Self-organizing map

A self-organizing map (SOM), also known as a Kohonen map, is an artificial neural network that learns a low-dimensional representation of high-dimensional input data through competitive, unsupervised learning. Invented by Teuvo Kohonen in 1982, the SOM arranges neurons in a fixed grid topology — typically two-dimensional — and adjusts their weights so that nearby neurons respond to similar inputs.

The learning process is iterative: for each input vector, the network finds the best matching unit (the neuron whose weights are closest to the input) and updates both that neuron and its neighbors. Over time, the map converges to a topology-preserving representation of the input space. SOMs are used for data visualization, clustering, and dimensionality reduction.