
Autoassociation
31
been described more rigorously starting on page 24. As can be seen in
Figure 3.3, results were quite good for this simple filter.
Of course, no competent scientist would use a neural filter for
a simple problem like this. Digital filters having all sorts of nice
optimality properties exist for this task, and any of them would most
likely be superior to an autoassociative neural filter. On the other
hand, what if the pattern sought is not a pure wave, but is very
complex? And what if there are several possible patterns, which may
have very different spectral characteristics? When one considers that
all one needs to do is trai ...