Self-Organising MAP or SOM

Self-organizing maps are special type of artificial neural networks. These networks are based upon competitive learning; the output neurons of the network compute in between themselves to be activated or fired, along with the result that only one output neuron, one neuron per group, is once at a time. An output neuron such wins the competition is named as a winner-takes-all neurons or merely a winning neuron. A self organizing map is characterized by the formation of a topographic map of the input patterns whether the spatial locations of the neurons in the lattice are inductive of intrinsic statistical features contained in the input patterns, thus the name "self-organizing map".

The principle goal of the self-organizing map or SOM is to transform an incoming signal pattern of arbitrary dimension into one or two-dimensional discrete map, and to perform such transformation adaptively in a topologically ordered fashion. Following diagram shows the schematic figure of a two-dimensional lattice of neurons commonly utilized as the discrete map. Each neuron in the lattice is completely connected to all the source nodes in the input layer. Such network represents a feed forward structure along with a single computational layer consisting of neurons arranged in columns and row. A one-dimensional lattice is a special case of the configuration depicted in following diagram, in this special case the computational layer consists merely of a single column or row neurons.

                                                            Diagram of: Two-Dimensional Lattice of Neuron

All input patterns presented to the network typically consists of a localized section or "spot" of activity against a quit background. The location and nature of such spot varies from one realization of the input pattern to another. Every neuron in the network should thus be exposed to a sufficient number of different realizations of the input pattern to ensure that the self-organization process has a possibility to mature properly.

The algorithm causes for the formation of self-organizing map proceeds first by initializing the synaptic weights in the network. It can be done by assigning them small values picked from a random number generator, and in doing so or doing, no prior order is imposed on the feature map. One time the network has been properly initialized, there are three necessary process involved in the formation of the self-organizing map, as summarized here are: