多维贝叶斯网络分类器加速学习算法

作为概率图模型,无限制多维贝叶斯网络分类器(GMBNC)是贝叶斯网络(BN)应用在多维分类应用时的精简模型,只包含对预测有效的局部结构.为了获得GMBNC,传统方法是先学习全局BN;为了避免全局搜索,提出了仅执行局部搜索的结构学习算法DOS-GMBNC.该算法继承了之前提出的IPC-GMBNC算法的主体框架,基于进一步挖掘的结构拓扑信息来动态调整搜索次序,以避免执行无效用的计算.实验研究验证了DOS-GMBNC算法的效果和效率:(1)该算法输出的网络质量与IPC-GMBNC一致,优于经典的PC算法;(2)在一个包含100个节点的问题中,该算法相对于PC和IPC-GMBNC算法分别节省了近89％和45％的计算量.

Accelerated Structure Learning for General Multi-dimensional Bayesian Network Classifier

General multi-dimensional Bayesian network classifier (GMBNC) is one kind of Bayesian network (BN) tailored for the application of multi-dimensional classification,hence it contains only features necessary for the prediction.To avoid global search,a novel algorithm called DOS-GMBNC was proposed.It inherits the framework of existing IPC-GMBNC,conducts a dynamic order of search by making use of the underlying topology information.Experimental stu-dies indicate the effectiveness and efficiency of DOS-GMBNC.It outputs networks with equal quality as PC and iPC-GMBNC algorithms,and it brings considerable reduction of computation complexity,e.g.about 89% and 45% less than PC and IPC-GMBNC respectively on a 100-node network problem.