A match-based clustering neural network for stable category learning of multiple-valued patterns

Cheng An Hung, Sheng-Fuu Lin*

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

Abstract

A match-based clustering neural network (MBCNN) for unsupervised pattern recognition is introduced. MBCNN is an extension of the adaptive Hamming net (AHN). which incorporates multiple-valued logic operations. AHN can only process binary-valued patterns, whereas MBCNN can process either multiple-valued or binary-valued patterns. In this paper, we connect AHN with MBCNN via the thermometer code. In addition, we also present a number of properties related to the learning process in MBCNN. In particular, an upper bound is derived for the number of list presentations required by an MBCNN to learn an arbitrary list of P-valued input patterns that is repeatedly presented to the architecture. Finally, a comparison between MBCNN and fuzzy ART is given.

Original languageEnglish
Pages (from-to)543-562
Number of pages20
JournalJournal of Information Science and Engineering
Volume13
Issue number4
StatePublished - 1 Dec 1997

Keywords

  • Adaptive Hamming net
  • Fuzzy ART
  • Match-based clustering neural network
  • Multiple-valued logic
  • Thermometer code

Fingerprint Dive into the research topics of 'A match-based clustering neural network for stable category learning of multiple-valued patterns'. Together they form a unique fingerprint.

Cite this