Visualisation of processes in ART neural networks


Author: Martin Elliffe
Source: GZipped PostScript (106kb); Adobe PDF (379kb)

Artificial Neural Networks (ANNs) are of current research interest in many institutions worldwide. Adaptive Resonance Theory (ART) forms the basis of a family of like-titled ANNs, of particular current research interest at the Computer Science Department of Victoria University of Wellington.

A problem common to many ANN paradigms is the difficulty users experience in understanding the processes by which the networks operate. Scientific visualisation techniques have been applied to produce visualisation tools which alleviate this problem.

This report discusses the inapplicability of current ANN visualisation techniques for use with ART. Seven new applicable techniques are presented.

The design of the irresistibly-titled SOTA (State Of The ART) as a single tool combining the new techniques is also presented, providing visualisation facilities to assist users in their understanding of the internal processes of ART ANNs.

[Up to Computer Science Technical Report Archive: Home Page]