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)

共收录 1148
2607.15396 2026-07-20 cs.CV cs.AI 新提交

Partial Information Decomposition as a Multi-Contrast 3D MRI Selection Strategy for Resource-Constrained Deep Neural Network Training in Brain Tumor Segmentation

部分信息分解作为一种多对比度3D MRI选择策略,用于脑肿瘤分割中资源受限的深度神经网络训练

Agamdeep Chopra, Mehmet Kurt

机构 * University of Washington(华盛顿大学)

AI总结 研究针对脑肿瘤分割中多对比度3D MRI分割计算量大的问题,采用部分信息分解框架对输入对排序选最优用于训练,实验表明该方法选出的T1c+T2-FLAIR是强双输入配置,证明了基于PID预训练选择的实用价值。

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.22759 2026-07-20 cs.AI 版本更新

Towards a General Intelligence and Interface for Wearable Health Data

迈向可穿戴健康数据的通用智能与接口

Girish Narayanswamy, Maxwell A. Xu, A. Ali Heydari, Samy Abdel-Ghaffar, Marius Guerard, Kara Vaillancourt, Zhihan Zhang, Jake Garrison, Levi Albuquerque, Dimitris Spathis, Hong Yu, Hamid Palangi, Xuhai "Orson" Xu, David G. T. Barrett, Joseph Breda, Jed McGiffin, Yubin Kim, Yuwei Zhang, Naghmeh Rezaei, Samuel Solomon, Karan Ahuja, Tim Althoff, Jake Sunshine, Ming-Zher Poh, Benjamin Yetton, Ari Winbush, Nicholas B. Allen, James M. Rehg, Isaac Galatzer-Levy, Yun Liu, John Hernandez, Anupam Pathak, Conor Heneghan, Yuzhe Yang, Ahmed A. Metwally, Pushmeet Kohli, Mark Malhotra, Shwetak Patel, Xin Liu, Daniel McDuff

机构 * Google Research(谷歌研究) Google DeepMind(谷歌DeepMind) University of Washington(华盛顿大学) University of Oregon(俄勒冈大学) University of Illinois Urbana-Champaign(伊利诺伊大学厄巴纳-香槟分校)

AI总结 提出一个基于超过一万亿分钟无标签传感器数据预训练的可穿戴健康基础模型,通过联合扩展模型容量和预训练数据量,在35项健康预测任务上实现系统性性能提升,并利用LLM代理自动搜索下游预测头,集成到个人健康代理中以提高相关性和安全性。

Comments Narayanswamy and Xu are co-first authors. McDuff and Liu are co-last authors

详情

展开后加载摘要…

URL PDF HTML 收藏
2607.15271 2026-07-17 cs.CV cs.GR cs.LG 新提交

Online Neural Space Time Memory for Dynamic Novel View Synthesis

用于动态新视角合成的在线神经时空记忆

Baback Elmieh, Lynn Tsai, Zeman Li, Srinivas Kaza, Tiancheng Sun, Gabor Csapo, Ali Behrouz, Yuan Deng, Stephen Lombardi, Steven M. Seitz, Xuan Luo

机构 * University of Washington(华盛顿大学) Google(谷歌)

AI总结 研究多视图流视频在线新视角合成问题,提出解耦记忆更新与应用频率的方法,通过跨视图注意力管理变形,引入辅助记忆损失和记忆缓存策略,实现实时、领先性能及微小尺度在线记忆。

Comments 15 pages. Preprint. Project page with demos and video results: https://nst-mem.github.io

详情

展开后加载摘要…

URL PDF HTML 收藏
2607.15267 2026-07-17 cs.AI cs.CL 新提交

Pretraining Data Can Be Poisoned through Computational Propaganda

预训练数据可通过计算宣传被下毒

Victoria Graf, Hannaneh Hajishirzi, Noah A. Smith, David Kohlbrenner, Kyle Lo

机构 * University of Washington(华盛顿大学) Allen Institute for Artificial Intelligence(艾伦人工智能研究所)

AI总结 研究发现预训练数据可通过公共讨论界面被下毒,引入HalfLife方法衡量恶意内容,探索在网络规模下毒预训练语料库的可行性,证明估计毒注入重要性,确立第三方网页内容为攻击语言模型预训练的可能载体。

