A. Ross and F. Doshi-Velez, “Improving the Adversarial Robustness and Interpretability of Deep Neural Networks by Regularizing Their Input Gradients”, AAAI, vol. 32, no. 1, Apr. 2018.