Massively Parallel Methods for Deep Reinforcement Learning

TL;DR

This paper introduces a massively distributed architecture for deep reinforcement learning, significantly improving performance and efficiency in training algorithms like DQN across multiple Atari games.

Contribution

It presents the first scalable distributed architecture for deep reinforcement learning, enabling faster training and better performance on complex environments.

Findings

Outperformed non-distributed DQN in 41 of 49 Atari games

Reduced training wall-time by an order of magnitude on most games

Demonstrated scalability and effectiveness of distributed RL architecture

Abstract

We present the first massively distributed architecture for deep reinforcement learning. This architecture uses four main components: parallel actors that generate new behaviour; parallel learners that are trained from stored experience; a distributed neural network to represent the value function or behaviour policy; and a distributed store of experience. We used our architecture to implement the Deep Q-Network algorithm (DQN). Our distributed algorithm was applied to 49 games from Atari 2600 games from the Arcade Learning Environment, using identical hyperparameters. Our performance surpassed non-distributed DQN in 41 of the 49 games and also reduced the wall-time required to achieve these results by an order of magnitude on most games.