详情

展开后加载摘要…

URL PDF HTML 收藏
2607.15265 2026-07-17 cs.CV cs.AI cs.MM cs.SD 新提交

SceneBind: Binding What and Where Across Vision, Audio and Language

SceneBind:跨视觉、音频和语言绑定“什么”与“哪里”

Mingfei Chen, Zijun Cui, Ruoke Zhang, Hyeonggon Ryu, Eli Shlizerman

机构 * University of Washington(华盛顿大学) University of Texas at Dallas(德克萨斯大学达拉斯分校) Hankuk University of Foreign Studies(韩国外国语大学)

AI总结 研究提出SceneBind全模态场景表示,结合全局语义与对象中心语义空间插槽解决空间结构缺失问题,还提出匹配方案。通过构建新数据集及训练协议进行训练评估,兼容预训练编码器,实现先进检索并能零样本转移到下游任务。

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

详情

展开后加载摘要…

URL PDF HTML 收藏
2607.15077 2026-07-17 cs.LG 新提交

An Introduction to Sparse Identification of Nonlinear Dynamics for Engineering Applications

工程应用中的非线性动力学稀疏识别介绍

Yao Cheng Li, Ana Larrañaga, Steven L. Brunton, Urban Fasel

机构 * Department of Aeronautics, Imperial College London(伦敦帝国理工学院航空系) Department of Mechanical Engineering, University of Washington(华盛顿大学机械工程系) NSF AI Institute in Dynamic Systems, University of Washington(华盛顿大学动态系统领域美国国家科学基金会人工智能研究所)

AI总结 介绍工程应用中非线性动力学稀疏识别(SINDy)方法,通过对候选非线性项库稀疏回归解决代理建模局限性,教程介绍该方法及扩展,经案例研究表明其易实现且灵活,是工程应用有价值的识别工具。

Comments 15 pages, 4 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
2607.14703 2026-07-17 cs.CV cs.AI 新提交

Pretraining Multiple Instance Learning Networks with Multi-Teacher Distillation from Pathology Slide Foundation Models

基于病理切片基础模型的多教师蒸馏预训练多实例学习网络

Mingxi Fu, Jiawen Li, Renao Yan, Jiali Hu, Qiehe Sun, Tian Guan, Yonghong He

机构 * Tsinghua Shenzhen International Graduate School, Tsinghua University(清华大学深圳国际研究生院) University of Washington(华盛顿大学) City University of Hong Kong (Dongguan)(香港城市大学(东莞)) Medical Optical Technology R&D Center, Research Institute of Tsinghua(清华研究院医学光学技术研发中心) Jinfeng Laboratory(金凤实验室)

AI总结 针对计算病理学中多实例学习存在的问题,提出基于蒸馏的预训练框架,利用两个基础模型作为教师,引入角分散归一化蒸馏损失,将蒸馏权重用于下游适应,实验表明该方法在少样本场景中优势明显,能提升计算效率。

详情

展开后加载摘要…

URL PDF HTML 收藏
2607.14611 2026-07-17 cs.CR cs.AI cs.MA 新提交

Bad Memory: Evaluating Prompt Injection Risks from Memory in Agentic Systems

糟糕的记忆:评估智能体系统中内存引发的提示注入风险

Soham Gadgil, David Alexander, Sai Sunku, Franziska Roesner

机构 * University of Washington(华盛顿大学)

AI总结 研究基于内存的智能体系统中提示注入攻击,利用沙盒合成工作区评估两个系统四个模型,发现虽难用外部内容重写内存文件,但已植入的有效载荷可攻击当前及未来会话,揭示持久内存改变威胁模型并推动相关防御研究。

Comments Preprint

详情

展开后加载摘要…

URL PDF HTML 收藏
2607.14418 2026-07-17 cs.LG econ.GN q-fin.EC 新提交

Adaptive Ad Load Design for Sponsored Search Markets: Evidence, Theory, and Deployment

