On ordinal comparison of policies in Markov reward processes

초록

An asymptotic exponential convergence rate of ordinal comparison from large deviations theory is well known for selecting the true best solution from the candidate solutions sample means. This note supplements the theories developed by Dai within the framework of ergodic Markov reward processes for epsilon-ordinal comparison of policies, establishing an asymptotic exponential convergence rate for the infinite-horizon average criterion.

키워드