Estimation of distribution programming based on Bayesian network

We propose estimation of distribution programming (EDP) based on a probability distribution expression using a Bayesian network. EDP is a population-based program search method, in which the population probability distribution is estimated, and individuals are generated based on the results. We focus our attention on the fact that the dependency relationship of nodes of the program (expressed as a tree structure) is explicit, and estimate the probability distribution of the program population using a Bayesian network. We compare EDP with GP (genetic programming) on several benchmark tests, i.e., a max problem and a Boolean function problem. We also discuss the trends of problems that are the forte of EDP.