Intellectics Group: Technical Report 01-05
Classification of Metaheuristics and Design of
Experiments for the Analysis of Components
Mauro Birrattari and Luis Paquete and Thomas
Stützle and Klaus Varrentrapp
Submitted to GECCO 2002
This report discusses two different approaches to the description of meta-heuristics. On one hand, we propose a number of different high-level cri-teria according to which metaheuristics can be described and classified. On the other hand, we discuss some method of design of experiments for studying the contribution and the relative importance of the different com-ponents of a metaheuristic. We maintain that, in order to be effective and complete, an analysis of metaheuristics must take into account both levels. In particular, such a two-fold investigation is necessary for coping with the fact that all metaheuristics, though described as clearly defined methods based on some intellectually appealing principles, when implemented in practice, they always appear in some hybrid form. Accordingly, with an high-level description, we wish to grasp the underlying principle, while by adopting statistical techniques of design of experiments we wish to pon-derate the contribution given to performance by each components in its possible instantiations.