Adversarially Robust Learning via Entropic Regularization