Characterizing the Generalization Error of Random Feature Regression with Arbitrary Data-Augmentation

Lucas Morisset, Alain Durmus, Adrien Hardy · 2026

Read Characterizing the Generalization Error of Random Feature Regression with Arbitrary Data-Augmentation by Lucas Morisset, Alain Durmus, Adrien Hardy (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-10290.txt.

Description

Lucas Morisset, Alain Durmus, and Adrien Hardy analyze how data augmentation regularizes supervised regression. Augmentation, such as rotating images to teach invariance, is routine in modern machine-learning pipelines, yet its regularization effect lacked a tight theoretical account. Their 2026 paper studies the proportional regime, where the number of covariates grows in proportion to the number of samples.

The authors give a tight account of test error, measured in mean squared error. It uses only population quantities of the true data plus first- and second-order statistics of the augmentation scheme. Results hold under misspecified feature maps and for any net architecture that trains only the last readout layer, with the rest frozen or randomly initialized. Specified for Gaussian data, the asymptotic account proves tight, giving practitioners a precise formula for how augmentation choices shape generalization.

Catalog record

CREATORLucas Morisset, Alain Durmus, Adrien Hardy
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE213 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.10290

Browse the full library · About the Tower of Babel