Universality of empirical risk minimization

Consider supervised learning from i.i.d. samples ${{\boldsymbol x}i,yi}{i\le n}$ 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).