赞助搜索市场的自适应广告加载设计:证据、理论与部署

Mohammad Rashid, Hema Yoganarasimhan

机构 * University of Washington(华盛顿大学)

AI总结 研究赞助搜索市场广告加载设计权衡,通过安卓应用商店实验发现增加广告加载量对收入、转化率和参与度的影响及异质性,设计并部署自适应算法e-LAAL,在生产部署中改善收益与转化率权衡,优于静态基准。

Comments 54 pages

详情

展开后加载摘要…

URL PDF HTML 收藏
2607.13938 2026-07-16 cs.RO 新提交

Discriminative Barrier Functions for Safe Adversarial Imitation Learning from Observation

用于从观察中进行安全对抗模仿学习的判别障碍函数

Anubhav Vishwakarma, Bhaumik Mehta, Caleb Hsu, Byron Boots, Karen Leung, Tyler Han

机构 * University of Washington(华盛顿大学)

AI总结 研究针对逆强化学习不安全及控制障碍函数设计难的问题,通过将奖励函数候选限制在CBF空间,实现安全在线控制与经验改进,能从无标签观察中恢复障碍函数,模拟实验显示其安全性能提升,并研究了不同IRL方法的权衡。

Comments 20 pages, 5 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
2607.13591 2026-07-16 cs.CL cs.AI 新提交

Memory as a Controlled Process: Learned Adaptive Memory Management for LLM Agents

作为受控过程的记忆:为大语言模型智能体学习自适应内存管理

Eric Hanchen Jiang, Zhi Zhang, Yuchen Wu, Levina Li, Dong Liu, Xiao Liang, Rui Sun, Yubei Li, Edward Sun, Haozheng Luo, Zhaolu Kang, Aylin Caliskan, Kai-Wei Chang, Ying Nian Wu

机构 * University of California Los Angeles(加利福尼亚大学洛杉矶分校) University of Washington(华盛顿大学) Northwestern University(西北大学)

AI总结 研究LLM智能体内存管理问题,提出MemCon框架将内存操作建模为马尔可夫决策过程,通过在线策略自适应管理内存,该框架与后端无关,实验表明其在多基准测试中优于基线,提升任务成功率并减少令牌消耗。

详情

展开后加载摘要…

URL PDF HTML 收藏
2607.13558 2026-07-16 cs.AI 新提交

Multi-Agent Collaborative Reasoning with Tool-Augmented Evidence for Urban Region Profiling

基于工具增强证据的多智能体协作推理用于城市区域剖析

Xixuan Hao, Yutian Jiang, Jiabo Liu, Yihang Yang, Guangyin Jin, Song Gao, Yuxuan Liang

机构 * The Hong Kong University of Science and Technology (Guangzhou)(香港科技大学(广州)) University of Washington(华盛顿大学) Chang’an University(长安大学) University of Wisconsin - Madison(威斯康星大学麦迪逊分校)

AI总结 研究针对城市区域剖析问题,提出UrbanAgent框架,通过多智能体协作推理解决跨模态不一致,将指标预测扩展为闭环过程,经实验验证其性能优于现有基线,在未见城市设置中有强泛化性。

Comments Accepted by KDD 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2607.13498 2026-07-16 cs.LG 新提交

Factorized Spectral Representations for Reinforcement Learning

用于强化学习的因式分解谱表示

Junyi Wu, Dan Li

机构 * University of Washington(华盛顿大学)

AI总结 该研究聚焦强化学习,提出FaStR方法,通过对转移核的三模张量CP分解,用噪声对比目标拟合,产生单独编码器形成谱表示。其因式分解形式缩小假设类,在高维运动任务中效果好,状态编码器可跨执行器移位转移。

详情

展开后加载摘要…

URL PDF HTML 收藏
2607.04113 2026-07-16 cs.LG cs.NA math.NA 版本更新

Asymptotic Preservation and Uniform Accuracy of Diffusion and Flow-Matching Samplers

扩散与流匹配采样器的渐近保持后验分析

Shiheng Zhang

机构 * University of Washington(华盛顿大学)

AI总结 研究将最小标准差视为奇异摄动参数,通过后验审计确定固定步长采样器的渐近保持性,分析不同时钟在终端层的稳定性及准确性,在特定模型上验证确定性和随机采样器特性,并用于EDM CIFAR-10检查点分析。

