Answers are grounded in this paper and cite their sources.
We propose a simulation-based algorithm for learning good policies for a Markov decision process with unknown transition law, with aggregated states. The state aggregation itself can be adapted on a slower time scale by an auxiliary learning algorithm. Rigorous justifications are provided for both algorithms.
No takes yet. Share an insight, caveat, or question.
Baras et al. (2002) studied this question.