Abstract: Despite the tremendous success of deep neural networks in various learning problems, it has been observed that adding intentionally designed adversarial perturbations to inputs of these ...