Propagation in CLG Bayesian networks based on semantic modeling

In this paper, we propose a new algorithm based on semantic model for inference in CLG Bayesian networks which is strongly inspired by the architecture of Madsen (in, Int J Approx Reason 49:503–521, 2008). By performing semantic modeling before physical computation, the proposed algorithm takes advantage of the semantic knowledge induced by the structure of the graph and the evidence. Thus, iteration between semantic modeling and physical computation can be avoided. Also,the presented architecture can exploit some remaining independencies in the relevant potentials which were ignored by the previous architecture. The correctness of the proposed algorithm has been proved and the resulting benefits are illustrated by examples. The results indicate a significant potential in semantic knowledge.