arXivDaily arXiv每日学术速递 周一至周五更新

高校专区

EPFL(洛桑联邦理工学院)

共收录 857
2511.01836 2025-11-25 cs.LG

Priors in Time: Missing Inductive Biases for Language Model Interpretability

时间中的先验:语言模型可解释性中缺失的归纳偏置

Ekdeep Singh Lubana, Can Rager, Sai Sumedh R. Hindupur, Valerie Costa, Greta Tuckute, Oam Patel, Sonia Krishna Murthy, Thomas Fel, Daniel Wurgaft, Eric J. Bigelow, Johnny Lin, Demba Ba, Martin Wattenberg, Fernanda Viegas, Melanie Weber, Aaron Mueller

机构 * Goodfire AI Independent(独立) SEAS, Harvard University(哈佛大学SEAS学院) EPFL(瑞士联邦理工学院) Kempner Institute at Harvard University(哈佛大学凯普内研究所) Department of Psychology, Stanford University(斯坦福大学心理学系) Department of Psychology, Harvard University(哈佛大学心理学系) Decode Research Boston University(波士顿大学)

AI总结 本文提出时序特征分析方法,通过引入时序归纳偏置,改进语言模型表示的可解释性,有效区分抽象与新颖信息,克服现有方法的局限。

Comments Preprint

详情

展开后加载摘要…

URL PDF HTML 收藏
2505.16570 2025-11-25 cs.CL

URLs Help, Topics Guide: Understanding Metadata Utility in LLM Training

URL助益,主题引导:理解元数据在LLM训练中的效用

Dongyang Fan, Vinko Sabolčec, Martin Jaggi

机构 * EPFL(苏黎世联邦理工学院)

AI总结 本研究探讨了元数据在LLM训练中的效用,发现URL上下文能提升训练效率,而主题和格式信息则能引导生成输出,但不加速训练。

Comments NeurIPS 2025, Camera Ready

详情

展开后加载摘要…

URL PDF HTML 收藏
2306.00833 2025-11-25 cs.SI cs.LG math.ST stat.ME stat.ML stat.TH

When Does Bottom-up Beat Top-down in Hierarchical Community Detection?

在何时自底向上方法胜过自顶向下方法在分层社区检测中?

Maximilien Dreveton, Daichi Kuroda, Matthias Grossglauser, Patrick Thiran

机构 * École Polytechnique Fédérale de Lausanne(瑞士联邦理工学院洛桑分校)

AI总结 本文研究了自底向上方法在分层社区检测中的优势,证明其在中间层次实现精确恢复的理论保证优于自顶向下方法。

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.17698 2025-11-25 stat.ML cs.LG

Quantum Fourier Transform Based Kernel for Solar Irrandiance Forecasting

基于量子傅里叶变换的太阳能辐照度预测核函数

Nawfel Mechiche-Alami, Eduardo Rodriguez, Jose M. Cardemil, Enrique Lopez Droguett

机构 * Ecole Polytechnique Fédérale de Lausanne (EPFL)(瑞士联邦理工学院) Mechanical Engineering Department, Universidad de Chile(智利大学机械工程系) Departamento de Ingeniería Mecánica y Metalúrgica, Escuela de Ingeniería, Pontificia Universidad Católica de Chile(智利天主教大学机械工程与冶金系) Garrick Institute for the Risk Sciences, University of California(加州大学风险科学研究所)

AI总结 本文提出基于量子傅里叶变换的核函数,用于提升太阳能辐照度预测的精度与鲁棒性。

Comments 33 pages, 13 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
2506.03425 2025-11-25 eess.AS cs.AI cs.LG

A Data-Driven Diffusion-based Approach for Audio Deepfake Explanations

一种数据驱动的扩散方法用于音频深度伪造解释

Petr Grinberg, Ankur Kumar, Surya Koppisetti, Gaurav Bharaj

机构 * EPFL(瑞士联邦理工学院) Reality Defender Inc.(Reality Defender公司)

