交

CoMet: Context and Multiplicity Decomposition for Multimodal Uncertainty Estimation

CoMet:多模态不确定性估计的上下文与多样性分解

Sanghyuk Chun, William Yang, Amaya Dharmasiri, Olga Russakovsky

机构 * Princeton University(普林斯顿大学)

AI总结 提出CoMet方法,将多模态大模型的不确定性分解为上下文项和多样性项,通过轻量级后验模块高效估计,无需自回归生成或重复采样,在多个基准上优于现有方法。

Comments 33 pages, 13.3MB

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.31121 2026-07-01 cs.AI 新提交

The Past Is Prologue: A Plug-in Controller for Selective Updates in Sequentially Evolving LLM Memory

过去即序幕:面向顺序演化的大语言模型记忆的选择性更新插件控制器

Zihan Chen, Songwei Dong, Chengshuai Shi, Peng Wang, Song Wang, Cong Shen, Jundong Li

机构 * University of Virginia(弗吉尼亚大学) Princeton University(普林斯顿大学) University of Central Florida(中佛罗里达大学)

AI总结 提出Janus插件控制器,通过记忆动量触发器检测异常并基于紧凑混合评估集决定是否接受记忆更新,避免有害覆盖,在六个数据集上提升准确率2.7-4.6点。

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.30774 2026-07-01 cs.AI 新提交

What Drives Interactive Improvement from Feedback?

什么驱动了来自反馈的交互式改进?

Bartłomiej Cupiał, Jan Łojek, Mikołaj Garstecki, Szymon Pobłocki, Alicja Ziarko, Piotr Miłoś

机构 * University of Warsaw(华沙大学) AKCES NCBR Princeton University(普林斯顿大学) Mistral AI Institute of Mathematics, Polish Academy of Sciences(波兰科学院数学研究所)

AI总结 通过受控的学生-教师协议,分离反馈、重试和格式修正对多轮语言智能体性能提升的影响,发现有用反馈必须提供超越通用重试的指导,且学生利用反馈的能力是交互改进的关键瓶颈。

Comments 9 pages, 7 figures, accepted to the RLxF Workshop at ICML 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.27103 2026-06-26 cs.CL 新提交

The Riddle Riddle: Testing Flexible Reasoning in Large Language Models and Humans

谜语谜题:测试大型语言模型和人类的灵活推理

Bella Fascendini, Kathryn McGregor, Max D. Gupta, Thomas L. Griffiths

机构 * Department of Psychology, Princeton University(普林斯顿大学心理学系) Department of Computer Science, Princeton University(普林斯顿大学计算机科学系)

AI总结 通过谜语谜题范式,发现LLM在真实谜语上准确率高(84.9%),但在谜语谜题上表现差(50.7%),而人类相反;错误分析表明LLM更倾向于过度使用创造性推理。

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.26158 2026-06-26 cs.AI 新提交

Life After Benchmark Saturation: A Case Study of CORE-Bench

基准测试饱和之后:CORE-Bench 案例研究

Nitya Nadgir, Sayash Kapoor, Kangheng Liu, Peter Kirgis, Matilda Orona, Stephan Rabanser, Tilman Bayer, Abhishek Shetty, Yue Ling, Derrick Chan-Sew, Rumi Nakagawa, Saiteja Utpala, Zachary S. Siegel, Arvind Narayanan

机构 * Independent(独立机构) Princeton University(普林斯顿大学) UC Berkeley(加州大学伯克利分校) MIT(麻省理工学院)

AI总结 针对基准测试准确率饱和问题,提出从构造效度、泛化性、效率、可靠性、模型与脚手架相对重要性及人机协作提升六个维度评估智能体,以CORE-Bench Hard为案例展示饱和后仍可获得有意义见解。

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.26498 2026-06-26 math.OC cs.LG 新提交

Mean-Field PhiBE: Continuous-Time Mean-Field Reinforcement Learning from Discrete-Time Data

平均场 PhiBE:基于离散时间数据的连续时间平均场强化学习

Erhan Bayraktar, Martin Hernandez, Qinxin Yan, Yuhua Zhu

机构 * Department of Mathematics, University of Michigan, Ann Arbor, MI, USA(密歇根大学数学系,安阿伯,密歇根州,美国) Department of Statistics and Data Science, University of California, Los Angeles, CA, USA(加州大学洛杉矶分校统计与数据科学系,加利福尼亚州,美国) Program in Applied and Computational Mathematics, Princeton University, Princeton, NJ, USA(普林斯顿大学应用与计算数学项目,普林斯顿,新泽西州,美国)

