A Parallel Algorithm for Learning Bayesian Networks

Kui Yu, Hao Wang, Xindong Wu

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, PL-SEM 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, PL-SEM parallel computes the expected statistics of each sample; and at the M-step, 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.

External IDs:dblp:conf/pakdd/YuWW07