In this question, many users have discussed online and offline learning in machine learning. But, in the context of reinforcement learning, what are exactly online and offline learning?

  • 3
    $\begingroup$ Not a complete answer, but off-line is simply using a sample of stored examples of agent behaviour in the task of refining agent's method to attain value to actions or directly select actions. Online learning is using a live stream of single examples coming from immediate now to do the same work: one example - one agent update. $\endgroup$ Apr 16, 2018 at 16:07
  • 2
    $\begingroup$ . . . or in short the terms have the same meaning in reinforcement learning as in supervised learning (or unsupervised learning). So the question you linked already answers your question. No need for another question here, this is really a duplicate unless perhaps you can explain in more detail what your problem is (because RL is a subset of ML, so the linked question is already an answer). Not to be confused with on-policy and off-policy learning, which is a different issue and specific to RL. $\endgroup$ Apr 16, 2018 at 19:17