|
|
|
|
Subject: TR : Uncertainty Analysis for MAS - conclusion and acknowledgement
From: Barreteau Olivier (olivier.barreteau@montpellier.cemagref.fr)
Date: Mon Jun 26 2006 - 16:24:11 CEST
-----Message d'origine-----
De : News and discussion about computer simulation in the social sciences [mailto:SIMSOC@JISCMAIL.AC.UK] De la part de Carsten Schulz
Envoyé : lundi 26 juin 2006 16:17
À : SIMSOC@JISCMAIL.AC.UK
Objet : Re: Uncertainty Analysis for MAS - conclusion and acknowledgement
Hi,
I'd like to thank all of you, who helped me, to get literature for my
problem. I got about a dozen mails with links and helpful hints.
I try to summarise those now:
There are two sites at which one should begin:
1. For validation in general:
http://www.econ.iastate.edu/tesfatsi/empvalid.htm
especially the articel of "Fagiolo, Windrum, and Moneta"
http://www.econ.iastate.edu/tesfatsi/EmpValid.Fagiolo.2006-14.pdf
An other good work is the JASSS-article of Richiardi, Leombruni et
al. (2006). They recommend "A Common Protocol for Agent-Based Social
Simulation" http://jasss.soc.surrey.ac.uk/9/1/15.html.
2. For the topic of error propagation and uncertainty (there is still no
hint for this special task for MAS)
www.nusap.net
especially the site with the guide and the tool catalogue
http://www.nusap.net/sections.php?op=viewarticle&artid=17
Unfortunately I formulated my question not very clear and so I got lots
of hints about the normal validation like cross-validation. Since that
is kind of basic and summed up very god at the site
http://www.econ.iastate.edu/tesfatsi/empvalid.htm
I will not summarise any of these articles.
Two techniques should be mentioned also. First the "Active Nonlinear
Tests (ANTs) of Complex Simulations Models" of John H. Miller. Because
the sensitivity analysis can become quite complex, he suggest a genetic
algorithm approach to test model boundaries.
http://zia.hss.cmu.edu/miller/papers/ant.pdf
Second: "An Architecture-Independent Cognitive-Trace Analysis Tool for
Real-World Human Behavior Models" Scott Neal Reilly, Karen A. Harper &
Andrew Young. It is a tracer tool for not too abstract agents and
their cognitions and behaviours.
Last but not least Doug Samuelson gave me the last hint, which I will
just cite to the list:
"I think it is fair to say that the profession has not come close to a
consensus on this subject. There are serious problems in trying to
assess the quality of models whose complexity approaches that of the
situations they purport to represent. Try searching for the literature
on "docking" simulation models -- it's scattered, and I wouldn't venture
to say I know all the key papers in this area, but it's worthwhile and
highly relevant to your question. Robert Axelrod, in particular, did
some good papers on this; perhaps you can get a pretty good set, for
starters, by finding the earliest of his papers and then finding
everything else that cited it."
Thanks again to all the active readers and writers of this list,
Carsten Schulz
Carsten Schulz wrote:
>
>
> Dear modellers,
>
> I am interested in the validation of multi-agent simulations, because
> in
> our project we have modelled a quite large and spatially explicit
> multi-agent simulation, which imports data from other models and exports
> data also to coupled models. Now we would like to validate the MAS and
> e.g. investigate the error propagation. But how?
>
> Is there already a known technique for validating agent-based models -
> like sensitivity analysis and Monte-Carlo simulation are techniques for
> validating statistical models?
>
> I thank you in advance for all the abstracts and literature you can
> provide me, I will write a short summary and send it back to the list.
>
>
> Carsten Schulz
> schulz@usf.uni-kassel.de
>