Dual Vigilance Fuzzy Adaptive Resonance Theory

Abstract

Clusters retrieved by generic Adaptive Resonance Theory (ART) networks are limited to their internal categorical representation. This study extends the capabilities of ART by incorporating multiple vigilance thresholds in a single network: stricter (data compression) and looser (cluster similarity) vigilance values are used to obtain a many-to-one mapping of categories-to-clusters. It demonstrates this idea in the context of Fuzzy ART, presented as Dual Vigilance Fuzzy ART (DVFA), to improve the ability to capture clusters with arbitrary geometry. DVFA outperformed Fuzzy ART for the datasets in our experiments while yielding a statistically-comparable performance to another more complex, multi-prototype Fuzzy ART-based architecture.

Department(s)

Electrical and Computer Engineering

Research Center/Lab(s)

Center for High Performance Computing Research

Comments

Research was sponsored in part by the Missouri University of Science and Technology Mary K. Finley Endowment and Intelligent Systems Center, USA ; the Coordenacao de Aperfeiçoamento de Pessoal de Nível Superior - Brazil (CAPES) - Finance code BEX 13494/13-9 ; and the Army Research Laboratory (ARL), USA , and was accomplished under Cooperative Agreement Number W911NF-18-2-0260.

Keywords and Phrases

Adaptive resonance theory; ART; Clustering; Topology; Unsupervised; Visual assessment of cluster tendency

International Standard Serial Number (ISSN)

0893-6080

Document Type

Article - Journal

Document Version

Citation

File Type

text

Language(s)

English

Rights

© 2019 Elsevier Ltd, All rights reserved.

Publication Date

01 Jan 2019

Share

 
COinS