This practical introduction describes the kinds of real-world problems neural network technology can solve. Surveying a range of neural network applications, the book demonstrates the construction and operation of artificial neural systems. Through numerous examples, the author explains the process of building neural-network applications that utilize recent connectionist developments, and conveys an understanding both of the potential, and the limitations of different network models. Examples are described in enough detail for you to assimilate the information and then use the accumulated experience of others to create your own applications. These examples are deliberately restricted to those that can be easily understood, and recreated, by any reader, even the novice practitioner. In some cases the author describes alternative approaches to the same application, to allow you to compare and contrast their advantages and disadvantages.
Organized by application areas, rather than by specific network architectures or learning algorithms, Builiding Neural Networks shows why certain networks are more suitable than others for solving specific kinds of problems. Skapura also reviews principles of neural information processing and furnishes an operations summary of the most popular neural-network processing models. Finally, the book provides information on the practical aspects of application design, and contains six topic-oriented chapters on specific applications of neural-network systems. These applications include networks that perform:
The book includes application-oriented excercises that further help you see how a neural network solves a problem, and that reinforce your understanding of modeling techniques.
Help With File Types
Simulators for Building Neural Networks are available:simulat.exe 48,652 bytes (binary) simulators.sit.hqx 46,331 bytes (ascii) simulators.tar.Z 38,145 bytes (binary)
These archives contain the complete set of files for the neural network simulators associated with the book. The identical material is archived in three different forms for the convenience of DOS/Windows (the binary, self-extracting .exe file), Mac (the ascii stuff, binhexed .sit.hqx file), and Unix (the binary, tarred, compressed .tar.Z file)
Files with the extension ".exe" are self-extracting. You need no extra software to unarchive these files on your DOS or Windows machine. Once transferred, use the syntax:filename -d
(the "-d" will preserve directory structure where directories have been created), where "filename" is the name of the file, minus the ".exe" extension.
Files with the extension ".sit.hqx" are archived using the Macintosh-native Stuffit program, and then bin-hexed, making them ASCII files (transferring them as binary won't hurt them, but is redundant). The easiest way to "unbinhex" and "unstuff" them in one go is to use the free program, StuffIt Expander from Aladdin Systems.
Files with the extension ".tar.Z" are archived using the Unix "tar" command, and then compressed. To make them available, first be sure that you are in binary mode before transferring the files (type "binary" at the "ftp>" prompt). Once transferred, use the syntax:uncompress filename.tar.Z tar xvf filename.tar
Single Neuron Computations.
The Backpropagation Network.
The Counterpropagation Network.
Adaptive Resonance Theory.
The Multidirectional Associative Memory.
The Hopfield Memory.
Developing a Data Representation.
Pattern Representation Methods.
Training and Performance Evaluation.
A Practical Example.
Predicting Commodity Futures.
Gender Recognition from Facial Images.
Imagery Feature Discovery.
Aircraft Tracking in Video Imagery.
Robotic Manipulator Control.
Implementation of a Fuzzy Network.
Fuzzy Neural Inference.
Fuzzy Control of BPN Learning.
Fuzzy Neural-System Summary.