AI总结 本文提出了一种数据驱动的扩散方法,通过时频差异信号训练模型,以更准确地识别音频深度伪造中的伪影区域。

Comments 5 pages, 3 figures, accepted at Interspeech 2025

Journal ref Interspeech 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2305.15118 2025-11-25 cs.LG cs.CY cs.DS

Fairness in Streaming Submodular Maximization over a Matroid Constraint

流式子模最大化中的公平性:在Matroid约束下

Marwa El Halabi, Federico Fusco, Ashkan Norouzi-Fard, Jakab Tardos, Jakub Tarnawski

机构 * Samsung - SAIT AI Lab, Montreal(三星 - SAIT AI实验室,蒙特利尔) Sapienza University of Rome(罗马萨皮恩扎大学) Google Zurich(谷歌苏黎世) Google Zurich EPFL(谷歌苏黎世EPFL) Microsoft Research(微软研究院)

AI总结 本文研究了在Matroid约束下流式子模最大化问题的公平性,提出流式算法和不可能结果,通过现实应用验证了效率、质量和公平性的权衡。

Comments Correcting error in Proposition C.6. This doesn't affect any other result in the paper

详情

展开后加载摘要…

URL PDF HTML 收藏
2502.01588 2025-11-24 cs.LG cs.SD eess.AS stat.ML

A Differentiable Alignment Framework for Sequence-to-Sequence Modeling via Optimal Transport

基于最优传输的序列到序列建模可微对齐框架

Yacouba Kaloga, Shashi Kumar, Petr Motlicek, Ina Kodrasi

机构 * Idiap Research Institute(Idiap研究机构) EPFL(瑞士联邦理工学院) BUT(布拉格技术大学)

AI总结 本文提出基于最优传输的可微对齐框架,改进序列到序列模型的对齐性能,同时在ASR性能上有所权衡。

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.16467 2025-11-21 cs.CL cs.AI cs.LG

Anatomy of an Idiom: Tracing Non-Compositionality in Language Models

成语的解剖:追踪语言模型中的非组合性

Andrew Gomes

机构 * Theoretical Particle Physics Laboratory, Institute of Physics, EPFL, Lausanne, Switzerland(理论粒子物理实验室,物理研究所,EPFL,洛桑,瑞士)

AI总结 本文研究了语言模型处理习语的非组合性机制,通过电路发现技术揭示了习语处理中的注意力模式和增强接收现象。

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.16346 2025-11-21 eess.SP cs.LG cs.SY eess.SY

VersaPants: A Loose-Fitting Textile Capacitive Sensing System for Lower-Body Motion Capture

VersaPants:一种用于下肢动作捕捉的宽松织物电容式传感系统

Deniz Kasap, Taraneh Aminosharieh Najafi, Jérôme Paul Rémy Thevenot, Jonathan Dan, Stefano Albini, David Atienza

机构 * École Polytechnique Fédérale de Lausanne (EPFL)(瑞士联邦理工学院洛桑分校)

AI总结 VersaPants是一种基于织物的电容式传感系统,通过轻量级Transformer模型实现下肢动作捕捉,具有高精度和低功耗的特点。

Comments 14 pages, 8 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.16186 2025-11-21 cs.CV

PrIntMesh: Precise Intersection Surfaces for 3D Organ Mesh Reconstruction

PrIntMesh:3D器官网格重建的精确交界面

Deniz Sayin Mercadier, Hieu Le, Yihong Chen, Jiancheng Yang, Udaranga Wickramasinghe, Pascal Fua

机构 * EPFL(瑞士联邦理工学院) UNC Charlotte(北卡罗来纳大学夏洛特分校) ELLIS Institute Finland(芬兰埃利斯研究所) Adis SA(Adis SA公司)

AI总结 PrIntMesh通过统一系统重建方法,实现高精度的3D器官网格重建,有效处理共享界面并保持结构一致性。

Comments 12 pages, 9 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.15854 2025-11-21 cs.LG

discretize_distributions: Efficient Quantization of Gaussian Mixtures with Guarantees in Wasserstein Distance

