Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!ncar!boulder!bill
From: bill@boulder.Colorado.EDU
Newsgroups: comp.ai.neural-nets
Subject: Re: Laymans info. on NN
Message-ID: <16572@boulder.Colorado.EDU>
Date: 5 Feb 90 18:57:04 GMT
References: <3058@munnari.oz.au> <1856@castle.ed.ac.uk> <3180@umn-d-ub.D.UMN.EDU> <1649@skye.ed.ac.uk> <1990Jan31.221330.12069@Neon.Stanford.EDU> <958@watserv1.waterloo.edu>
Sender: news@boulder.Colorado.EDU
Reply-To: bill@synapse.Colorado.EDU (Bill Skaggs)
Organization: University of Colorado, Boulder
Lines: 28

In article <958@watserv1.waterloo.edu> ssingh@watserv1.waterloo.edu 
($anjay "lock-on" $ingh - Indy Studies) writes:
>Another question that has gone unanswered since I began studying NNs. I have
>been to two tutorials, one by Rumelhart and Sejnowski, and the other by
>Hinton, and both imply that all the mathematics needed to make an
>original contribution to NN work is simple calculus. Is this really
>the case? [ . . . ]

  Well, the answer is, it depends on the kind of contribution you want
to make.  If you want to do theoretical work on questions like "Are
neural networks computationally equivalent to Turing machines?" or
"For what problems is a Boltzmann machine guaranteed to converge on
the best solution?", then it helps to have a pretty strong mathematical
background.

  However, there are a number of extremely important open questions
for which ingenuity, intuition, and hard work are likely to mean more
than mathematical knowledge.  For example:  a large portion of the
PDP books (edited by Rumelhart & McClelland) is devoted to setting up
simple network models of various psychological processes and examining
their behavior.  A lot more work can be done in that direction, probably
using little more than back-prop or competitive learning.  Perhaps the
most important basic question in the whole domain is the extent to
which massively distributed parallelism can make up for crude and
simplistic algorithms, and that is something which probably can only
be investigated by looking at lots of examples.

	-- Bill Skaggs