Integrating Resource Planning with Bayesian Optimization

Graduate Thesis Or Dissertation

Public Deposited

Citeable URL: https://ir.library.oregonstate.edu/concern/graduate_thesis_or_dissertations/5q47rv131

Descriptions

Attribute Name	Values
Creator	Dolatnia, Nima
Abstract	Bayesian optimization (BO) aims to optimize costly-to-evaluate functions by running a limited number of experiments that each evaluates the function at a selected input. Typical BO formulations assume that experiments are selected sequentially, or in fixed batches. Moreover, these experiments can be executed immediately upon request and have the same duration. In the following work, we study two key settings that the current BO framework fails to capture. 1) scenarios where the execution of an experiment requires setup and preparation time, which may vary according to the type of experiment. 2) scenarios where the possibility of parallelism is unbounded and, furthermore, the duration of each experiment can vary. We first define separate BO formulations that model these scenarios. We then propose an algorithm to address each of these problems. Finally, we evaluate the proposed algorithms on a diverse benchmark problems and show that they produce high-quality results compared to a number of natural baselines.
License	All rights reserved
Resource Type	Dissertation
Date Issued	2018-09-07
Degree Level	Doctoral
Degree Name	Doctor of Philosophy (Ph.D.)
Degree Field	Statistics
Degree Grantor	Oregon State University
Commencement Year	2019
Advisor	Emerson, Sarah C. Fern, Alan P.
Academic Affiliation	Statistics
Rights Statement	In Copyright
Publisher	Oregon State University
Peer Reviewed	No
Language	English [eng]
Embargo date range	2018-09-20 to 2019-04-19

Parents:

This work has no parents.

In Collection:

Thumbnail	Title	Date Uploaded	Visibility	Actions
	DolatniaNima2018.pdf	2018-09-19	Public	Download