Self-organising maps for pattern recognition

Self-organisation is a fundamental pattern recognition process, in which intrinsic inter-and intra-pattern relationships are learnt without the presence of a potentially biased external influence. In this paper, we present and review the statistical and optimal properties of this important learning model from some pattern recognition aspects through a number of examples and extended applications. The SOM is optimal for vector quantisation. Its topographical ordering provides a mapping with enhanced fault and noise tolerant abilities. It is also applicable to many other applications such as dimensionality reduction and classification. A Bayesian SOM has been devised to extend the mapping as an optimal estimator for probability density functions. This enhanced SOM further explains the functional role of the neighbourhood in the mapping process and reveals that the neighbourhood function form imposes an underlying distribution prototype to the neurons.