ICLR16: DeepCloak: Masking Deep Neural Network Models for Robustness Against Adversarial Samples - View it on GitHub