Neural net computation is a technique of data handling that is quickly gaining importance. A neural network can be thought of as a mathematical function with generalized estimation and prediction capabilities. It is a computational system made up of a number of simple, yet highly connected, layered processing elements called nodes or neurons. The neurons process information by their dynamic response to external information.

Neural networks, as the name implies, were originally biologically inspired to perform in a manner analogous to the basic functions of neurons. The network is specified by its architecture, transfer functions, and learning laws. Figure A.4.1 illustrates that the input from the net to a given neuron (NET), is the sum of the inputs, o _i, each multiplied with its corresponding weight, w _i. Before the NET is passed on to the subsequent network layers, it is multiplied with a nonlinear transfer function, F, as indicated in the figure. The architecture of a simple network is given in the Fig. A.4.2. If needed, additional layers can be added to the network, complicating the schematic of the structure.

Figure A.4.1

Figure A.4.2

Before the network can be used, it must undergo a learning step. The network learns in a computation-intensive step by establishing the weights that, when applied to the inputs of the nodes, will yield the required output. Note that the derivative of F(NET) with respect to the output, OUT, is rather simple in the chosen example (F' in Fig. A.4.1). Since only the...