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.20689 2026-07-16 cs.CL cs.AI cs.IR cs.LG 版本更新

DIVE: Embedding Compression via Self-Limiting Gradient Updates

DIVE: 通过自限制梯度更新实现嵌入压缩

Dongfang Zhao

机构 * University of Washington Tacoma School of Engineering and Technology(华盛顿大学塔可姆分校工程与技术学院)

AI总结 本文提出DIVE方法,通过自限制的三元组损失和头级NT-Xent对比损失解决嵌入压缩中因标注数据稀缺导致的过拟合问题,提升了检索性能。

详情

展开后加载摘要…

URL PDF HTML 收藏
2607.12886 2026-07-15 cs.AI 新提交

A Multi-Agent System for Autonomous, Fine-Tuning-Free Clinical Symptom Detection: Development and Validation Study

一种用于自主、无需微调的临床症状检测的多智能体系统:开发与验证研究

Cameron Cagan, Pedram Fard, Jiazi Tian, Jingya Cheng, Shawn N. Murphy, Hossein Estiri

机构 * Massachusetts General Hospital(麻省总医院) University of Washington(华盛顿大学)

AI总结 研究针对临床症状检测中信息难结构化及现有方法不足的问题,提出多智能体系统Pythia,无需人工提示工程或微调,能自主优化提取提示。通过与词汇表比较,验证其在临床记录症状提取上的有效性及推广性,优于部分传统方法。

详情

展开后加载摘要…

URL PDF HTML 收藏
2607.12520 2026-07-15 cs.AI 新提交

The Model Knows Your Project, Not You: Measuring Recognition in LLMs with NameRank

模型了解你的项目,而非你本人:使用NameRank衡量大语言模型中的识别度

Bojie Li, Noah Shi

机构 * Pine AI(松树人工智能公司) University of Washington(华盛顿大学)

AI总结 研究用NameRank衡量大语言模型对实体的识别度,通过对多实体多模型探测及独立评判获取分数,发现识别关注可索引工件,奥运资质与知名奖项情况不同,独立创作者中工具与创造者排名有别等,并指出文献计量法难测识别度等结论。

详情

展开后加载摘要…

URL PDF HTML 收藏
2607.12227 2026-07-15 cs.AI 新提交

Rethinking the Evaluation of Harness Evolution for Agents

重新思考智能体的 harness 进化评估

Yike Wang, Huaisheng Zhu, Zhengyu Hu, Yige Yuan, Zhengyu Chen, Shakti Senthil, Hannaneh Hajishirzi, Yulia Tsvetkov, Pradeep Dasigi, Teng Xiao

机构 * Allen Institute for AI(艾伦人工智能研究所) University of Washington(华盛顿大学)

AI总结 研究重新审视大语言模型智能体的自动 harness 进化评估,通过在可比条件下与基线比较及在保留任务上测试,发现其不总优于简单方法且泛化有限,对其有效性提出质疑,强调需更公平评估协议和基准。

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.24267 2026-07-15 cs.CL cs.AI 版本更新

Pigeonholing: how bad prompts hurt models, causing collapse and mistakes

鸽笼效应:不良提示导致模型崩溃和犯错

Hyunji Nam, Keertana Chidambaram, Dorottya Demszky, Natasha Jaques

机构 * Stanford University(斯坦福大学) University of Washington(华盛顿大学)

AI总结 研究不良上下文导致大语言模型性能下降和模式崩溃的“鸽笼效应”,发现重复错误答案、收敛于狭窄答案集等问题,并提出RLVR合成错误缓解方法。

Comments 10 pages

详情

展开后加载摘要…

URL PDF HTML 收藏
2312.17670 2026-07-15 cs.CV cs.LG q-bio.QM q-bio.TO 版本更新

The TopCoW Challenge -- Topology-Aware Circle of Willis Segmentation for CT and MR Angiography

