Improved BiGAN training with marginal likelihood equalization
This work addresses training challenges in generative adversarial networks, particularly for bidirectional GANs, offering incremental improvements in sample quality and diversity for applications in image generation.
The paper tackles the issue of overrepresentation of certain sample types in bidirectional GANs by proposing a novel training procedure that enforces matching between the inverse inference network's distribution and the prior, and uses non-uniform mini-batch sampling, resulting in improved quality and variety in generated samples as measured on datasets like CIFAR10, Fashion MNIST, and CelebA.
We propose a novel training procedure for improving the performance of generative adversarial networks (GANs), especially to bidirectional GANs. First, we enforce that the empirical distribution of the inverse inference network matches the prior distribution, which favors the generator network reproducibility on the seen samples. Second, we have found that the marginal log-likelihood of the samples shows a severe overrepresentation of a certain type of samples. To address this issue, we propose to train the bidirectional GAN using a non-uniform sampling for the mini-batch selection, resulting in improved quality and variety in generated samples measured quantitatively and by visual inspection. We illustrate our new procedure with the well-known CIFAR10, Fashion MNIST and CelebA datasets.