SOM demo
Click on the [Run SOM] button several times (about 10 times: click once,
then it will train for 5000 iterations and stop. Click once again, etc.)
to see how SOM map forms
on uniformly distributed 2D input. The applet body (Base.java) was from
the book "Java in a Nutshell" from O'reilley - the Scribble applet,
so you can actually draw something on this canvas :-).
Oh, and currently, the simulation schedule is fixed, but I will make
it dynamically changable. If you do not get good results,
go back and come back again and try it. It is sensitive to the initial
state of the map. Each node in the mesh represent the 2 afferent weight
of each neuron in the map layer. The edge connects nodes of the neurons
neighboring each other on the map layer. The black-red-yellow-white coded
activity display shows the euclidean distance between the neuron's
afferent weights and the current 2D input. The brighter, the closer
distance between the weights and the input.
The source code is in