10 Unsupervised Networks

Unsupervised neural networks employ training algorithms that do not make use of desired output data. They are used to find structures in the data, such as clusters of data points, or one- or two-dimensional relationships among the data. When such structures are discovered, they can help describe the data in a more compact ways.

A short tutorial on unsupervised networks is given in Section 2.8, Unsupervised and Vector Quantization (VQ) Networks. Section 10.1, Unsupervised Network Functions and Options, describes the functions and their options to work with unsupervised networks. Examples of the use of the commands are given in Section 10.2, Examples without SOM, and Section 10.3, Examples with SOM. Section 10.4, Change Step Length and Neighbor Influence, describes how you can change the training algorithm by changing the step length and the neighbor feature.