Glossary term
Glossary term
Agentic Systems
In DQN-like algorithms, the memory used by the agent to store state transitions for use in experience replay.
Created for this library
A robotics RL team uses a replay buffer to reuse logged interactions efficiently during off-policy training.
A trading research team uses a replay buffer to reuse historical market trajectories during training of its execution policy.
An ad-bidding RL team uses a replay buffer to break temporal correlations in training data when learning from logged interactions.
Definition source: Google for Developers Machine Learning Glossary | Creative Commons Attribution 4.0 License