离散化分布:具有Wasserstein距离保证的高斯混合分布高效量化

Steven Adams, Elize Alwash, Luca Laurenti

机构 * TU Delft(代尔夫特理工大学) EPFL(苏黎世联邦理工学院)

AI总结 discretize_distributions通过高效量化方法和Wasserstein距离保证,实现高斯混合分布的准确离散近似。

详情

展开后加载摘要…

URL PDF HTML 收藏
2506.01497 2025-11-20 cs.NE cs.AR cs.LG

SPICEMixer - Netlist-Level Circuit Evolution

Stefan Uhlich, Andrea Bonetti, Arun Venkitaraman, Chia-Yu Hsieh, Yağız Gençer, Mustafa Emre Gürsoy, Ryoga Matsuo, Lorenzo Servadei

机构 * SonyAI, Switzerland(索尼AI(瑞士)) Sony Semiconductor Solutions Europe, Germany(索尼半导体解决方案欧洲(德国)) EPFL, Switzerland(苏黎世联邦理工学院(瑞士)) TU Munich, Germany(慕尼黑技术大学(德国))

详情

展开后加载摘要…

URL PDF HTML 收藏
2504.13768 2025-11-20 cs.LG cs.CE physics.comp-ph

Equi-Euler GraphNet: An Equivariant, Temporal-Dynamics Informed Graph Neural Network for Dual Force and Trajectory Prediction in Multi-Body Systems

Vinay Sharma, Rémi Tanguy Oddon, Pietro Tesini, Jens Ravesloot, Cees Taal, Olga Fink

机构 * EPFL, Intelligent Maintenance and Operations Systems(苏黎世联邦理工学院,智能维护与操作系统) SKF, Research and Technology Development(SKF,研究与技术开发)

Comments Reuploaded with new version-- equation 16 was incorrect

详情

展开后加载摘要…

URL PDF HTML 收藏
2507.04416 2025-11-20 cs.CL

RAT: Bridging RNN Efficiency and Attention Accuracy via Chunk-based Sequence Modeling

Xiuying Wei, Anunay Yadav, Razvan Pascanu, Caglar Gulcehre

机构 * CLAIRE, EPFL(EPFL的CLAIRE) Google DeepMind(谷歌DeepMind)

Comments Accepted by NeurIPS 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.14569 2025-11-19 cs.LG

Task Addition and Weight Disentanglement in Closed-Vocabulary Models

Adam Hazimeh, Alessandro Favero, Pascal Frossard

机构 * EPFL(瑞士联邦理工学院) LTS4(LTS4实验室) PCSL(PCSL实验室)

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.14358 2025-11-19 eess.SY cs.RO cs.SY

Identifying Time-varying Costs in Finite-horizon Linear Quadratic Gaussian Games

Kai Ren, Maryam Kamgarpour

机构 * SYCAMORE Lab, École Polytechnique Fédérale de Lausanne (EPFL), Switzerland(苏黎世联邦理工学院(EPFL))

详情

展开后加载摘要…

URL PDF HTML 收藏
2506.16029 2025-11-19 cs.CL cs.AI cs.LG

EvoLM: In Search of Lost Language Model Training Dynamics

Zhenting Qi, Fan Nie, Alexandre Alahi, James Zou, Himabindu Lakkaraju, Yilun Du, Eric Xing, Sham Kakade, Hanlin Zhang

机构 * Harvard(哈佛大学) Stanford(斯坦福大学) EPFL(苏黎世联邦理工学院) CMU(卡内基梅隆大学)

Comments NeurIPS 2025 (Oral)

详情

展开后加载摘要…

URL PDF HTML 收藏
2510.04108 2025-11-18 cs.LG cs.NA math.NA math.ST stat.TH

Can Linear Probes Measure LLM Uncertainty?

