A Kohonen network is a neural network algorithm that creates a self-organising map. A grid of nodes (usually 2D regular mesh) is initialised with a vector a eaqch node. Then one by one the training data is compared with the nodes and allocated to the node that is closest. The node itself is swung to be closer to the new example, as are the nodes immediately surrounding it. Sometimes those a little way off are negatively weighted (pushed away from) the training set item. Iterating this process leads to a set of node values on the grid such that close cells tend to have some sort of semantic unity in terms of how they classify the data.
Used on pages 121, 123, 172
Also known as kohonen nets
Kohonen Self-Organising Network