A ´non-model building´ approach to solving hierarchical functions

The hierarchical Bayesian optimization algorithm (hBOA) by M. Pelikan and D.E. Goldberg (2001), used diversity preservation along with the original Bayesian optimization algorithm BOA by M. Pelikan et al. (1999) to tackle boundedly difficult hierarchical functions. However, model building can be an expensive process, and a pertinent question is the possibility of developing operators that can solve certain classes of hierarchical functions in the traditional GA domain. This study shows, that by following a three-step approach to hierarchical problem solving - effective linkage learning, merging of low-order BBs, and diversity preservation - it is possible to use competent (non-model building) selec-to-recombinative GAs to solve certain classes of hierarchical functions. Experimental bounds were found on the type of hierarchical problems that could be solved, and perturbation based linkage detection was found to be the limiting factor.