Analyzing the Components of Distributed Coevolutionary GAN Training

TL;DR

This paper investigates how diversity maintenance mechanisms in distributed coevolutionary GAN training, such as selection and migration, affect the quality of generated images, demonstrating that combined strategies yield the best results.

Contribution

It analyzes the impact of selection and migration components on diversity and performance in distributed coevolutionary GAN training, highlighting their combined effectiveness.

Findings

Combined selection and migration improve GAN quality.

Migration alone achieves competitive results.

Selection without migration reduces performance.

Abstract

Distributed coevolutionary Generative Adversarial Network (GAN) training has empirically shown success in overcoming GAN training pathologies. This is mainly due to diversity maintenance in the populations of generators and discriminators during the training process. The method studied here coevolves sub-populations on each cell of a spatial grid organized into overlapping Moore neighborhoods. We investigate the impact on the performance of two algorithm components that influence the diversity during coevolution: the performance-based selection/replacement inside each sub-population and the communication through migration of solutions (networks) among overlapping neighborhoods. In experiments on MNIST dataset, we find that the combination of these two components provides the best generative models. In addition, migrating solutions without applying selection in the sub-populations achieves competitive results, while selection without communication between cells reduces performance.