• The goal of SOM is to find prototype vectors that represent the input vectors and realize a continuous mapping from the high-dimensional input space to a grid of nodes.
  • The cluster is represented by a node with an associated weight vector.
  • A similarity measure is computed to determine the matching nodes of input vectors. Along with those of neighboring nodes, the weight-vectors of matching nodes are updated by the input vectors which are different from K-means.
  • By iterative training, the model vectors learn to represent the probability distribution of the data in an orderly fashion.
  • When the training is complete, clusters identified by mapping all data points to the output neurons and similar clusters are placed near each other.
  • Please be attention that a random initialization of the weight vectors may result in instability of the clustering results.
  • SOM Toolbox
    - Esa Alhoniemi, Johan Himberg, Juha Parhankangas and Juha Vesanto, 2000-2005