arXivDaily arXiv每日学术速递 周一至周五更新

高校专区

University of Washington(华盛顿大学)

2026-05-18 至 2026-05-18 共收录 6
2605.15787 2026-05-18 cs.LG cs.AI

Grokking as Structural Inference: Transformers Need Bayesian Lottery Tickets

通过结构推断理解Grokking:Transformer需要贝叶斯彩票

Kai Hidajat, Solden Stoll, Joseph An

机构 * Department of Computer Science(计算机科学系) University of Washington(华盛顿大学) Seattle, WA 98195(西雅图, WA 98195)

AI总结 研究探讨了Transformer在延迟泛化现象中的结构推断机制,提出贝叶斯彩票理论,解释了泛化延迟与结构学习的关系。

详情

展开后加载摘要…

URL PDF HTML 收藏
2504.18522 2026-05-18 stat.ML cs.LG

Extrapolation Guarantees for Perturbation Modeling Under the Additive Latent Shift Assumption

在加性潜在位移假设下对扰动建模的外推保证

Julius von Kügelgen, Jakob Ketterer, Michael Vollenweider, Michael Scholkemper, Xinwei Shen, Nicolai Meinshausen, Jonas Peters

机构 * Seminar for Statistics, ETH Zurich(统计系,苏黎世联邦理工学院) ETH Zurich(苏黎世联邦理工学院) DZNE, Bonn, Germany(波恩德国DZNE) Department of Statistics, University of Washington, Seattle, USA(华盛顿大学统计系,美国西雅图)

AI总结 本文研究了在加性潜在位移假设下,通过扰动建模预测新扰动组合的分布,提出PDAE模型并证明了外推保证。

Comments Updated preprint with new material and empirical results; previous version presented at the ICLR'25 Workshop on Learning Meaningful Representations of Life

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.15663 2026-05-18 cs.LG

On the Power of Adaptivity for $\varepsilon$-Best Arm Identification in Linear Bandits

在线性老虎机中ε-最佳臂识别的适应性功率研究

Arnab Maiti, Yunbei Xu, Kevin Jamieson

机构 * University of Washington(华盛顿大学) National University of Singapore(新加坡国立大学)

AI总结 本文研究了在线性老虎机中ε-最佳臂识别的最小样本复杂度,提出非适应性固定设计方法及适应性采样策略,揭示了适应性在不同动作集中的效果差异。

Comments Accepted at COLT 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.15549 2026-05-18 cs.LG cs.AI cs.CE

CTF4Nuclear: Common Task Framework for Nuclear Fission and Fusion Models

CTF4Nuclear: 用于核裂变和核聚变模型的通用任务框架

Stefano Riva, Carolina Introini, Antonio Cammi, Dean Price, Alexey Yermakov, Yue Zhao, Philippe M. Wyder, Judah Goldfeder, Jan Williams, Amy Sara Rude, Matteo Tomasetto, Joe Germany, Joseph Bakarji, Georg Maierhofer, Miles Cranmer, J. Nathan Kutz

机构 * Autodesk Research(Autodesk研究院) Department of Energy, Nuclear Engineering Division, Politecnico di Milano(能源部,核工程系,米兰理工学院) Nuclear Science and Engineering, Massachusetts Institute of Technology(核科学与工程,麻省理工学院) Department of Applied Mathematics, University of Washington(应用数学系,华盛顿大学) Department of Electrical and Computer Engineering, University of Washington(电气与计算机工程系,华盛顿大学) High Performance Machine Learning, SURF(高性能机器学习,SURF) Distyl AI Department of Computer Science, Columbia University(计算机科学系,哥伦比亚大学) Department of Mechanical Engineering, University of Washington(机械工程系,华盛顿大学) Department of Mechanical Engineering, Politecnico di Milano(机械工程系,米兰理工学院) Department of Mathematics, American University in Beirut(数学系,贝鲁特美国大学) Department of Mechanical Engineering, American University in Beirut(机械工程系,贝鲁特美国大学) Department of Applied Mathematics and Theoretical Physics, University of Cambridge(应用数学与理论物理系,剑桥大学)

AI总结 本文提出CTF4Nuclear框架,用于核工程中机器学习方法的标准化评估,通过12个指标和稀疏测量系统监控,提升核工业科学ML的严谨性和可重复性。

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.19399 2026-05-18 cs.CL cs.AI cs.LG

DR Tulu: Reinforcement Learning with Evolving Rubrics for Deep Research

DR Tulu:基于演进标准的深度研究强化学习

Rulin Shao, Akari Asai, Shannon Zejiang Shen, Hamish Ivison, Varsha Kishore, Jingming Zhuo, Xinran Zhao, Molly Park, Samuel G. Finlayson, David Sontag, Tyler Murray, Sewon Min, Pradeep Dasigi, Luca Soldaini, Faeze Brahman, Wen-tau Yih, Tongshuang Wu, Luke Zettlemoyer, Yoon Kim, Hannaneh Hajishirzi, Pang Wei Koh

机构 * University of Washington(华盛顿大学) Allen Institute for AI(人工智能研究院) Carnegie Mellon University(卡内基梅隆大学) Massachusetts Institute of Technology(麻省理工学院) Seattle Children's Hospital(西雅图儿童医院) University of California, Berkeley(加州大学伯克利分校)

AI总结 本文提出RLER方法,通过演进标准与策略模型共进化,开发出首个开源深度研究模型DR Tulu,其在多个长文深度研究基准上表现优异,且成本显著低于现有模型。

Comments ICML 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2510.10454 2026-05-18 cs.AI

Traj-CoA: Patient Trajectory Modeling via Chain-of-Agents for Lung Cancer Risk Prediction

Traj-CoA:通过链式代理进行患者轨迹建模用于肺癌风险预测

Sihang Zeng, Yujuan Fu, Sitong Zhou, Zixuan Yu, Lucas Jing Liu, Jun Wen, Matthew Thompson, Ruth Etzioni, Meliha Yetisgen

机构 * University of Washington(华盛顿大学) Fred Hutch Cancer Center(Fred Hutch癌症中心) Harvard University(哈佛大学) Google(谷歌)

AI总结 Traj-CoA通过链式代理系统处理电子健康记录数据,减少噪声并保留完整时间线,从而在肺癌风险预测中优于基线方法。

Comments Accepted by NeurIPS 2025 GenAI4Health Workshop

详情

展开后加载摘要…

URL PDF HTML 收藏