The number of classes to be distinguished is not known, the method must adapt the weights without external information. Only the most active neurons for a given input are modified.

This process really assumes the usual learning approach.

Each group of neurons takes an average configuration for the class to recognize, and it is a statistical approach. The complexity of the network is related to the number of classes to distinguish.

To avoid confusions, this process must be able to distinguish at least the number of classes for which it is performed.