Bayesian Conditional Generative Adverserial Networks

TL;DR

This paper introduces Bayesian Conditional GANs (BC-GANs), a novel framework that employs a Bayesian approach to generative modeling, enhancing flexibility and performance across various learning paradigms.

Contribution

The paper presents BC-GANs, which incorporate a Bayesian framework into GANs, allowing for a random generator function and improved handling of different learning settings.

Findings

BC-GANs outperform state-of-the-art methods in experiments.

The Bayesian approach improves flexibility in learning tasks.

BC-GANs effectively handle unsupervised, supervised, and semi-supervised learning.

Abstract

Traditional GANs use a deterministic generator function (typically a neural network) to transform a random noise input $z$ to a sample $\mathbf{x}$ that the discriminator seeks to distinguish. We propose a new GAN called Bayesian Conditional Generative Adversarial Networks (BC-GANs) that use a random generator function to transform a deterministic input $y'$ to a sample $\mathbf{x}$. Our BC-GANs extend traditional GANs to a Bayesian framework, and naturally handle unsupervised learning, supervised learning, and semi-supervised learning problems. Experiments show that the proposed BC-GANs outperforms the state-of-the-arts.