Efficient unsupervised training algorithm for self-organizing networks

Abstract

We develop a very efficient training algorithm for self-organizing networks (SON) which have run times that are linear with both the number of desired output vectors and training set vectors, respectively. The algorithm is very efficient compared with older ones which have run-times that are quadratic (~N2) with the number of desired output vectors. Run-times for both cases were predicted from models derived from the algorithms and were also confirmed by actual programs.