Dropout-GAN: Learning from a Dynamic Ensemble of Discriminators

TL;DR

Dropout-GAN introduces a dynamic ensemble of discriminators by randomly dropping them during training, which enhances generator diversity, prevents mode collapse, and stabilizes GAN training.

Contribution

This paper presents Dropout-GAN, a novel framework that uses adversarial dropout to improve diversity and stability in GAN training by dynamically varying discriminators.

Findings

Promotes sample diversity within and across epochs

Eliminates mode collapse in GANs

Stabilizes training process

Abstract

We propose to incorporate adversarial dropout in generative multi-adversarial networks, by omitting or dropping out, the feedback of each discriminator in the framework with some probability at the end of each batch. Our approach forces the single generator not to constrain its output to satisfy a single discriminator, but, instead, to satisfy a dynamic ensemble of discriminators. We show that this leads to a more generalized generator, promoting variety in the generated samples and avoiding the common mode collapse problem commonly experienced with generative adversarial networks (GANs). We further provide evidence that the proposed framework, named Dropout-GAN, promotes sample diversity both within and across epochs, eliminating mode collapse and stabilizing training.