论文部分内容阅读
时间复杂性是基于 EM 框架的贝叶斯网络学习算法应用的一个瓶颈问题.本文首先提出一种并行的参数EM 算法来学习具有缺省数据的贝叶斯网络参数,实验表明该算法可有效降低参数学习的时间复杂性.进而将该算法应用到结构 EM 算法中,提出一种并行的结构 EM 算法(PL-SEM),PL-SEM 算法并行地计算各个样本的期望充分因子和贝叶斯网络的参数,降低结构学习的时间复杂性.
Time complexity is a bottleneck problem of applying Bayesian network learning algorithm based on EM framework.This paper first proposes a parallel parameter EM algorithm to learn Bayesian network parameters with default data, experiments show that the algorithm can effectively reduce Then the algorithm is applied to the structure EM algorithm, a parallel structure EM algorithm (PL-SEM) is proposed. The PL-SEM algorithm calculates the expected sufficient factors of each sample and the Bayesian network The parameters reduce the time complexity of structure learning.