Learning Bayesian Network from Structure Boundaries

Bayesian network is an important theoretical tool in the artificial algorithm field, and learning structure from data is considered as NP-hard. In this article, a hybrid learning method is proposed by starting from analysis of information provided by low-order conditional independence testing. The methods of constructing boundaries of the structure space of the target network are given, as well as the complete theoretical proof. A search scoring algorithm is operated to find the final structure of the network. Simulation results show that the hybrid learning method proposed in this article has higher learning precision and is more efficient than similar algorithms.