Robust Generalization

Build upon the notion of adversarial certainty, we develop a general training method to generate adversarial examples with reduced certainty for improving robust generalization.