A novel normalization technique for unsupervised learning in ANN

Unsupervised learning is used to categorize multidimensional data into a nu mber of meaningful classes on the basis of the similarity or correlation be tween individual samples, In neural-network implementation of various unsup ervised algorithms such as principal component analysis (PCA), competitive learning or self-organizing map (SOM), sample vectors are normalized to equ al lengths so that similarity could be easily and efficiently obtained by t heir dot products. In general, sample vectors span the whole multidimension al feature space and existing normalization methods distort the intrinsic p atterns present in the sample set. In this work, a novel method of normaliz ation by mapping the samples to a new space of one more dimension has been proposed. The original distribution of the samples in the feature space is shown to be almost preserved in the transformed space. Simple rules are giv en to map from original space to the normalized space and vice versa.