Time: 15:00-16:00, Jun 17th, 2009
Place: Seminar Room 2 (1509), 15F, National Institute of Informatics
(map)
Fee: Free
You need to register your name, affiliation and e-mail address in
advance. Please send a mail titled "22nd Grace Seminar" including
the information to event-info@grace-center.jp.
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Details：

Speaker: Dr. Stéphane Bonneaud, European Center for Virtual Reality

Title:
"Model-agents for the modeling and simulation of complex systems:
application to the ecosystemic of fisheries"

Abstract:

We study the modeling process for the simulation of
complex systems. The modeling and simulation processes
are complex, therefore the modeling has to be rationalized
and control gained over it. Our proposition is to conceive
the complex model as a society of models in interaction.
We apply the pattern oriented approach to rationalize the
modeling by focusing it on the data. We use the agent
approach to distribute the observation and control in the
global model. The society of models thus enables the
emergence of the patterns through the agent's activities.
The artefacts, applied to the modeling, support the data,
the patterns and the agent's interactions. The coupling of
the agents is therefore done through the artefacts, which
is different from an approach based on events.
We identify three types of agents:
(1) the model-agents that execute the model of a phenomenon;
(2) the observator-agents that build up observables,
(3) the controller-agents that locally control the global model.
The agents are also able to adapt to their environment using
operations to modify their perceptions and actions.
The complex model can be therefore incrementally built,
the agent's behaviors can easily be modified, and an agent
can be added or removed from the global model without impact
on the structure of the other agents.
At last, we identify four types of behaviors for model-agents:
(1) based on differential equations (intensional);
(2) individual based (emergent) ;
(3) based on raw data or pre-calculated (extensional ) ;
(4) played by an expert (participatory ).
To go even further in the rationalization of the modeling,
we exhibit computational biases induced by the translation of
conceptual models of agent populations into computational models.
We show that our models are very sensitive to the studied biases.
It is necessary for thematicians --domain experts-- to know the
sources of those biases and the simulation tool must be
specialized in order to:
(1) explicit the model,
(2) guarantee the identified properties and
(3) make systematic analysis.
This is what we have done in our simulator.
Finally, this work takes place in the ANR project Chaloupe.
The challenge here is to envision the co-viability of ecological
populations and of their exploitations by social groups (e.g. fisheries).
We show our simulation results, confirming by this mean that
the society of models enables a flexible and rigorous modeling.