AI总结 针对仅离散时间数据可用但种群连续演化的模型无关连续时间平均场控制问题,提出平均场PhiBE方法,将离散时间信息融入Wasserstein空间上的连续时间PDE,并推导策略梯度定理,实现模型无关的演员-评论家算法,证明一阶一致性估计。

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.25450 2026-06-26 cs.LG cs.CL 新提交

The Generalization Spectrum: A Chromatographic Approach to Evaluating Learning Algorithms

泛化谱:一种评估学习算法的色谱方法

Jinghan Zhang, Zerui Cheng, Shiqi Chen, Ge Zhang, Wenhao Huang, Jiashuo Liu, Junxian He, Tianle Cai

机构 * ByteDance Seed(字节跳动Seed) Hong Kong University of Science and Technology(香港科技大学) Princeton University(普林斯顿大学) University of Oxford(牛津大学)

AI总结 提出泛化谱框架,通过构建从精确回忆到跨语言实现转移等不同转移距离的测试变体,揭示算法从单个样本泛化到其他样本的程度,并应用于竞争编程任务评估不同学习范式。

Comments Accepted at ICML 2026. 30 pages, 6 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.25206 2026-06-25 cs.RO cs.AI cs.CL 新提交

RAVEN: Long-Horizon Reasoning & Navigation with a Visuo-Spatio-Temporal Memory

RAVEN: 基于视觉-空间-时间记忆的长期推理与导航

Yixun Hu, Zhicheng Zheng, Lihan Zha, Chunwei Xing, Rajdeep Singh, Omar Hossain, Antonio Loquercio, Dhruv Shah

机构 * Princeton University(普林斯顿大学) University of Pennsylvania(宾夕法尼亚大学)

AI总结 提出RAVEN记忆系统,通过存储视觉嵌入与位姿时间信息,实现长期机器人问答与导航,在多个基准上超越字幕记忆系统,以10倍更低检索成本匹配前沿VLM。

Comments Project website: https://ravenmem.github.io/

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.25178 2026-06-25 cs.AI 新提交

Transferability for General Reasoning: An Automated Curriculum for Multi-Domain RLVR

通用推理的可迁移性:多领域RLVR的自动课程

Yongjin Yang, Jiarui Liu, Yinghui He, Lechen Zhang, Bernhard Schölkopf, Zhijing Jin

机构 * Jinesis Lab, University of Toronto & Vector Institute(Jinesis实验室,多伦多大学与向量研究所) Carnegie Mellon University(卡内基梅隆大学) Princeton University(普林斯顿大学) University of Illinois Urbana-Champaign(伊利诺伊大学厄巴纳-香槟分校) ELLIS Institute Tübingen(ELLIS研究所图宾根) Max Planck Institute for Intelligent Systems(马克斯·普朗克智能系统研究所) EuroSafeAI

AI总结 提出迁移感知课程(TAC),利用优势信号和投影梯度估计跨领域可迁移性,在多领域推理套件上实现最佳平均准确率。

Comments 32 pages, including supplementary material; code available at https://github.com/YangYongJin/transfer-aware-curriculum

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.24884 2026-06-24 cs.RO cs.AI cs.LG 新提交

InSight: Self-Guided Skill Acquisition via Steerable VLAs

InSight: 通过可引导的VLA实现自主技能获取

Maggie Wang, Lars Osterberg, Stephen Tian, Ola Shorinwa, Jiajun Wu, Mac Schwager

机构 * Stanford University(斯坦福大学) Princeton University(普林斯顿大学)

AI总结 提出InSight框架,通过将VLA模型在基本动作层面变得可引导,实现自主技能获取,包括自动分割演示为基本动作和VLM引导的数据飞轮,无需人类演示即可学习新技能。

Comments Project website: https://insight-vla.github.io

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.23085 2026-06-23 cs.RO 新提交

Foresight: Failure Detection for Long-Horizon Robotic Manipulation with Action-Conditioned World Model Latents

Foresight: 基于动作条件世界模型潜在变量的长时程机器人操作故障检测

Haoran Zhang, Yifu Lu, Boyang Wang, Xuhui Kang, Yen-Ling Kuo, Zezhou Cheng, Mengdi Wang, Odest Chadwicke Jenkins

机构 * University of Michigan(密歇根大学) Princeton University(普林斯顿大学) University of Virginia(弗吉尼亚大学)

