基于模型融合的分布式贝叶斯网络学习算法

Distributed Bayesian network learning algorithm based on model fusion

提出了一个从同构数据集中学习贝叶斯网络结构的分布式算法。该算法首先使用搜索评分的方法学习每个局部贝叶斯网络结构,然后取节点对互信息变量和条件互信息变量的数学期望作为全局学习的评价标准,融合所有局部结构得到全局结构。由于只使用了数据集中变量间的互信息和条件互信息,没有直接获取局部个体数据信息,从而可以实现有效的隐私保护。该算法在A larm数据集上进行测试,边的误差率小于6%,运行时间比集中学习的算法的运行时间短,验证了算法的有效性。

For learning Bayesian network structure from homogeneous datasets,this paper proposed a distributed algorithm.It firstly larned each local tructure using score method,then with the expectations of mutual information and conditional mutual information as evaluated criterion,it fused these local structures to obtain global structure.For using only mutual information and conditional mutual information of variation,without obtaining directly sample data,it might effectively protect privacy.Simulating the algorithm on Alarm dataset,the ratio of error is less than 6% ,the running time of the algorithm is shorter than the running time of collective algorithms,the algotithm is valid.