Wasserstein GANs Work Because They Fail - to Approximate the Wasserstein Distance

Explore a thought-provoking lecture on the counterintuitive success of Wasserstein GANs in machine learning. Delve into Jan Stanczuk's analysis from the University of Cambridge, presented at the Simons Institute, which challenges conventional understanding of these generative models. Discover why Wasserstein GANs are effective despite their failure to accurately approximate the Wasserstein distance. Gain insights into the dynamics and discretization of PDEs, sampling, and optimization in this 33-minute talk that offers a fresh perspective on the underlying mechanisms of GANs in the field of artificial intelligence and computational mathematics.