Provably Robust Deep Learning via Adversarially Trained Smoothed Classifiers

Recent works have shown the effectiveness of randomized smoothing as a scalable technique for building neural network-based classifiers that are provably robust to ℓ₂-norm adversarial perturbations. In this paper, we employ adversarial training to improve the performance of randomized smoothing. We design an adapted attack for smoothed classifiers, and we show how this attack can be used in an adversarial training setting to boost the provable robustness of smoothed classifiers. We demonstrate through extensive experimentation that our method consistently outperforms all existing provably ℓ2-robust classifiers by a significant margin on ImageNet and CIFAR-10, establishing the state-of-the-art for provable ℓ2-defenses. Moreover, we find that pre-training and semi-supervised learning boost adversarially trained smoothed classifiers even further. Our code and trained models are available on github (opens in new tab).