Improving the Model Guided Sampling Optimization by Model Search and Slice Sampling

Authors: Lukáš Bajer, Martin Holeňa, V. Charypar
Publication year: 2013
Publication type: Konferenční příspěvek (mezinárodní konf.)
Periodical: ITAT 2013: Information Technologies - Applications and Theory Workshops, Posters, and Tutorials
Projects: Learning of functional relationships from high-dimensional data
Constructing Advanced Comprehensible Classifiers
Bibliographic citation: BAJER, Lukáš, HOLEŇA, Martin, CHARYPAR, V. Improving the Model Guided Sampling Optimization by Model Search and Slice Sampling. In: VINAŘ, T., HOLEŇA, M., LEXA, M., PEŠKA, L., VOJTÁŠ, P., eds. ITAT 2013: Information Technologies - Applications and Theory Workshops, Posters, and Tutorials. North Charleston: CreateSpace Independent Publishing Platform, 2013, S. 86-91. ISBN 978-1-4909-5208-6.
ID: 396832

Abstract

Model Guided Sampling Optimization (MGSO) was recently proposed as an alternative for Jones’ Krigingbased EGO algorithm for optimization of expensive blackbox functions. Instead of maximizing a chosen criterion (e.g., expected improvement), MGSO samples probability of improvement of the Gaussian process model forming multiple candidates – a whole population of suggested solutions. This paper further develops this algorithm using slice sampling method and continuous local optimization of the Gaussian process model.