Re: tlug: Karl-Max has cool dreams [was: dual-pentium processors]

["Stephen J. Turnbull" <turnbull@example.com>]

Well, we've got Karl-Max using whitespace to make his text readable
...

>>>>> "Karl-Max" == Karl-Max Wagner <karlmax@example.com> writes:

    >> of work.  And then there are more problems where we don't know
                                    ^^^^
    >> any useful parallel algorithm than problems where such

    Karl-Max> Right. But they are in the minority.
              ~~~~~                      ^^^^^^^^
... now maybe we can get him to read what he's responding to?

I am not a specialist, but I do read about this for philosophical
interest (and because as soon as we have a good technical analysis of
abstract distributed processing	I plan to apply it to microeconomic
theory), and it seems to me that I've read Manuel's statement
(relatively few problems are parallelizable) many times, and I've
never seen it denied---until Karl-Max just did so.

It seems to me that our choice of problems may be biased by our much
better understanding of sequential algorithms.  It is clear from my
(best characterized as "abortive", if not "abortions") attempts to
apply automata theory to economics that societies and economies choose 
problems that can be solved in parallel (that's why the market economy 
works better than any other) and end up with "efficient" solutions (no 
single processor can achieve a better processor-local solution without 
lowering some other processor's solution) rather than "global maxima" 
(inasmuch as the latter requires lots of expensive communication).

But I don't think this is universally applicable.

    >> I don't think, there will be ubiquitous parallel computing
    >> without a paradigm shift in programming languages.  Please

    Karl-Max> Let's say: you need the same paradigm like a hardware
    Karl-Max> description language.

Raaaaight.  Not!  I mean, I think that's inappropriate.  What we know
neural networks (and similar) to be good at are excellent approximate
solutions to pattern recognition problems, "trainability," and the
like.  To apply parallel hardware discription languages to arbitrary
problems will require the willingness to accept approximate answers
from our machines.

Uh-uh.  People are cheaper and more flexible.  What we want from our
machines is better reliability than people can give you.  We know very
little about getting exact solutions to arbitrary problems based on
parallel computation.

-- 
University of Tsukuba                Tennodai 1-1-1 Tsukuba 305-8573 JAPAN
Institute of Policy and Planning Sciences        Tel/fax: +1 (298) 53-5091
--------------------------------------------------------------
Next Nomikai: 18 September, 19:30 Tengu TokyoEkiMae 03-3275-3691
Next Meeting: 10 October, Tokyo Station Yaesu central gate 12:30
--------------------------------------------------------------
Sponsor: PHT, makers of TurboLinux http://www.pht.co.jp