CDUL - CLASS DIRECTED UNSUPERVISED LEARNING

A novel neural network called Class Directed Unsupervised Learning (CD UL) is introduced. The architecture, based on a Kohonen self-organisin g network, uses additional input nodes to feed class knowledge to the network during Graining, in order to optimise the final positioning of Kohonen nodes in feature space. The structure and training of CDUL ne tworks is detailed, showing that (a) networks cannot suffer from the p roblem of single Kohonen nodes being trained by vectors of more than o ne class, (b) the number of Kohonen nodes necessary to represent the c lasses is found during training, and (c) the number of training setpas ses CDUL requires is low in comparison to similar networks. CDUL is su bsequently applied to the classification of chemical excipients from N ear Infrared (NIR) reflectance spectra, and its performance compared w ith three other unsupervised paradigms. The results thereby obtained d emonstrate a superior performance which remains relatively constant th rough a wide range of network parameters.