A-HEP: Adaptive Hybrid Evolutionary Programming for Learning Bayesian Networks - Liang Yong

This paper describes an optimized algorithm for learning Bayesian Network structure by using adaptive population sized evolutionary programming. Bayesian network (BN) is a popular knowledge discovery model which can represent the causal relationship of different events or attributes with uncertainty. Learning the structure solely by dependency analysis or search-and-score approach is not effective. The hybrid algorithm on evolutionary programming, HEP, has been shown to be effective and efficient to solve this learning problem [8]. By introducing the concept of adjusting the population size according to the individuals’ dissimilarity, HEP is further optimized on the execution time with comparable performance. The empirical results illustrate that the optimized algorithm has reduced the running time by half.