Adaptive Density Estimation for Generative Models

TL;DR

This paper introduces an adaptive density estimation method using deep invertible transformations in latent space, effectively combining the strengths of GANs and likelihood-based models for improved sample quality and likelihood evaluation.

Contribution

It proposes a novel hybrid generative model that overcomes parametric conflicts by employing invertible transformations, enabling efficient likelihood computation and adversarial training.

Findings

Achieves GAN-like sample quality with competitive FID scores.

Provides likelihood scores that surpass existing hybrid models.

Balances sample quality and likelihood evaluation effectively.

Abstract

Unsupervised learning of generative models has seen tremendous progress over recent years, in particular due to generative adversarial networks (GANs), variational autoencoders, and flow-based models. GANs have dramatically improved sample quality, but suffer from two drawbacks: (i) they mode-drop, i.e., do not cover the full support of the train data, and (ii) they do not allow for likelihood evaluations on held-out data. In contrast, likelihood-based training encourages models to cover the full support of the train data, but yields poorer samples. These mutual shortcomings can in principle be addressed by training generative latent variable models in a hybrid adversarial-likelihood manner. However, we show that commonly made parametric assumptions create a conflict between them, making successful hybrid models non trivial. As a solution, we propose to use deep invertible transformations in the latent variable decoder. This approach allows for likelihood computations in image space, is more efficient than fully invertible models, and can take full advantage of adversarial training. We show that our model significantly improves over existing hybrid models: offering GAN-like samples, IS and FID scores that are competitive with fully adversarial models, and improved likelihood scores.