AI总结 提出Foresight框架,利用动作条件世界模型的潜在表征监测操作轨迹,仅用最终任务级标签训练,结合函数共形预测自适应校准阈值,在仿真和真实机器人长时程任务中实现高效故障检测。

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.22938 2026-06-23 cs.LG cs.AI 新提交

Provable Benefits of RLVR over SFT for Reasoning Models: Learning to Backtrack Efficiently

RLVR 相对于 SFT 在推理模型中的可证明优势:学习高效回溯

Stanley Wei, Juno Kim

机构 * Princeton University(普林斯顿大学) University of California, Berkeley(加州大学伯克利分校)

AI总结 本文从理论上证明,在推理任务中,基于可验证奖励的强化学习(RLVR)相比监督微调(SFT)能学习高效回溯,从而在推理时计算上实现指数级优势。

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.21045 2026-06-23 cs.CR cs.LG 新提交

OVIG: Optimistic Verification of AI Training Integrity via Gradient Signals

OVIG: 通过梯度信号乐观验证AI训练完整性

Hongxu Su, Jianzhu Yao, Huan Zhang, Xuechao Wang, Pramod Viswanath

机构 * Princeton University(普林斯顿大学) University of Illinois Urbana-Champaign(伊利诺伊大学厄巴纳-香槟分校)

AI总结 提出OVIG框架,利用异构重放校准的梯度差异经验边界,通过乐观采样和步长参数审计外包训练完整性,在多种攻击下保持0%攻击成功率,存储和传输开销降低1996倍。

Comments 18 pages, 7 figures, 11 tables. Submitted to IEEE Symposium on Security and Privacy (S&P)

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.22859 2026-06-23 cs.AI astro-ph.IM physics.soc-ph 新提交

AI Scientists as Engines of Discovery: A Case for Development within Reformed Institutions

作为发现引擎的AI科学家:在改革机构内发展的案例

Raul Jimenez, Boris Bolliet, Francisco Villaescusa-Navarro, Rabih Zbib, Benjamin Wandelt, David N. Spergel, Thomas Meier, Jessica Montgomery, Hana Aliee, Licia Verde

机构 * Institute of Cosmos Sciences (ICCUB), University of Barcelona(巴塞罗那大学宇宙科学研究所) ICREA Cavendish Astrophysics, University of Cambridge(剑桥大学卡文迪什天体物理学) Kavli Institute for Cosmology, University of Cambridge(剑桥大学卡维里宇宙学研究所) Center for Computational Astrophysics, Flatiron Institute(弗拉蒂隆研究所计算天体物理中心) Department of Astrophysical Sciences, Princeton University(普林斯顿大学天体物理科学系) Avature Department of Physics and Astronomy, Johns Hopkins University(约翰霍普金斯大学物理与天文学系) Department of Applied Mathematics and Statistics, Johns Hopkins University(约翰霍普金斯大学应用数学与统计学系) Flatiron Institute(弗拉蒂隆研究所) Munich Center for Machine Learning, LMU Munich(慕尼黑大学慕尼黑机器学习中心) Department of Computer Science and Technology, University of Cambridge(剑桥大学计算机科学与技术系) School of Clinical Medicine, University of Cambridge(剑桥大学临床医学院)

AI总结 本文论证多智能体AI系统将从被动工具进化为“AI科学家”,通过原型框架Denario加速发现周期,并提出机构需为验证、问责、可解释性和双重用途安全进行改革。

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.19641 2026-06-23 cs.RO cs.CV 新提交

Scaling Self-Play for End-to-End Driving

扩展端到端驾驶的自我对弈

Luke Rowe, Roger Girgis, Rodrigue de Schaetzen, Daphne Cornelisse, Alaap Grandhi, Felix Heide, Eugene Vinitsky, Christopher Pal, Liam Paull

机构 * Mila(米拉研究所) Université de Montréal(蒙特利尔大学) Polytechnique Montréal(蒙特利尔理工学院) Torc Robotics NYU Tandon School of Engineering(纽约大学坦登工程学院) McMaster University(麦克马斯特大学) Princeton University(普林斯顿大学)

AI总结 提出大规模自我对弈训练策略,通过高效模拟器Gigapixel实现像素级自我对弈,结合DAgger蒸馏和感知适应,提升端到端驾驶模型性能。

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.17289 2026-06-23 cs.AI cs.CL 新提交

Nothing from Something: Can a Language Model Discover 0?

无中生有:语言模型能否发现0?

