Adaptively Resizing Populations: Algorithm, Analysis, and First Results

Abs tract . Deciding on an appropriate population size for a given genet ic algor it hm (GA) applicat ion can oft en be crit ical to the success of the algorit hm . Too small, and t he GA can fall vict im to sampling erro r , affect ing the efficacy of it s search . Too large, and t he GA wastes computational resour ces. Although advice exists for sizing GA popula tions, much of this adv ice involves theoret ical aspects that are not access ible to the novice. T his paper sugges ts an algori thm for adapt ively resizing GA populat ions. The algorit hm is suggested based on recent theoret ical developments that relate population size to schema fitness variance. T he algorithm is developed theoret ically, simulated wit h expecte d value equat ions , and test ed on a pro blem where populat ion sizing can mislead the GA. The pos it ive results pr esented suggest that adapt ively sizing GA populations may be a practi cal extension to the typ ical GA . Such an extension frees the user from a crit ical param eter decision , an d expands the usefulness of GA search. Moreover , t his extension creates a new, interest ing class of genetic search syste ms, where adaptive cha nges in population size reflect problem complexity.