Catálogo de publicaciones - libros

The conversion and extension of the Incremental Pareto-Coevolution Archive algorithm (IPCA) into the domain of Genetic Programming classification is presented. In particular, the coevolutionary aspect of the IPCA algorithm is utilized to simultaneously evolve a subset of the training data that provides distinctions between candidate classifiers. Empirical results indicate that such a scheme significantly reduces the computational overhead of fitness evaluation on large binary classification data sets. Moreover, unlike the performance of GP classifiers trained using alternative subset selection algorithms, the proposed Pareto-coevolutionary approach is able to match or better the classification performance of GP trained over all training exemplars. Finally, problem decomposition appears as a natural consequence of assuming a Pareto model for coevolution. In order to make use of this property a voting scheme is used to integrate the results of all classifiers from the Pareto front, post training.

We define a set of measures that capture some different aspects of neutrality in evolutionary algorithms fitness landscapes from a qualitative point of view. If considered all together, these measures offer a rather complete picture of the characteristics of fitness landscapes bound to neutrality and may be used as broad indicators of problem hardness. We compare the results returned by these measures with the ones of negative slope coefficient, a quantitative measure of problem hardness that has been recently defined and with success rate statistics on a well known genetic programming benchmark: the multiplexer problem. In order to efficaciously study the search space, we use a sampling technique that has recently been introduced and we show its suitability on this problem.

We propose expression simplification and tree compression as aids in understanding the evolution of regular structure in Genetic Programming individuals. We apply the analysis to two previously-published algorithms, which aimed to promote regular and repeated structure. One relies on subtree duplication operators, the other uses repeated evaluation during a developmental process. Both successfully generated solutions to difficult problems, their success being ascribed to promotion of regular structure. Our analysis modifies this ascription: the evolution of regular structure is more complex than anticipated, and the success of the techniques may have arisen from a combination of promotion of regularity, and other, so far unidentified, effects.

Embedded Cartesian Genetic Programming (ECGP) is an extension of Cartesian Genetic Programming (CGP) capable of acquiring, evolving and re-using partial solutions. In this paper, we apply for the first time CGP and ECGP to the ones-max and order-3 deceptive problems, which are normally associated with Genetic Algorithms. Our approach uses CGP and ECGP to evolve a sequence of commands for a tape-head, which produces an arbitrary length binary string on a piece of tape. Computational effort figures are calculated for CGP and ECGP and our results compare favourably with those of Genetic Algorithms.

We explore a homeostatic approach to program execution in computer systems: the “concentration” of computation services is regulated according to their fitness. The goal is to obtain a self-healing effect so that the system can resist harmful mutations that could happen during on-line evolution. We present a model in which alternative program variants are stored in a repository representing the organism’s “genotype”. Positive feedback signals allow code in the repository to be (in analogy to gene expression in biology), meaning that it is injected into a reaction vessel (execution environment) where it is executed and evaluated. Since execution is equivalent to a chemical reaction, the program is consumed in the process, therefore needs more feedback in order to be re-expressed. This leads to services that constantly regulate themselves to a stable condition given by the fitness feedback received from the users or the environment. We present initial experiments using this model, implemented using a chemical computing language.

We have applied a range of data mining techniques to a data base of log file records created from genetic programming runs on twelve different problems. We have looked for unexpected patterns, or golden nuggets in the data. Six were found. The main discoveries were a surprising amount of evaluation of duplicate programs across the twelve problems and one case of pathological behaviour which suggested a review of the genetic programming configuration. For problems with expensive fitness evaluation, the results suggest that there would be considerable speedup by caching evolved programs and fitness values. A data mining analysis performed routinely in a GP application could identify problems early and lead to more effective genetic programming applications.

Object Oriented Evolutionary Programming is used to evolve programs that calculate some statistical measures on a set of numbers. We compared this technique with a more standard functional representation. We also studied the effects of scalar and Pareto-based multi-objective fitness functions to the induction of multi-task programs. We found that the induction of a program residing in an OO representation space is more efficient, yielding less fitness evaluations, and that scalar fitness performed better than Pareto-based fitness in this problem domain.

A fundamental issue in evolutionary learning is the definition of the solution representation language. We present the application of Object Oriented Genetic Programming to the task of coevolving general recursive sorting algorithms along with their primitive representation alphabet. We report the computational effort required to evolve target solutions and provide a comparison between crossover and mutation variation operators, and also undirected random search. We found that the induction of evolved method signatures (typed parameters and return type) can be realized through an evolutionary fitness-driven process. We also found that the evolutionary algorithm outperformed undirected random search, and that mutation performed better than crossover in this problem domain. The main result is that modular sorting algorithms can be evolved.

This work analyzes fitness landscapes for the image filter design problem approached using functional-level Cartesian Genetic Programming. Smoothness and ruggedness of fitness landscapes are investigated for five genetic operators. It is shown that the mutation operator and the single-point crossover operator generate the smoothest landscapes and thus they are useful for practical applications in this area. In contrast to the gate-level evolution, a destructive behavior of a simple crossover operator has not been confirmed.

In this paper we present a method for creating scheduling heuristics for parallel proportional machine scheduling environment and arbitrary performance criteria. Genetic programming is used to synthesize the priority function which, coupled with an appropriate meta-algorithm for a given environment, forms the priority scheduling heuristic. We show that the procedures derived in this way can perform similarly or better than existing algorithms. Additionally, this approach may be particularly useful for those combinations of scheduling environment and criteria for which there are no adequate scheduling algorithms.