Self Organising Maps

• The basic idea of a SOM is that you have a bunch of units that represent the (typically Euclidean) space.
• Initially, the units are put down randomly.
• In response to training data they are moved so that they represent the training data.
• It's an unsupervised system.
• They're used for clustering, and thus commonly used for analysing data.
• If your training data is, say, the shape of an 8 in two dimensional space, the nodes will move into an 8.