A supervised fuzzy adaptive resonance theory with distributed weight update

Authors:
Aisha Yousuf;Yi Lu Murphey
Affiliations:
Department of Electrical and Computer Engineering, University of Michigan – Dearborn, Dearborn, MI;Department of Electrical and Computer Engineering, University of Michigan – Dearborn, Dearborn, MI
Venue:
ISNN'10 Proceedings of the 7th international conference on Advances in Neural Networks - Volume Part I
Year:
2010

Citing 4
Cited 0

Fuzzy ART: Fast stable learning and categorization of analog patterns by an adaptive resonance system

Neural Networks
Fuzzy ART properties

Neural Networks
Optimal Algorithmic Complexity of Fuzzy ART

Neural Processing Letters
ART properties of interest in engineering applications

IJCNN'09 Proceedings of the 2009 international joint conference on Neural Networks

Quantified Score

Hi-index	0.00

Visualization

Abstract

The Fuzzy Adaptive Resonance Theory is an unsupervised clustering algorithm that solves stability plasticity dilemma The existing winner-take-all approach to updating weights in Fuzzy ART has two flaws: (i) it only updates one cluster while an input might belong to more than one cluster and (ii) the winner-take-all approach is costly in training time since it compares one weight to the input at a time We propose an algorithm that compares all weights to the input simultaneously and allows updating multiple matching clusters that pass the vigilance test To mitigate the effects of possibly updating clusters belonging to the wrong class we introduced weight scaling depending on the “closeness” of the weight to the input In addition, we introduced supervision to penalize the weight update for weights that have the wrong class The results show that our algorithm outperformed original Fuzzy ART in both classification accuracy and time consumption.