Fuzzy associative memories based on subsethood and similarity measures with applications to speaker identification

  • Authors:

  • Estevão Esmi;Peter Sussner;Marcos Eduardo Valle;Fábio Sakuray;Laécio Barros

  • Affiliations:

  • Department of Applied Mathematics, University Campinas, Campinas, São Paulo, Brazil;Department of Applied Mathematics, University Campinas, Campinas, São Paulo, Brazil;Department of Mathematics, University of Londrina, Londrina, Paraná, Brazil;Department of Computer Science, University of Londrina, Londrina, Paraná, Brazil;Department of Applied Mathematics, University Campinas, Campinas, São Paulo, Brazil

  • Venue:
  • HAIS'12 Proceedings of the 7th international conference on Hybrid Artificial Intelligent Systems - Volume Part II
  • Year:
  • 2012

Quantified Score

Hi-index 0.00

Visualization

Abstract

Recently, we presented a non-distributive fuzzy associative memory (FAM) called the Kosko subsethood FAM, for short KS-FAM. This model can be classified as a morphological neural network because it is based on computing the degree of fuzzy inclusion or subsethood of patterns and this operation can be considered an erosion in fuzzy mathematical morphology. In this paper, we introduce a whole range of extensions of the KS-FAM called S-FAMs, dual S-FAMs, and SM-FAMs. Here, the acronyms S-FAM and SM-FAM stand for respectively subsethood FAM and similarity measure FAM. The new models share some properties with the KS-FAM such as unlimited absolute storage capacity and a small number of spurious memories. The paper finishes some experimental results concerning the problem of text-independent speaker identification. For comparative purposes, we included the recognition rates obtained by some well-known classifiers from the literature.