Reference Work Entry

Encyclopedia of Machine Learning

pp 886-888

Self-Organizing Maps

  • Samuel Kaski


Kohonen maps; Self-organizing feature maps; SOM


Self-organizing map (SOM), or Kohonen Map, is a computational data analysis method which produces nonlinear mappings of data to lower dimensions. Alternatively, the SOM can be viewed as a clustering algorithm which produces a set of clusters organized on a regular grid. The roots of SOM are in neural computation (see neural networks); it has been used as an abstract model for the formation of ordered maps of brain functions, such as sensory feature maps. Several variants have been proposed, ranging from dynamic models to Bayesian variants. The SOM has been used widely as an engineering tool for data analysis, process monitoring, and information visualization, in numerous application areas.
Self-Organizing Maps. Figure 1

A schematic diagram showing how the SOM grid of units (circles on the left, neighbors connected with lines) corresponds to an “elastic net” in the data space. The mapping from the grid loca ...

This is an excerpt from the content