episodic-mdp

This paper studies decentralized multi-player Q-learning in episodic Markov decision processes under three forms of information asymmetry, proposing algorithms that achieve regret bounds matching the single-agent Q-learning rate up to logarithmic factors.