Parental and cyclic-rate mutation in genetic algorithms: an initial investigation
Theodore P. Hoehn, Chrisila C. Pettey · 1999
Genetic algorithms (GAs) have traditionally combined and confounded the effects of gametic and somatic mutation. In an attempt to disentangle these effects, a novel, modified GA is proposed in which mutation occurs not only following selection (i.e., in offspring)— as is the tradition—but also at the parental level, between evaluation and selection of population members. Two experiments were conducted that compared the relative performances of several mutational-rate variants of the proposed parental mutation GA with corresponding standard GAs matched with them in terms of overall probability of mutation during an iteration of the algorithm. Results indicate that inclusion of parental mutation within a GA does not in general adversely affect its performance. For both types of GA, a novel, cyclic-rate mutation variant outperformed the low and (very) high constant-rate variants. Results of a third experiment indicate that, for the most part, it is irrelevant at which algorithmic locus (parental vs. offspring) the cyclic-rate mutation occurs or whether there is also mutation at the other locus.