ART1: Similarity Measures

  • Authors:

  • Ei Ei Khin;Amarottam Shrestha;R. Sadananda

  • Affiliations:

  • Computer Science and Information Management Program, Asian Institute of Technology, GPO Box 2754, Bangkok 10501, Thailand;Dept. of Information Technology, Sirindhorn International Institute of Technology, Thammasat University, P.O. Box 22, Thammasat Rangsit P.O., Pathumthani 12121, Thailand E-mail:amar@siit.tu.ac.th ...;Computer Science and Information Management Program, Asian Institute of Technology, GPO Box 2754, Bangkok 10501, Thailand

  • Venue:
  • Neural Processing Letters
  • Year:
  • 1997

Quantified Score

Hi-index 0.01

Visualization

Abstract

This paper concerns with the ART1 (Adaptive ResonanceTheory 1) in Neural Network. Important features ofART1 are similarity measure (criterion), vigilanceparameter (&rgr;), and their function to classify theinput patterns. Experimental results show that thesimilarity measure as designed originally does notincrease the number of categories with the increasedvalue of &rgr; but decreases, too. This is againstthe claim of ’stability-plasticity‘ dilemma. A numberof researchers have considered this and suggestedalternative similarity measures. Here, we propose anew similarity criterion which eliminates this problemand also possesses the property of lowest listpresentations needed for self stabilization of thenetwork. We compare the results of differentsimilarity criteria experimentally and present them ingraphs. Analysis of the network under noisyenvironment is also carried out.