Abstract

A review of recent development of the self-organising map (SOM) for applications related to data mapping and visualisation is presented. Neural networks are biological inspired learning and mapping methods, which can learn complex nonlinear relationships between variables from supplied data samples. The algorithm is being exploited as a useful and increasingly promising mapping method for structure visualisation, data mining, knowledge discovery and retrieval. The SOM is a simple mathematical model of the mappings which exist in parts of the mammalian cortex (especially visual and auditory cortex). The inputs to a SOM are often drawn from high dimensional space and the algorithm has been used, in an innovative approach, as visualisation tool for dimensionality reduction projections. One of the most important properties of the SOM is its topological preservation, i.e. neighbouring points in the input space will be mapped to nearby neuron nodes in the map space. Such a property can be employed for the visualisation of mutual semantic relationships between inputs. The SOM is also a reducing process and can approximate a vast quantity of data in a much diminished set of representatives