Improving Generalization of Universal Adversarial Perturbation via Dynamic Maximin Optimization

Open in new window