Concept
Experience replay
Experience replay is a technique where a learning agent stores past interactions in a buffer and samples from them repeatedly during training, rather than learning only from the most recent data and discarding it. It originated in off-policy reinforcement learning (DQN being…
The rest of “Experience replay” is a premium feature: every concept in the library gets a precise, practitioner-focused write-up like this one, cross-linked straight from the paper summaries that use it.
Log in to unlock→