Evaluating Bayes nets with concurrent process networks

The computation complexity of the total probability mass of a leaf node of a general Bayes network can be exponential in the number of ancestor nodes of that leaf. It is a well known result that for a large class of networks, a number of minterms only linear in the number of ancestor nodes contributes about 67% of the total probability mass. The problem of Bayes net search is to generate only these high mass minterms. We introduce a concurrent algorithm for attempting this, based on converting the net into a concurrent process network. Each parent node sends messages containing partial minterms to child nodes. The novel idea is to prioritize these messages to give higher weight to partial terms that are likely candidates for inclusion in the final high mass minterms. We have implemented this algorithm in KL1 and discuss its attributes