Extending the scalability of linkage learning genetic algorithms. Theory \& practice. (Q2571315)

From the preface of the book: ``There are two primary ojectives of this monograph. The first goal is to identify certain limits of genetic algorithms that use only fitness for learning genetic linkage. Both an explanatory theory and experimental results to support the theory are provided. The other goal is to propose a better design of the linkage learning genetic algorithm. After understanding the cause of the observed performance barrier, the design of the linkage learning genetic algorithm is modified accordingly to improve its performance on the problems of uniformly scaled building blocks. This book starts with presenting the background of the linkage learning genetic algorithm. Then, it introduces the use of promoters on chromosomes to improve the performance of the linkage learning genetic algorithm on uniformly scaled problems. The convergence time model is constructed by identifying the sequencial behavior, developing the tightness time model, and establishing the connection in between. The use of subchromosome representations is to avoid the limit implied by the convergence time model. The experimental results suggest that the use of chromosome representations may be a promising way to design a better linkage learning genetic algorithm. The study depicted in this monograph finds that using promoters on the chromosome can improve nucleation potential and promote correct building-block formation. It also observes that the linkage learning genetic algorithm has a consistent, sequential behavior instead of different behaviors on different problems as was previously believed. Moreover, the competion among building blocks of equal salience is the main cause of exponencial growth of convergence time. Finally, adopting subchromosome representations can reduce the competition among the building blocks, and therefore, scalable genetic linkage learning for a unimetric approach is possible.'' The book has a foreword by David E. Goldberg, is divided into 9 chapters and presents a list of 98 references.