Poste Post-doctoral fellowship: Machine learning techniques for computing viability limits of dynamical systems

Post-doctoral fellowship: Machine learning techniques for computing viability limits of dynamical systems Post-Doc DeadLine: 30/10/2009 guillaume.deffuant@cemagref.fr http:// Post-doctoral fellowship: Machine learning techniques for computing viability limits of dynamical systems Duration: 18 months, to begin January 1st 2010. Location: Clermont-Ferrand, France, Laboratoire d'Ingenierie des Systemes Complexes (Cemagref), node of PASCAL2 network (together with LIMOS Universite Blaise Pascal Clermont-Ferrand, Sequel Project INRIA Lille, LITIS INSA Rouen) Supervisors: Guillaume Deffuant (LISC) and Gaelle Loosli (LIMOS) Gross salary: 2150 Euros per month. Deadline for applications: 30 October 2009 Subject: LISC and LIMOS collaborate in the elaboration of efficient algorithms for determining viability limits (or viability kernels) of dynamical systems. The problem is to approximate the boundary of a manifold in the state space of the system as precisely as possible. This manifold corresponds to the states from which there exists a control policy that maintains the viability of the system (i.e. keeps it within the boundaries of an a priori given constraint set). The first algorithms, based on "support vector machines", provide good results in general. However for some geometric configurations, they can create artificial boundaries which make the global process of approximation fail. Moreover, SVMs are not very convenient to compute a distance to the boundary, which is important in the algorithm. The research subject of the post-doctorate is to develop new learning algorithms which avoid these problems, while being as parsimonious as possible in order to insure compact representations in spaces of large dimension. This approach will be tested on different dynamical systems representing ecological or social phenomena. Candidate profile: PhD in machine learning or applied mathematics. Contact: Send CV and motivation letter to: guillaume.deffuant@cemagref.fr