At the forefront of Artificial Intelligence

Home » Articles » Neural Networks » Applications/Code

Kohonen Demonstrator

This program is a simple demonstration of a Kohonen self-organizing neural network. The program merely maps itself to a set of coordinates ranging from -0.5 to 0.5 on both the x and y-axis.

The program layout is very simple - the Run button will start the network. Note, this may take some time, so be patient! Note that phase is shown in the title bar. The two edit controls at the right are the Phase 1 (top) and Phase 2 (bottom) iterations. Alter these values to see how it affects the learning of the program.

During Phase 1, the learning coefficient, k, goes from 0.9 to 0.01, linearly decreasing with the iterations. The neighbourhood, N_x is set at half the diameter of the net, and also linearly decreases. During Phase 2, k decreases from 0.1 to 0.0, with N_x fixed at 1.

For a better explanation of Kohonen organizing networks (and a screenshot of this program!) see the Self-organizing Networks essay.

Version 1.2

Flicker-free drawing of the network, a thread bug fixed, and settings saved to the registry. This program seems to be very popular - in fact, it is by far the most popular download from Generation5 at the moment. If there are any additional features people might want to see to Kohonen, please mail me!

Version 1.1

Now uses a thread to run the network, so the GUI is freed up. Using threads also meant I could add a 'Stop' button. I also added a reset button so that you can run the network again and again and again!!

Last Updated: 15/01/2000

Article Toolbar

BibTeX entry