ABSTRACT
Arti¡cial neural networks are a family of techniques for numerical learning, like the optimization algorithms reviewed in Chapters 6 and 7, but in contrast to the symbolic learning techniques reviewed in Chapter 5. They consist of many nonlinear computational elements that form the network nodes or neurons, linked by weighted interconnections. They are analogous in structure to the neurological system in humans and animals, which is made up of real rather than arti¡cial neural networks. Practical arti¡cial neural networks are much simpler than biological ones, so it is unrealistic to expect them to produce the sophisticated behavior of humans or animals. Nevertheless, they are effective at a range of tasks based on pattern matching. Throughout the rest of this book we will use the expression neural network to mean an arti¡cial neural network. The technique of using neural networks is described as connectionism.
