Conduct Evolution Using Induction Learning
Ruan Huai · 2001
Genetic algorithms must try to keep the balance between the positive effect of the selector and the disruptiveness of crossover or mutation, otherwise it will suffer from premature convergence or low convergence speed. Unlike other improvement works, this paper proposes a method that conducts evolution to keep the balance through induction learning. Rules were drawn from previous evolutions which reflect past successes or failures. They were then used to conduct evolutions to avoid the repeated failures and speed up the evolution. Experiments on function optimization and facility layout have validated its effectiveness.