【摘要】 对贝叶斯网络的在线参数学习进行了研究。分析了ML和Voting EM算法的特点。发现它们在快速适应样本特征变化、预测与确定算法参数方面存在的不足,并提出基于上述两种方法的混合在线学习算法。改进算法根据修正参数误差以及调节数据量权重动态获取与确定贝叶斯网络。研究结果表明,改进算法在快速获取知识参数与知识检验正确率方面,比Voting EM方法具有更好的特点。更多还原

【Abstract】 Online learning Bayesian network parameter is studied.Characteristics of ML and Voting EM algorithm are analyzed.Shortage of adapting to the characteristics of samples and lack of predicting and determining algorithm parameters are found in these methods,and a novel algorithm based on the combination of former two methods is proposed.The modified algorithm obtains and determines Bayesian network according to amended parameter errors and adjustments of the weight of data.The results show that the proposed algorithm is better than Voting EM method in the speed of learning parameters and the accuracy of the learned knowledge.更多还原