Hans-Georg Beyer: The Theory of Evolution Strategies, Kartoniert / Broschiert

Beschreibung

Evolutionary algorithms, such as evolution strategies, genetic algorithms, or evolutionary programming, have found broad acceptance in the last ten years. In contrast to its broad propagation, theoretical analysis in this subject has not progressed as much. This monograph provides the framework and the first steps toward the theoretical analysis of Evolution Strategies (ES). The main emphasis is deriving a qualitative understanding of why and how these ES algorithms work. Evolutionary Algorithms, in particular Evolution Strategies, Genetic Algorithms, or Evolutionary Programming, have found wide acceptance as robust optimization algorithms in the last ten years. Compared with the broad propagation and the resulting practical prosperity in different scientific fields, the theory has not progressed as much. This monograph provides the framework and the first steps toward the theoretical analysis of Evolution Strategies (ES). The main emphasis is on understanding the functioning of these probabilistic optimization algorithms in real-valued search spaces by investigating the dynamical properties of some well-established ES algorithms. The book introduces the basic concepts of this analysis, such as progress rate, quality gain, and self-adaptation response, and describes how to calculate these quantities. Based on the analysis, functioning principles are derived, aiming at a qualitative understanding of why and how ES algorithms work.

Inhaltsangabe

1. Introduction.- 2. Concepts for the Analysis of the ES.- 3. The Progress Rate of the (1
% MathType!MTEF!2!1!+-
% feaagCart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn
% hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr
% 4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq-Jc9
% vqaqpepm0xbba9pwe9Q8fs0-yqaqpepae9pg0FirpepeKkFr0xfr-x
% fr-xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaaeaaaaaaaaa8
% qadaqadaWdaeaapeGaaGyma8aadaWfGaqaa8qacaGGSaaal8aabeqa
% a8qacqGHRaWkaaGccqaH7oaBaiaawIcacaGLPaaaaaa!3C4E!
$$\left( {1\mathop ,\limits^ + \lambda } \right)$$
?)-ES on the Sphere Model.- 5. The Analysis of the (?, ?)-ES.- 6. The (?/?, ?) Strategies or Why Sex May be Good.- 7. The (1, ?)-?-Self-Adaptation.- Appendices.- A. Integrals.- A. 1 Definite Integrals of the Normal Distribution.- A. 2 Indefinite Integrals of the Normal Distribution.- A. 3 Some Integral Identities.- B. Approximations.- B. 1 Frequently Used Taylor Expansions.- B. 3 Cumulants, Moments, and Approximations.- B. 3.1 Fundamental Relations.- B. 3.2 The Weight Coefficients for the Density Approximation of a Standardized Random Variable.- B. 4 Approximation of the Quantile Function.- C. The Normal Distribution.- C. 3 Product Moments of Correlated Gaussian Mutations.- C. 3.1 Fundamental Relations.- C. 3.2 Derivation of the Product Moments.- D. (1, ?)-Progress Coefficients.- D. 2 Table of Progress Coefficients of the (1, ?)-ES.- References.

Klappentext

Evolutionary Algorithms, in particular Evolution Strategies, Genetic Algorithms, or Evolutionary Programming, have found wide acceptance as robust optimization algorithms in the last ten years. Compared with the broad propagation and the resulting practical prosperity in different scientific fields, the theory has not progressed as much.
This monograph provides the framework and the first steps toward the theoretical analysis of Evolution Strategies (ES). The main emphasis is on understanding the functioning of these probabilistic optimization algorithms in real-valued search spaces by investigating the dynamical properties of some well-established ES algorithms. The book introduces the basic concepts of this analysis, such as progress rate, quality gain, and self-adaptation response, and describes how to calculate these quantities. Based on the analysis, functioning principles are derived, aiming at a qualitative understanding of why and how ES algorithms work.