Wasserstein Distributional Robustness and Regularization in Statistical Learning