Flow-GAN: Combining Maximum Likelihood and Adversarial Learning in Generative Models

TL;DR

Flow-GANs combine adversarial training with exact likelihood evaluation, enabling high-quality sample generation and accurate likelihood estimation, bridging the gap between GANs and explicit probabilistic models.

Contribution

This paper introduces Flow-GANs, a novel model that supports both adversarial and maximum likelihood training for improved generative modeling.

Findings

Adversarial training yields high-quality samples but poor likelihood scores.

Maximum likelihood training results in better likelihood but lower sample quality.

Hybrid training achieves both high likelihood and good sample fidelity.

Abstract

Adversarial learning of probabilistic models has recently emerged as a promising alternative to maximum likelihood. Implicit models such as generative adversarial networks (GAN) often generate better samples compared to explicit models trained by maximum likelihood. Yet, GANs sidestep the characterization of an explicit density which makes quantitative evaluations challenging. To bridge this gap, we propose Flow-GANs, a generative adversarial network for which we can perform exact likelihood evaluation, thus supporting both adversarial and maximum likelihood training. When trained adversarially, Flow-GANs generate high-quality samples but attain extremely poor log-likelihood scores, inferior even to a mixture model memorizing the training data; the opposite is true when trained by maximum likelihood. Results on MNIST and CIFAR-10 demonstrate that hybrid training can attain high held-out likelihoods while retaining visual fidelity in the generated samples.