Optimal transport based distance between probability distributions; stabilizes GAN training and evaluates distributional shifts more robustly than JS divergence.