Theory and Methods of Neural Network Structure Design

Author: Wei Haikun
Publisher:
Publish Date: 2005-02-01
Features: This book is written by the author based on years of teaching and research in neural network theory and its applications at Southeast University. It serves as a graduate-level textbook for neural network design. Building upon the introduction of common neuron models and weight learning algorithms, the book systematically and comprehensively covers neural networks, with a particular emphasis on the principles and methods of designing feedforward neural network structures. The specific content includes: various factors influencing the generalization ability of neural networks, such as structural complexity, samples, initial weights, learning time, and prior knowledge; how these factors affect neural network performance; common neural network structure optimization design methods, including pruning algorithms (weight decay method, sensitivity calculation method, correlation pruning method, etc.), construction algorithms (cascade correlation algorithm, resource allocation network, etc.), and evolutionary methods; common neural network parameter optimization design methods, including optimal stopping methods, active learning methods, adding noise to sample inputs, and neural network ensembles. Most of these methods are widely used typical feedforward neural network design approaches in China. The book introduces their implementation principles and simulation examples, and provides MATLAB implementation code for most of the algorithms. This book is suitable for engineering technicians, senior students, graduate students, and teachers in disciplines such as automation and signal processing.