Welcome to Slashdot News Technology Encryption Linux United States
 faq
 code
 awards
 privacy
 slashNET
 older stuff
 rob's page
 preferences
 andover.net
 submit story
 advertising
 supporters
 past polls
 topics
 about
 jobs
 hof

 Sections
books
ask slashdot
features
radio

1000-Pentium Beowulf Cluster for Genetic Programming
Technology Posted by Roblimo on Tuesday August 10, @06:17AM EDT
from the survival-of-the-fittest-code dept.
Kimberley Burchett writes "Genetic Programming Inc. has built a 1000-processor Beowulf cluster for research into genetic programming. GP still has a long way to go, but it's the coolest technology ever for programmers." Update: For more info about genetic programming, please see this article in Salon.

AES Finalists, Round 2 | New Power-of-Two Prefixes?  >

 

Slashdot Login
Nickname:

Password:

Don't have an account yet? Go Create One. A user account will allow you to customize all these nutty little boxes, tailor the stories you see, as well as remember your comment viewing preferences.

Related Links
  • Kimberley Burchett
  • Genetic Programming Inc.
  • 1000-processor Beowulf cluster
  • genetic programming
  • this article
  • Salon
  • More on Technology
  • Also by Roblimo
  • "1000-Pentium Beowulf Cluster for Genetic Programming" | Login/Create an Account | Top | 38 comments | 13 siblings
    Threshold:
    The Fine Print: The following comments are owned by whoever posted them. Slashdot is not responsible for what they say.
    ( We can't even spell bayta!)
    Beowulf cluster for SGI ? (Score:1)
    by bain (bain@no-spoof-reaper.org) on Tuesday August 10, @06:21AM EDT (#1)
    (User Info)
    Does anybody know if anybody has done clusters using SGI MIPS boxes ...

    That would be interresting to see..

    bain
    [ Reply to This | Parent ]
    i want one
    by Anonymous Coward on Tuesday August 10, @06:43AM EDT (#3)
    Anyone figure how much one of these would cost from the specs they gave us?
    [ Reply to This | Parent ]
    What's so cool about it?
    by Anonymous Coward on Tuesday August 10, @06:54AM EDT (#4)
    Correct me if I'm wrong, but genetic programming seems to be limited to a small set of problems where a well-define goal state exists, and where metrics permit the valuation of the distance between the goal and the program's current fitness to achieve that goal.

    In other words, it's pretty unlikely, impossible even, to breed something like MS Office or Photoshop by application of the genetic programming paradigm.

    --
    I don't use Linux.
    [ Reply to This | Parent ]
    Slashdot me baby.
    by Anonymous Coward on Tuesday August 10, @06:57AM EDT (#6)
    Slashdot me baby.
    [ Reply to This | Parent ]
    Beowulf? (Score:1)
    by Stiletto on Tuesday August 10, @07:13AM EDT (#9)
    (User Info) http://netstorm.gamestats.com/
    Maybe someone could make a Beowulf out of it... Oh... uh... no wait...
    [ Reply to This | Parent ]
    top500 supercomputers (Score:1)
    by semis on Tuesday August 10, @07:16AM EDT (#11)
    (User Info)
    How does this rate with Avalon and cplant on top500.org ?
    [ Reply to This | Parent ]
    Pardon my paranoia (Score:1)
    by unitron (unitron@familycom.com) on Tuesday August 10, @07:35AM EDT (#14)
    (User Info)
    "1000-Pentium Beowulf Cluster for Genetic Programming"
    Pardon my paranoia, but just whose genetics are going to be programmed, for what, and why? :)


    "I wish I could change my sig file without the change being retroactive" unitron
    [ Reply to This | Parent ]
    Conway's life? (Score:1)
    by Erich (nobody@wreck.org) on Tuesday August 10, @07:46AM EDT (#15)
    (User Info)
    Based on the way the nodes are arranged (a torroid) and the fact that each node talks with it's neighbors, and the fact that the server waits for the entire generation to complete...

