
344
Chapter 19
Overview
When a neural network is trained, the measure of performance that is
optimized is usually the mean square error of the outputs. There are
many theoretical and practical advantages to using it, and it will be
briefly reviewed here. However, the mean square error has little
intuitive meaning to most people. More pragmatic measures of
performance are needed if the ability of a network actually to perform
its job is to be evaluated. Naturally, the best measure for a particular
network depends on the job duties of that network. Therefore, the
topics of this chapter may each be specific to a particular task and be
independen ...