Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news4.ner.bbnplanet.net!news3.near.net!paperboy.wellfleet.com!news-feed-1.peachnet.edu!usenet.eel.ufl.edu!col.hp.com!sdd.hp.com!swrinde!howland.reston.ans.net!news.nic.surfnet.nl!rug.nl!rolf
From: rolf@cs.rug.nl (Rolf P. W"urtz)
Subject: Re: Searching for NN reference
Message-ID: <DEHq8n.9wn@rug.nl>
To: rolf
Sender: root@rug.nl (Operator)
Nntp-Posting-Host: rug102.cs.rug.nl
Organization: Dept. of Computing Science, Groningen University
References:  <DEHHAB.A3w@cee.hw.ac.uk>
Date: Wed, 6 Sep 1995 15:41:11 GMT
Lines: 23

In article <DEHHAB.A3w@cee.hw.ac.uk>, ceemw@cee.hw.ac.uk (Marcus John Williams) writes:
> 
> Apparently
> all the network had done was learn that american tank photos had good
> lighting and russian photos (taken from spy planes?) were poor quality.
> 

Well, this seems to be the fundamental problem of neural networks applied
to vision (and probably to much more). There is simply no chance to cover
the huge variety in input with training examples. The net will always
come up with the simplest (in its own terms) description of the training
set, and that might have nothing to do with what whoever set up that set
had in mind.

Anyway, this sounds like a good story, so please post more about it, don't
just respond by email.

Rolf

+---------------------------------------------------------------------------+
| Rolf P. W"urtz | mailto:rolf@cs.rug.nl | URL: http://www.cs.rug.nl/~rolf/ |
| Department of Computing Science, University of Groningen, The Netherlands |
+---------------------------------------------------------------------------+
