English

Universality of max-margin classifiers

Statistics Theory 2023-10-03 v1 Machine Learning Statistics Theory

Abstract

Maximum margin binary classification is one of the most fundamental algorithms in machine learning, yet the role of featurization maps and the high-dimensional asymptotics of the misclassification error for non-Gaussian features are still poorly understood. We consider settings in which we observe binary labels yiy_i and either dd-dimensional covariates zi{\boldsymbol z}_i that are mapped to a pp-dimension space via a randomized featurization map ϕ:RdRp{\boldsymbol \phi}:\mathbb{R}^d \to\mathbb{R}^p, or pp-dimensional features of non-Gaussian independent entries. In this context, we study two fundamental questions: (i)(i) At what overparametrization ratio p/np/n do the data become linearly separable? (ii)(ii) What is the generalization error of the max-margin classifier? Working in the high-dimensional regime in which the number of features pp, the number of samples nn and the input dimension dd (in the nonlinear featurization setting) diverge, with ratios of order one, we prove a universality result establishing that the asymptotic behavior is completely determined by the expected covariance of feature vectors and by the covariance between features and labels. In particular, the overparametrization threshold and generalization error can be computed within a simpler Gaussian model. The main technical challenge lies in the fact that max-margin is not the maximizer (or minimizer) of an empirical average, but the maximizer of a minimum over the samples. We address this by representing the classifier as an average over support vectors. Crucially, we find that in high dimensions, the support vector count is proportional to the number of samples, which ultimately yields universality.

Keywords

Cite

@article{arxiv.2310.00176,
  title  = {Universality of max-margin classifiers},
  author = {Andrea Montanari and Feng Ruan and Basil Saeed and Youngtak Sohn},
  journal= {arXiv preprint arXiv:2310.00176},
  year   = {2023}
}

Comments

60 pages

R2 v1 2026-06-28T12:36:47.951Z