
Neural Network Adaptation 179
The mechanism that allows such complex mappings to be developed is not fully
understood for each type of multilayer neural network, but in general the network
partitions the input space into regions, and a mapping from the partitioned regions
to the next space is perfor med by the set of weights to the next layer of PEs, eventually
producing an output response. This capability allows some very complex decision
regions to be formed for classification and pattern matching problems, as well as for
applications that require function approximation.
Several issues must be addressed when working with multilayer neural networks. ...