基于局部优化具有连续变量的贝叶斯网络结构学习  

Learning Bayesian network structure with continuous variables based on local optimization

作　　者：张剑飞[1] 王辉[2] 周颜军[2] 王双成[3] 

机构地区：[1]齐齐哈尔大学计算机与控制工程学院,黑龙江齐齐哈尔161006 [2]东北师范大学计算机学院,吉林长春130024 [3]上海立信会计学院信息科学院,上海201600

出　　处：《东北师大学报（自然科学版）》2006年第1期27-30,共4页Journal of Northeast Normal University(Natural Science Edition)

基　　金：吉林省自然科学基金资助项目(20030517-1)

摘　　要：概述了具有连续变量的贝叶斯网络结构学习存在的主要问题,给出了基于局部优化的具有连续变量的贝叶斯网络结构学习方法.通过构造局部最优回归模式、局部回归模式的条件组合及环路处理,建立了具有连续变量的贝叶斯网络结构,既可以避免复杂的结构打分运算及结构空间搜索,同时又不会出现由于离散化而导致过多的信息丢失及假依赖现象.Main problems are presented in learning Bayesian network structure with continuous variables. To solve these problems, a new method of learning Bayesian network structure with continuous variables based on local optimization is developed. In this method, the Bayesian network structure with continuous variables is set up by constructing local optimization regression models, conditional combination of these models and dealing with loop. This method can avoid complicated structure scoring and structure search. At the same time, the case of leading to miss information and false dependency can be got over.

关 键 词：连续变量 贝叶斯网络 结构学习 回归 

分 类 号：TP311.134.1[自动化与计算机技术—计算机软件与理论]