ENSEEIHT, Toulouse, 21 et 22 mars 2016

Intervenants

• Frédéric Bonnans, CMAP, Ecole Polytechnique, Palaiseau
• Eric Féron, Georgia Tech, USA
• Pierre Martinon, CMAP, Ecole Polytechnique, Palaiseau

Programme

Cours 1 (6h) : Eric Féron, Georgia Tech, USA
Formal methods in optimization engineering : Optimization algorithms have been traditionally used as desktop applications for non
real-time tasks. The advent of fast computers allows optimization algorithms to be used in a real-time setting. They allow large, but safety-critical systems to be controlled much more efficiently than traditional control methods can. For that to happen, there is the necessity to guarantee the safety of these algorithms; only then can such algorithms compete with well-established, older methods.

This course will discuss how formal methods can contribute to supporting safety claims for optimization algorithms, convex optimization algorithms being first among them. First off, formal methods will be briefly described. Then some applications to optimization software will be discussed, in the context of recent advances in convex optimization and its implementation on fast computers. Last, the relation of these topics to industrial applications and development standards will be introduced.

Cours 2 (6h) : Frédéric Bonnans et Pierre Martinon, CMAP, Ecole Polytechnique, Palaiseau
Part 1 :
Bases théoriques et algorithmiques.Problèmes de contrôle déterministe et stochastique
Systèmes avec coûts de transition
Equation de Hamilton-Jacobi-Bellman
Discrétisation semi lagrangienne