Importance Resampling for Off-policy Prediction
This addresses variance issues in off-policy learning for reinforcement learning practitioners, but it is incremental as it builds on existing resampling and variance-reduction strategies.
The paper tackles the high variance problem in off-policy prediction in reinforcement learning by proposing Importance Resampling (IR) as an alternative to importance sampling, demonstrating improved sample efficiency and lower variance updates in microworlds and a racing car simulator.
Importance sampling (IS) is a common reweighting strategy for off-policy prediction in reinforcement learning. While it is consistent and unbiased, it can result in high variance updates to the weights for the value function. In this work, we explore a resampling strategy as an alternative to reweighting. We propose Importance Resampling (IR) for off-policy prediction, which resamples experience from a replay buffer and applies standard on-policy updates. The approach avoids using importance sampling ratios in the update, instead correcting the distribution before the update. We characterize the bias and consistency of IR, particularly compared to Weighted IS (WIS). We demonstrate in several microworlds that IR has improved sample efficiency and lower variance updates, as compared to IS and several variance-reduced IS strategies, including variants of WIS and V-trace which clips IS ratios. We also provide a demonstration showing IR improves over IS for learning a value function from images in a racing car simulator.