Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!uflorida!gatech!psuvax1!omega!ian
From: ian@omega (Ian Parberry)
Newsgroups: comp.ai.neural-nets
Subject: Re: request for philosophic reactions to connectionism
Keywords: connectionism theory and experiment
Message-ID: <4548@psuvax1.cs.psu.edu>
Date: 4 May 89 14:34:31 GMT
References: <370@eurtrx.UUCP> <935@syma.sussex.ac.uk> <10139@burdvax.PRC.Unisys.COM>
Sender: news@psuvax1.cs.psu.edu
Reply-To: ian@theory.cs.psu.edu (Ian Parberry)
Organization: Penn State University
Lines: 35

In article <10139@burdvax.PRC.Unisys.COM> pastor@bigburd.PRC.Unisys.COM (Jon Pastor) writes:

>Similarly, mathematical rigor (e.g., proofs of convergence) is undeniably a
>Good Thing, but many of us got into AI because rigorous solution techniques
>often require assumptions and restrictions that do not hold in the real world.

> Only a fool would claim that
>mathematical rigor is unimportant, but practitioners will gladly use a tool
>that has strong empirical support while the theoreticians continue looking
>for formal results vindicating the empirical results.
>

I want to put in a plug for theoreticians here.  I see too much theory-bashing
at neural network conferences.  I am told that this is the legacy of Minsky
and Papert.  I have heard "distinguished" neural networkers say things like
(I'm paraphrasing from a faulty memory here ) "we don't need theory",
"combinatorics is harder than analysis", "those discrete theory people don't
know what they're doing", even in invited talks.  This does not set a good
example.

I am a theoretician.  I believe that theory and practice are complementary.
Experiments may show that there is something deep going on.  Theory trys to
explain it.  Theoretical results steer future experiments (pruning the search
tree, if you like).  Each side feeds the other.

Theory is hard.  We make assumptions and restrictions because the mathematical
tools do not yet (and may never) exist for dealing with the real problems.
This doesn't mean that theory is impractical.  Experimentation is hard too,
but each takes some of the uncertainty out of the other.  Experiments tell
theoreticians what to try to prove (and sometimes how to go about it).  Theory
tells experimenters what approaches are more likely to be profitable than
others.

Some people like to do theory, and some hate it.  That's fine.  But each
side should at least learn what the other side is doing, and what it means.