TopCoW挑战——用于CT和MR血管造影的拓扑感知Willis环分割

Kaiyuan Yang, Fabio Musio, Yihui Ma, Norman Juchler, Johannes C. Paetzold, Rami Al-Maskari, Luciano Höher, Hongwei Bran Li, Ibrahim Ethem Hamamci, Anjany Sekuboyina, Suprosanna Shit, Houjing Huang, Chinmay Prabhakar, Ezequiel de la Rosa, Bastian Wittmann, Diana Waldmannstetter, Florian Kofler, Fernando Navarro, Martin J. Menten, Ivan Ezhov, Daniel Rueckert, Iris N. Vos, Ynte M. Ruigrok, Birgitta K. Velthuis, Hugo J. Kuijf, Pengcheng Shi, Wei Liu, Ting Ma, Maximilian R. Rokuss, Yannick Kirchhoff, Fabian Isensee, Klaus Maier-Hein, Chengcheng Zhu, Huilin Zhao, Philippe Bijlenga, Julien Hämmerli, Catherine Wurster, Laura Westphal, Jeroen Bisschop, Elisa Colombo, Hakim Baazaoui, Hannah-Lea Handelsmann, Andrew Makmur, James Hallinan, Amrish Soundararajan, Benedikt Wiestler, Jan S. Kirschke, Evamaria O. Riedel, Roland Wiest, Emmanuel Montagnon, Laurent Letourneau-Guillon, Kwanseok Oh, Dahye Lee, Orhun Utku Aydin, Adam Hilbert, Jana Rieger, Dimitrios Rallios, Satoru Tanioka, Alexander Koch, Dietmar Frey, Abdul Qayyum, Moona Mazher, Steven Niederer, Nico Disch, Julius C. Holzschuh, Dominic LaBella, Francesco Galati, Daniele Falcetta, Maria A. Zuluaga, Chaolong Lin, Haoran Zhao, Zehan Zhang, Minghui Zhang, Xin You, Hanxiao Zhang, Guang-Zhong Yang, Yun Gu, Sinyoung Ra, Jongyun Hwang, Hyunjin Park, Junqiang Chen, Marek Wodzinski, Henning Müller, Nesrin Mansouri, Florent Autrusseau, Cansu Yalcin, Rachika E. Hamadache, Clara Lisazo, Joaquim Salvi, Adrià Casamitjana, Xavier Lladó, Uma Maria Lal-Trehan Estrada, Valeriia Abramova, Luca Giancardo, Arnau Oliver, Paula Casademunt, Adrian Galdran, Matteo Delucchi, Oscar Camara, Jialu Liu, Haibin Huang, Yue Cui, Zehang Lin, Yusheng Liu, Shunzhi Zhu, Tatsat R. Patel, Adnan H. Siddiqui, Vincent M. Tutino, Maysam Orouskhani, Huayu Wang, Mahmud Mossa-Basha, Yuki Sato, Sven Hirsch, Susanne Wegener, Bjoern Menze

