Pattern Recognition Using Neural Networks covers traditional linear pattern recognition and its nonlinear extension via neural networks. The approach is algorithmic for easy implementation on a computer, which makes this a refreshing what-why-and-how text that contrasts with the theoretical approach and pie-in-the-sky hyperbole of many books on neural networks. It covers the standard decision-theoretic pattern recognition of clustering via minimum distance, graphical and structural methods, and Bayesian discrimination.
Pattern recognizers evolve across the sections into perceptrons, a layer of perceptrons, multiple-layered perceptrons, functional link nets, and radial basis function networks. Other networks covered in the process are learning vector quantization networks, self-organizing maps, and recursive neural networks. Backpropagation is derived in complete detail for one and two hidden layers for both unipolar and bipolar sigmoid activation functions. The more efficient fullpropagation, quickpropagation, cascade correlation, and various methods such as strategic search, conjugate gradients, and genetic algorithms are described. Advanced methods are also described, including the full training algorithms for radial basis function networks and random vector functional link nets, as well as competitive learning networks and fuzzy clustering algorithms.
Special topics covered include:
feature engineering
data engineering
neural engineering of network architectures
validation and verification of the trained networks
This textbook is ideally suited for a senior undergraduate or graduate course in pattern recognition or neural networks for students in computer science, electrical engineering, and computer engineering. It is also a useful reference and resource for researchers and professionals.