Universality of empirical risk minimization

TL;DR

This paper demonstrates that in high-dimensional supervised learning, the training and test errors exhibit universality, depending only on the covariance structure of features, regardless of their specific distribution, including random features and neural tangent models.

Contribution

It establishes universality results for empirical risk minimization in high dimensions without requiring strong convexity or independence assumptions on features.

Findings

Training error depends only on feature covariance structure.

Test error of near-minimizers shows similar universality.

Results apply to random features and neural tangent models.

Abstract

Consider supervised learning from i.i.d. samples $\{{\boldsymbol x}_i,y_i\}_{i\le n}$ where ${\boldsymbol x}_i \in\mathbb{R}^p$ are feature vectors and ${y} \in \mathbb{R}$ are labels. We study empirical risk minimization over a class of functions that are parameterized by $\mathsf{k} = O(1)$ vectors ${\boldsymbol \theta}_1, . . . , {\boldsymbol \theta}_{\mathsf k} \in \mathbb{R}^p$ , and prove universality results both for the training and test error. Namely, under the proportional asymptotics $n,p\to\infty$, with $n/p = \Theta(1)$, we prove that the training error depends on the random features distribution only through its covariance structure. Further, we prove that the minimum test error over near-empirical risk minimizers enjoys similar universality properties. In particular, the asymptotics of these quantities can be computed $-$to leading order$-$ under a simpler model in which the feature vectors ${\boldsymbol x}_i$ are replaced by Gaussian vectors ${\boldsymbol g}_i$ with the same covariance. Earlier universality results were limited to strongly convex learning procedures, or to feature vectors ${\boldsymbol x}_i$ with independent entries. Our results do not make any of these assumptions. Our assumptions are general enough to include feature vectors ${\boldsymbol x}_i$ that are produced by randomized featurization maps. In particular we explicitly check the assumptions for certain random features models (computing the output of a one-layer neural network with random weights) and neural tangent models (first-order Taylor approximation of two-layer networks).