Why do we feel a need to write a book about pattern recognition when many excellent books are already available on this classical topic? The answer lies in the depth of our coverage of neural networks as natural pattern classifiers and clusterers. Artificial neural network computing has emerged as an extremely active research area with a central focus on manipulation of pattern-formatted information, information containing an underlying pattern. This has given rise to a new coherent approach to pattern recognition which builds upon both the contributions of the past and the rapid progress in neural network research. Pattern recognition has grown to encompass a wider scope of methodology than is available in the traditional domain of statistical pattern recognition. The addition of artificial neural network computing to traditional pattern recognition has given rise to a new, different, and more powerful methodology which we intended to present in this book for the practitioner.

Pattern recognition systems are systems which automatically classify or cluster complex patterns or objects based on their measured properties or on features derived from these properties. With this viewpoint a neural network can be seen as a system that recognizes patterns. The discovery of underlying regularities is, however, precisely the task at which neural networks excel. In this very real sense the study of neural networks and the study of pattern recognition converge. Neither subject is truly complete in the absence of the other. We suggest that many of the effective applications of neural networks in domains not generally thought of explicitly as pattern recognition (e.g., control) can be viewed as pattern recognition in the sense that they still depend on the network ability to detect and identify subtle underlying regularities in the input space.

The extent to which neural networks are or should be reflective of biological systems has been a contentious subject. Primarily, we take an engineering approach, foregoing any extensive treatment of biological plausibility. Nevertheless, we recognize that when designing a system it can be useful to observe other systems which perform the desired function well. Biological systems are superb pattern recognizers. By the way of analogy, in designing the first airplane one might do well to observe birds in an attempt to isolate characteristics enabling flight (e.g., wing shape, mass/volume ratio, etc.). However, at some point in the process, one must break free of slavishly following the biological metaphor. Otherwise, there could be no aircraft capable of supersonic flight. We therefore believe that while we may look to biological systems for inspiration, artificial neural systems must ultimately take on their own identity to be truly effective.

Our book is an attempt to cover pattern classification and neural network approaches within the same framework geared toward the practitioner. Neural networks should not be considered a black box, governed by complicated mathematics, with answers that may surprise or disappoint us. Armed with an understanding of underlying theory and practical examples the practitioner will be better able to make judicious design choices which will render neural application predictable and effective.