J. C. García García, R. García Ródenas

In computational management and engineering issues, there are tasks which involve the optimization of functions defined as black boxes. These functions are computationally expensive, have noise and do not feature explicit gradients. One of the most used techniques is the response surface method (RSM). In this case, a surrogate model related to the expensive objective function is obtained. This model will be used to define a sampling strategy in order to add new points of the feasible region. The surrogate model is updated and the procedure is iterated. In this work, we propose a parallel strategy of the previous scheme based on qualSolve algorithm [Jakobsson et al. (2012)]. In addition, we carry out a numerical comparative with the classic EGO (Efficient Global Optimization) and its parallel versions with multi-point designs (Constant Liar and Kriging Believer) in order to show that the proposed algorithm is highly promising.

Keywords: Parallel, surrogate models, black box function, simulation and optimization

Scheduled

TE1 Computational Management Methods
May 31, 2016  4:45 PM
Salón de actos

The effect of preference similarity on the formation of clusters and the cohesion of social networks

D. Di Caprio, F. J. Santos Arteaga, M. Tavana

Reduced bases function approach to option valuation

V. Gatón Bustillo, J. de Frutos