Browse Books

Reviewer: H. Van Dyke Parunak

Many scientific and computational problems can be cast in the language of searching. Theory builders search for patterns that parsimoniously explain large amounts of data. Computer programmers search for sequences of operations that will transform inputs into outputs in accordance with some specification. Businesspeople search for sequences of operations, or mixes of products, or investment portfolios that will optimize profit. Genetic algorithms (GAs) are a novel way to search that combines random probing of the space of possible solutions with mechanisms to favor and recombine promising partial solutions over time. They are one of several computational mechanisms inspired by the idea of survival of the fittest (the most prominent others being evolutionary programming and evolution strategies). Invented by John Holland 30 years ago, and for nearly two decades cultivated mostly among his immediate students, genetic algorithms have recently attracted widespread attention as a way to attack problems that are too poorly defined to support more top-down approaches. Mitchell's book grew out of notes for a short course on GAs at the Santa Fe Institute, where she is a research professor. Unlike many encyclopedic texts with the word “introduction” in the title, this slender volume offers a concise, selective, readable overview of the technology and its uses, targeted at self-study by a newcomer to the field. It is too short to be the primary textbook for a full-semester course, but the questions and exercises at the end of each chapter strongly support a classroom environment, and the volume would be an excellent adjunct to a more traditional AI text in a one- or two-semester course. The exercises include extended projects as well as fodder for weekly homework assignments. Chapter 1, “Genetic Algorithms: An Overview,” introduces the concept of genetic computation, relates its early history, and walks the reader through some simple examples. It deserves wide attention as an excellent standalone survey of the field. Chapter 2, “Genetic Algorithms in Problem Solving,” shows how GAs serve as an engineering tool to solve problems, such as devising computer programs, analyzing and predicting data, and designing neural networks. Chapter 3, “Genetic Algorithms in Scientific Models,” highlights the use of GAs as models from which researchers hope to learn more about biological evolutionary processes, such as the interaction between learning and evolution, sexual selection, ecosystems, and the dynamics of evolutionary activity. The case studies in these chapters not only demonstrate the range of problems for which GAs can be used, but illustrate the wide range of decisions that an implementor must make in a new GA application, thus laying the foundation for chapter<__?__Pub Fmt hardspace>5. Chapter 4, “Theoretical Foundations of Genetic Algorithms,” surveys some of the extensive work that has been done to develop a mathematical understanding of how and why GAs work. Chapter 5, “Implementing a Genetic Algorithm,” outlines the practical questions implementors must ask: What sort of problems are best suited for GAs__?__ How should the problem be encoded into a chromosome__?__ How should the fittest individuals from one generation be selected for the next__?__ What genetic operators should be used__?__ How should one go about setting the many parameters that guide the operation of a GA__?__ In both these chapters, Mitchell conveys a sense of a field with more questions than answers, yet one that has attracted considerable attention from a diverse array of approaches. Chapter 6, “Conclusions and Future Directions,” is the author's assessment of the state of the field and her recommendations for topics most urgently requiring research attention. Mitchell solidly delivers what her title promises. This book is clearly the best single place to start to learn what GAs are, the kinds of problems to which they have been applied, and how the field is likely to develop in the next five years. While its coverage is deliberately selective, it includes a list of 15 general <__?__Pub Caret>references; a compendium of journals, conferences, and network sites; and a bibliography of over 200 works through 1995, as well as a brief index.