
24
Chapters
When a neural network has exactly as many output neurons as input
neurons and is trained so that its outputs attempt to match its inputs
for every member of a training set, it is said to be an autoassociative
network. Of what use would such a network be? It seems silly,
because if we already know the inputs, why would we want a network
to reproduce that input? The answer is that these networks tend not
to reproduce just any input pattern applied
to
them. Rather, they tend
to reproduce only those input patterns for which they have been
trained. The effect is that if a trained network is presented with an
input pattern that resemble ...