About exchanging expectation and supremum for conditional Wasserstein GANs

TL;DR

This paper provides a mathematical justification for exchanging supremum and expectation in conditional Wasserstein GANs, especially when discriminators are only partially Lipschitz-1, addressing a gap in the theoretical understanding.

Contribution

It offers a rigorous mathematical rationale for the exchange of supremum and expectation in conditional Wasserstein GANs with partially Lipschitz-1 discriminators.

Findings

Justifies the exchange of supremum and expectation mathematically.

Clarifies conditions under which weaker Lipschitz constraints are valid.

Supports the use of partially Lipschitz-1 discriminators in practice.

Abstract

In cases where a Wasserstein GAN depends on a condition the latter is usually handled via an expectation within the loss function. Depending on the way this is motivated, the discriminator is either required to be Lipschitz-1 in both or in only one of its arguments. For the weaker requirement to become usable one needs to exchange a supremum and an expectation. This is a mathematically perilous operation, which is, so far, only partially justified in the literature. This short mathematical note intends to fill this gap and provides the mathematical rationale for discriminators that are only partially Lipschitz-1 for cases where this approach is more appropriate or successful.