Maximum size for Kohonen Layer is 20x20. (Learning rate decrement size) * (# of epochs) must not reach the initial learning rate, because in each epoch the learning rate is decremented by the learn rate decrement size. Also the radius is decreased in each epoch by a value of 1. If it reaches 0 before the training process stops, it continues by 0 radius. While testing the network, the red square will move to the winner node on the memory map.

1) First select some input output pairs. In fact, output pattern has no importance on the training. It is only for visualisation on the Kohonen layer. You may also create your own patterns by changing the values on the grid (1=white, -1=black).

2) Enter Kohonen network parameters. Suggested paramer values are: 20-10-0.3-0.03-10. Then construct the network and train.

3) Select one of the input pattern you trained and apply it to observe the winner. Then distort the pattern by changing the values -1 and 1. For your change to be effective you should click the "change" button. Apply your distorted pattern and observe how the winner is changing. Repeat testing step selecting other input patterns.

4) Repeat experiment by selecting other pattern pairs and by setting network parameters to some other values, for example to 20-10-0.3-0.03-10 or 10-5-0.3-0.06-5 or 10-5-0.1-0.02-5