A parallel algorithm for learning Bayesian networks

Abstract

Computing the expected statistics is the main bottleneck in learning Bayesian networks in large-scale problem domains. This paper presents a parallel learning algorithm, PL-SEM, for learning Bayesian networks, based on an existing structural EM algorithm (SEM). Since the computation of the expected statistics is in the parametric learning part of the SEM algorithm, PLSEM exploits a parallel EM algorithm to compute the expected statistics. The parallel EM algorithm parallelizes the E-step and M-step. At the E-step, PLSEM parallel computes the expected statistics of each sample; and at the Mstep, with the conditional independence of Bayesian networks and the expected statistics computed at the E-step, PL-SEM exploits the decomposition property of the likelihood function under the completed data to parallel estimate each local likelihood function. PL-SEM effectively computes the expected statistics, and greatly reduces the time complexity of learning Bayesian networks. © Springer-Verlag Berlin Heidelberg 2007.