Neural Network applications using Beowulf

[Robert G. Brown]

On Mon, 11 Nov 2002, Thomas Zheng wrote:

> Hi,
> 
> I am studying the feasibility of using Beowulf as a computing platform for 
> Neural Network applications, more specifically, Self-Organizing Map 
> applications.  Has anyone done similar research on this topic?  Any 
> comments are greatly welcomed.

I haven't done SO maps, but I've dallied a bit with genetically
optimized FFBP NN's.  They are definitely parallelizable (a variety of
the work can be done in parallel with equally various organizations) but
they aren't quite trivially parallel.  For example, you can distribute
the training data to all nodes and then split/distribute a population to
the nodes for assessment against the training set, but there is a fair
bit of communication involved.  Or you can distribute the entire
population to the nodes and split up the assessment.  One way or
another, though, one ends up shipping lots of networks around.

There are other possibilities, as well -- maintaining independent
genetic threads on the nodes, limiting communications to specific
intervals, running champion/challenger higher order optimizations of the
network parameters themselves.

I actually think that there is room to do a whole lot of interesting
research on this in the realm of Real Computer Science.

Too bad I'm a physicist...;-)

   rgb

> 
> Thanks in advance.
> 
> Regards,
> Thomas Zheng
> 
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
> 

-- 
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 at phy.duke.edu