2020/12/01 by Geoffrey Chinot, Chinot, Geoffrey, Matthias Löffler +3 · 1 citation
Engineering · Mathematics · #62J05 #Applied mathematics #Artificial intelligence #Combinatorics #Computer science #Covariate #Estimator #FOS: Computer and information sciences #FOS: Mathematics #Gaussian #Information Theory (cs.IT) #Logarithm #Machine Learning (stat.ML) #Mathematical analysis #Mathematical optimization #Mathematics #Norm (philosophy) #Numerical Analysis (math.NA) #Rank (graph theory) #Regularization (linguistics) #Robustness (evolution) #Sparse and Compressive Sensing Techniques #Statistical Methods and Inference #Statistics #Statistics Theory (math.ST) #Upper and lower bounds
paper · pdf · doi:10.48550/arxiv.2012.00807
openalex publication_date 2020/12/01 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/05
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 ℓ1, ℓ2, 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.