Metaheuristics are a relatively new but already established approachto c- binatorial optimization. A metaheuristic is a generic algorithmic template that can be used for ?nding high quality solutions of hard combinatorial – timization problems. To arrive at a functioning algorithm, a metaheuristic needs to be con?gured: typically some modules need to be instantiated and someparametersneedto betuned.Icallthese twoproblems”structural”and “parametric” tuning, respectively. More generally, I refer to the combination of the two problems as “tuning”. Tuning is crucial to metaheuristic optimization both in academic research andforpracticalapplications.Nevertheless,relativelylittle researchhasbeen devoted to the issue. This book shows that the problem of tuning a me- heuristic can be described and solved as a machine learning problem. Using the machine learning perspective, it is possible to give a formal de?nitionofthetuningproblemandtodevelopagenericalgorithmfortuning metaheuristics.Moreover,fromthemachinelearningperspectiveitispossible tohighlightsome?awsinthecurrentresearchmethodologyandtostatesome guidelines for future empirical analysis in metaheuristics research. This book is based on my doctoral dissertation and contains results I have obtained starting from 2001 while working within the Metaheuristics Net- 1 work. During these years I have been a?liated with two research groups: INTELLEKTIK, Technische Universität Darmstadt, Darmstadt, Germany and IRIDIA, Université Libre de Bruxelles, Brussels, Belgium. I am the- fore grateful to the research directors of these two groups: Prof. Wolfgang Bibel, Dr. Thomas Stützle, Prof. Philippe Smets, Prof. Hugues Bersini, and Prof. Marco Dorigo.
Editorial Reviews
From the Back Cover
The importance of tuning metaheuristics is widely acknowledged in scientific literature. However, there is very little dedicated research on the subject. Typically, scientists and practitioners tune metaheuristics by hand, guided only by their experience and by some rules of thumb. Tuning metaheuristics is often considered to be more of an art than a science.
This book lays the foundations for a scientific approach to tuning metaheuristics. The fundamental intuition that underlies Birattari’s approach is that the tuning problem has much in common with the problems that are typically faced in machine learning. By adopting a machine learning perspective, the author gives a formal definition of the tuning problem, develops a generic algorithm for tuning metaheuristics, and defines an appropriate experimental methodology for assessing the performance of metaheuristics.
Wow! eBook
