Ian Clarke ([EMAIL PROTECTED]) wrote:

> The idea would be that nodes in the network would constantly evaluate
> the overall performance of their neighbors (average response times etc).
>  Embedded in their IDs, nodes will have a compact description of their
> settings (randomly chosen at startup, and periodically modified at
> random too).  After a while, the top 30% (yet another arbitrary value!)
> can then be averaged, and the user's node can reset its own values to
> those.  

Settings that work well for a node on a Pentium 4 with 2 GB of RAM
and Sun Java on Windows XP Pro, might not work so well on a node on
a Pentium-120 with 128 MB running Kaffe on OpenBSD.

If you want a genetic learning scheme, I'd suggest that a node
should randomly tweak its *own* settings and keep statistical results
over long periods of time.

-- 
Greg Wooledge                  |   "Truth belongs to everybody."
[EMAIL PROTECTED]              |    - The Red Hot Chili Peppers
http://wooledge.org/~greg/     |

Attachment: pgp00000.pgp
Description: PGP signature

Reply via email to