Paper argues deployed RL should be continual

Arxiv CS.LG (machine learning)

The paper critiques the train-then-fix paradigm in deployed RL, where agents stop learning after initial training. It advocates for continual learning approaches to maintain performance over time.

reinforcement-learning

Arxiv CS.LG (machine learning)·Parnian BehdinKevin RoiceGolnaz Mesbahi

7 days ago