机构 * Department of Quantitative Biomedicine, University of Zurich, Zurich, Switzerland Institute of Computational Life Sciences, Zurich University of Applied Sciences (ZHAW), Waedenswil, Switzerland Department of Neuroradiology, University Hospital of Zurich, Zurich, Switzerland Department of Neurosurgery, Zhongnan Hospital of Wuhan University, Wuhan, China Department of Radiology at Weill Cornell Medicine, Cornell University, New York, USA Institute for Tissue Engineering School of Computation, Information Technology, Technical University of Munich, Germany Athinoula A. Martinos Center for Biomedical Imaging, Harvard Medical School, Boston, USA School of Medicine Health, TUM Klinikum, Technical University of Munich, Germany Munich Center for Machine Learning, Munich, Germany Department of Computing, Imperial College London, London, UK Image Sciences Institute, UMC Utrecht, Utrecht, The Netherlands Department of Neurology Neurosurgery, University Medical Center Utrecht, Utrecht, The Netherlands Department of Radiology, University Medical Center Utrecht, Utrecht, The Netherlands Electronic \& Information Engineering School, Harbin Institute of Technology (Shenzhen), China Peng Cheng Laboratory, Shenzhen, China Division of Medical Image Computing, German Cancer Research Center (DKFZ), Heidelberg, Germany Faculty of Mathematics Computer Science, Heidelberg University, Germany Helmholtz Imaging, German Cancer Research Center, Heidelberg, Germany Data Science School for Health, Karlsruhe/Heidelberg, Germany Learning Group, Department of Radiation Oncology, Heidelberg University Hospital Department of Radiology, University of Washington, Seattle, WA, USA Department of Radiology, Ren Ji Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China Department of Clinical Neurosciences, Division of Neurosurgery, Geneva University Hospitals, Geneva, Switzerland Department of Neurology, University Hospital of Zurich, Zurich, Switzerland Department of Physiology, University of Toronto, Canada Department of Neurosurgery, University Hospital of Zurich, Zurich, Switzerland Department of Diagnostic Imaging, National University Hospital, Singapore University of Chicago, USA Department of Diagnostic Interventional Neuroradiology, University Hospital Berne University of Berne, Berne, Switzerland Centre de Recherche du Centre Hospitalier de l’Université de Montréal (CRCHUM), Montréal, Québec, Canada DEEPNOID Inc., Seoul, South Korea Department of Artificial Intelligence, Korea University, Seoul, South Korea Charité Lab for AI in Medicine (CLAIM), Charité Universitätsmedizin Berlin, Berlin, Germany Lung Institute, Faculty of Medicine, Imperial College London, London, UK Centre for Medical Image Computing, Department of Computer Science, University College London, London, UK Department of Radiation Oncology, Duke University Medical Center, Durham, NC, USA Institute of Medical Technology, Peking University Health Science Center, Beijing, China Hangzhou Genlight MedTech Co., Ltd., China Institute of Medical Robotics, Shanghai Jiao Tong University, Shanghai, China Department of Automation, Shanghai Jiao Tong University, Shanghai, China Department of Artificial Intelligence, Sungkyunkwan University, Seoul, South Korea Department of Electrical Computer Engineering, Sungkyunkwan University, Seoul, South Korea Shanghai MediWorks Precision Instruments Co., Ltd., China Institute of Informatics, HES-SO Valais-Wallis, Switzerland Department of Measurement Electronics, AGH University of Krakow, Poland Laboratoire de Thermique et Energie de Nantes (LTeN), Université Nantes, Polytech’Nantes, Nantes, France Research Institute of Computer Vision Center for Precision Health, McWilliams School of Biomedical Informatics, University of Texas Health Science Center at Houston, USA Physense, BCN-Medtech, Department of Communication Information Technologies, Universitat Pompeu Fabra, Barcelona, Spain Department of Mathematical Modeling Machine Learning, University of Zurich, Zurich, Switzerland Laboratory of Brain Atlas Brain-inspired Intelligence, Institute of Automation, Chinese Academy of Sciences, Beijing, China School of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing, China School of Computer Information Engineering, Xiamen University of Technology, Xiamen, China Vascular Research Center, University at Buffalo, NY, USA Department of Pathology Anatomical Sciences, University at Buffalo, NY, USA Department of Neurosurgery, University at Buffalo, NY, USA LPIXEL Inc., Tokyo, Japan

AI总结 组织TopCoW基准挑战,发布含125对MRA和CTA扫描的注释数据集,参与者提交CoW分割和变体分类算法,经评估,最佳算法在多任务中表现出色,证明CoW分割算法对下游临床应用有可解释性效用。

Comments Summary paper for the TopCoW Challenge: 4 figures, 1 table, and supplementary material in appendix. Accepted for publication in NEJM AI. Datasets and best-performing algorithm Dockers are available at https://zenodo.org/records/15692630 and https://zenodo.org/records/15665435

Journal ref NEJM AI 2026;3(8)

详情

展开后加载摘要…

URL PDF HTML 收藏
2607.11598 2026-07-14 cs.AI 新提交

Interaction Scaling: Grounding the Third Axis of Test-Time Compute

交互缩放:奠定测试时计算的第三轴基础

Bojie Li, Noah Shi

机构 * Pine AI(松树人工智能公司) University of Washington(华盛顿大学)

