Visualising the internal components of networks

Authors:
Clinton Woodward;Gerard Murray
Affiliations:
Centre for Intelligent Systems and Complex Processes, School of Biophysical Sciences and Electrical Engineering, Swinburne University of Technology, Australia;Centre for Intelligent Systems and Complex Processes, School of Biophysical Sciences and Electrical Engineering, Swinburne University of Technology, Australia
Venue:
IEA/AIE'2003 Proceedings of the 16th international conference on Developments in applied artificial intelligence
Year:
2003

Citing 2
Cited 0

FERNN: An Algorithm for Fast Extraction of Rules fromNeural Networks

Applied Intelligence
Enhanced Artificial Neurons for Network Applications

Proceedings of the 14th International conference on Industrial and engineering applications of artificial intelligence and expert systems: engineering of intelligent systems

Quantified Score

Hi-index	0.00

Visualization

Abstract

Feed forward artificial neural networks (FFANN), trained by supervision, are useful modelling tools. Data models are developed by statistical regression of some measure of error in network performance over the period of training. The success of training can only be gauged after the testing phase. Valuable information concerning how a network develops a representative model, the act of modelling, is lost as only the final weight state of the network, the model, is considered relevant. A group of visualisation techniques are presented that allows network performance to be studied in real time, from any output perspective of any network component during the training phase. Although confined to data subsets with a two-dimensional input vector, these techniques offer fast functional elucidation in image form, without computational expense or exhaustive mathematical analysis.