Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!math.ohio-state.edu!uwm.edu!news.alpha.net!news.mathworks.com!uunet!munnari.oz.au!newshost.anu.edu.au!csc.canberra.edu.au!isepc3.canberra.edu.au!DavidC
From: DavidC@ise.canberra.edu.au (David Clark)
Subject: Re: [Q] Pattern Recognition
Message-ID: <DavidC.34.2F419178@ise.canberra.edu.au>
Lines: 28
Sender: news@csc.canberra.edu.au
Nntp-Posting-Host: isepc3.canberra.edu.au
Organization: University of Canberra, Information Sciences and Engineering
X-Newsreader: Trumpet for Windows [Version 1.0 Rev A]
References: <3hehc5$lor@sgi.iunet.it> <Ruw4L5p.predictor@delphi.com>
Date: Wed, 15 Feb 1995 05:30:32 GMT

In article <Ruw4L5p.predictor@delphi.com> Will Dwinnell <predictor@delphi.com> writes:
>From: Will Dwinnell <predictor@delphi.com>
>Subject: Re: [Q] Pattern Recognition
>Date: Sat, 11 Feb 95 12:40:17 -0500

>Luca Venturini <Mc7880@Mclink.it> writes:
> 
part of posting removed
> 
>Now, taking a step back, we may divide classification systems into a
>few broad groups: those who give a single continuous output (such as
>CMACs), those which give multiple outputs (such as feedforward
>neural networks with more than one output node), and those which give
>a single categorical output (such as a decision tree, or some of those
>class-oriented neural networks like PNNs).  For your problem, you
>could use a multiple output system (continuous or categorical doesn't
>matter, since each output in a continuous system might represent belief
>in one category
could you amplify a little here, please.
I am under the impression that if you train for classification, you can't also 
use the output for confidence in the classifcation.
>: use some strategy for disambiguation, like choosing
>the category with the highest belief value) or a single output
>categorical system.  Using a single output continuous system would
>require having multiple such systems, one for each category.
> 
thanks,
David
