Randomized Strategies are Useless in Markov Decision Processes

Hugo Gimbert (1)

1 LaBRI - Laboratoire Bordelais de Recherche en Informatique

We show that in a Markov decision process with arbitrary payoff mapping, restricting the set of behavioral strategies from randomized to deterministic does not influence the value of the game nor the existence of almost-surely or positively winning strategies. This result still holds for Markov decision processes with partial observation.

Domaines: Informatique et théorie des jeux [cs.GT]

Hugo Gimbert. Randomized Strategies are Useless in Markov Decision Processes. 2009. ⟨hal-00403463v4⟩