Highly Adaptive Principal Component Regression

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Wang, Mingxun, Schuler, Alejandro, van der Laan, Mark, Meixide, Carlos García
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866917460035239936
author Wang, Mingxun
Schuler, Alejandro
van der Laan, Mark
Meixide, Carlos García
author_facet Wang, Mingxun
Schuler, Alejandro
van der Laan, Mark
Meixide, Carlos García
contents The Highly Adaptive Lasso (HAL) is a nonparametric regression method that achieves almost dimension-free convergence rates under minimal smoothness assumptions, but its implementation can be computationally prohibitive in high dimensions due to the large design matrix it requires. The Highly Adaptive Ridge (HAR) has been proposed as a related ridge-regularized analogue. Building on both procedures, we introduce the Principal Component Highly Adaptive Lasso (PCHAL) and Principal Component Highly Adaptive Ridge (PCHAR). These estimators use an outcome-blind principal-component reduction of the HAL basis, offering substantial computational gains over HAL while achieving empirical performance comparable to HAL and HAR. We also describe an early-stopped gradient descent variant, which provides a convenient form of smooth spectral regularization without explicitly selecting a hard principal-component cutoff. Finally, we uncover that under special circumstances, the HAL kernel is identical to the covariance function of Brownian motion.
format Preprint
id arxiv_https___arxiv_org_abs_2602_10613
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Highly Adaptive Principal Component Regression
Wang, Mingxun
Schuler, Alejandro
van der Laan, Mark
Meixide, Carlos García
Machine Learning
The Highly Adaptive Lasso (HAL) is a nonparametric regression method that achieves almost dimension-free convergence rates under minimal smoothness assumptions, but its implementation can be computationally prohibitive in high dimensions due to the large design matrix it requires. The Highly Adaptive Ridge (HAR) has been proposed as a related ridge-regularized analogue. Building on both procedures, we introduce the Principal Component Highly Adaptive Lasso (PCHAL) and Principal Component Highly Adaptive Ridge (PCHAR). These estimators use an outcome-blind principal-component reduction of the HAL basis, offering substantial computational gains over HAL while achieving empirical performance comparable to HAL and HAR. We also describe an early-stopped gradient descent variant, which provides a convenient form of smooth spectral regularization without explicitly selecting a hard principal-component cutoff. Finally, we uncover that under special circumstances, the HAL kernel is identical to the covariance function of Brownian motion.
title Highly Adaptive Principal Component Regression
topic Machine Learning
url https://arxiv.org/abs/2602.10613