A combination of exact algorithms for inference on Bayesian belief networks

Cutset conditioning and clique-tree propagation are two popular methods for exact probabilistic inference in Bayesian belief networks. Cutset conditioning is based on decomposition of a subset of network nodes, whereas clique-tree propagation depends on aggregation of nodes. We characterize network structures in which the performances of these methods differ. We describe a means to combine cutset conditioning and clique-tree propagation in an approach called aggregation after decomposition (AD), which can perform inference relatively efficiently for certain network structures in which neither cutset conditioning nor clique-tree propagation performs well. We discuss criteria to determine when AD will perform more efficient belief-network inference than will clique-tree propagation.