    I think that these guys are just really into Conway's life. (you know, the one that xlock -mode life plays). They got a grant from some big company, and were like ``Hey... if we could get 1000 processors, we could re-generate the board really fast.

    Seriously, though.. genetic programming can provide some really nifty results for strange things... Speech / sight recognition, for instance. Produce a program that tries to recognize the difference between wav files of ``yes'' and ``no''... then mutate it a bunch of times, and see which programs recognise better, and which worse... sort of interesting.

    -- Erich

    Paranoia is just Reality at a higher resolution!

    [ Reply to This | Parent ]
    GP - Ideal candidate for RC5 or SETI style effort (Score:1)
    by Sanity (I.Clarke@strs.co.uk) on Tuesday August 10, @08:20AM EDT (#23)
    (User Info) http://www.dcs.ed.ac.uk/~iic
    I have often thought that Genetic Programing would make a perfect candidate for a SETI or distributed.net style system. Simple define a simple language in which the programs can be written, define a simple language in which a fitness function can be defined, write a client with some kind of distribution mechanism (which can even be arranged as a hierarchy to avoid server bottle-necks) and Bobs your uncle!

    Any takers...?

    --
    A subversive is anyone who can out-argue their government.

    [ Reply to This | Parent ]
    "Coolest thing ever for programmers"? (Score:2)
    by Salamander (jdarcy@mediaone.net) on Tuesday August 10, @08:44AM EDT (#28)
    (User Info)
    Not exactly. If they ever truly get it right, there won't be programmers. I can hardly think of anything less cool.
    [ Reply to This | Parent ]
    A good intro book for Genetic Programming ... (Score:1)
    by Zach Frey (zfrey@bright.net) on Tuesday August 10, @08:59AM EDT (#30)
    (User Info) http://www.bright.net/~zfrey/

    ... is David E. Goldberg's Genetic Algorithms in Search, Optimization and Machine Learning. I used this book in a class I took (mumble) years ago (it was a new release then), and it's a very clear introduction to both the theory and practice of genetic algorithms. The book includes sample code (in Pascal, unfortunately).

    Goldberg was a graduate student of John Holland, who invented the idea of using algorithms modeled on populations of genomes to do effective computer searches. Not only is Goldberg well entrenched in the academic side of GAs/GP, he has also used them in the Real World&tm;. His doctoral thesis was on using GAs to determine the optimal routing of gas pipelines.

    GAs are useful in problems where you have a large search space, and you want to find an optimal (or nearly-optimal) solution. (Think of "Travelling Salesman"-type problems.) While it's tough (maybe impossible) to prove that you have the perfectly optimal solution with a GA, they are very good and finding solutions within some small delta of optimal, which is often all you need. Set up properly, they are also very good at avoiding the local minima problems that affect other types of machine search, such as classical rule-based AI or neural networks.

    Of course, finding a good schema encoding and a fitness function that reflects what you're trying to optimize for are non-trival exercises ...


    [ Reply to This | Parent ]
    ops/sec (Score:1)
    by vyesue on Tuesday August 10, @09:37AM EDT (#35)
    (User Info)
    {The 1,000-Pentium system operates at an aggregate clock rate of 0.350 Tera-Hertz. Because multiple machine operations are typically performed on each clock cycle on the Pentium II processor, the machine performs somewhere in the neighborhood of a tera-ops per second.}

    Is this right? ~3 operations per cycle seems awfully high...
    [ Reply to This | Parent ]
    Pictures! (Score:1)
    by malice95 (malice@exit109.com) on Tuesday August 10, @10:06AM EDT (#38)
    (User Info) http://www.jeepin.org/
    We Wanna see pictures!:) Well I do at least..


    [ Reply to This | Parent ]
  • 1 reply beneath your current threshold.
  •   APL hackers do it in the quad.  
    All trademarks and copyrights on this page are owned by their respective owners. Comments are owned by the Poster. The Rest 1997-99 Andover.Net.
    [ home | awards | supporters | rob's homepage | contribute story | older articles | Andover.Net | advertising | past polls | about | faq ]