Instance-based Algorithm Selection on Quadratic Assignment Problem Landscapes

Outline

A. Beham, M. Affenzeller, S. Wagner - Instance-based Algorithm Selection on Quadratic Assignment Problem Landscapes - GECCO '17: Proceedings of the Genetic and Evolutionary Computation Conference Companion, Berlin, Germany, Deutschland, 2017, pp. 1471-1478

Abstract

Among the many applications of fitness landscape analysis a prominent example is algorithm selection. The no-free-lunch (NFL) theorem has put a limit on the general applicability of heuristic search methods. Improved methods can only be found by specialization to certain problem characteristics which limits their application to other problems. This creates a very interesting and dynamic field for algorithm development. However, this also leads to the definition of a large range of different algorithms that are hard to compare exhaustively. An additional challenge is posed by the fact that algorithms have parameters and thus to each algorithm there may be a large number of instances. In this work the application of algorithm selection to problem instances of the quadratic assignment problem (QAP) is discussed.