Interpolation can hurt robust generalization even when there is no noise
Machine Learning
2021-12-20 v2 Machine Learning
Abstract
Numerous recent works show that overparameterization implicitly reduces variance for min-norm interpolators and max-margin classifiers. These findings suggest that ridge regularization has vanishing benefits in high dimensions. We challenge this narrative by showing that, even in the absence of noise, avoiding interpolation through ridge regularization can significantly improve generalization. We prove this phenomenon for the robust risk of both linear regression and classification and hence provide the first theoretical result on robust overfitting.
Keywords
Cite
@article{arxiv.2108.02883,
title = {Interpolation can hurt robust generalization even when there is no noise},
author = {Konstantin Donhauser and Alexandru Ţifrea and Michael Aerni and Reinhard Heckel and Fanny Yang},
journal= {arXiv preprint arXiv:2108.02883},
year = {2021}
}