Re: [SwarmFest2004] Re: [Directors] swarmfest paper evaluations: can't all be talks!

[gepr]

Rick Riolo writes:
 > 
 > re the PSO paper:
 > 
 > actually, my objection is not that its not agent-based,
 > but that its not a *modeling* paper.
 > i thought the unifying theme is, to quote the cfp:
 >    ...multi-agent modeling including...
 > 
 > basically this paper is proposing a change to a
 > search algorithm to make it maybe search better.
 > i would similarly not look favorably on papers
 > that were about improving NN's, EAs, etc.
 > do we really want to move toward accepting any
 > papers about improved versions of NNs, GAs, etc?

Well, _I_ do. [grin]

But, I disagree with the position that this "algorithms paper" is not
a modeling paper.

At least a half of modeling is getting the right behavior out
of the individuals.  And this change to the particle swarm model
is an attempt to coerce the collective behavior of the individuals
from using global/perfect information into using local/imperfect
information.  And that is a _very_ common thing other ABMers have
to do.

This is _definitely_ a modeling paper.  It's at least as welcome
as the previous papers on things like XPCOM, model profiling, 
etc that we've done throughout the years.

 > if not any, how do we decide which to accept and which not?

Personally, I think quality is mostly measured through how 
_well_ someone describes what they've done and whether what
they've done is _useful_ in any way.

Content, imho, for SwarmFest should be judged by fine-grained
agent-based modeling, which includes any _network_ model,
including connectionist models like ANNs, discrete dynamics
like Wuenche's ddlab or cellular automata, panmictic algorithms
like GA/GP, etc.  What I would be less likely to accept would
be _cognate_ agents like BDI agents or expert systems.

I also think that concurrency and distribution play a large
role in "swarm style modeling", which would include things 
like hybrid languages like Mozart/Oz and adaptive algorithms
like load balancers and such, as well as topics like CCS, 
rewriting logics, discrete-event scheduling, etc.

These are all things that inspired and are relevant to Swarm.

 > All that said...
 > if people feel it fits in with the collective view of SwarmFest,
 > we can move it to a talk status.
 > if, as the (few) responses we have seen indicate,
 > the feeling is we go with all 20min talks,
 > we do have a bit of time to play with.
 > 
 > (tho having this be a talk does use up some of that free time
 > you were pushing for, glen! ;-)

But, I'm just doing a little friendly pushing.  Do what you think is
best.  If nobody agrees with me (which happens alot), then I'll
happily log my concientious objection and stay silent from that point
on. [grin]

glen