Wasserstein GAN

TL;DR

This paper introduces WGAN, a novel GAN training method that enhances stability, eliminates mode collapse, and offers meaningful learning metrics, supported by theoretical insights into distribution distances.

Contribution

WGAN presents a new training algorithm for GANs that improves stability and theoretical understanding, addressing key issues like mode collapse.

Findings

Enhanced training stability in GANs

Elimination of mode collapse

Theoretical connection to distribution distances

Abstract

We introduce a new algorithm named WGAN, an alternative to traditional GAN training. In this new model, we show that we can improve the stability of learning, get rid of problems like mode collapse, and provide meaningful learning curves useful for debugging and hyperparameter searches. Furthermore, we show that the corresponding optimization problem is sound, and provide extensive theoretical work highlighting the deep connections to other distances between distributions.