Ramzi Dakhmouche, Adrien Letellier, Hossein Gorji

机构 * Institute of Mathematics, EPFL, Switzerland(瑞士联邦理工学院数学研究所) Computational Engineering Lab, Empa, Switzerland(瑞士Empa计算工程实验室) ENSAE, IP Paris, France(法国IP巴黎高等商学院)

详情

展开后加载摘要…

URL PDF HTML 收藏
2510.04102 2025-11-18 cs.LG cs.NA math.NA math.PR

Why Cannot Neural Networks Master Extrapolation? Insights from Physical Laws

Ramzi Dakhmouche, Hossein Gorji

机构 * Institute of Mathematics, EPFL, Switzerland(瑞士联邦理工学院数学研究所) Computational Engineering Lab, Empa, Switzerland(瑞士Empa计算工程实验室)

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.09287 2025-11-18 cs.AI cs.CY cs.LG

From Model Training to Model Raising

Roland Aydin, Christian Cyron, Steve Bachelor, Ashton Anderson, Robert West

机构 * Hamburg University of Technology(汉堡理工大学) Helmholtz-Zentrum Hereon(海德堡研究中心) University of Toronto(多伦多大学) EPFL(苏黎世联邦理工学院)

Comments Accepted for publication in Communications of the ACM (CACM), Opinion section

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.12181 2025-11-18 cs.CV cs.LG

MixAR: Mixture Autoregressive Image Generation

Jinyuan Hu, Jiayou Zhang, Shaobo Cui, Kun Zhang, Guangyi Chen

机构 * Tsinghua University(清华大学) Mohamed bin Zayed University of Artificial Intelligence (MBZUAI)(Mohamed bin Zayed人工智能大学) École Polytechnique Fédérale de Lausanne (EPFL)(洛桑联邦理工学院) Carnegie Mellon University(卡内基梅隆大学)

详情

展开后加载摘要…

URL PDF HTML 收藏
2508.07208 2025-11-18 cs.LG cs.AI

What One Cannot, Two Can: Two-Layer Transformers Provably Represent Induction Heads on Any-Order Markov Chains

Chanakya Ekbote, Marco Bondaschi, Nived Rajaraman, Jason D. Lee, Michael Gastpar, Ashok Vardhan Makkuva, Paul Pu Liang

机构 * Massachusetts Institute of Technology(麻省理工学院) École Polytechnique Fédérale de Lausanne(瑞士联邦理工学院) University of California, Berkeley(加州大学伯克利分校) Princeton University(普林斯顿大学)

Comments Accepted at NeurIPS 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.11163 2025-11-17 cs.LG

Training Neural Networks at Any Scale

Thomas Pethick, Kimon Antonakopoulos, Antonio Silveti-Falls, Leena Chennuru Vankadara, Volkan Cevher

机构 * École Polytechnique Fédérale de Lausanne (EPFL), Laboratory for Information and Inference Systems (LIONS)(瑞士联邦理工学院(EPFL)、信息与推断系统实验室)

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.11078 2025-11-17 cs.CV eess.IV eess.SP

SplineSplat: 3D Ray Tracing for Higher-Quality Tomography

Youssef Haouchat, Sepand Kashani, Aleix Boquet-Pujadas, Philippe Thévenaz, Michael Unser

机构 * Biomedical Imaging Group(生物医学成像组) Center for Imaging -- EPFL(成像中心 -- EPFL) Paul Scherrer Institute (PSI)(保罗·施特默研究所(PSI))

详情

展开后加载摘要…

URL PDF HTML 收藏
2509.04966 2025-11-17 cs.LG

Neuro-Spectral Architectures for Causal Physics-Informed Networks

Arthur Bizzi, Leonardo M. Moreira, Márcio Marques, Leonardo Mendonça, Christian Júnior de Oliveira, Vitor Balestro, Lucas dos Santos Fernandez, Daniel Yukimura, Pavel Petrov, João M. Pereira, Tiago Novello, Lucas Nissenbaum

机构 * École Polytechnique Fédérale de Lausanne (EPFL)(联邦理工学院洛桑校区) Instituto de Matemática Pura e Aplicada (IMPA)(数学与应用数学研究所) Universidade do Estado do Rio de Janeiro (UERJ)(里约热内卢州立大学) Laboratório Nacional de Computação Científica (LNCC)(国家科学计算实验室) University of Georgia (UGA)(佐治亚大学)

