摘要
提出一种在数据缺失下增量学习贝叶斯网络的有效算法IBN-M。IBN-M用结构化的EM算法来补全数据集中缺失的数据,并且能在并行和启发式搜索策略提供的较大的搜索空间里搜索,有效地避免了采用结构化EM算法而导致的局部极值。同时采用增量学习的方法,解决了大规模数据学习存在的内存空间不足的问题。实验结果表明IBN-M算法在数据缺失下贝叶斯网络的增量学习中确实能够学出相对精确的网络模型。
This article presents an efficient algorithm--IBN-M for incremental learning in Bayesian Network with missing values. IBN-M uses the structural EM（ expectation maximisation） algorithm to complement the missing value in dataset,and it can search in a bigger searching space provided by the parallel and heuristic strategies, which effectively avoids local maximal caused by structural EM algorithm. It deploys the method of incremental learning at the same time which resolves the deficient memory space caused by large dataset learning. Experimental result indicates that the IBN-M algorithm do derive from incremental learning the relatively accurate network in the Bayesian Network with missing values.
出处
《计算机应用与软件》
CSCD
2010年第2期73-75,共3页
Computer Applications and Software
基金
国家教育部博士点基金(20060285008)
江苏省自然科学基金(BK2003030)
关键词
贝叶斯网络
增量学习
缺失数据
Bayesian network Incremental learning Missing values
