6th Workshop on Computational Optimization, Modelling & Simulation (COMS) Session 1

Time and Date: 10:35 - 12:15 on 1st June 2015

Room: V201

Chair: Leifur Leifsson

261 Surrogate-Based Airfoil Design with Space Mapping and Adjoint Sensitivity [abstract]
Abstract: This paper presents a space mapping algorithm for airfoil shape optimization enhanced with adjoint sensitivities. The surrogate-based algorithm utilizes low-cost derivative information obtained through adjoint sensitivities to improve the space mapping matching between a high-fidelity airfoil model, evaluated through expensive CFD simulations, and its fast surrogate. Here, the airfoil surrogate model is constructed though low-fidelity CFD simulations. As a result, the design process can be performed at a low computational cost in terms of the number of high-fidelity CFD simulations. The adjoint sensitivities are also exploited to speed up the surrogate optimization process. Our method is applied to a constrained drag minimization problem in two-dimensional inviscid transonic flow. The problem is solved for several low-fidelity model termination criteria. The results show that when compared with direct gradient-based optimization with adjoint sensitivities, the proposed approach requires 49-78% less computational cost while still obtaining a comparable airfoil design.
Yonatan Tesfahunegn, Slawomir Koziel, Leifur Leifsson, Adrian Bekasiewicz
317 How to Speed up Optimization? Opposite-Center Learning and Its Application to Differential Evolution [abstract]
Abstract: This paper introduces a new sampling technique called Opposite-Center Learning (OCL) intended for convergence speedup of meta-heuristic optimization algorithms. It comprises an extension of Opposition-Based Learning (OBL), a simple scheme that manages to boost numerous optimization methods by considering the opposite points of candidate solutions. In contrast to OBL, OCL has a theoretical foundation – the opposite center point is defined as the optimal choice in pair-wise sampling of the search space given a random starting point. A concise analytical background is provided. Computationally the opposite center point is approximated by a lightweight Monte Carlo scheme for arbitrary dimension. Empirical results up to dimension 20 confirm that OCL outperforms OBL and random sampling: the points generated by OCL have shorter expected distances to a uniformly distributed global optimum. To further test its practical performance, OCL is applied to differential evolution (DE). This novel scheme for continuous optimization named Opposite-Center DE (OCDE) employs OCL for population initialization and generation jumping. Numerical experiments on a set of benchmark functions for dimensions 10 and 30 reveal that OCDE on average improves the convergence rates by 38% and 27% compared to the original DE and the Opposition-based DE (ODE), respectively, while remaining fully robust. Most promising are the observations that the accelerations shown by OCDE and OCL increase with problem dimensionality.
H. Xu, C.D. Erdbrink, V.V. Krzhizhanovskaya
281 Visualizing and Improving the Robustness of Phase Retrieval Algorithms [abstract]
Abstract: Coherent x-ray diffractive imaging is a novel imaging technique that utilizes phase retrieval and nonlinear optimization methods to image matter at nanometer scales. We explore how the convergence properties of a popular phase retrieval algorithm, Fienup’s HIO, behave by introducing a reduced dimensionality problem allowing us to visualize convergence to local minima and the globally optimal solution. We then introduce generalizations of HIO that improve upon the original algorithm’s ability to converge to the globally optimal solution.
Ashish Tripathi, Sven Leyffer, Todd Munson, Stefan Wild
257 Fast Optimization of Integrated Photonic Components Using Response Correction and Local Approximation Surrogates [abstract]
Abstract: A methodology for a rapid design optimization of integrated photonic couplers is presented. The proposed technique exploits variable-fidelity electromagnetic (EM) simulation models, additive response correction for accommodating the discrepancies between the EM models of various fidelities, and local response surface approximations for a fine tuning of the final design. A specific example of a 1,555 nm coupler is considered with an optimum design obtained at a computational cost corresponding to about 24 high-fidelity EM simulations of the structure.
Adrian Bekasiewicz, Slawomir Koziel, Leifur Leifsson
197 Model Selection for Discriminative Restricted Boltzmann Machines Through Meta-heuristic Techniques [abstract]
Abstract: Discriminative learning of Restricted Boltzmann Machines has been recently introduced as an alternative to provide a self-contained approach for both unsupervised feature learning and classification purposes. However, one of the main problems faced by researchers interested in such approach concerns with a proper selection of its parameters, which play an important role in its final performance. In this paper, we introduced some meta-heuristic techniques for this purpose, as well as we showed they can be more accurate than a random search, that is commonly used by some works.
Joao Paulo Papa, Gustavo Rosa, Aparecido Marana, Walter Scheirer and David Cox