Differentiable Abstract Interpretation for Provably Robust Neural Networks

[edit]

Matthew Mirman, Timon Gehr, Martin Vechev ;
Proceedings of the 35th International Conference on Machine Learning, PMLR 80:3578-3586, 2018.

Abstract

We introduce a scalable method for training robust neural networks based on abstract interpretation. We present several abstract transformers which balance efficiency with precision and show these can be used to train large neural networks that are certifiably robust to adversarial perturbations.

Related Material