Learning Methods For Sequential Decision Making With Imperfect Representations