X-Message-Number: 7849
From:  (Thomas Donaldson)
Subject: Re: CryoNet #7840 - #7847
Date: Thu, 13 Mar 1997 09:45:19 -0800 (PST)

Hi again!

For Garrett Smyth:

I'm not arguing with you and basically I'd say you're correct. I will add one
more point though, and its one that people involved in the theory of computing
will like: neural nets are NOT von Neumann machines, either the ones working
in us or the ones we make for our computers. The ones working in us are not
DIGITAL, in addition to the second problem our neural nets share with those
used in computing: there are many more than just one processor --- or should
be, to take advantage of the features of neural nets (yes, small ones have
been emulated in PCs, just as small parallel computers can be emulated 
(sort of) in PCs ... though you lose speed, and have very strong limits on
the number of processors you can emulate!). 

Forgetting for a moment the issue of analog responses, neural nets have 
another feature. Certainly someone may claim that their behavior might be
IMITATED by a von Neumann machine --- just as small neural net programs can
run in PCs. But the definition of von Neumann machine omitted one major issue
important to us as human beings (and it would remain important even if we 
can become immortal): the TIME required for these machines to work. We want
parallel machines so that we can solve our problems in a reasonable time. Any
kind of computer, theoretical or not, which fails to do so becomes at most
a theoretical curiosity.

And so we see the foundations shifting underneath us as we live. Theorists 
have started to look at nondigital computers too. Personally I think that 
we'll see von Neumann's ideas become historical curiosities. Sun now makes
parallel machines, and Intel with its MMX chips uses parallelism right on
the chip... so far, noticeable only to the assembly-language programmer. 
But we'll see.

			Long long life,

				Thomas Donaldson


Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=7849