A Parallel Sampling Method for Bayesian Networks

This paper describes the method to estimate probabilities on Bayesian belief networks (BNs). A BN has nodes showing random variables and shows cause and effect relationships among nodes as a graph. We calculate posterior probabilities, and then estimate the uncertain plural events. As one of the method for estimating probabilities in BN is stochastic sampling. The method has known as taking time to calculate as BN become larger and more complex. Therefore, this paper proposes a parallel sampling method on BN. Then a BN need efficiently dividing to generate samples in parallel, therefore we use community detection. The number of nodes, which have mutual dependence among nodes, is the least to use community detection. In addition, pipeline processing as generating samples, which have mutual dependence reduces waiting time causing by existing them.