Connecting Generative Adversarial Networks and Actor-Critic Methods

TL;DR

This paper reveals a formal connection between GANs and actor-critic methods, showing how insights from one can inform the other to improve stability and scalability in training deep networks.

Contribution

It establishes a theoretical link between GANs and actor-critic methods, facilitating cross-community strategies for more stable and scalable optimization.

Findings

GANs can be viewed as actor-critic methods with a fixed environment.

Strategies for stabilizing training are applicable across both models.

Extensions to GANs and RL involve complex information flow.

Abstract

Both generative adversarial networks (GAN) in unsupervised learning and actor-critic methods in reinforcement learning (RL) have gained a reputation for being difficult to optimize. Practitioners in both fields have amassed a large number of strategies to mitigate these instabilities and improve training. Here we show that GANs can be viewed as actor-critic methods in an environment where the actor cannot affect the reward. We review the strategies for stabilizing training for each class of models, both those that generalize between the two and those that are particular to that model. We also review a number of extensions to GANs and RL algorithms with even more complicated information flow. We hope that by highlighting this formal connection we will encourage both GAN and RL communities to develop general, scalable, and stable algorithms for multilevel optimization with deep networks, and to draw inspiration across communities.