Provably robust deep generative models

Filipe Condessa, Zico Kolter

Recent work in adversarial attacks has developed provably robust methods for training deep neural network classifiers. However, although they are often mentioned in the context of…