Comments Accepted at NeurIPS 2025 (poster). 24 pages, 10 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
2502.04339 2025-11-17 math.ST cond-mat.dis-nn cs.IT cs.LG math.IT math.PR stat.TH

Analysis of Diffusion Models for Manifold Data

Anand Jerry George, Rodrigo Veiga, Nicolas Macris

机构 * EPFL, School of Computer and Communication Sciences(苏黎世联邦理工学院计算机与通信科学学院) Department of Electrical Engineering University 1 City 1(电气工程系大学1城市1)

详情

展开后加载摘要…

URL PDF HTML 收藏
2410.20856 2025-11-17 cs.LG cs.AI

Strada-LLM: Graph LLM for traffic prediction

Seyed Mohamad Moghadas, Bruno Cornelis, Alexandre Alahi, Adrian Munteanu

机构 * Department of Electronics and Informatics, Vrije Universiteit Brussel(布鲁塞尔自由大学电子与信息系) VITA Lab, EPFL(EPFL VITA实验室)

Comments 13 pages

详情

展开后加载摘要…

URL PDF HTML 收藏
2503.14121 2025-11-13 stat.ML cond-mat.dis-nn cs.IT cs.LG math.IT math.PR

Fundamental Limits of Matrix Sensing: Exact Asymptotics, Universality, and Applications

Yizhou Xu, Antoine Maillard, Lenka Zdeborová, Florent Krzakala

机构 * Statistical Physics of Computation Laboratory (SPOC), EPFL, Switzerland(统计物理计算实验室(SPOC),瑞士联邦理工学院) Inria Paris, DI ENS, PSL University, Paris, France(巴黎Inria,ENS DI,巴黎大学,法国) Information, Learning, and Physics Laboratory (IDEPHICS), EPFL, Switzerland(信息、学习和物理实验室(IDEPHICS),瑞士联邦理工学院)

Journal ref COLT 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2502.14389 2025-11-13 cs.CL cs.HC

Leveraging Small LLMs for Argument Mining in Education: Argument Component Identification, Classification, and Assessment

Lucile Favero, Juan Antonio Pérez-Ortiz, Tanja Käser, Nuria Oliver

机构 * ELLIS Alicante, Spain(阿利坎特ELLIS研究所,西班牙) Universitat d’Alacant, Spain(阿利坎特大学,西班牙) École Polytechnique Fédérale de Lausanne, EPFL, Switzerland(洛桑联邦理工学院,瑞士)

详情

展开后加载摘要…

URL PDF HTML 收藏
2502.00901 2025-11-13 cs.LG cond-mat.dis-nn

Fundamental limits of learning in sequence multi-index models and deep attention networks: High-dimensional asymptotics and sharp thresholds

Emanuele Troiani, Hugo Cui, Yatin Dandi, Florent Krzakala, Lenka Zdeborová

机构 * Emanuele Troiani Statistical Physics Of Computation Laboratory, École Polytechnique Fédérale de Lausanne (EPFL)(埃因斯坦理工大学统计物理计算实验室) Hugo Cui Center of Mathematical Sciences and Applications, Harvard University(哈佛大学数学科学中心) Yatin Dandi Statistical Physics Of Computation Laboratory, École Polytechnique Fédérale de Lausanne (EPFL) Information Learning and Physics Laboratory, École Polytechnique Fédérale de Lausanne (EPFL)(埃因斯坦理工大学统计物理计算实验室) Florent Krzakala Information Learning and Physics Laboratory, École Polytechnique Fédérale de Lausanne (EPFL)(埃因斯坦理工大学信息学习与物理实验室) Lenka Zdeborová Statistical Physics Of Computation Laboratory, École Polytechnique Fédérale de Lausanne (EPFL)(埃因斯坦理工大学统计物理计算实验室)

Journal ref ICML 2025

详情

展开后加载摘要…

URL PDF HTML 收藏