Revisiting Fundamentals of Experience Replay

TL;DR

This paper systematically analyzes experience replay in deep RL, revealing new insights about replay capacity and ratio effects, and challenging conventional beliefs with extensive empirical evidence.

Contribution

It provides a comprehensive analysis of experience replay properties, demonstrating the impact of capacity and replay ratio on algorithm performance, and highlights the benefits of uncorrected n-step returns.

Findings

Greater replay capacity can significantly boost performance for some algorithms.

Uncorrected n-step returns are uniquely beneficial despite lacking theoretical grounding.

Controlling replay ratio is crucial for understanding and optimizing deep RL performance.

Abstract

Experience replay is central to off-policy algorithms in deep reinforcement learning (RL), but there remain significant gaps in our understanding. We therefore present a systematic and extensive analysis of experience replay in Q-learning methods, focusing on two fundamental properties: the replay capacity and the ratio of learning updates to experience collected (replay ratio). Our additive and ablative studies upend conventional wisdom around experience replay -- greater capacity is found to substantially increase the performance of certain algorithms, while leaving others unaffected. Counterintuitively we show that theoretically ungrounded, uncorrected n-step returns are uniquely beneficial while other techniques confer limited benefit for sifting through larger memory. Separately, by directly controlling the replay ratio we contextualize previous observations in the literature and empirically measure its importance across a variety of deep RL algorithms. Finally, we conclude by testing a set of hypotheses on the nature of these performance benefits.