Linkage learning using graphical markov model structure: An experimental study

Abstract

Linkage learning to identify the interaction structure of an optimization problem helps the evolutionary algorithms to search the optimal solution. Learning the structure of a distribution representing the interactions of the optimization problem is equivalent to learning the variables of the problem linkage. The objective of this paper is to test the efficiency of an EDA that use Boltzmann selection and a cliqued Gibbs sampler (named Adaptive Extended Tree Cliqued - EDA (AETCEDA)) to learn the linkage of the problem and generate samples. Some optimization problems difficult for the Genetic Algorithms are used to test the proposed algorithm. As results of the experiment is to emphasize that the difficulty of the optimization, as assessed by the number of evaluations, is proportional to the sizes of the cliques of the learned models, that in time, is proportional to the structure of the test problem. © Springer-Verlag Berlin Heidelberg 2013.