Structure Learning of Bayesian Networks Based on Vertical Segmentation Data

Abstract

A distributed approach in learning a Bayesian networks from vertical segmentation data was promoted in the paper. The approach includes four sequential steps: local learning, sample selection, cross learning, and combination of the results. The main improvement of the algorithm brings forward in the second step. The complex sub-structure of local BN is considered that exist a hidden node which contacts with the sub-structure. The hidden node exist in the other local BN. The experiment proved that the distributed learning method can learn almost the same structure as the result obtained by a centralized learning method.