Evolutionary Algorithms in Theory and Practice: : Evolution Strategies, Evolutionary Programming, Genetic Algorithms

Citation

Galletly, J. (1998), "Evolutionary Algorithms in Theory and Practice: : Evolution Strategies, Evolutionary Programming, Genetic Algorithms", Kybernetes, Vol. 27 No. 8, pp. 979-980. https://doi.org/10.1108/k.1998.27.8.979.4

Bäck’s book provides an in‐depth examination and comparison of three classes of so‐called evolutionary algorithms, viz. evolution strategies (ESes), evolutionary programming (EP) and genetic algorithms (GAs). For the uninitiated, evolutionary strategies are a rather hot topic in optimization problem research and development. They are a specialized, heuristic class of probabilistic search and optimization algorithms based on the metaphor of natural selection and evolution. The major representatives of evolutionary algorithms, according to Bäck, are ESes, EP and GAs, each of which were developed independently. Each of these novel computing paradigms are dealt with by Bäck, but the emphasis is put on GAs, albeit extensions and analysis of GAs.

Part I presents a comparison of evolutionary algorithms. One chapter details the biological background to evolutionary algorithms, explaining how biological and genetic concepts such as populations of individuals, and operations such as selection, recombination and mutation can be mapped into evolutionary algorithm paradigms for optimization problem solving. Bäck gives a readable description, short but full enough, of the natural biological processes involved. He then explains how these concepts are used as metaphors for the novel computational search and optimization process: a population of trial solutions instead of a single one; and computational operators, based on their biological counterparts, to manipulate, transform and refine the trial solutions. Another chapter examines and analyses each of the three evolutionary algorithm representatives, in turn, highlighting their different characteristics. Of interest, in its own right, is the historical development of these different approaches by Rechenberg and Schwefel, Fogel and Fogel, and Holland, respectively. Some new theoretical results concerning recombination in ESes, the convergence velocity and selection algorithm of EP, and convergence properties of GAs are presented. Two more chapters are devoted a description of a number of theoretical test functions (“artificial landscapes”, as Bäck terms them) and an empirical comparison of the evolutionary algorithms’ performance with these topologies.

Part II (entitled “Extending genetic algorithms”) concentrates on GAs and describes how the ES view (and, to a lesser degree, the EP view) of selection and mutation are transferred to GAs. The impact of the selection operators and the mutation rate is investigated. In particular, Bäck analyses the selective pressure imposed on the evolutionary search by proportional selection, ranking, tournament selection, (μ, λ)‐selection and (μ + λ)‐selection. Back also confirms that mutation is not the simplistic operator as some would argue, but has complex properties and a non‐trivial interaction with the selection operator. The section finishes with a description of a meta‐evolutionary algorithm ‐ one designed to solve the problem of determining the best values for the GA parameters.

This book represents a excellent treatise on evolutionary algorithms, and may well become a classic text for the field, given Bäck’s solid mathematical treatment of the subject matter and clear descriptions. The style and organization of the book means it is more suitable for reference, rather than a course text book. The work is careful and well‐researched. Bäck reveals some theoretically interesting results based on convergence velocity analysis which deserve a wider recognition.