Fitness Landscape Analysis: From Problem Understanding to Design of Evolutionary Algorithms

As an effective kind of optimization technique, evolutionary algorithms (EAs) have been widely used in many fields. In the context of EAs, to determine which algorithm would be the most appropriate for the specific problem, the techniques of fitness landscape analysis (FLA) are often used to identify the landscape features of the optimization problem, and select the best algorithm for the problem according to the features. In fact, the landscape features can also be considered as useful feedback for guiding the search procedure of EAs. In recent years, there are some seminal works which use FLA techniques to design EAs, the key concept of them is to adaptively configure algorithms according to the landscape features. In this work, we provide a briefly survey on this emerging research topic, including two aspects: 1) Some representative FLA techniques and the applications to the design of EAs are introduced, with the aim of highlighting the possible ways of extending the FLA techniques from problem understanding to the design of EAs. 2) The constraint of the existing FLA techniques is discussed, and the possible solution, the online FLA techniques, is also presented. So, in this survey, the related FLA techniques and corresponding EAs variants are reviewed according to the answers to the two questions.