摘要
基于Bayes统计理论,提出了一种从数据样本中学习Bayes网络的Markov链Monte Carlo(MCMC)方法。首先通过先验概率和数据样本的结合得到未归一化的后验概率,然后使用此后验概率指导随机搜索算法寻找“好”的网络结构模型。通过对Alarm网络的学习表明了本算法具有较好的性能。
A Bayesian network is a graphical model that encodes probabilistic relationships among variables of interest. In many cases, the authors hoped to learn Bayesian networks from data. Using the Markov chain Monte Carlo (MCMC) approach, this paper proposed a Bayesian statistical method for learning Bayesian networks from data, in terms of network structures and parameters. Prior specification and stochastic search were two important components of this approach. The combination of prior probability and data samples induced a posterior distribution that would guide the stochastic search towards the network structures having the maximal posterior probability. The performance of this approach is illustrated by the learning of the Alarm network from data.
出处
《控制理论与应用》
EI
CAS
CSCD
北大核心
2003年第4期582-584,588,共4页
Control Theory & Applications
基金
国家自然科学基金(60073053)
