Choice Review

If one takes seriously Darwin's theory of evolution by natural selection, then it is reasonable to base optimization methods on selection. Genetic algorithms are a caricature of real genetic systems--chromosomes are replaced by bit vectors representing possible solutions to a problem. A measure of fitness is assigned to a vector and the probability of a vector's participating in the production of the next generation of vectors is proportional to this fitness measure. Goldberg presents genetic algorithms as they have been developed by John Holland's group at the University of Michigan; this is the first accessible, systematic introduction to their work. Goldberg clearly describes the methods for designing and running genetic algorithms, giving Pascal programs for a number of simple algorithms. He reports on better software environments that have been developed to allow easy creation of genetic algorithms, but the small Pascal programs should allow the interested reader to build and test some simple genetic algorithms. The book's major shortcoming is that it concentrates on the work of one group and ignores the work of others, in particular, the whole line of work using genetic algorithms in numerical function optimization. The difficulty of problem representation is also shortchanged. Why are genetic algorithms important? One major reason is that these algorithms parallelize automatically. For this reason genetic and neural-net algorithms may be the wave of the future. Because of this possibility and because Goldberg clearly presents genetic algorithms, this book is a must for every academic library. -P. Cull, Oregon State University