On the robustness of minimum norm interpolators and regularized empirical risk minimizers

TL;DR

This paper develops a comprehensive theory for minimum norm interpolators and regularized empirical risk minimizers in linear models, providing bounds on prediction error that account for adversarial errors and various norms.

Contribution

It introduces a general framework for analyzing the robustness of minimum norm interpolators and RERM without error restrictions, with concrete bounds and optimality results.

Findings

Prediction error is bounded by Rademacher complexity and noise norms.

Minimum norm interpolators perform well under overparameterization and low noise levels.

Theoretical bounds are validated for Gaussian features and multiple norms.

Abstract

This article develops a general theory for minimum norm interpolating estimators and regularized empirical risk minimizers (RERM) in linear models in the presence of additive, potentially adversarial, errors. In particular, no conditions on the errors are imposed. A quantitative bound for the prediction error is given, relating it to the Rademacher complexity of the covariates, the norm of the minimum norm interpolator of the errors and the size of the subdifferential around the true parameter. The general theory is illustrated for Gaussian features and several norms: The $\ell_1$, $\ell_2$, group Lasso and nuclear norms. In case of sparsity or low-rank inducing norms, minimum norm interpolators and RERM yield a prediction error of the order of the average noise level, provided that the overparameterization is at least a logarithmic factor larger than the number of samples and that, in case of RERM, the regularization parameter is small enough. Lower bounds that show near optimality of the results complement the analysis.