Practical Neural Network Recipies in C++

232

Chapter 14

Some problems are so complex, involving so much data of such diverse

variety, and/or in such complex relationships, that a single neural

network may not be practical. If we are using a multilayer feed-

forward network because of its run-time speed, the training time may

be too long. If we are using a probabilistic neural net to avoid long

training, or perhaps because we desire Bayesian confidences, run-time

speed may be unacceptably slow. It may even be that our problem

simply breaks down into several parts in a natural way, and we are

interested in exploiting that physical structure. These are the

principal reasons for considerin ...