Swarm Intelligence

Through this process, generations tend to improve; this amounts to a

kind of stochastic hill climbing, as the more fit members of the popula-

tion are more likely to be retained and mutated.

De Jong, Potter, and Spears also tested a GA implemented with cross-

over only—no mutation—and found that this kind tended to flounder in

the early generations, but once the population started to converge on

one particular peak or another, improvement came relatively fast. These

crossover-only GAs, called GA-C, were very successful at finding one

of the optima, if you waited long enough. The same was true of tradi-

tional GAs with both crossover and mutation. Mutation-only ...