Probabilistic Generative Adversarial Networks

TL;DR

This paper introduces PGAN, a novel GAN variant that incorporates a probabilistic model with a likelihood-based loss, improving training stability and providing meaningful output quality measures.

Contribution

The paper presents PGAN, integrating a Gaussian Mixture Model into GANs, enabling likelihood-based evaluation and enhanced training stability.

Findings

PGAN generates realistic MNIST images.

Likelihoods correlate with image quality.

Improved training stability in GANs.

Abstract

We introduce the Probabilistic Generative Adversarial Network (PGAN), a new GAN variant based on a new kind of objective function. The central idea is to integrate a probabilistic model (a Gaussian Mixture Model, in our case) into the GAN framework which supports a new kind of loss function (based on likelihood rather than classification loss), and at the same time gives a meaningful measure of the quality of the outputs generated by the network. Experiments with MNIST show that the model learns to generate realistic images, and at the same time computes likelihoods that are correlated with the quality of the generated images. We show that PGAN is better able to cope with instability problems that are usually observed in the GAN training procedure. We investigate this from three aspects: the probability landscape of the discriminator, gradients of the generator, and the perfect discriminator problem.