基于广义信息论的贝叶斯分类器动态建模  

Dynamic modeling of bayesian classifier based on general information theory

作　　者：王利民[1,2] 李雄飞[1,2] 张海龙[1,2] 

机构地区：[1]吉林大学计算机科学与技术学院,长春130012 [2]吉林大学符号计算与知识工程教育部重点实验室,长春130012

出　　处：《吉林大学学报（工学版）》2009年第3期776-780,共5页Journal of Jilin University:Engineering and Technology Edition

基　　金：国家自然科学基金项目(60275026)

摘　　要：基于广义互信息描述变量之间的条件独立和条件相关性,并根据信息非负递增性原则去除冗余变量,由测试样本而非整个样本空间构造贝叶斯分类子模型。结合局部计算近似推理进行概率密度和条件概率分布估计。在UCI机器学习数据集上的实验结果证明了该算法的合理性和有效性。By reasoning the conditional independence and dependence between attribute values based on General Information theory,redundant attributes are removed naturally and a submodel is constructed for each test sample rather that the whole sample space.The joint probability density and conditional probability is estimated based on marginal computation.The experimental study on the UCI data set shows that,this algorithm can describe the marginal dependency of mixed-mode data more intuitively and accurately.

关 键 词：人工智能 贝叶斯网络 条件独立 局部计算 

分 类 号：TP182[自动化与计算机技术—控制理论与控制工程]