Dwarfs, 2003

As a part of the Neural Networks presentation I implemented the example of Kohonen neural network.

Kohonen neural networks

Kohonen's neural networks are widely used as an unsupervised (without teacher) learning method capable of capturing the underlying structure of the training data.

To demonstrate its behavior, I developed the following example:

Example description

There are dwarfs sitting in a row on the bench in the park. These dwarfs would really like to look like people walking by. There are boys (with trousers and shirts) and girls (with skirts). Their trousers may be either longer, or shorter. They wear the light or the dark T-shirt. The problem of these dwarfs is, that they do not clearly recognize, which animal walking around is a human (there are also dogs, squirrels, ...). Only the dwarf, that mostly resembles the walking figure, realizes: "Oh, that's a person" nd adapts itself to look little bit more like the walking person. This is spotted by the two nearest dwarfs and they also adapt themselves.

In the example bellow (Java plugin is required), we can see dwarfs in the upper row. Walking people appear in the bottom row. The person walks to the dwarf, that looks most similar. This dwarf and its neighbors change their look a little bit to be more like the walker. Then the next person comes.