Matrix-Free Two-to-Infinity and One-to-Two Norms Estimation

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Tsyganov, Askar, Frolov, Evgeny, Samsonov, Sergey, Rakhuba, Maxim
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866915736502403072
author Tsyganov, Askar
Frolov, Evgeny
Samsonov, Sergey
Rakhuba, Maxim
author_facet Tsyganov, Askar
Frolov, Evgeny
Samsonov, Sergey
Rakhuba, Maxim
contents In this paper, we propose new randomized algorithms for estimating the two-to-infinity and one-to-two norms in a matrix-free setting, using only matrix-vector multiplications. Our methods are based on appropriate modifications of Hutchinson's diagonal estimator and its Hutch++ version. We provide oracle complexity bounds for both modifications. We further illustrate the practical utility of our algorithms for Jacobian-based regularization in deep neural network training on image classification tasks. We also demonstrate that our methodology can be applied to mitigate the effect of adversarial attacks in the domain of recommender systems.
format Preprint
id arxiv_https___arxiv_org_abs_2508_04444
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Matrix-Free Two-to-Infinity and One-to-Two Norms Estimation
Tsyganov, Askar
Frolov, Evgeny
Samsonov, Sergey
Rakhuba, Maxim
Machine Learning
Numerical Analysis
65F35
In this paper, we propose new randomized algorithms for estimating the two-to-infinity and one-to-two norms in a matrix-free setting, using only matrix-vector multiplications. Our methods are based on appropriate modifications of Hutchinson's diagonal estimator and its Hutch++ version. We provide oracle complexity bounds for both modifications. We further illustrate the practical utility of our algorithms for Jacobian-based regularization in deep neural network training on image classification tasks. We also demonstrate that our methodology can be applied to mitigate the effect of adversarial attacks in the domain of recommender systems.
title Matrix-Free Two-to-Infinity and One-to-Two Norms Estimation
topic Machine Learning
Numerical Analysis
65F35
url https://arxiv.org/abs/2508.04444