关键词: 贝叶斯网 结构学习 马尔科夫等价类 链图

Abstract:

Bayesian networks （BNs） is a compact representation of complex joint probability distribution. BNs have sound probabilistic semantics, explicit enco
          ding of relevance relationships which led BNs to be one of the best methods for dealing with uncertainty in the AI domain. Learning BNs from data has be come an important problem, and there are some problems in the existing learning algorithms： （1） The number of illegal structures is exponential, whic    ch infects the efficiency of structural learning ; （2） Comparing the structures in the same equivalent class slows down the speed of convergence; （3 ） If the prior distribution of each structure is equal, the more the structures contained in the equivalent class, the higher the probability of the cl   lass. This paper presents an algorithm for learning the Markov equivalence class of the Bayesian Network, which maps the search space from the skeleton  space to the Markov equivalent class space. Experiments show that the searching space is decreased, compared with the algorithm searching in the Directe  d Acyclic Graph space, the convergence speed and the efficiency are improved

Key words: Bayesian network, structural learning, Markov equivalence class;chain graph