Re: swarm agents with genetic algorithms

[Norberto Eiji Nawa]

Hello Paul,

I am very interested in the topic of learning agents and collective
behavior, especially in the context of socioeconomic systems. There is
a lot being done in the field of multi-agent systems, but somehow I
feel that the issues dear to the Swarm community (model
validation/verification, reproducing others results) do not receive
enough attention there (someone correct me if I am wrong). A Swarm
interest group in learning agents sounds a very nice idea to me!

I have recently started fiddling around with Breeder, as you
noticed. Actually I was on hurry to do some simple experiments, so I
have to admit that I didn't take a careful look in all the inner
meanders of the library code; I basically used the same framework in
the demo app (gademo) and changed the fitness/evaluation function. The
point that I want to stress here is that I used a simple binary
encoding in the chromosomes, so I had very little to change in the
basic framework. The pertinent question is, how easy would have it
been to use Breeder or GM if the encoding scheme were more complex,
and the evaluation function of the individuals involved some
non-straightforward process simulation, as it is often the case in
multi-agent simulations?

For instance, in the evolutionary computation community, there are
several packages/libraries that support the implementation of a GA,
but my impression is that only a small minority makes use of them
(again, correct me if I am wrong). One of the reasons for being so, I
guess, is that people conclude that their _special mechanisms/encoding
schemes_ wouldn't fit in the precoded packages, and end up taking the
burden of coding everything by themselves, despite all the problems. I
am not quite sure whether this is only due to the widespread belief
that "it is easier/faster to do it by yourself than learn someone
else's code" or if there are strong inherent limitations in the usage
of GA libraries. (If I am not wrong, J.J.Merelo is also the author of
GAGS, a GA library; I am very sure he thought about all these issues
long ago, could you illuminate us with your knowledge :-)?)

Having showed my pessimistic side, I still think that a group on
learning mechanisms (GA/NN) for agents would be very interesting.

Cheers,

Eiji

ps.: I recently downloaded ASM-2.0 and started reading its
documentation. I didn't know about http://ArtStkMkt.sourceforge.net,
was there an announcement on that? 

                  ==================================
   Swarm-Support is for discussion of the technical details of the day
   to day usage of Swarm.  For list administration needs (esp.
   [un]subscribing), please send a message to <address@hidden>
   with "help" in the body of the message.