Fostering Diversity in Spatial Evolutionary Generative Adversarial Networks

TL;DR

This paper introduces Mustangs, a spatially distributed CoE-GAN that enhances diversity during training by employing different loss functions, leading to more accurate generators and addressing common GAN training issues.

Contribution

The paper presents Mustangs, a novel spatially distributed CoE-GAN that improves diversity and training stability through varied loss functions, outperforming traditional methods.

Findings

Mustangs trains statistically more accurate generators on MNIST and CelebA.

Using different loss functions fosters diversity and stability in GAN training.

Mustangs reduces mode collapse and training instability.

Abstract

Generative adversary networks (GANs) suffer from training pathologies such as instability and mode collapse, which mainly arise from a lack of diversity in their adversarial interactions. Co-evolutionary GAN (CoE-GAN) training algorithms have shown to be resilient to these pathologies. This article introduces Mustangs, a spatially distributed CoE-GAN, which fosters diversity by using different loss functions during the training. Experimental analysis on MNIST and CelebA demonstrated that Mustangs trains statistically more accurate generators.