Skip navigation
Please use this identifier to cite or link to this item: http://arks.princeton.edu/ark:/88435/dsp014f16c526k
Title: Scaling High-Dimensional Heuristic Search Techniques
Authors: Zaslavsky, Maxim
Advisors: Engelhardt, Barbara
Contributors: Pillow, Jonathan
Department: Computer Science
Class Year: 2016
Abstract: The optimization problem of hyperparameter search is an important problem in tuning machine learning models. Crucially, even heuristic search techniques, like Bayesian optimization, fail in parameter spaces of large dimensionality. We discuss improvements to the complexity of Bayesian optimization and strategies to accelerate high-dimensional search. Tested on one particular high-dimensional function, our method brings the search 2.05% closer to the true optimum per extra second of computation time devoted to learning from previous observations of the black-box objective.
Extent: 19 pages
URI: http://arks.princeton.edu/ark:/88435/dsp014f16c526k
Type of Material: Princeton University Senior Theses
Language: en_US
Appears in Collections:Computer Science, 1988-2016

Files in This Item:
File SizeFormat 
Zaslavsky_Maxim_2016_Thesis.pdf1.52 MBAdobe PDF    Request a copy


Items in Dataspace are protected by copyright, with all rights reserved, unless otherwise indicated.