Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!nntp.club.cc.cmu.edu!goldenapple.srv.cs.cmu.edu!das-news2.harvard.edu!fas-news.harvard.edu!newspump.wustl.edu!news.ecn.bgu.edu!news-spur1.maxwell.syr.edu!news.maxwell.syr.edu!news.mathworks.com!newsgate.duke.edu!interpath!news.interpath.net!news.interpath.net!sas!newshost.unx.sas.com!saswss
From: saswss@hotellng.unx.sas.com (Warren Sarle)
Subject: Re: Training a Probabilistic Neural Network
Originator: saswss@hotellng.unx.sas.com
Sender: news@unx.sas.com (Noter of Newsworthy Events)
Message-ID: <E796Ap.LJA@unx.sas.com>
Date: Tue, 18 Mar 1997 19:04:01 GMT
X-Nntp-Posting-Host: hotellng.unx.sas.com
References: <19970304160700.LAA18911@ladder01.news.aol.com> <Pine.SOL.3.91.970304213021.2947D-100000@miles> <Pine.SOL.3.91.970306033045.3342F-100000@miles> <tanaka-1703970910450001@itserv.it.okayama-u.ac.jp>
Organization: SAS Institute Inc.
Lines: 29


In article <tanaka-1703970910450001@itserv.it.okayama-u.ac.jp>, tanaka@mathpro.it.okayama-u.ac.jp (Masahiro Tanaka) writes:
|> In article <Pine.SOL.3.91.970306033045.3342F-100000@miles>, Greg Heath
|> <heath@ll.mit.edu> wrote:
|> 
|> > In general, however, the Gaussian widths must be determined by trial
|> and error. 
|> 
|> Since I haven't understood the question (it's lost in my server), I may
|> be misunderstanding, but the above statement is not usually true. It can
|> be easily estimated by maximum likelihood method. The following article
|> may be useful.
|> 
|> R.L. Streit and T.E.Luginbuhl,"Maximum likelihood training of
|> probabilistic neural networks", IEEE Trans. Neurel Networks, 5, 5,
|> 764-783, 1994.

No. A PNN is usually defined as kernel discriminant analysis, and kernel
density estimates cannot be obtained by maximum likelihood.  The article
cited above uses normal mixtures, not kernels estimates, and therefore
the title is very misleading.

-- 

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.
 *** Do not send me unsolicited commercial or political email! ***

