Generational GA Steady - State GAProblem Fitness Evaluations Fitness

Traditional genetic algorithms use operator settings such as the crossover rate or number of crossover points that are xed throughout a given run. The choice of settings can have a major eeect on performance, but nding good settings can be hard. One option is to encode the operator settings onto each member of the GA population, and allow them to evolve too. This paper describes an empirical investigation into co-evolving operator settings in genetic algorithms. The results indicate that the problem representation and the choice of operators that are applied to the encoded operator settings is important for useful adaptation to take place.