Coevolutionary, Distributed Search for Inducing Concept Descriptions

This paper presents a highly parallel genetic algorithm, designed for concept induction in propositional and first order logics. The parallel architecture is an adaptation for set covering problems, of the diffusion model developed for optimization. The algorithm exhibits other two important methodological novelties related to Evolutionary Computation. First, it combines niches and species formation with coevolution, in order to learn multimodal concepts. This is done by integrating the Universal Suffrage selection operator with the coevolution model recently proposed in the literature. Second, it makes use of a new set of genetic operators, which maintain diversity in the population. The experimental comparison with previous systems, not using coevolution and based on traditional genetic operators, shows a substantial improvement in the effectiveness of the genetic search.