Salvato in:
Dettagli Bibliografici
Autori principali: Zhang, Yikai, Lin, Jiahe, Li, Fengpei, Zheng, Songzhu, Raj, Anant, Schneider, Anderson, Nevmyvaka, Yuriy
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:https://arxiv.org/abs/2501.02353
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866912177574641664
author Zhang, Yikai
Lin, Jiahe
Li, Fengpei
Zheng, Songzhu
Raj, Anant
Schneider, Anderson
Nevmyvaka, Yuriy
author_facet Zhang, Yikai
Lin, Jiahe
Li, Fengpei
Zheng, Songzhu
Raj, Anant
Schneider, Anderson
Nevmyvaka, Yuriy
contents In this work, we study the weighted empirical risk minimization (weighted ERM) schema, in which an additional data-dependent weight function is incorporated when the empirical risk function is being minimized. We show that under a general ``balanceable" Bernstein condition, one can design a weighted ERM estimator to achieve superior performance in certain sub-regions over the one obtained from standard ERM, and the superiority manifests itself through a data-dependent constant term in the error bound. These sub-regions correspond to large-margin ones in classification settings and low-variance ones in heteroscedastic regression settings, respectively. Our findings are supported by evidence from synthetic data experiments.
format Preprint
id arxiv_https___arxiv_org_abs_2501_02353
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Reweighting Improves Conditional Risk Bounds
Zhang, Yikai
Lin, Jiahe
Li, Fengpei
Zheng, Songzhu
Raj, Anant
Schneider, Anderson
Nevmyvaka, Yuriy
Machine Learning
G.3; I.3
In this work, we study the weighted empirical risk minimization (weighted ERM) schema, in which an additional data-dependent weight function is incorporated when the empirical risk function is being minimized. We show that under a general ``balanceable" Bernstein condition, one can design a weighted ERM estimator to achieve superior performance in certain sub-regions over the one obtained from standard ERM, and the superiority manifests itself through a data-dependent constant term in the error bound. These sub-regions correspond to large-margin ones in classification settings and low-variance ones in heteroscedastic regression settings, respectively. Our findings are supported by evidence from synthetic data experiments.
title Reweighting Improves Conditional Risk Bounds
topic Machine Learning
G.3; I.3
url https://arxiv.org/abs/2501.02353