Two Remarkable Computational Competencies of the Simple Genetic Algorithm

When applied to one-off instances of a wide range of combinatorial optimization problems the Simple Genetic Algorithm (SGA) frequently generates solutions of high fitness. Till date, however, the discovery of one or more seemingly hard problems (consisting of an infinite set of problem instances) that can be solved robustly and scalably by an SGA has eluded researchers. This is unfortunate because the identification of such problems is highly germane to the formulation and verification of explanations for the adaptive capacity of the SGA. The search for such problems has largely been conducted under one or both of the following assumptions: i) that the computational competencies of the SGA arise from its purported proficiency at identifying and composing building blocks, and ii) that the computational competencies of the SGA are to be found in its capacity for global optimization on classes of fitness functions yet to be pinpointed. In this paper we identify two seemingly hard problems that the SGA can solve scalably and robustly. In doing so we showcase computational competencies of the SGA that repudiate both of the assumptions mentioned above. Remarkably the two problems that we identify are closely related to a hard data-mining problem at the cutting edge of genetics having to do with the identification of epistatically interacting quantitative trait loci.