贝叶斯网络适应性学习

Adaptability Learning for Bayesian Network

在现实中,随着对领域问题认识的深入,往往需要对贝叶斯网络进行调整,以使贝叶斯网络模型能够更好地反映实际问题.但调整后的贝叶斯网络中一些新参数需要根据原有贝叶斯网络来确定,目前缺乏对新参数学习方法的研究.本文基于专家知识调整贝叶斯网络结构,将原贝叶斯网络和新贝叶斯网络相结合,通过推理进行新参数的迭代学习,可实现贝叶斯网络的适应性学习.

With further understanding to the domanial problems, it is often necessary to regulate the Bayesian network to meet the demand. But in the regulated Bayesian network the new parameters need to be computed according to the old Bayesian network. At present, however, lack the research of the methods of learning new parameters. In this paper, Bayesian network structure is regulated based on experts knowledge. The old Bayesian network is combined with the new one to ascertain the new parameters through reasoning. The adaptability learning of Bayesian network can be realized by above method.