Neural Network Systems

In the last few years, devices inspired by the architecture of the brain have become much more powerful. An upsurge of interest has brought together workers and ideas from many fields including neuroscience, physics, computer science, electrical engineering, psychology, and mathematics. A lot of effort has been concentrated on networks using very rough models of neuron cells (formal neurons). These networks are simple enough to permit mathematical investigation, numerical simulation and hardware implementation. The ability of such systems to learn from examples is a particularly attractive feature. Research is still in its infancy, but it is expected that these models will be useful both as models of real brain function and as computational devices for many applications including optimization, pattern recognition, speech analysis, and signal processing. We shall review the architectures and the associated learning algorithms that have been proposed. We shall explain the notion of generalization from the training examples, and give various examples of problems and applications of practical interest that can be handled by neural networks.