Faster Deep Reinforcement Learning with Slower Online Network

TL;DR

This paper introduces a simple modification to deep reinforcement learning algorithms that encourages the online network to stay close to the target network, resulting in improved robustness and performance on Atari benchmarks.

Contribution

The paper proposes a novel update mechanism that keeps the online network near the target network, enhancing stability and performance in deep RL algorithms.

Findings

DQN Pro outperforms DQN on Atari benchmarks.

Rainbow Pro surpasses Rainbow in robustness and accuracy.

The proposed method improves training stability in noisy environments.

Abstract

Deep reinforcement learning algorithms often use two networks for value function optimization: an online network, and a target network that tracks the online network with some delay. Using two separate networks enables the agent to hedge against issues that arise when performing bootstrapping. In this paper we endow two popular deep reinforcement learning algorithms, namely DQN and Rainbow, with updates that incentivize the online network to remain in the proximity of the target network. This improves the robustness of deep reinforcement learning in presence of noisy updates. The resultant agents, called DQN Pro and Rainbow Pro, exhibit significant performance improvements over their original counterparts on the Atari benchmark demonstrating the effectiveness of this simple idea in deep reinforcement learning. The code for our paper is available here: Github.com/amazon-research/fast-rl-with-slow-updates.