Orateur
Dr
Javier Gil-Quijano
(Laboratoire d'Informatique,Paris VI)
Description
The modelling by simulation of complex systems is a cyclic process: the modeller incorporates his/her knowledge into the model, runs simulations, discovers bugs or unwanted effects, corrects the model and eventually his/her knowledge, and the cycle restarts. The process ends when it is not possible to further improve the model because of technical or knowledge limitations. That cyclic process is particularly hard when modelling multi-level complex phenomena mainly because of the emergence of high level structures : the behaviour of lower level agents can be strongly influenced by the existence of emergent structures. Those structures must then be detected and considered as agents in simulations. The detection of structures is particularly difficult because of their dynamic nature. To consider the structures as agents implies to provide them with a behaviour. The latter is not an easy task because of the interdependence of the behaviours of agents that are placed at different levels. In this talk I propose to adress those problems by using automated learning mechanisms. In the first part of this talk I propose the use of statistical learning to discover the emergent structures. In the second part I propose the use of technics of automatic composition of programs to build the agents’ behaviours.
0
Javier Gil-Quijano