AI总结 研究测试时增加计算量的新方法,提出交互缩放概念,通过模型与外部仪器交互突破传统方法局限,在硬编码任务和视觉工件处理上展现优势,表明交互缩放真实且区别于推理和采样,需反馈和度量基于实际。

详情

展开后加载摘要…

URL PDF HTML 收藏
2607.11368 2026-07-14 cs.DC cs.LG cs.PF 新提交

Decomposing Runtime, Kernel, and Quantization Speedups via a Matched FP16 Intermediate: A Hardware-Conditioned Case Study on Four NVIDIA RTX A5000 GPUs

通过匹配的 FP16 中间层分解运行时、内核和量化加速:基于四块 NVIDIA RTX A5000 GPU 的硬件条件案例研究

Weijia Han, Lisha Qu

机构 * University of Washington(华盛顿大学)

AI总结 研究通过匹配的 FP16 中间层分解运行时、内核和量化加速,以四块 NVIDIA RTX A5000 GPU 为案例,分析加速各部分占比、分片影响因素等,得出运行时占主要加速比例,量化可扩展并发用户,还探讨了实例选择及有效性威胁。

Comments 36 pages, 8 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
2607.09739 2026-07-14 cs.AI cs.CL 新提交

Coresets Before Score Sets: Evaluation-Unsupervised Prompt Subset Selection for LLM Benchmarks

分数集之前的核心集:用于大语言模型基准测试的评估无监督提示子集选择

Jihan Yao, Gantavya Bhatt, Arnav Das, Peter Jin, Ke Bao, Qiaolin Yu, Khushi Bhardwaj, Chang Su, Jialei Wang, Yikai Zhu, Sugam Devare, Damon Mosk-Aoyama, Zhen Dong, Venkat Krishna Srinivasan, Yineng Zhang, Oleksii Kuchaiev, Jiantao Jiao, Banghua Zhu, Jeff Bilmes

机构 * University of Washington(华盛顿大学) University of California, Berkeley(加利福尼亚大学伯克利分校) Oracle(甲骨文公司) Together AI(Together AI公司) LMSYS NVIDIA(英伟达公司)

AI总结 研究大语言模型基准测试的核心集选择,采用评估无监督方法,利用次模子集选择,开发多种次模函数。在新大规模套件上实验发现设施选址函数效果好,该目标不限于特定模式,在相关排行榜上表现优且计算成本低,证明次模性对基准压缩有用。

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.28622 2026-07-14 cs.CV cs.GR 版本更新

Meshtryoshka: Differentiable Rendering of Real-World Scenes via Mesh Rasterization

Meshtryoshka: 通过网格光栅化实现真实场景的可微分渲染

David Charatan, Daniel Xu, Richard Szeliski, George Kopanas, Vincent Sitzmann

机构 * Massachusetts Institute of Technology(麻省理工学院) University of Washington(华盛顿大学) Google DeepMind(谷歌DeepMind)

AI总结 提出Meshtryoshka框架,利用嵌套网格壳表示和现成三角形光栅化器,通过符号距离函数间接更新顶点位置,实现大规模无界场景的高质量可微分渲染。

Comments Daniel Xu and David Charatan contributed equally; author order decided by coin flip. Project website: https://danielxu9393.github.io/meshtryoshka-website/

详情

展开后加载摘要…

URL PDF HTML 收藏
2509.19129 2026-07-14 cs.CV 版本更新

KAMERA: Enhancing Aerial Surveys of Ice-associated Seals in Arctic Environments

KAMERA:增强北极环境中与冰相关海豹的航空调查

Adam Romlein, Benjamin X. Hou, Yuval Boss, Cynthia L. Christman, Stacie Koslovsky, Erin E. Moreland, Jason Parham, Anthony Hoogs

机构 * NOAA NMFS AFSC MML(国家海洋管理局NMFS AFSC MML) CICOES(国际极地科学组织) University of Washington(华盛顿大学)

AI总结 KAMERA系统用于北极环境中与冰相关海豹的航空调查,通过多相机多光谱同步及实时检测,减少数据集处理时间,能利用多光谱检测目标,数据带元数据,图像和检测结果可映射到世界平面,软件等完全开源。