Phoebe Zeng, Thomas L. Griffiths, Brenden M. Lake

机构 * Department of Computer Science, Princeton University(普林斯顿大学计算机科学系)

AI总结 研究语言模型能否独立发现“零”的概念,通过算术任务测试,发现GPT-2规模模型无法在测试时泛化,但少量示例训练后显著提升,且语言预训练减少所需示例约50%。

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.14150 2026-06-23 cs.LG cs.CL 新提交

Small LLMs: Pruning vs. Training from Scratch

小型LLM:剪枝 vs. 从头训练

Yufeng Xu, Taiming Lu, Kunjun Li, Jiachen Zhu, Mingjie Sun, Zhuang Liu

机构 * Princeton University(普林斯顿大学) New York University(纽约大学) Carnegie Mellon University(卡内基梅隆大学)

AI总结 本文通过六种剪枝方法在Llama-3.1-8B上比较剪枝与从头训练,发现有限预算下剪枝更优,预算充足时粗粒度剪枝可被超越。

Comments Our code is available at https://github.com/zlab-princeton/pruning-vs-scratch

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.19370 2026-06-19 cs.LG cs.AI cs.MA 新提交

Human-like autonomy emerges from self-play and a pinch of human data

类人自主性从自我对弈和少量人类数据中涌现

Daphne Cornelisse, Julian Hunt, Zixu Zhang, Waël Doulazmi, Kevin Joseph, Jaime Fernández Fisac, Eugene Vinitsky

机构 * NYU Tandon School of Engineering(纽约大学坦登工程学院) NYU Courant(纽约大学库朗数学科学研究所) Princeton University(普林斯顿大学) Centre for Robotics, Mines Paris(巴黎矿业大学机器人中心) Valeo(法雷奥)

AI总结 提出一种结合自我对弈强化学习与少量人类演示的正则化方法,仅用30分钟人类数据即可训练出与人类协调的驾驶策略,训练时间仅15小时。

Comments 10 pages

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.19539 2026-06-19 astro-ph.SR cs.AI 新提交

Review of Machine Learning Models for Solar Energetic Particle Prediction

太阳高能粒子预测的机器学习模型综述

Spiridon Kasapis, Pouya Hosseinzadeh, Kathryn Whitman, Ricky Egeland, Manolis Georgoulis, Angelos Vourlidas, Athanasios Papaioannou, Eleni Lavasa, Anastasios Anastasiadis, Giorgos Giannopoulos, Andres Munoz-Jaramillo, Bala Poduval, Irina N. Kitiashvili, Alexander G. Kosovichev, Viacheslav Sadykov, Soukaina Filali Boubrahimi, Tate T. Hutchins, Hameedullah A. Farooki, Manuel E. Cuesta, Leng Y. Khoo, Sungmin Pak, Robert Czarnota, Jamie S. Rankin, Jamey Szalay, Mitchell M. Shen, Georgios Livadiotis, Zigong Xu, David J. McComas, Nikolaos Sarlis, Dionissios Hristopulos, Arik Posner, Alec J. Engell, Mohammed AbuBakr Ali, Ali G. A. Abdelkawy, Abdelrazek M. K. Shaltout, M. M. Beheary, Christina O. Lee, Sigiava Aminalragia-Giamini, Constantinos Papadimitriou, Ingmar Sandberg, Savvas Raptis, Shah Muhammad Hamdi, Monica Laurenza, Mirko Stumpo, Sumanth A. Rotti, India Jackson, Aatiya Ali, Atilim Gunes Baydin, Nathan Schwadron, Subhamoy Chatterjee, Maher A. Dayeh, Gelu M. Nita, Patrick M. O'Keefe, Chun Jie Chong, Paul Kosovich, Russell D. Marroquin, Berkay Aydin, Petrus C. Martens, Lulu Zhao, Yang Chen, Yian Yu, Monica G. Bobra, Ward Manchester, Tamas Gombosi, Ming Zhang, Jesse Torres, Philip K. Chan, Mohamed Nedal, Kamen Kozarev, Peijin Zhang, Kimberly Moreland, Hazel M. Bain, Samuel Hart, Michael J. Starkey, Alan G. Ling, Simone Benella

