用遗传算法实现信度网结构学习:一种新的编码方案

Structure Learning of Belief Network by Genetic Algorithms: A New Network Encoding Method

近年来,信度网已经成为表达一组随机变量间的概率关系的常见方法。在大的变量域中信度网的结构生成是信度网应用的难点。为了解决信度网的结构学习问题,一些研究者研究如何从数据集自动学习结构。本文研究采用遗传算法进行信度网结构学习时存在的问题和编码方法,并给出了一种新的信度网编码方案。数值试验显示遗传算法能够给出理想的结果。

In the last few years Belief networks have become a popular way of modeling probabilistic relationships among a set of variables for a given domain. It's a very hard task that treats a good belief network for large domains. Therefore, some researchers have studied how this construction can be automated. This work introduces how to do structure learning by genetic algorithms and discuss the encoding method using in GA. A new encoding method has presented. A case study has shown that GA is good for structure learning.