Distributionally and Adversarially Robust Logistic Regression via Intersecting Wasserstein Balls