Evolutionary Search

Art Sedighi

Adaptive Learning of Polynomial Networks: Genetic Programming, Backpropagation and Bayesian Methods by Nikolay Y. Nikolaev and Hotoshi Iba, Springer, ISBN-10: 0387312390, ISBN-13: 978-0387312392, 316 pp.

One of the main challenges that scientists—computer or otherwise—face today is sorting through and making sense of the vast amount of data available. Whether it’s genomics, experimental physics, astronomy, Web search results, or customer-buying patterns at a local grocery store, the sheer volume of information we can access can be overwhelming. And because we don’t know what information we’ll need in the future, we end up saving everything and anything we get our hands on.

As the amount of data grows, we must find ways to sort through it and extract relevant information. Genetic programming, a technique for using past information and data to predict future patterns, has recently seen tremendous growth in sorting-related applications.

Nikolay Y. Nikolaev and Hotoshi Iba target their book to both academia and practitioners in data search and analysis. They provide theoretical and practical methods for developing algorithms that infer linear and nonlinear models for learning. They also describe numerous neural network methodologies that can help improve statistical data models. Their book covers three main fields of research: evolutionary computation, neural networks, and Bayesian inference. Genetic programming is the text’s basic theme, and researchers in the field will appreciate clearly how the authors introduce operators and methodologies that make searching simple and efficient.

The content includes materials that come directly from courses Nikolaev and Iba have been teaching for several years at the University of London and University of Tokyo, respectively. The text is broken into two subsections, depending on the intended student audience:

• Undergraduate students will learn how to design and implement a genetic programming system’s basic mechanisms, which include the selection scheme, mutation mechanism, and crossover learning operators.
• Graduate students will focus mostly on improving search control of genetic algorithms and analyzing the performance of various operators and algorithms discussed throughout the text.

The authors’ examples come from experiments they conducted over the past few years. They used genetic programming techniques in most of the experiments but improved on these techniques by introducing numerous neural network and Bayesian algorithms into the mix. They demonstrate these experiments and include many other examples throughout the text. Unlike many texts in this topic area, the authors pay special attention to the usability of the models given. They want readers to be able to use this text in the real world and find the models useful there.

I recommend this book for anyone interested in the fast-growing field of genetic programming and neural networks. Nikolaev and Iba have done an amazing job compiling a text that’s both easy to follow and filled with techniques and methods useful to practitioners in the field.

Art Sedighi is the founder and CTO of SoftModule. He’s also an editorial board member of IEEE Software and editor of the Bookshelf department. Contact him at sediga@alum.rpi.edu .