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

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

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

    >> 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

    Karl-Max> I didn't think so much of neural networks. Most paralell
    Karl-Max> processing schemes are a lot more primitive than
    Karl-Max> that. Take image processors as an example. It is easy to
    Karl-Max> cut an image into small pieces to be processed sepately
    Karl-Max> ( for the most part ).  There are other problems of that
    Karl-Max> kind, like weather forecasting, code cracking ( remember
    Karl-Max> the RC5 cracking initiative ! ). In fact, that kind of
    Karl-Max> things are what most computers are processing.

Can I have a reference for that fact?  I think Bill Gates would be
surprised to hear that, as would the top guys at Intel.

    >> like.  To apply parallel hardware discription languages to
    >> arbitrary problems will require the willingness to accept
    >> approximate answers from our machines.

    Karl-Max> There ain't anything like and "exact" answer. Your
    Karl-Max> answers are only as good as your input values. So in
    Karl-Max> fact approximate answers are the order of the
    Karl-Max> day. Anybody using circuit analyzer software knows
    Karl-Max> that......

By "approximate" I mean "a worse approximation than the input values".
Eg, a local optimum can be arbitrarily far from the global optimum in
a non-convex optimization problem.  (How far from Mt. Everest is
Fuji-san?  Mt Ranier?)

I should have defined approximate.  Sorry.

    >> Uh-uh.  People are cheaper and more flexible.  What we want
    >> from our

    Karl-Max> Hmmm......actually people are pretty expensive.....

Not in the realm of economic decision-making, which is why I chose the 
phrase "arbitrary problems."

    >> machines is better reliability than people can give you.  We
    >> know very

    Karl-Max> It's another question whether they give that to youm
    Karl-Max> Machines are great in messing things up. No human being
    Karl-Max> can make such a mess than a computer running Amok.

As you are so fond of pointing out, that's not a computer problem,
that's a management problem.

    >> little about getting exact solutions to arbitrary problems
    >> based on parallel computation.

    Karl-Max> Actually - do we need them ? In practice approximate
    Karl-Max> solutions are good enough.

Not in horseshoes, hand grenades, and air traffic control.

Implicit in your confidence is the fact that you are reserving the
right to choose which problems you will apply your methods to.  That's
OK, in fact that is the essence of good management.  However you
should address the fact that I have problems you can't solve, and that
Manuel is considering ALL problems.  If you want, you can point out
that we started at "Symmetric Multiprocessing," but I still maintain
that the collection of nearly identical processors we call "the
economy" is a pretty good approximation to a symmetric multiprocessing 
system.

-- 
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