Neural networks : an introduction

The concepts of neural-network models and techniques of parallel distributed processing are comprehensively presented in a three-step approach: - After a brief overview of the neural structure of the brain and the history of neural-network modeling, the reader is introduced to "neural" information processing, i.e. associative memory, perceptrons, feature-sensitive networks, learning strategies, and practical applications. - Part 2 covers more advanced subjects such as spin glasses, the mean-field theory of the Hopfield model, and the space of interactions in neural networks. - The self-contained final part discusses seven programs that provide practical demonstrations of neural-network models and their learning strategies. Ample opportunity is given to improve and modify the source codes. The software is included on a 5 1/4 inch MS DOS diskette and can be run using Borland's TURBO C 2.0 compiler, the Microsoft C compiler (5.0), or compatible compilers. The mysteries of the human mind have fascinated scientists and philosophers alike for centuries. Descartes identified our ability to think as the foundation stone of ontological philosophy. Others have taken the human mind as evidence of the existence of supernatural powers, or even of God. Serious scientific in­ vestigation, which began about half a century ago, has partially answered some of the simpler questions (such as how the brain processes visual information), but has barely touched upon the deeper ones concerned with the nature of consciousness and the possible existence of mental features transcending the biological substance of the brain, often encapsulated in the concept "soul". Besides the physiological and philosophical approaches to these questions, so impressively presented and contrasted in the recent book by Popper and Ec­ cles [P077), studies of formal networks composed of binary-valued information­ processing units, highly abstracted versions of biological neurons, either by mathematical analysis or by computer simulation, have emerged as a third route towards a better understanding of the brain, and possibly of the human mind. Long remaining - with the exception of a brief period in the early 1960s - a rather obscure research interest of a small group of dedicated scientists scattered around the world, neural-network research has recently sprung into the limelight as a "fashionable" research field Front Matter....Pages I-XIII Front Matter....Pages 1-1 The Structure of the Central Nervous System....Pages 2-11 Neural Networks Introduced....Pages 12-22 Associative Memory....Pages 23-36 Stochastic Neurons....Pages 37-44 Cybernetic Networks....Pages 45-50 Multilayered Perceptrons....Pages 51-61 Applications....Pages 62-72 Network Architecture and Generalization....Pages 73-86 Associative Memory: Advanced Learning Strategies....Pages 87-103 Combinatorial Optimization....Pages 104-112 VLSI and Neural Networks....Pages 113-118 Symmetrical Networks with Hidden Neurons....Pages 119-125 Coupled Neural Networks....Pages 126-131 Unsupervised Learning....Pages 132-144 Front Matter....Pages 145-145 Statistical Physics and Spin Glasses....Pages 146-155 The Hopfield Network for p / N → o....Pages 156-164 The Hopfield Network for Finite p / N ....Pages 165-186 The Space of Interactions in Neural Networks....Pages 187-202 Front Matter....Pages 203-203 Numerical Demonstrations....Pages 204-207 ASSO : Associative Memory....Pages 208-217 Front Matter....Pages 203-203 ASSCOUNT : Associative Memory for Time Sequences....Pages 218-221 PERBOOL : Learning Boolean Functions with Back-Propagation....Pages 222-228 PERFUNC : Learning Continuous Functions with Back-Propagation....Pages 229-232 Solution of the Traveling-Salesman Problem....Pages 233-244 KOHOMAP : The Kohonen Self-organizing Map....Pages 245-249 Back Matter....Pages 250-266