Adversarial Lipschitz Regularization

TL;DR

This paper introduces Adversarial Lipschitz Regularization, a new method for enforcing the Lipschitz constraint in Wasserstein GANs through explicit adversarial penalties, improving training stability and sample quality.

Contribution

It proposes a novel explicit Lipschitz regularization technique inspired by Virtual Adversarial Training, demonstrating its effectiveness in Wasserstein GANs.

Findings

Explicit Lipschitz penalty is viable and effective.

The method achieves competitive performance with existing approaches.

Highlights the connection between Lipschitz regularization and adversarial training.

Abstract

Generative adversarial networks (GANs) are one of the most popular approaches when it comes to training generative models, among which variants of Wasserstein GANs are considered superior to the standard GAN formulation in terms of learning stability and sample quality. However, Wasserstein GANs require the critic to be 1-Lipschitz, which is often enforced implicitly by penalizing the norm of its gradient, or by globally restricting its Lipschitz constant via weight normalization techniques. Training with a regularization term penalizing the violation of the Lipschitz constraint explicitly, instead of through the norm of the gradient, was found to be practically infeasible in most situations. Inspired by Virtual Adversarial Training, we propose a method called Adversarial Lipschitz Regularization, and show that using an explicit Lipschitz penalty is indeed viable and leads to competitive performance when applied to Wasserstein GANs, highlighting an important connection between Lipschitz regularization and adversarial training.