机构 * Department of Astrophysical Sciences, Princeton University, Princeton, NJ, USA Computational Physics Branch, NASA Ames Research Center, Moffett Field, CA, USA Department of Computer Science, Utah State University, Logan, UT, USA Space Radiation Analysis Group, NASA Johnson Space Center, Houston, TX, USA Johns Hopkins Applied Physics Lab, 11100 Johns Hopkins Rd, Laurel, MD 20723, United States Research Center for Astronomy Applied Mathematics of the Academy of Athens, 4 Soranou Efesiou Street, Athens 11527, Greece Institute for Astronomy, Astrophysics, Space Applications Southwest Research Institute, Boulder, CO, USA Space Science Center, University of New Hampshire, Durham, NH, USA Department of Physics, New Jersey Institute of Technology, Newark, NJ, USA Astronomy Department, Georgia State University, Atlanta, GA, USA Department of Computer Science, Princeton University, Princeton, NJ, USA Department of Mathematics, Rowan University, Glassboro, NJ, USA Astronomy, California Institute of Technology, Pasadena, CA, USA Department of Physics, National Kapodistrian University of Athens, Athens, Greece School of Electrical Computer Engineering, Technical University of Crete, Chania, Greece Department of Astronomy Meteorology, Faculty of Science, Al-Azhar University, Cairo, Egypt Space Sciences Lab, University of California, Berkeley, CA, USA Research Consultancy, Athens, Greece Institute for Space Astrophysics Department of Physics Astronomy, Georgia State University, Atlanta, GA 30303, USA Aryabhatta Research Institute of Observational Sciences (ARIES), Manora Peak, Nainital-263001, Uttarakhand, India Department of Computer Science, Oxford University, Oxford, England Southwest Research Institute, San Antonio, TX, USA Computer Science Department, New Jersey Institute of Technology, Newark, NJ, USA Department of Physics, University of California San Diego, La Jolla, CA 92093, USA Department of Computer Science, Georgia State University, Atlanta, GA 30303, USA Department of Climate Engineering, University of Michigan, Ann Arbor, MI, USA Department of Statistics, University of Michigan, Ann Arbor, MI, USA Department of Electrical Engineering Computer Science, Florida Institute of Technology, Melbourne, FL, USA Astrophysics Section, School of Cosmic Physics, Dublin Institute for Advanced Studies, DIAS Dunsink Observatory, Dublin D15 XR2R, Ireland Institute of Astronomy of the Bulgarian Academy of Sciences, Sofia, Bulgaria Center for Solar-Terrestrial Research, New Jersey Institute of Technology, Newark, NJ 07102, USA Cooperative Programs for the Advancement of Earth System Science, University Corporation for Atmospheric Research, Boulder, CO, USA CIRES, University of Colorado Boulder, Boulder, CO, USA Space Weather Prediction Center, NOAA, Boulder, CO, USA Astronomy, College of Science, The University of Texas at San Antonio, San Antonio, TX, USA Space Weather Prediction Center, National Oceanic The University of Texas at San Antonio, San Antonio, TX, USA Environmental Research, Inc., MA, USA

AI总结 综述了用于太阳高能粒子预测的机器学习模型,包括数据集、架构、输入输出比较,并提出了未来研究建议。

Comments Review Paper, Maine text: 23 pages, References: 5 pages, Appendix: 42 pages

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.18247 2026-06-17 cs.RO cs.AI 新提交

Visual Verification Enables Inference-time Steering and Autonomous Policy Improvement

视觉验证实现推理时引导与自主策略改进

Mingtong Zhang, Dhruv Shah

机构 * Princeton University(普林斯顿大学)

AI总结 提出VERITAS框架,利用预训练通用机器人策略作为生成器,结合无梯度视觉验证器在推理时评估动作,实现无需额外训练的推理时策略引导和离线策略改进。

Comments Website: https://veritas-improvement.github.io

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.17887 2026-06-17 cs.HC cs.AI 新提交

AI Adoption Across a Multinational Workforce: Sociotechnical Conditions for GenAI Acceptance in Human Resources

AI在跨国劳动力中的采纳:人力资源中GenAI接受的社会技术条件

Dalia Ali, Maria José Rodríguez Velázquez, Manoel Horta Ribeiro, Vera Liao, Orestis Papakyriakopoulos

机构 * Technical University of Munich(慕尼黑技术大学) University of Michigan(密歇根大学) Princeton University(普林斯顿大学)

AI总结 研究跨国科技公司从传统HR系统转向GenAI系统过程中,员工采纳受情境适配、搜索素养和信任校准等社会技术条件影响,并提出了包容性部署的设计建议。

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.17657 2026-06-17 cs.AI 新提交

Using Cognitive Models to Improve Language Model Simulation of Human Persuasion Games

