Kshitij Chandna: Improving Adversarial Robustness by Penalizing Natural Accuracy. ECCV Workshops (1) 2022: 517-533