Weitere Informationen

Dokumente

Einladung zum Vortrag "Algorithmen zur diskreten stochastischen Optimierung mit stetigen Hilfsvariablen" (11.03.2015, ab 16:15 Uhr)

Discrete stochastic optimization with continuous auxiliary variables

 

Rodolphe Le Riche* (speaker), Alexis Lasseigne**, François-Xavier Irisarri**

* CNRS LIMOS and Ecole des Mines de St-Etienne, France

** ONERA Composite Systems and Materials, France

 

In this talk, we will consider stochastic discrete optimization algorithms in a unified, versatile, framework : it consists in alternating the sampling of a probability density function, p(x), evaluating the objective functions of the created points, the x's, and updating the density. Algorithms such as Monte Carlo Markov Chains, Simulated Annealing, Estimation of Density Algorithms, Bayesian optimization and Evolutionary algorithms can fall under this description. Then, taking a model identification perspective, we will argue that the simpler the structure of p(), the least objective function evaluations are necessary to learn it. However, a minimum degree of complexity of p() is needed to optimize coupled functions. We propose to create coupled densities at no extra cost in terms of calls to the objective function by using another definition of the optimization variables, the auxiliary variables. Examples will be given in optimum design of composite structures where the primary optimization variables are fibers orientations and the auxiliary variables are the structural stiffnesses.

Der Vortragende ist Dr. Matthias Hamburg. Er ist Managing Consultant bei der Sogeti Deutschland GmbH.

 

 

Veranstaltungsdetails:

Mittwoch, 11.03.2015, 16:15 Uhr
Emil-Figge-Straße 42, Raum A.E.02
Rodolphe Le Riche (speaker) (CNRS LIMOS and Ecole des Mines de St-Etienne, France)
Alexis Lasseigne, François-Xavier Irisarri (ONERA Composite Systems and Materials, France)

Die  Veranstaltung ist offen für alle. Studenten willkommen.