Revisiting Fundamentals of Experience Replay
arXiv:2007.06700
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.
Published at ICML 2020. First two authors contributed equally and code available at https://github.com/google-research/google-research/tree/master/experience_replay
Cited by in corpus (7)
- Learning offline: memory replay in biological and artificial reinforcement learning
- Training Larger Networks for Deep Reinforcement Learning
- Rank the Episodes: A Simple Approach for Exploration in Procedurally-Generated Environments
- Qgraph-bounded Q-learning: Stabilizing Model-Free Off-Policy Deep Reinforcement Learning
- Solving Heterogeneous General Equilibrium Economic Models with Deep Reinforcement Learning
- DCUR: Data Curriculum for Teaching via Samples with Reinforcement Learning
- High Throughput Training of Deep Surrogates from Large Ensemble Runs