Generalization and Equilibrium in Generative Adversarial Nets (GANs)

TL;DR

This paper investigates the generalization properties of GANs, revealing that while they may not generalize well in standard metrics, they do in neural net distance, and introduces MIX+GAN to improve training stability.

Contribution

It demonstrates the existence of approximate pure equilibria in GAN training for certain generator classes and proposes MIX+GAN to enhance existing GAN training methods.

Findings

GAN training may not generalize well in standard metrics.

Generalization occurs in neural net distance metric.

MIX+GAN improves stability and performance of GAN training.

Abstract

We show that training of generative adversarial network (GAN) may not have good generalization properties; e.g., training may appear successful but the trained distribution may be far from target distribution in standard metrics. However, generalization does occur for a weaker metric called neural net distance. It is also shown that an approximate pure equilibrium exists in the discriminator/generator game for a special class of generators with natural training objectives when generator capacity and training set sizes are moderate. This existence of equilibrium inspires MIX+GAN protocol, which can be combined with any existing GAN training, and empirically shown to improve some of them.