Ross, A.; Doshi-Velez, F. Improving the Adversarial Robustness and Interpretability of Deep Neural Networks by Regularizing Their Input Gradients. AAAI 2018, 32.