Adversarial Training and Robustness for Multiple Perturbations

For other perturbations, these defenses offer no guarantees and, at times, even increase the model's vulnerability.