Revisiting Fundamentals of Experience Replay

Yoshua Bengio, Hugo Larochelle, Rishabh Agarwal, William Fedus, Prajit Ramachandran, Mark Rowland, Will Dabney

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 1 functions out of this paper's own repositories and ran 1 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

Repositories linked to this paper

• google-research/google-research — marked official in Papers with Code
• DrJupiter/Deep-Reinforcement-Rubiks-Cube-Solver

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

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.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2007.06700")
get_code_for_paper("2007.06700")
have("2007.06700")

Connect an agent — have() is free.