Provably Robust Adversarial Examples

Dimitar I. Dimitrov, Gagandeep Singh, Timon Gehr +1

We introduce the concept of provably robust adversarial examples for deep neural networks - connected input regions constructed from standard adversarial examples which are guarant…