[.uk] Foundations of Genetic Programming (ISBN 3540424512)

Exciting New Developments in EC Theory:
Langdon and Poli are both internationally recognized experts in Evolutionary Computation (EC) and, in particular, Genetic Programming. They have both contributed extensively to the theoretical "foundations" of GP and hence may speak with no small degree of authority about GP theory. As a physicist working in EC I like the balance that the authors have struck between mathematical rigor and understandable intuition. The book is not as rigorous as Vose's well known GA book. However, it is much easier to read. Neither does it take the "engineering" rule of thumb approach, as does Goldberg's book for instance. It covers very well recent important developments in the theory of GP and in that sense makes very good reading for anyone with a serious interest in EC theory. It is not for the novice, even though technically it is not a difficult book. It is really a research monograph and not a textbook. In that sense the title is a little bit misplaced. With the exciting direction the authors are pointing in I believe that in five years time another book of the same title should truly be able to lay out what are the foundations of GP theory and also show the theoretical unity that exists between the different branches of EC.

A survey of what was new in 2002:
This book was published in 2002 to provide a survey of the direction research had taken in the field of Genetic Programming. There is an explanation of what genetic programming is and how it is different from genetic algorithms in chapter 1(GP is a "generalization" of GA). Chapter 2 discusses the problems with the fitness landscape. Chapter 3 - 6 discusses various schema theory approaches and proofs. Chapter 6 has a great explanation of effective fitness. There are numerous theorems and proofs in the book. There are informative examples of the max problem and the artificial ant (Santa Fe Trail) problems. Chapter 11 is about how GP convergences are a tricky matter and how subtrees can hide interesting incidences of convergence. This is not an introductory text, it is intended for graduate level or higher readers. There is much theoretical work here and a limited background in this area will result in limited understanding of the material.

Good introduction to GP theory:
Langdon and Poli do a fantastic job of summarizing the major theoretical results of genetic programming. The first chapter gives a quick and clear introduction to genetic programming. They continue with a comprehensive summary of previous research in schema theory, and then they present their exciting theoretical results. Their description of an exact schema theorem (microscopic and macroscopic) for GP is a bit dense, but they provide a good discussion of how to interpret these results. As a whole, this book is generally easy to follow, even with little prior exposure to genetic programming. Of course, this book is not intended to be a general introduction to genetic programming (one of John Koza's books would be more appropriate), but instead it is intended to present some of the theoretical foundations of the field.

The modern revolution:
Currently working as an undergraduate student in Ann Arbor, Michigan as a Computer Science major I'm an intrigued by Genetic Programming alongside all motives of this in-depth field. I found this book to be a modest account of what is new and theoretical within this field. Expressing advanced features with a short introduction; this book is profoundly for somebody with somewhat of a background. A recommended start in the computer evolutionary field is: An Introduction to Genetic Algorithms (1996), by Melanie Mitchell.

specialised maths treatment of GP:
This book can be usefully read along with a companion text by the same publisher - "Introduction to Evolutionary Computing". Langdon and Poli provide a focused look, on the specifics of genetic programming. The maths treatment here is significantly more involved than the other book. Foundations starts with what I suppose in this field is an obligatory section on the concept of a fitness landscape. A very useful metaphor of what you'll be attempting to do, as a researcher. However, the authors carefully point out the limitations of this idea. Notably that some spaces might have no natural metric. The book then rapidly goes into the ideas of GP schemas and hyperschemas. Accompanied by a nice theoretical analysis of key performance goals like the rate of convergence in the GP search space. A solid offering to the GP researcher.