theory

Surprising benefits of ridge regularization for noiseless regression
Numerous recent works show that overparameterization implicitly reduces variance for minimum-norm interpolators, suggesting vanishing …
Konstantin Donhauser, Alexandru Țifrea, Michael Aerni, Reinhard Heckel, Fanny Yang
PDF Cite Code
Maximizing the robust margin provably overfits on noiseless data
We show that the robust max-margin solution overfits even on noiseless data yielding a worse performance than ridge regularized and early stopped robust logistic regression.
Konstantin Donhauser, Alexandru Țifrea, Michael Aerni, Reinhard Heckel, Fanny Yang
PDF Cite Code