Value Function Prospective Learning