Generative Adversarial Network Training is a Continual Learning Problem

TL;DR

This paper frames GAN training as a continual learning problem, proposing methods to mitigate discriminator forgetting, which improves training stability and sample quality with minimal additional computation.

Contribution

It introduces the idea of using continual learning techniques to address GAN training challenges and demonstrates their effectiveness in improving performance.

Findings

Enhanced discriminator memory preserves previous generator samples.

Improved GAN training stability and sample diversity.

Minimal computational overhead for the proposed methods.

Abstract

Generative Adversarial Networks (GANs) have proven to be a powerful framework for learning to draw samples from complex distributions. However, GANs are also notoriously difficult to train, with mode collapse and oscillations a common problem. We hypothesize that this is at least in part due to the evolution of the generator distribution and the catastrophic forgetting tendency of neural networks, which leads to the discriminator losing the ability to remember synthesized samples from previous instantiations of the generator. Recognizing this, our contributions are twofold. First, we show that GAN training makes for a more interesting and realistic benchmark for continual learning methods evaluation than some of the more canonical datasets. Second, we propose leveraging continual learning techniques to augment the discriminator, preserving its ability to recognize previous generator samples. We show that the resulting methods add only a light amount of computation, involve minimal changes to the model, and result in better overall performance on the examined image and text generation tasks.