This paper studies a value function iteration algorithm that can be applied to almost all stationary dynamic programming problems. Using nonexpansive function approximation and Monte Carlo integration, we develop a randomized fitted Bellman operator and a corresponding algorithm that is globally...