Overcoming the Curse of Dimensionality in Reinforcement Learning Through Approximate Factorization