Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news.mathworks.com!udel!news.sprintlink.net!howland.reston.ans.net!darwin.sura.net!news.Vanderbilt.Edu!NewsWatcher!user
From: goldenjb@ctrvax.vanderbilt.edu (jim golden)
Subject: Re: "Trading... - Oh... really?" Part II (last)
Message-ID: <goldenjb-311094113108@129.59.170.62>
Followup-To: comp.ai.neural-nets
Sender: news@news.vanderbilt.edu
Nntp-Posting-Host: 129.59.170.62
Organization: vanderbilt
References: <83591.arcon@dial.illinois.net> <goldenjb-261094142042@129.59.170.62> <1994Oct27.115504.26956@oxvaxd> <goldenjb-281094120415@129.59.170.62> <1994Oct30.140320.1@tnclus>
Date: Mon, 31 Oct 1994 16:14:01 GMT
Lines: 31

In article <1994Oct30.140320.1@tnclus>, tarmoleinone@tnclus.tele.nokia.fi
(Mr Tarmo Leinonen) wrote:


>         Could you please explain, which of your rhetoric questions are
>         relevant and which irrelevant?
> 
>         "Why not some easily implementable statistical method"
> 
>         Statistics itself is highly reliable. Statistians are mortals
>         and highly unreliable. Easily implementable methods are equally
>         inefficient to find what you are looking for.
> 
>         The easiest way to find out, if your data contains something
>         essential, is to feed it into a neural network. If the NN
>         is able to demonstrate some kind of predictive ability (you
>         need statistics for this), it is another decision how to
>         implement working version. It can be made by tayloring NN
>         or by statistical methods, if you can find any suitable  easily.
>         

Yes and no.  Statistics are reliable as are some statisticians.  The point
being, with certain statistical methods you can explain how you arrived at
your answer.  If I feed my data to some classifier system and I get
clustering, it may or may not mean anything at all.  And, you can not say
what it did without some fundamental knowledge of your network.  My
argument is not that NN are useful, its that they are useful *only* if you
understand how that network works.  GIGO and all that...  My first question
to anyone implementing solutions from a multilayer BP is "why did you
choose that architecture and what happens to your solutions if you fidle
with it?"
