NON-LINEAR DATA MAPPING AND DIMENSIONALITY REDUCTION SYSTEM

摘要

The invention relates to a system for organizing n-dimensional data in a representation of a smaller dimension in a non-linear and non-supervised manner. The types of methods presented herein are commonly known as self-organizing maps and resemble, but are not identical to, the well known Kohonen's self-organizing maps. The basic idea of the invention consists of a combination of data clustering and soft projection thereof in a space of a smaller dimension (usually a bidimensional mesh). The disclosed system comprises two modified versions of the functional of the well known "Fuzzy c-means" clustering algorithm, wherein the cluster centers or dictionary vectors are distributed on a regular, low dimensionality mesh. To this end, a penalty term is added to the functional with the aim of ensuring soft distribution of dictionary vector values on said mesh. In one of said cases, fidelity to data is achieved by minimizing the differences between the data and the dictionary vectors while in the other case the new functional is based upon the estimation of input data probability density.