Learning the Structure of Dynamic Bayesian Network with Hybrid Data and Domain Knowledges

Dynamic Bayesian Networks (DBNs) is a powerful graphical model for representing temporal stochastic processes. Learning the structure of DBNs is the fundamental step for parameter learning, inference, application etc. In some cases, such as computational systems biology, learning the structure of DBNs facing the two challenges (1) experimental settings only capture few time series and steady state measurements. (2) the knowledge about DBNs is uncertainty, rare and even with conflict. The paper considers the time series data, steady state and domain knowledge simultaneously, presents a novel algorithm for learning the structure of DBNs. Compare with single source learning, empirical experiment shows that learning with hybrid data and domain knowledges improved the accuracy and effectiveness of the DBNs structure learning.