Adaptive Neural Network Topology in Tree-Structured Self-Organizing Feature Map

Abstract

Self-Organizing Feature Map (SOM) is a layered neural network consisting of an input layer and a competitive layer for the data visualization and vector quantization. The accuracy of SOM vector quantization depends on the number of competitive layer's neurons because the codebook vectors correspond to the competitive layer's neurons. Therefore, when an unknown data set is given, it is difficult to decide the sufficient competitive layer size. In this paper, we propose a Tree-Structured SOM (TS-SOM) based method in order to adaptively change the competitive layer size and structure. TS-SOM is a faster SOM method applying a tree search algorithm. We applied the pruning of neurons and layer creation to the tree structure of TS-SOM by using the error among neighboring neurons.