Marta Soto

A Factorized Distribution Algorithm Using Single Connected Bayesian Networks

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.

Linking Entropy to Estimation of Distribution Algorithms

This chapter presents results on the application of the concept of entropy to estimation of distribution algorithms (EDAs). Firstly, the Boltzmann mutual information curves are introduced. They are shown to contain a lot of information about the difficulty of the functions. Next, a design method of discrete benchmark functions is presented. The newly developed approach allows the construction of both single and random classes of functions that obey a given collection of probabilistic constraints. This application and the next — the construction of low cost search distributions — are based on the principle of maximum entropy. The last proposal is the linear entropic mutation (LEM), an approach that measures the amount of mutation applied to a variable as the increase of its entropy. We argue that LEM is a natural operator for EDAs because it mutates distributions instead of single individuals.

A Maximum Entropy Approach to Sampling in EDA – The Single Connected Case

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.

Vine Estimation of Distribution Algorithms with Application to Molecular Docking

Four undirected graphical models based on copula theory are investigated in relation to their use within an estimation of distribution algorithm (EDA) to address the molecular docking problem. The simplest algorithms considered are built on top of the product and normal copulas. The other two construct high-dimensional dependence models using the powerful and flexible concept of vine-copula. Empirical investigation with a set of molecular complexes used as test systems shows state-of-the-art performance of the copula-based EDAs in the docking problem. The results also show that the vine-based algorithms are more efficient, robust and flexible than the other two. This might suggest that the use of vines opens new research opportunities to more appropriate modeling of search distributions in evolutionary optimization.

Theory and practice of cellular UMDA for discrete optimization

A new class of estimation of distribution algorithms (EDAs), known as cellular EDAs (cEDAs), has recently emerged. In these algorithms, 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 simplest 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 algorithms to the field of EDAs. We also give in this work empirical arguments in favor of using the cUMDAs instead of its centralized equivalent.

A fitness function model for detecting ellipses with Estimation of Distribution Algorithms

This paper introduces a novel method for ellipse detection that is based on Estimation of Distribution Algorithms. The main contribution is the construction of a new fitness function model that in contrast to existing methods can assign positive evaluations to ellipses that do not exist in the image. This approach produces much smoother fitness landscapes increasing the efficiency of the search. A preliminary study of the role of mutation in dealing with the natural multimodality of the ellipse detection problem is also presented.

Too busy to learn [individual learning interaction with evolutionary algorithm in Busy Beaver problem]

The goal of this research is to analyze how individual learning interacts with an evolutionary algorithm in its search for best candidates for the Busy Beaver problem. To study this interaction, two learning models, implemented as local search procedures, are proposed. Experimental results show that, in highly irregular search spaces that are prone to premature convergence, local search methods are not an effective help to evolution. In addition, one interesting effect related to learning is reported: when the mutation rate is too high, learning acts as a repair, reintroducing some useful information that was lost.

On the use of Factorized Distribution Algorithms for problems defined on graphs

Abstract 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