一种新的Bayes网络条件概率学习方法

针对大规模Bayes网络的条件概率赋值问题,提出一种学习方法.首先使用类层次结构定义一种新的基于层次的Bayes网络模型,用于表示大规模Bayes网络.然后将训练数据集由单个数据表的形式转化成多表数据库,其中每个数据库表对应一个Bayes网络模块.在此基础上导出条件概率计算公式,从每个数据库表中算出相应的Bayes网络模块的条件概率表,由此实现对整个层次Bayes网络的概率赋值.通过适当增加数据库表的数目来控制每个表中属性的个数,保证计算的可行性.将层次Bayes网络及计算公式用于解决图像中文本的自动检测与定位问题,实验结果表明了它们的有效性.

A New Approach to Learning Conditional Probabilities in Bayesian Networks

A learning approach is proposed to assignation in large scale Bayesian networks. model is defined based on class hierarchical solve the problems of conditional probability Firstly, a new hierarchical Bayesian Network structure, which is used to represent large scale Bayesian networks. Then, the train data set is changed from a single table to a database composed of some database tables. And each database table corresponds to a Bayesian network block. Based on that, a formula of conditional probability is developed. And each conditional probabilistic table of Bayesian network block can be calculated from the database tables respectively. Proper adjustment of the attribute number in each database table can assure the validity of this learning approach. Experiments in automatic detection and location of texts in images show the feasibility of this hierarchical Bayesian network and learning approach.