Comments 10 pages, 9 figures, 4 tables. Code: https://github.com/Kitware/kamera

Journal ref 2025 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW), 2025, pp. 2183-2192

详情

展开后加载摘要…

URL PDF HTML 收藏
2607.08946 2026-07-13 cs.LG cs.CL 新提交

Training, Reading, and Editing Legible Transformers

训练、读取和编辑清晰可读的Transformer

Mark Oskin

机构 * University of Washington(华盛顿大学) School of Computer Science and Engineering(计算机科学与工程学院)

AI总结 研究如何让Transformer清晰可读,提出用通道方差下限等方法,构建出最清晰可读的Transformer,使单元能分离检测与命名,编辑更局部,还揭示清晰度调节旋钮,质量与传统基线相当。

详情

展开后加载摘要…

URL PDF HTML 收藏
2607.08877 2026-07-13 cs.RO cs.LG 新提交

FlowDAgger: Human-in-the-Loop Adaptation of Generative Robot Policies in Latent Space

FlowDAgger:在潜在空间中对生成式机器人策略进行人工参与的自适应调整

Michael Murray, Daphne Chen, Simran Bagaria, Dean Fortier, Tess Hellebrekers, Galen Mullins, Harshavardhan Gajarla, Oier Mees, Maya Cakmak, Andrey Kolobov

机构 * Microsoft Research(微软研究院) University of Washington(华盛顿大学) ETH Zurich(苏黎世联邦理工学院)

AI总结 研究针对预训练生成式机器人策略在实际部署中出现的问题,提出FlowDAgger方法,通过动作反转从人工干预获取监督,在模拟及现实操纵中评估,该方法优于基线且能保留预训练技能,为机器人基础模型自适应调整提供实用路径。

详情

展开后加载摘要…

URL PDF HTML 收藏
2604.17267 2026-07-13 cs.AI stat.AP 版本更新

Rectification Difficulty and Optimal Sample Allocation in LLM-Augmented Surveys

LLM增强调查中的校正难度与最优样本分配

Zikun Ye, Hema Yoganarasimhan

机构 * Michael G. Foster School of Business, University of Washington(华盛顿大学Michael G. Foster商学院)

AI总结 本文研究在LLM预测可用时如何分配人类样本以提高估计效率,提出校正难度分析、最优分配规则及元学习方法,通过实验证明其在不同领域和LLM上的有效性。

详情

展开后加载摘要…

URL PDF HTML 收藏
2603.00395 2026-07-13 cs.SD cs.LG eess.AS 版本更新

Fine-grained Soundscape Control for Augmented Hearing

细粒度声音景观控制用于增强听觉

Seunghyun Oh, Malek Itani, Aseem Gauri, Shyamnath Gollakota

机构 * Paul G. Allen School of Computer Science and Engineering, University of Washington(保罗·G·阿伦计算机科学与工程学院,华盛顿大学) Hearvana AI(Hearvana人工智能)

AI总结 Aurchestra通过细粒度声音控制实现可穿戴设备上多声音源的独立调节与混合,提升环境声音的增强与抑制效果。

Comments 15 pages, 11 figures, 4 tables, published at ACM MobiSys 2026

Journal ref MobiSys '26: Proceedings of the 24th Annual International Conference on Mobile Systems, Applications and Services (2026) 371-391

详情

展开后加载摘要…

URL PDF HTML 收藏
2607.08724 2026-07-10 cs.LG cs.RO 新提交

Latent Memory Palace: Reasoning for Control as Autoregressive Variational Inference

潜在记忆宫殿:作为自回归变分推理的控制推理

Chuning Zhu, Eva Xu, Jose Barreiros, Krishnan Srinivasan, Paarth Shah, Abhishek Gupta

机构 * University of Washington(华盛顿大学) Toyota Research Institute(丰田研究院)

AI总结 研究将语言模型的推理能力应用于连续控制策略的问题,提出潜在记忆宫殿(LMP)方法,通过自回归潜在空间组织信息进行变分推理,推导强化学习技术优化下限,该方法在模拟和现实领域表现良好,还产生高性能动作分词器,为控制的潜在推理提供新视角。

详情

展开后加载摘要…

URL PDF HTML 收藏