[SunRescue] I could run a small country...

[James Lockwood]

On Sat, 23 Dec 2000, Kurt Huhn wrote:

> > Or if you want to have useful fun, try to solve problems using
> > neural nets.  Got one going now that trains in about a week between
> > 16 450MHz UltraSparc CPU's (gradient backpropagation, 16-way
> > annealing).  Figures out clothing fashionability, of all things.
> > 
> > -James
> 
> Got any URLs for this type of thing - looks like some good 'geek-out'
> material for Sunday mornings...

For a start, search on the net for "libneural".  It's a simple C++
gradient backprop library that's fairly well written, if not optimal.
I've written a similar (but fuller featured and faster) toolkit in a mix
of C and sparcv9 assembly but I don't have redistribution rights (paying
the bills at the day job right now).

The beauty of a net is that you just have to give it input, expected
output, and a staggering amount of CPU time.  :)

The very large boardsize "8-queens" solver is based on a really neat
genetic algorithm that a friend of mine came up with a couple weeks ago.
This code _will_ be made public shortly, we want as many people running it
as possible to get a better picture of statistical distributions at very
large sizes.  It's in many ways an ideal problem to distribute, a large
amount of work is done for a small number of results which are trivially
checkable.

-James