Optimizing the Neural Architecture of Reinforcement Learning Agents

TL;DR

This paper evaluates neural architecture search methods for optimizing reinforcement learning agents, demonstrating that NAS can discover architectures that outperform manually designed ones on Atari benchmarks.

Contribution

It provides empirical evidence that modern NAS techniques can effectively improve neural network architectures for RL agents, surpassing manual designs.

Findings

NAS methods outperform manual architectures on Atari benchmarks

Modern NAS techniques are effective for RL architecture optimization

Automated architecture search can lead to better RL agent performance

Abstract

Reinforcement learning (RL) enjoyed significant progress over the last years. One of the most important steps forward was the wide application of neural networks. However, architectures of these neural networks are typically constructed manually. In this work, we study recently proposed neural architecture search (NAS) methods for optimizing the architecture of RL agents. We carry out experiments on the Atari benchmark and conclude that modern NAS methods find architectures of RL agents outperforming a manually selected one.