Regularized least squares learning with heavy-tailed noise is minimax optimal

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Mollenhauer, Mattes, Mücke, Nicole, Meunier, Dimitri, Gretton, Arthur
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866914140925198336
author Mollenhauer, Mattes
Mücke, Nicole
Meunier, Dimitri
Gretton, Arthur
author_facet Mollenhauer, Mattes
Mücke, Nicole
Meunier, Dimitri
Gretton, Arthur
contents This paper examines the performance of ridge regression in reproducing kernel Hilbert spaces in the presence of noise that exhibits a finite number of higher moments. We establish excess risk bounds consisting of subgaussian and polynomial terms based on the well known integral operator framework. The dominant subgaussian component allows to achieve convergence rates that have previously only been derived under subexponential noise - a prevalent assumption in related work from the last two decades. These rates are optimal under standard eigenvalue decay conditions, demonstrating the asymptotic robustness of regularized least squares against heavy-tailed noise. Our derivations are based on a Fuk-Nagaev inequality for Hilbert-space valued random variables.
format Preprint
id arxiv_https___arxiv_org_abs_2505_14214
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Regularized least squares learning with heavy-tailed noise is minimax optimal
Mollenhauer, Mattes
Mücke, Nicole
Meunier, Dimitri
Gretton, Arthur
Machine Learning
Statistics Theory
62G08 (Primary) 62G35, 62J07 (Secondary)
This paper examines the performance of ridge regression in reproducing kernel Hilbert spaces in the presence of noise that exhibits a finite number of higher moments. We establish excess risk bounds consisting of subgaussian and polynomial terms based on the well known integral operator framework. The dominant subgaussian component allows to achieve convergence rates that have previously only been derived under subexponential noise - a prevalent assumption in related work from the last two decades. These rates are optimal under standard eigenvalue decay conditions, demonstrating the asymptotic robustness of regularized least squares against heavy-tailed noise. Our derivations are based on a Fuk-Nagaev inequality for Hilbert-space valued random variables.
title Regularized least squares learning with heavy-tailed noise is minimax optimal
topic Machine Learning
Statistics Theory
62G08 (Primary) 62G35, 62J07 (Secondary)
url https://arxiv.org/abs/2505.14214