Saved in:
Bibliographic Details
Main Authors: Stark, Thomas, Steinberger, Lukas
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2502.10578
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918303978487808
author Stark, Thomas
Steinberger, Lukas
author_facet Stark, Thomas
Steinberger, Lukas
contents In this paper we investigate the generalization error of gradient descent (GD) applied to an $\ell_2$-regularized OLS objective function in the linear model. Based on our analysis we develop new methodology for computationally tractable and statistically efficient linear prediction in a high-dimensional and massive data scenario (large-$n$, large-$p$). Our results are based on the surprising observation that the generalization error of optimally tuned regularized gradient descent approaches that of an optimal benchmark procedure $monotonically$ in the iteration number $t$. On the other hand standard GD for OLS (without explicit regularization) can achieve the benchmark only in degenerate cases. This shows that (optimal) explicit regularization can be nearly statistically efficient (for large $t$) whereas implicit regularization by (optimal) early stopping can not. To complete our methodology, we provide a fully data driven and computationally tractable choice of the $\ell_2$ regularization parameter $λ$ that is computationally cheaper than cross-validation. On this way, we follow and extend ideas of Dicker (2014) to the non-gaussian case, which requires new results on high-dimensional sample covariance matrices that might be of independent interest.
format Preprint
id arxiv_https___arxiv_org_abs_2502_10578
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Implicit vs. explicit regularization for high-dimensional gradient descent
Stark, Thomas
Steinberger, Lukas
Statistics Theory
62J07, 62-08 (Primary) (Secondary) 60B20
In this paper we investigate the generalization error of gradient descent (GD) applied to an $\ell_2$-regularized OLS objective function in the linear model. Based on our analysis we develop new methodology for computationally tractable and statistically efficient linear prediction in a high-dimensional and massive data scenario (large-$n$, large-$p$). Our results are based on the surprising observation that the generalization error of optimally tuned regularized gradient descent approaches that of an optimal benchmark procedure $monotonically$ in the iteration number $t$. On the other hand standard GD for OLS (without explicit regularization) can achieve the benchmark only in degenerate cases. This shows that (optimal) explicit regularization can be nearly statistically efficient (for large $t$) whereas implicit regularization by (optimal) early stopping can not. To complete our methodology, we provide a fully data driven and computationally tractable choice of the $\ell_2$ regularization parameter $λ$ that is computationally cheaper than cross-validation. On this way, we follow and extend ideas of Dicker (2014) to the non-gaussian case, which requires new results on high-dimensional sample covariance matrices that might be of independent interest.
title Implicit vs. explicit regularization for high-dimensional gradient descent
topic Statistics Theory
62J07, 62-08 (Primary) (Secondary) 60B20
url https://arxiv.org/abs/2502.10578