Question: Should there be a tag for offline-reinforcement-learning on AI Stack Exchange?
Background: It seems that the current relevant tags for offline RL are [tag:off-policy methods], offline-learning, and batch-learning (the second of those tags being a synonym of the third). When I looked at the batch-learning tag, the questions seemed to be split between offline RL and questions regarding batches of data (e.g. for neural network training) - two very distinct types of questions. I feel that a tag that distinguishes offline RL from batch learning may be helpful for the AI Stack Exchange community.
From my own experience, it looks like offline RL is emerging as a more specific term in the research community throughout the past few years. For example, a survey by Sergei Levine et al. in 2020 had a fairly substantial paragraph (second paragraph of section 2.2 here) that discusses the distinctions between the terms off-policy learning, batch learning, and offline RL. In summary, offline RL is a subset of off-policy learning that doesn't involve any more interaction with the environment, a major difference. Also, they explicitly state that using the term batch learning in place of offline RL can be confusing.
While it isn't discussed in the survey, I find that the offline learning seems to be too general of a tag for offline RL questions, for example, due to the vastly different algorithms and intuitions used to find an optimal policy versus training a classifier. In addition, the Wikipedia page on offline learning explicitly states that "Offline learning is working in batch mode", again causing possible confusion if offline learning is used as a tag instead of offline RL. Finally, I think using a generic reinforcement learning tag is also too general due to the different algorithms and concerns in the field of offline RL.
I haven't made a tag yet, and I'm mainly wondering if my thought process is correct since I have never created a tag. Also, I'm unsure if there are any unwritten rules regarding tags that I am unaware of. For those wondering, I asked this question because I was surprised to find that there was no offline-reinforcement-learning tag when answering a question on offline RL.
