Practical Neural Network Recipies in C++

220

Chapter 12

Autoassociative Versions

As should be obvious by now, the probabilistic neural network is

fundamentally a classifier. However, there are at least two simple but

often effective ways in which it can be used in an autoassociative

mode. Which of these methods is chosen depends on the nature of the

training data.

Of course, the class assigned to the input is no longer consid-

ered the output of the network. Now, the output is a vector of the

same length as the input. We compute that output vector after the

classification has been done.

If the training samples for each class are noisy versions of a

single prototype for that class,