Algorithms for computationally expensive optimization problems

Optimization problems arise in many application areas such as engineering and management. Cleaning up contaminated groundwater at minimal costs, maximizing the produced energy from renewable sources, or minimizing the costs of managing the agricultural land use of watersheds to assure the quality of drinking water are just few examples. These problems have in common that for computing a single objective function value a black-box simulation code has to be run that may require a considerable amount of computation time (from several minutes to several hours or even days).

The goal is thus to find a good approximation of the optimum of the objective function within very few function evaluations in order to keep the total computation time as low as possible. Due to the black-box characteristic of the objective function, its structure can, however, not be exploited in an optimization scheme.

In her doctoral dissertation, MSc Juliane Muller describes optimization algorithms that are able to tackle such problems. So-called surrogate models are used in place of the computationally expensive objective function during the optimization in order to reduce the number of costly function evaluations. Algorithms for continuous, integer, and mixed-integer optimization problems have been developed.

Public defence of a doctoral dissertation on Wednesday, 9 November

The doctoral dissertation of MSc Juliane Muller in the field of Applied Mathematics "Surrogate Model Algorithms for Computationally Expensive Black-Box Global Optimization Problems" will be publicly examined at the Faculty of Science and Environmental Engineering of Tampere University of Technology (TUT) in room S4 in the Sähkötalo building (Korkeakoulunkatu 3, Tampere) on Wednesday, November 28, 2012 at 12:00.

The opponent will be Jorge More (Argonne National Laboratory). Robert Piche from TUT’s Mathematics department will act as Chairman.

Juliane Muller lives in Tampere and works as a researcher at TUT.

Further Information: Juliane Muller, juliane.mueller@tut.fi