Timezone: »

 
Poster
Interpolation can hurt robust generalization even when there is no noise
Konstantin Donhauser · Alexandru Tifrea · Michael Aerni · Reinhard Heckel · Fanny Yang

Tue Dec 07 08:30 AM -- 10:00 AM (PST) @
in Poster Session 1 »

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 \emph{robust overfitting}.

Keywords: Theory · Machine Learning · Robustness

Author Information

Konstantin Donhauser (ETH Zurich)
Alexandru Tifrea (Swiss Federal Institute of Technology)
Michael Aerni (Swiss Federal Institute of Technology)
Reinhard Heckel (TUM)
Fanny Yang (ETH Zurich)

More from the Same Authors