
Hybrid Networks
239
4) Use the above outputs to train a probabilistic neural
network.
5) To classify an unknown, present it to the trained feed-
forward network. Then present the outputs of that
network to the inputs of the probabilistic neural network.
If the decision reached is different from that indicated by
the maximally activated output neuron of the feedforward
net, suspiciously examine both training sets.
Attention-based Processing
When human beings look at a printed page, they do not spend much
time taking in the whole gestalt. Rather, they flit about, focusing on
different areas and taking in different levels of detail. When they firs ...