In recent years, genetic algorithms (GA) have gained some attention as general purpose problem optimizers. The simple interface of GAs make them easily applicable to a wide spectrum of problems. Most of the problem-specific knowledge is hidden in the coding of the chromosomes and in the fitness function. Additionally, special operators and a problem-dependent parameterization might be required.
In Chapter 1, we focus on selection schemes of GAs. We investigate the accuracy of various sampling schemes and propose an algorithm with an adjustable accuracy. Tournament selection is a prominent selection method in the field of evolutionary computation. We extend the basic idea of tournament selection by mixed size tournaments. The selection pressure of mixed size tournament selection can be fine-tuned comparable to linear or exponential ranking selection.
Chapter 2 and 3 are dedicated to applications of GAs. In image processing, the wavelet transformation is a state of the art technology and is therefore often part of modern compression software. In Chapter 2, we concentrate on two- dimensional wavelet packet decomposition obtained by the best-basis- and near-best- basis-algorithm. We compare the performance of the latter algorithms and a GA. A new crossover-operator has been designed to aim for improved results of the GA.
Pseudo random number generators are ubiquitous in the field of computer science. In Chapter 3, we concentrate on the identification of good multipliers for linear congruential generators (LCG). As figures of merit, we apply the spectral test and the continued-fraction-test. A GA was applied for the multiplier search and, as a byproduct, provided ideas for an efficient encoding of multipliers. In a parallel environment, we utilized the efficient encoding for a large scale search for good LCG- parameters.