New geometric concepts in fuzzy-ART and fuzzy-ARTMAP: category regions

Author

Anagnostopoulos, Georgios C. ; Georgiopoulos, Michael

Author_Institution

Sch. of Electr. Eng. & Comput. Sci., Univ. of Central Florida, Orlando, FL, USA

Volume

1

fYear

2001

fDate

2001

Firstpage

32

Abstract

We introduce new geometric concepts regarding categories in fuzzy ART (FA) and fuzzy ARTMAP (FAM), which add a geometric facet to the process of node selection in the F₂ layer by patterns. Apart from providing the means to better understand the training and performance phase of these two architectures, the new concepts, namely the category regions, lead us to interesting theoretical results, when training either architecture. First, we define the commitment test as a novelty detection mechanism similar to the vigilance test. Next, we define various category regions. Through those definitions and 3 derived lemmas we identify areas in the vigilance-choice parameter space, for which 4 results are stated that are applicable to both FA and the FAM classifier training phase

Keywords

ART neural nets; fuzzy neural nets; learning (artificial intelligence); category regions; fuzzy-ART; fuzzy-ARTMAP; geometric concepts; learning phase; parameter space; performance phase; vigilance test; Computer architecture; Computer science; Computer vision; Neural networks; Organizing; Resonance; Subspace constraints; Supervised learning; Testing; Unsupervised learning;

fLanguage

English

Publisher

ieee

Conference_Titel

Neural Networks, 2001. Proceedings. IJCNN '01. International Joint Conference on

Conference_Location

Washington, DC

ISSN

1098-7576

Print_ISBN

0-7803-7044-9

Type

conf

DOI

10.1109/IJCNN.2001.938987

Filename

938987