非确定先验信息的贝叶斯网结构学习方法

Structure Learning Method of Bayesian Network with Uncertain Prior Information

针对非确定先验结构信息下的贝叶斯网络学习问题,提出一种非确定先验结构信息贝叶斯网络的结构学习方法。为更好地利用不确定性信息,对MDL测度进行改进,提出SMDL测度,使之能在学习过程中考虑先验信息的不确定性,使用模拟退火算法对问题进行求解。通过实验对算法的可行性和效率进行验证。

This paper presents a structure learning method of Bayesian network to solve the problem of structure learning with uncertain prior information. A description method of the uncertain prior information is given. An improved MDL score method named SMDL is proposed to fit the uncertain prior information in learning process. Simulated annealing method is used to solve the problem. This method is validated by experiments.