Skip to yearly menu bar Skip to main content

Workshop: A Blessing in Disguise: The Prospects and Perils of Adversarial Machine Learning

Maximizing the robust margin provably overfits on noiseless data

Fanny Yang · Reinhard Heckel · Michael Aerni · Alexandru Tifrea · Konstantin Donhauser


Numerous recent works show that overparameterization implicitly reduces variance, suggesting vanishing benefits for explicit regularization in high dimensions. However, this narrative has been challenged by empirical observations indicating that adversarially trained deep neural networks suffer from robust overfitting. While existing explanations attribute this phenomenon to noise or problematic samples in the training data set, we prove that even on entirely noiseless data, achieving a vanishing adversarial logistic training loss is suboptimal compared to regularized counterparts.

Chat is not available.