Discrete Optimization and Data Analysis: An Empirical Approach to Landscapes

Metaheuristics for discrete optimization can be thought of as dynamics on combinatorial landscapes, i.e., discrete energy landscapes. Since the features of a landscape determine the performance of a search algorithm, landscape analysis is a key tool in the selection and tuning of metaheuristics. While landscapes have been analyzed in this context, we believe there is room for further synthesis. We are currently developing a general framework for landscape analysis in order to perform a broader, more systematic study. Our main goal, rather than direct application to meta-optimization, is to understand any and all structure underlying the classes of combinatorial landscapes that appear in the world by considering a wide range of problem instances with diverse origins.