Learning of two-choice, differential reward problems with informational constraints on payoff combinations