Model Complexity vs. Performance in the Bayesian Optimization Algorithm

Abstract

The Bayesian Optimization Algorithm (BOA) uses a Bayesian network to estimate the probability distribution of promising solutions to a given optimization problem. This distribution is then used to generate new candidate solutions. The objective is to improve the population of candidate solutions by learning and sampling from good solutions. A Bayesian network (BN) is a graphical representation of a probability distribution over a set of variables of a given problem domain. The number of topological states that a BN can create depends on a parameter called maximum allowed indegree. We show that the value of the maximum allowed indegree given to the Bayesian network used by the BOA strongly affects the performance of this algorithm. Furthermore, there is a limited set of values for this parameter for which the performance of the BOA is maximized.KeywordsBayesian NetworkCandidate SolutionPartial SolutionConjunctive Normal FormSpurious CorrelationThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.