Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!MathWorks.Com!news.kei.com!ub!acsu.buffalo.edu!jn
From: jn@cs.Buffalo.EDU (Jai Natarajan)
Subject: Re: Visualizing Neural Networks
Message-ID: <CxtqHJ.F0E@acsu.buffalo.edu>
Originator: jn@hadar.cs.Buffalo.EDU
Sender: nntp@acsu.buffalo.edu
Nntp-Posting-Host: hadar.cs.buffalo.edu
Organization: State University of New York at Buffalo/Computer Science
References: <HkLekqLVTBCC065yn@login.dknet.dk> <37s1uh$ogp@cantaloupe.srv.cs.cmu.edu> <CxsBtK.5zn@acsu.buffalo.edu> <STOET.94Oct17100006@lili7.uni-bielefeld.de>
Date: Mon, 17 Oct 1994 15:46:30 GMT
Lines: 25


In article <STOET.94Oct17100006@lili7.uni-bielefeld.de>, stoet@lili7.uni-bielefeld.de (Gijsbert Stoet) writes:
|> >>>>> "Jai" == Jai Natarajan <jn@cs.Buffalo.EDU> writes:
|> In article <CxsBtK.5zn@acsu.buffalo.edu> jn@cs.Buffalo.EDU (Jai Natarajan) writes:
|> 
|> 
|>     Jai> If i understand your second question right, you want to know how to
|>     Jai> judge whether your training is progressing properly. One strict rule
|>     Jai> you must observe (and which is tempting to violate) is that you must
|>     Jai> NEVER show your test set to the net while training. That means you
|>     Jai> can't just pause occasionally, feed some test samples into the net &
|>     Jai> see that the recognition rate is incresing. This process introduces a
|>     Jai> bias into your training process (both human and neural).
|> 
|> Of course can you pause occasionally. When showing the "test set", i.e. the
|> input-output pattern which are not part of the set to be learned, you should
|> not change the connection-strenghts. Then, the test-set doesn't influence the
|> performance. (As far as I know).
|> 
|> Gijsbert
What I meant was that you shouldn't judge completion of training from these
results. Otherwise you may end up tailoring to the test set. That's what I meant
by "introducing a human bias". BTW I think there is a general paper by Richard
Lippmann available which talks about this - I'm not sure of the ref. though.
Jai
