Alberto Ochoa’s research while affiliated with Institute of Cybernetics and other places

A new class of estimation of distribution algorithms (EDAs), known as cellular EDAs (cEDAs), has recently emerged. In these algo- rithms, the population is decentralized by partitioning it into many small collaborating subpopulations, arranged in a toroidal grid, and interacting only with its neighboring subpopulations. In this work, we study the sim- plest cEDA —the cellular univariate marginal distribution algorithm (cUMDA). In an attempt to explain its behaviour, we extend the well known takeover time analysis usually applied to other evolutionary algo- rithms to the field of EDAs. We also give in this work empirical arguments in favor of using the cUMDAs instead of its centralized equivalent.

The success of evolutionary algorithms, in particular Factorized Distribution Algorithms (FDA), for many pattern recognition tasks heavily depends on our ability to reduce the number of function evaluations. This paper introduces a method to reduce the population size overhead. We use low order marginals during the learning step and then compute the maximum entropy joint distributions for the cliques of the graph. The maximum entropy distribution is computed by an Iterative Proportional Fitting embedded in a junction tree message passing scheme to ensure consistency. We show for the class of single connected FDA that our method outperforms the commonly-used PLS sampling.

This short paper surveys current work on the use of Factorized Distribution Algorithms for the solution of combinatorial optimization problems denned on graphs. We also advance a number of approaches for future work along this line

Single connected Factorized Distribution Algorithms (FDA-SC) use factorizations of the joint distribution, which are trees, forests or polytrees. At each stage of the evolution they build a polytree from which new points are sampled. We study empirically the relation between the accuracy of the learned model and the quality of the new search points generated. We show that a change of the learned model before sampling might reduce the population size requirements of sampling.

The class of factorized distribution algorithms (FDA) uses factorizations of the joint distribution of the best points. At each stage of the evolution, PDA algorithms build a model from which new points are efficiently sampled. This paper explores the class of single connected factorizations: polytrees. Using this class, we gain in efficiency and simplicity in the procedures for learning the networks. The price we have to pay is a less expressive power. However, sometimes the representation power of polytrees is adequate for optimization purposes