Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!rochester!udel!gatech!news.mathworks.com!tank.news.pipex.net!pipex!oleane!simtel!news.sprintlink.net!redstone.interpath.net!sas!mozart.unx.sas.com!saswss
From: saswss@hotellng.unx.sas.com (Warren Sarle)
Subject: Fuzzy ARTMAP (Was Re: Requested ART2 Reference ...)
Originator: saswss@hotellng.unx.sas.com
Sender: news@unx.sas.com (Noter of Newsworthy Events)
Message-ID: <DFMx3E.632@unx.sas.com>
Date: Thu, 28 Sep 1995 21:30:50 GMT
X-Nntp-Posting-Host: hotellng.unx.sas.com
References:  <442a67$5r6@news.magi.com>
Organization: SAS Institute Inc.
Lines: 38


In article <442a67$5r6@news.magi.com>, youngdp@magi.com (David P Young) writes:
|> ...
|> When you get disappointed with ART2, try something better .. Fuzzy ART Map !!!
|>
|> Gail A. Carpenter, Stephen Grossberg, John H. Reynolds, Natalya Markuzon and
|> David B. Rosen, " Fuzzy ARTMAP: A Neural Network Architecture for Incremental
|> Supervised Learning of Analog Multidimensional Maps", IEEE Transactions on
|> Neural Networks, vol.3, pp 698-713, Sep 1992.
|>
|> (and now for this commercial message...)
|> Additional ART references can be found in the bibliography of my M.Eng thesis
|> entitled "A Fuzzy ART Map Neural Network Speech Recognition System Based On
|> Formant Ratios" available by FTP at the Neuroprose site:
|>
|> FTP// archive.cis.ohio-state.edu /pub/neuroprose/Thesis/ young.thesis.tar.Z

I finally got a copy of the IEEE TONN paper, and I have looked through
Captain Young's thesis, and I am still baffled by match tracking in
Fuzzy ARTMAP. According to the former reference (p. 704):

   At the start of each input presentation the ART_a vigilance
   parameter rho_a equals a baseline vigilance rho-bar_a. ... If
      |x^ab| < rho_ab |y^b|
   then rho_a is increased until it is slightly larger than ...

That's all very well during training, but the expression above depends
on the _target_ value via y^b. So what do you do during the _test_
phase? Using target values during testing would be cheating.

I noticed that Captain Young uses different baseline vigilances for
training and testing. What is the rationale for this?

-- 

Warren S. Sarle       SAS Institute Inc.   The opinions expressed here
saswss@unx.sas.com    SAS Campus Drive     are mine and not necessarily
(919) 677-8000        Cary, NC 27513, USA  those of SAS Institute.
