Aaron Davidson writes:
> In the mean time, can anyone think of useful distributed applications?
> I can imagine a large scale project using Genetic Algorithms could
> potentially work. Each node could evaluate various specimens, then report
> to the central server with the top percentile of tested specimens.
The Grand Challenge in GP is find a mutation function which can mutate machine instructions directly without producing invalid code too frequently. The easiest way to solve this is to allow the fitness function to mutate, feedbacking the individuals into the pool. With time the fitness function will crystallize with enough built-in knowledge to handle the specifics of the given machine instruction set.
> Tierra was more difficult because all of the specimens must be tested
> against each other in a competetive arena, which does not parallelize well.
> GA projects that can be tested in isolation would be ideal however.
Isolation is counterproductive, occasionally swapping individua with a IP address randomly selected from participant pool will do the trick. Bandwidth requirements are almost negligeable.
Email me privately for details. This can be an immensely worthwhile project if done right.