Learning Structure Illuminates Black Boxes – An Introduction to Estimation of Distribution Algorithms J¨ orn Grahl 1 , Stefan Minner 1 , and Peter A. N. Bosman 2 1 Department of Logistics University of Mannheim Schloss, 68131 Mannheim joern.grahl/minner@bwl.uni-mannheim.de 2 Centre for Mathematics and Computer Science P.O. Box 94079 1090 GB Amsterdam Peter.Bosman@cwi.nl Summary. This chapter serves as an introduction to estimation of distribution algorithms (EDAs). Estimation of distribution algorithms are a new paradigm in evolutionary computation. They combine statistical learning with population-based search in order to automatically identify and exploit certain structural properties of optimization problems. State-of-the-art EDAs consistently outperform classical genetic algorithms on a broad range of hard optimization problems. We review fundamental terms, concepts, and algorithms which facilitate the understanding of EDA research. The focus is on EDAs for combinatorial and continuous non-linear optimization and the major differences between the two fields are discussed. 1 Introduction In this chapter, we give an introduction to estimation of distribution algorithms (EDA, see [60]). Estimation of distribution algorithms constitute a novel paradigm in evolutionary computation (EC). The EDA principle is still being labelled differently in the literature: estimation of distribution algorithms, probabilistic model building genetic algorithms (PMBGA), iterated density estimation evolutionary algorithms (IDEA) or optimization by building and using probabilistic models (OBUPM). For the sake of brevity we call this class of algorithms EDAs. EDAs have emerged in evolutionary computation from research into the dynam- ics of the simple genetic algorithm (sGA, see [45] and [32]). It has been found in this research that using standard variation operators, e.g., two-parent recombination or mutation operators, easily leads to an exponential scale-up behavior of the sGA. This means, that the required time measured by the number of fitness evaluations to reliably solve certain optimization problems grows exponentially with the size of