使用认知模型改进语言模型对人类说服博弈的模拟

Zirui Cheng, Zeyu Shen, Thomas L. Griffiths, Peter Henderson

机构 * Princeton University(普林斯顿大学)

AI总结 提出方程到行为提示和强化学习方法,使语言模型匹配认知模型(如贝叶斯更新、动机推理),在说服博弈中提升模拟人类决策多样性的能力。

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.17410 2026-06-17 cs.CV 新提交

Attention Alignment Between Humans and Vision-Language Models

人类与视觉语言模型之间的注意力对齐

Isaac R. Christian, Udith Haputhanthrige, Hanna Hornfeld, Declan Campbell, Samuel Nastase, Taylor Webb, Michael Graziano

机构 * Princeton Neuroscience Institute, Princeton University(普林斯顿大学普林斯顿神经科学研究所) Department of Psychology, Princeton University(普林斯顿大学心理学系) Department of Computer Science, Princeton University(普林斯顿大学计算机科学系) Department of Psychology and Center for Computational Language Sciences, University of Southern California(南加州大学心理学系与计算语言科学中心) Department of Psychology, Université de Montréal(蒙特利尔大学心理学系)

AI总结 本研究比较了六种视觉语言模型的空间注意力图与人类注视热图,发现解码器架构(LSTM vs Transformer)主导对齐程度,LSTM解码器对齐度更高但空间分散且任务区分度低,而Transformer解码器注意力更集中且任务区分度强。

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.17053 2026-06-16 cs.CL cs.CV 新提交

Context-Aware RL for Agentic and Multimodal LLMs

上下文感知强化学习用于智能体与多模态大语言模型

Peiyang Xu, Bangzheng Li, Sijia Liu, Karthik R. Narasimhan, Pramod Viswanath, Prateek Mittal, Xingyu Fu

机构 * Princeton University(普林斯顿大学) UC Davis(加州大学戴维斯分校)

AI总结 提出ContextRL方法,通过间接辅助目标(上下文选择奖励)增强大模型在长上下文和多模态任务中的细粒度推理能力,在5个长程基准和12个视觉问答基准上分别提升+2.2%和+1.8%。

Comments 29 pages, 9 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.16899 2026-06-16 cs.LG 新提交

Fantastic Pretraining Optimizers and Where to Find Them II: Hyperball Optimization

奇妙预训练优化器及其发现之处 II:超球优化

Kaiyue Wen, Xingyu Dang, Kaifeng Lyu, Tengyu Ma, Percy Liang

机构 * Stanford University(斯坦福大学) Princeton University(普林斯顿大学) Tsinghua University(清华大学)

AI总结 针对Muon等优化器在大模型预训练中增益随规模增大而减弱的问题,提出Hyperball包装器,固定权重矩阵及其更新的Frobenius范数,在1.2B参数模型上实现20-30%的token等效加速,并改善学习率迁移。

Comments Corresponding blog post: https://psychedelic-sunstone-851.notion.site/Fantastic-Pretraining-Optimizers-and-Where-to-Find-Them-2-1-Hyperball-Optimization-2e924306e6f280e7a5ffee00eb40a0dd

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.16694 2026-06-16 cs.LG cs.AI physics.app-ph q-bio.NC 新提交

Adaptive inference and function vectors in deep transformers

深度变换器中的自适应推理与函数向量

Ravin Raj, Gautam Reddy

机构 * Joseph Henry Laboratories of Physics, Princeton University(普林斯顿大学约瑟夫·亨利物理实验室)

AI总结 提出深度变换器作为平均场交互系统实现分布式推理的理论,利用函数向量逐层推断潜在上下文变量,在上下文回归任务中预测非高斯分层结构与深度的关系,并通过约束线性注意力变换器验证。

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.16242 2026-06-16 cs.LG cs.CL 新提交

Rapid Poison: Practical Poisoning Attacks Against the Rapid Response Framework

快速投毒:针对快速响应框架的实用投毒攻击

David Huang, Jaewon Chang, Avidan Shah, Prateek Mittal, Chawin Sitawarin

机构 * Princeton University(普林斯顿大学)

AI总结 揭示针对快速响应框架的投毒攻击,通过提示注入在训练集中植入恶意样本,实现目标性投毒和概念后门攻击,仅1%投毒率即可导致高达100%误报率和96%漏报率。

Comments Spotlight at ICML 2026

详情

展开后加载摘要…

URL PDF HTML 收藏