The Self-Organizing Map

Abstract

Among the architectures and algorithms suggested for artificial neural networks, the Self-Organizing Map has the special property of effectively creating spatially organized “internal representations” of various features of input signals and their abstractions. One novel result is that the self-organization process can also discover semantic relationships in sentences. In this respect the resulting maps very closely resemble the topographically organized maps found in the cortices of the more developed animal brains. After supervised fine tuning of its weight vectors, the Self-Organizing Map has been particularly successful in various pattern recognition tasks involving very noisy signals. In particular, these maps have been used in practical speech recognition, and work is in progress on their application to robotics, process control, telecommunications, etc. This paper contains a survey of several basic facts and results. © 1990 IEEE