Genetic Algorithms

[Robert G. Brown]

On Thu, 24 Jun 1999, Steve Hill wrote:

> Where a Beowulf might well be extremely useful is where the evaluation of
> fitness is non trivial. For example, neural network design using GA's.

I demonstrated our (commercial) parallelized neural network that uses a
very extensive GA for initialization at Linux Expo in Raleigh a month
ago.  Although there is some advantage to beowulfery in parallel
evaluations of the fitness, there is enough serial work that the most
gain one is likely to be able to realize -- as determined by Amdahl's
Law -- appears (for our code, at any rate) to be between 10 and 20,
depending on the size and scale of the training set.  I'm hoping to
parallelize the GA itself more directly and up the gain to a higher
fraction, but there are definitely some serial parts to the code.

   rgb

Robert G. Brown	                       http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb@phy.duke.edu