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)

2025-12-12 至 2025-12-12 共收录 5
2512.10172 2025-12-12 cs.HC cs.AI cs.CL

Offscript: Automated Auditing of Instruction Adherence in LLMs

Offscript: LLMs指令遵循自动审计

Nicholas Clark, Ryan Bai, Tanu Mitra

机构 * University of Washington Information School Seattle Washington USA University of Washington\ G. Allen School of Computer Science \& Engineering Seattle Washington USA University of Washington Information School University of Washington\ G. Allen School of Computer Science \& Engineering

AI总结 Offscript通过自动化审计检测LLM指令遵循问题,揭示86.4%的对话中存在潜在违规行为,其中22.2%经人工确认为实质性违规。

详情

展开后加载摘要…

URL PDF HTML 收藏
2506.15881 2025-12-12 cs.LG

T-SHRED: Symbolic Regression for Regularization and Model Discovery with Transformer Shallow Recurrent Decoders

T-SHRED:基于Transformer浅层递归解码器的符号回归用于正则化和模型发现

Alexey Yermakov, David Zoro, Mars Liyao Gao, J. Nathan Kutz

机构 * Electrical and Computer Engineering, University of Washington(华盛顿大学电气与计算机工程系) Applied Mathematics, University of Washington(华盛顿大学应用数学系) Computer Science & Engineering, University of Washington(华盛顿大学计算机科学与工程系)

AI总结 T-SHRED通过结合Transformer和符号回归,提升模型正则化和可解释性,适用于不同尺度的混沌系统预测。

Comments 17 pages, 5 figures, submitted to Transactions of the Royal Society (Symbolic Regression in the Physical Sciences)

详情

展开后加载摘要…

URL PDF HTML 收藏
2412.16326 2025-12-12 cs.CV cs.LG

When Worse is Better: Navigating the compression-generation tradeoff in visual tokenization

当更差的是更好的:在视觉分块化中的压缩-生成权衡导航

Vivek Ramanujan, Kushal Tirumala, Armen Aghajanyan, Luke Zettlemoyer, Ali Farhadi

机构 * University of Washington(华盛顿大学) Meta FAIR

AI总结 本文提出CRT方法,通过正则化潜在空间提升生成性能,实现更高效的图像生成模型。

Comments Spotlight at NeurIPS 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2512.10123 2025-12-12 physics.comp-ph cs.LG

A Model-Guided Neural Network Method for the Inverse Scattering Problem

一种指导模型的神经网络方法用于反散射问题

Olivia Tsang, Owen Melia, Vasileios Charisopoulos, Jeremy Hoskins, Yuehaw Khoo, Rebecca Willett

机构 * Department of Computer Science, University of Chicago(计算机科学系,芝加哥大学) Center for Computational Mathematics, Flatiron Institute(计算数学中心,Flatiron研究所) National Institute for Theory and Mathematics in Biology(生物理论与数学国家研究所) Department of Electrical & Computer Engineering, University of Washington(电气与计算机工程系,华盛顿大学) Computational and Applied Mathematics, Department of Statistics, University of Chicago(计算与应用数学,统计系,芝加哥大学) Data Science Institute, University of Chicago(数据科学研究所,芝加哥大学)

AI总结 本文提出了一种指导模型的神经网络方法,通过可微求解器显式整合物理规律,以提高反散射问题的重建质量与效率。

Comments 28 pages

详情

展开后加载摘要…

URL PDF HTML 收藏
2512.10102 2025-12-12 cs.CV

Hierarchical Instance Tracking to Balance Privacy Preservation with Accessible Information

层级实例跟踪以平衡隐私保护与可获取信息

Neelima Prasad, Jarek Reynolds, Neel Karsanbhai, Tanusree Sharma, Lotus Zhang, Abigale Stangl, Yang Wang, Leah Findlater, Danna Gurari

机构 * University of Colorado Boulder(科罗拉多大学博尔德分校) Pennsylvania State University(宾夕法尼亚州立大学) University of Washington(华盛顿大学) Georgia Institute of Technology(佐治亚理工学院) University of Illinois at Urbana-Champaign(伊利诺伊大学厄巴纳-香槟分校)

AI总结 本文提出层级实例跟踪任务,构建首个支持该任务的基准数据集,通过评估多种模型展示数据集的挑战性,旨在平衡隐私保护与信息可获取性。

Comments Accepted at WACV 2026

详情

展开后加载摘要…

URL PDF HTML 收藏