In this paper we present three different architectures for the evaluation of influence diagrams: HUGIN, Shafer-Shenoy (S-S), and Lazy Propagation (LP). HUGIN and LP are two new architectures introduced in this paper. The computational complexity using the three architectures are compared on the same structure, the LImited Memory Influence Diagram (LIMID), where only the requisite information for the computation of optimal policies is depicted. Because the requisite information is explicitly represented in the diagram, the evaluation procedure can take advantage of it. Previously, it has been shown that significant savings in computational time can be obtained by performing the calculation on the LIMID rather than on the traditional influence diagram. In this paper we show how the obtained savings is considerably increased when the computations are performed according to the LP scheme.
Anders L. Madsen, Dennis Nilsson