Stochastic search in metaheuristics

Abstract

Stochastic search is a key mechanism underlying many metaheuristics. The chapter starts with the presentation of a general framework algorithm in the form of a stochastic search process that contains a large variety of familiar metaheuristic techniques as special cases. Based on this unified view, questions concerning convergence and runtime are discussed at the level of a theoretical analysis. Concrete examples from diverse metaheuristic fields are given. In connection with runtime results, important topics such as instance difficulty, phase transitions, parameter choice, No-Free-Lunch theorems or fitness landscape analysis are addressed. Furthermore, a short sketch of the theory of black-box optimization is given, and generalizations of results to stochastic search under noise and to robust optimization are outlined.