
32
Chapter 3
hundred noisy random samples of each of the 4 prototypes were used
to train the network, giving a total of 400 signals in the training set.
It is vital to remember that no pure prototypes were presented to the
network. The network saw only noisy examples, closely resembling the
bottom signal in each of Figures 3.4 through 3.7.
After training was complete, the network was presented with
noisy samples of each of the four prototypes. The test samples did not
duplicate any of the training samples; they were contaminated with
new random noise. This input, along with the filtered output, is
shown in Figures 3.8 through 3.11.
Exposin