Kohonen SOM Demo

This NeuronDotNet application graphically demonstrates how a Kohonen SOM organizes itself during learning.

• Kohonen SOM Demo
• Kohonen SOM Demo Source

This sample applications demonstrates the behaviour of a Kohonen SOM. It provides a GUI to visualize all features of a SOM provided by NeuronDotNet. Using this application, one can visualize lattice topologies (rectangular and hexagonal) and layer shapes (linear, planar, circular, cylindrical and toroidal surfaces). It also demonstrates how a neighborhood functions influences the behaviour of a SOM.

The training data consists of a set of points in two-dimensional space. The SOM on training, organizes itself to match the training set (this can be used in clustering applications). It can be observed that,

• using a linear layer (a layer with unit height) finds an approximate diameter of the complete graph containing the input points
• using a circular layer (a layer with unit height and circular rows) provides an approximate solution to the Traveling Salesman Problem for the input points

One can extend this application to have three dimensional training data, and test how a SOM can be used in dimensionality reduction. (Arbitrary color can be seen as a three-dimensional point along RGB axes).

Zedgraph, an open source library is used to plot the graphs.

A screenshot of the application is shown below.

More Samples