Can overfitted deep neural networks in adversarial training generalize? – An approximation viewpoint

Overview

In this talk, I will discuss whether overfitted DNNs in adversarial training can generalize from an approximation viewpoint. We prove by construction the existence of infinitely many adversarial training classifiers on over-parameterized DNNs that obtain arbitrarily small adversarial training error (overfitting), whereas achieving good robust generalization error under certain conditions concerning the data quality, well separated, and perturbation level. This construction is optimal and thus points out the fundamental limits of DNNs under adversarial training with statistical guarantees. Part of this talk comes from our recent work.

Details

Status

Released

Release date

March 1, 2024

Original language

English

Countries

United Kingdom

Production

University of Warwick

Budget

Not recorded

Revenue

Not recorded

Homepage

Official site