Methods to compare expensive stochastic optimization algorithms with random restarts

We consider the challenge of numerically comparing optimization algorithms that employ random-restarts under the assumption that only limited test data is available. We develop a bootstrapping technique to estimate the incumbent solution of the optimization problem over time as a stochastic process....

Full description

Saved in:
Bibliographic Details
Published in:Journal of global optimization Vol. 72; no. 4; pp. 781 - 801
Main Authors: Hare, Warren, Loeppky, Jason, Xie, Shangwei
Format: Journal Article
Language:English
Published: New York Springer US 01.12.2018
Springer
Springer Nature B.V
Subjects:
Algorithms
Asymptotic properties
Computer Science
Mathematical optimization
Mathematics
Mathematics and Statistics
Methods
Operations Research/Decision Theory
Optimization
Optimization algorithms
Real Functions
Stochastic processes
Test procedures
Random restarts
Benchmarking
Nonconvex optimization
Stochastic optimization
ISSN:0925-5001, 1573-2916
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:We consider the challenge of numerically comparing optimization algorithms that employ random-restarts under the assumption that only limited test data is available. We develop a bootstrapping technique to estimate the incumbent solution of the optimization problem over time as a stochastic process. The asymptotic properties of the estimator are examined and the approach is validated by an out-of-sample test. Finally, three methods for comparing the performance of different algorithms based on the estimator are proposed and demonstrated with data from a real-world optimization problem.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0925-5001
1573-2916
DOI:10.1007/s10898-018-0673-7