What is Wasserstein loss and why is it important in generative models?

Wasserstein loss measures optimal transport distance between probability distributions.

Wasserstein distance defines the minimum cost required to transform one probability distribution into another. In generative models like Wasserstein GANs, this replaces divergence-based losses such as KL or Jensen-Shannon divergence. It provides smoother gradients even when distributions do not overlap, solving a key instability problem in GAN training.