Deep Generative Modelling: A Comparative Review of VAEs, GANs, Normalizing Flows, Energy-Based and Autoregressive Models

TL;DR

This paper provides a comprehensive comparison of major deep generative models, including VAEs, GANs, normalizing flows, energy-based, and autoregressive models, highlighting their principles, differences, and recent advances.

Contribution

It offers a unified review and comparison of various deep generative modeling approaches, clarifying their interrelations and recent developments.

Findings

Energy-based models and VAEs have distinct training stability profiles.

GANs excel in generating high-quality, diverse samples.

Normalizing flows enable exact likelihood computation.

Abstract

Deep generative models are a class of techniques that train deep neural networks to model the distribution of training samples. Research has fragmented into various interconnected approaches, each of which make trade-offs including run-time, diversity, and architectural restrictions. In particular, this compendium covers energy-based models, variational autoencoders, generative adversarial networks, autoregressive models, normalizing flows, in addition to numerous hybrid approaches. These techniques are compared and contrasted, explaining the premises behind each and how they are interrelated, while reviewing current state-of-the-art advances and implementations.