Learning Essential Graph with Immune Co-evolutionary Algorithm

Essential graph is a graphical representation for Markov equivalence classes of Bayesian networks. Learning essential graph can avoid some problems in traditional Bayesian networks learning algorithms: (1) the number of illegal structures is exponential, which infect the efficiency of structure learning; (2) comparing the structures in same equivalent class slow down the speed of convergence;(3) if the prior distribution for each structure is equal, the more structures contain in the equivalent class the higher prior probability of the class has. This paper employs two competitive bio-inspired algorithms, immune algorithm and co-evolutionary algorithm, for learning Essential graph. The algorithm combines dependency analysis and search-scoring approach together. Experiments show that the searching space was decreased, compare with prior works, the convergence speed and the efficiency was improved.