Practical Neural Network Recipies in C++

Classification

Networks

21

elsewhere. At this point, we are simply dealing with a black box

having three photographic inputs and sufficient outputs to determine

a blood type. In problems of this type, it is better to use the more

compact binary encoding for output classes.

There is no reason to limit compact encoding to binary

subclasses, although the situation can rapidly become quite complex.

Serious consideration should be given to separate subclasses whenever

the classes are separable. For example, consider a network that must

listen to a spoken sound and provide some classification information.

Its task might be to decide whether the startin ...