Dropout-GAN: Learning from a Dynamic Ensemble of Discriminators
This addresses a common issue in GAN training for researchers and practitioners, though it appears incremental as it builds on existing multi-adversarial frameworks.
The paper tackles the mode collapse problem in generative adversarial networks (GANs) by introducing adversarial dropout in multi-adversarial networks, which forces the generator to satisfy a dynamic ensemble of discriminators. The result is a more generalized generator that promotes sample diversity and stabilizes training, eliminating mode collapse.
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.