Large-Scale Retrieval for Reinforcement Learning Peter C. Humphreys

Neural Information Processing Systems 

In addition, new information can be attended to by the agent, without retraining, by simply augmenting the retrieval dataset.