Hill Climbing Beats Genetic Search on a Boolean Circuit Synthesis Problem of Koza's

摘要

An experiment described in chapter 9 of the book "Genetic Programming" shows that the method is more efficient than random-generate-and-test on a boolean circuit synthesis task. Here we show that hill climbing is more efficient than genetic programming on this problem. It is interesting to note that our improved results were obtained by mating the current best hypothesis with completely random S-expressions, rather than with members of a high-fitness population. Perhaps, for this task, fragments of high fitness individuals have no special value when transplanted into other individuals.