Selective Experience Replay for Lifelong Learning

TL;DR

This paper introduces a selective experience replay method for deep reinforcement learning that mitigates catastrophic forgetting by choosing experiences based on different criteria, with distribution matching being most effective.

Contribution

It proposes a novel experience replay strategy with four selection methods, demonstrating that distribution matching effectively prevents forgetting across various tasks.

Findings

Distribution matching prevents catastrophic forgetting.

Selective experience replay improves lifelong learning performance.

Coverage maximization benefits in specific task importance scenarios.

Abstract

Deep reinforcement learning has emerged as a powerful tool for a variety of learning tasks, however deep nets typically exhibit forgetting when learning multiple tasks in sequence. To mitigate forgetting, we propose an experience replay process that augments the standard FIFO buffer and selectively stores experiences in a long-term memory. We explore four strategies for selecting which experiences will be stored: favoring surprise, favoring reward, matching the global training distribution, and maximizing coverage of the state space. We show that distribution matching successfully prevents catastrophic forgetting, and is consistently the best approach on all domains tested. While distribution matching has better and more consistent performance, we identify one case in which coverage maximization is beneficial - when tasks that receive less trained are more important. Overall, our results show that selective experience replay, when suitable selection algorithms are employed, can prevent catastrophic forgetting.