Model predictive control is almost optimal for restless bandits.

Nicolas Gast, Dheeraj Narasimha

Browse the full COLT paper archive.