Abstract: To improve the efficiency of Bayesian Network(BN) structure learning, this paper developes a novel algorithm based on chain-model and particle swarm. It measures the quality of topological orders with a regular chain-model which contains information of causality between nodes, so that the quality of the searched topological orders is improved. And then the position-selectable updating PSO is optimized with dynamical weighting coefficient, which balances global search and local search and improves search capability. Experimental results show that the novel algorithm can get a better solution, and convergence rate is enhanced with a comparison to the I-ACO-B algorithm.
Bayesian Network(BN) structure learning,
Particle Swarm Optimization(PSO) algorithm,
conditional independence test