Linking Generative Adversarial Learning and Binary Classification

TL;DR

This paper reveals a fundamental connection between generative adversarial training and binary classification, showing that discriminators in GANs effectively measure divergence between real and generated data, offering new insights into training methods.

Contribution

It establishes a basic link between GAN training and binary classification, suggesting discriminator loss functions can be designed based on divergence measures.

Findings

Discriminators compute an f-divergence between data distributions.

Re-derivation in decision theory supports the link.

Implications for designing discriminator loss functions in GANs.

Abstract

In this note, we point out a basic link between generative adversarial (GA) training and binary classification -- any powerful discriminator essentially computes an (f-)divergence between real and generated samples. The result, repeatedly re-derived in decision theory, has implications for GA Networks (GANs), providing an alternative perspective on training f-GANs by designing the discriminator loss function.