基于子图分解和pso 的贝叶斯网络结构学习算法 - atlantis press.pdf

International Conference on Education, Management and Computer Science (ICEMC 2016) A Bayesian Network Structure Learning Algorithm Based on the Combination of PSO and Sub-graph Decomposition Shaorong Feng School of Information Science and Engineering, Xiamen University, 361005 Xiamen, China shaorong@xmu.edu.cn Keywords: Bayesian network; Structure learning; Maximal prime decomposition technology; Particle swarm optimization algorithm Abstract. In this paper we mainly focus on how to reduce the searching spaces when accelerating the efficiency of searching for the best Bayesian network structure and a corresponding algorithm named MPD-PSO based on the combination of MPD(maximal prime decomposition technology) and PSO(particle swarm optimization algorithm) was proposed. In the algorithm we proposed, we firstly use the MBDA (Markov Boundary Discovery Algorithm) algorithm to achieve the Markov boundary, and construct the undirected independent graph, then we decomposed the large undirected independent graph into several maximal prime sub-graphs by using the MPD algorithm, therefore we transform a high-dimensional structure learning problem into a low-dimensional structure (sub-graph structure) learning problem and reduce the searching space relatively. After that we use the PSO algorithm to learn the sub-graph, and we merge the learned sub-structure by correcting the wrong edges to achieve the best Bayesian structure. We validate the proposed algorithm by using Asia and Alarm network, and the results verify the superiority of the proposed algorithm in learning effect and running time over the GA (Genetic Algorithm) and PSO algorithms. 基于子图分解和PSO 的贝叶斯网络结构学习算法