While adversarial training methods have resulted in significant improvements in the deep
neural nets' robustness against norm-bounded adversarial perturbations, their …