Universally, the best ways to search, classify, and learn follow a general pattern with two key ingredients: don’t search for the best, but for one best acceptable solution, and perform random actions in the solution space of a given problem where these random actions (operations) obey well-defined constraints.

Genetic algorithms (GAs) also follow these two key rules, and, in addition, they can satisfy multiple objectives concurrently. This fine book discusses all of these aspects of GAs.

A GA is a simple cyclic procedure that acts on a collection of objects called chromosomes, which represent potential solutions to a problem. It selects some of them (the others die), operates on them with the genetic processes of crossover and mutation, and then the cycle repeats. The biological origin of the inspiration behind GAs is more than apparent. Because the selection procedure is based on the fitness of the living chromosomes, a hidden reinforcement mechanism exists. In addition, since an infinity of representation strategies exists per problem (usually the chromosomes appear as long binary strings, where the bit of each string position is linked to the representation/encoding strategy), many fitness functions can be defined in a single or multiobjective sense. Since many termination criteria can be formulated, the ecology of the XGA procedures is plentiful, and most of these procedures are discussed in the book. The book also emphasizes variable-length chromosome GAs (VGAs).

Furthermore, the book contains a wealth of information on applications. It focuses extensively on two application areas: Web mining and bioinformatics. In two long chapters, the authors provide detailed information on how GAs have been used in each application area in the recent past, and where they can further flourish.

The large number of references (to additional material in various journals) make this book ideal for a graduate student.

I remember many years ago reading an article [1] in which the author discussed some limitations of perceptrons--another algorithmic approach to classifying, which has many similarities to and differences from GAs (perceptrons are also discussed in the book). The insight that was being searched for rested in the inherent parallelism, adaptability, and encodability that happen simultaneously in GAs (and other biological systems).

Obviously, the key to making good solvers, not only through GAs but through any algorithmic approach, is the representation coding. That is, one represents the potential solutions and operators that act on them in a way that exposes the truly important components. This displays the insight of the algorithm designers--a skill enhanced through reading this book.