Improving the Adversarial Robustness and Interpretability of Deep Neural Networks by Regularizing Their Input Gradients.

Andrew Slavin Ross, Finale Doshi-Velez

Browse the full AAAI paper archive.