Equivalences between maximum a posteriori inference in Bayesian networks and maximum expected utility computation in influence diagrams