Neuro-Computers: Optimization Based Learning

The brain-like architecture of artificial neural networks makes them ideal for tackling problems that are too difficult for conventional architectures, specifically problems that involve pattern recognition or other perceptual tasks. Neuro-Computers: Optimization Based Learning provides an intermediate-level exposition of the exciting world of neuro-computers. It presents the importance of neuro-computing to artificial intelligence, giving historical background and present-day implementation options. The book demonstrates the superiority of the adaptive search strategy over conventional fixed parameter searches performed by backpropagation algorithms. It then explores global optimization strategy and presents genetic algorithms as viable methods to train neuro computers on non-trivial problems. This self-contained volume is delivered in a format that is suitable for graduate students, as well as researchers who want to begin work in neuro-computing or related artificial intelligence applications.

Table of Contents:
Foreword Preface Notations and Symbols Introduction Neurocomputing Organization of the Brain The Neuron Model Importance of the Connectionist Approach Motivation Historical Background The Adaptive Linear Combiner Artificial Neural Network Models Learning Algorithms for Feedforward Networks Implementation Applications Composite Optimization Strategy Pattern Processing for Intelligent Behavior The Power of Hidden Units Supervised Learning The Optimization Approach Steepest Descent Optimal Step Length Adaptation Heuristics The Search for Global Solution Multimodal Performance Surfaces Simulated Annealing The Method of Covering Implementation of the Method of Covering Generalization and Fault Tolerance ANN Performance Issues Better Performance Metrices Constrained Optimization for Better Generalization Fault Tolerance Constraints Development of Fault Tolerant ANNs Genetic Training in Neuro-Computers Global-Minimization for Feedforward Neural Networks Genetic Algorithm: A Global Optimizer Genetic Learning in Neural Networks Neuro-Computer Training Based on Advanced Genetic Operators Two-Parents Multipoint Restricted Crossover (Double-MRX) Three-Parents Multipoint Restricted Crossover (Triple-MRX) Elitist Selection Mutation Scheduling Hybrid Learning Simulation and Case Studies The Simulation System Learning Binary Mapping Fault Tolerance and Generalization Performance of Genetically Trained Neural Network Performance of Hybrid Learning Epilogue Bibliography Index

Review :
"The book contains seven chapters, which include many results of simulation that demonstrate the usefulness of the approach. It also contains an extensive bibliography that will be of great value to students as well as researchers. The style of presentation is lucid as well as thorough. I commend Dr. Shukla for his excellent work." N.K. Sinha Fellow IEEE, Professor Emeritus, Department of Electrical and Computer Engineering, McMaster University, Hamilton, Ontario, Canada