G-STAR: A new kriging-based trust region method for global optimization

Giulia Pedrielli, Szu Hui Ng

Research output: Chapter in Book/Report/Conference proceedingConference contribution

4 Citations (Scopus)

Abstract

Trust region methods are an efficient technique to identify good solutions when the sampling effort needs to be controlled due to the cost of running simulation. Meta-model based applications of trust region methods have already been proposed and their convergence has been characterized. Nevertheless, these approaches keep the strongly local characteristic of the original trust region method. This is not desirable in that information generated at local level are "lost" as the search progresses. A first consequence is that the search technique cannot guarantee global convergence. We propose a global version of the trust region method, the Global Stochastic Trust Augmented Region (G-STAR). The trust region is used to focus the simulation effort and balance between exploration and exploitation. Such an algorithm focuses the sampling effort in trust regions sequentially generated by adopting an extended Expected Improvement criterion. This paper presents the algorithm and the preliminary numerical results.

Original languageEnglish (US)
Title of host publication2016 Winter Simulation Conference: Simulating Complex Service Systems, WSC 2016
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages803-814
Number of pages12
ISBN (Electronic)9781509044863
DOIs
StatePublished - Jan 17 2017
Externally publishedYes
Event2016 Winter Simulation Conference, WSC 2016 - Arlington, United States
Duration: Dec 11 2016Dec 14 2016

Other

Other2016 Winter Simulation Conference, WSC 2016
CountryUnited States
CityArlington
Period12/11/1612/14/16

Fingerprint

Trust Region Method
Kriging
Global optimization
Global Optimization
Sampling
Trust Region
Metamodel
Global Convergence
Exploitation
Costs
Simulation
Model-based
Numerical Results

ASJC Scopus subject areas

  • Software
  • Modeling and Simulation
  • Computer Science Applications

Cite this

Pedrielli, G., & Ng, S. H. (2017). G-STAR: A new kriging-based trust region method for global optimization. In 2016 Winter Simulation Conference: Simulating Complex Service Systems, WSC 2016 (pp. 803-814). [7822143] Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/WSC.2016.7822143

G-STAR : A new kriging-based trust region method for global optimization. / Pedrielli, Giulia; Ng, Szu Hui.

2016 Winter Simulation Conference: Simulating Complex Service Systems, WSC 2016. Institute of Electrical and Electronics Engineers Inc., 2017. p. 803-814 7822143.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Pedrielli, G & Ng, SH 2017, G-STAR: A new kriging-based trust region method for global optimization. in 2016 Winter Simulation Conference: Simulating Complex Service Systems, WSC 2016., 7822143, Institute of Electrical and Electronics Engineers Inc., pp. 803-814, 2016 Winter Simulation Conference, WSC 2016, Arlington, United States, 12/11/16. https://doi.org/10.1109/WSC.2016.7822143
Pedrielli G, Ng SH. G-STAR: A new kriging-based trust region method for global optimization. In 2016 Winter Simulation Conference: Simulating Complex Service Systems, WSC 2016. Institute of Electrical and Electronics Engineers Inc. 2017. p. 803-814. 7822143 https://doi.org/10.1109/WSC.2016.7822143
Pedrielli, Giulia ; Ng, Szu Hui. / G-STAR : A new kriging-based trust region method for global optimization. 2016 Winter Simulation Conference: Simulating Complex Service Systems, WSC 2016. Institute of Electrical and Electronics Engineers Inc., 2017. pp. 803-814
@inproceedings{1d3852edcb234bf9acf3b1573ecf2cf7,
title = "G-STAR: A new kriging-based trust region method for global optimization",
abstract = "Trust region methods are an efficient technique to identify good solutions when the sampling effort needs to be controlled due to the cost of running simulation. Meta-model based applications of trust region methods have already been proposed and their convergence has been characterized. Nevertheless, these approaches keep the strongly local characteristic of the original trust region method. This is not desirable in that information generated at local level are {"}lost{"} as the search progresses. A first consequence is that the search technique cannot guarantee global convergence. We propose a global version of the trust region method, the Global Stochastic Trust Augmented Region (G-STAR). The trust region is used to focus the simulation effort and balance between exploration and exploitation. Such an algorithm focuses the sampling effort in trust regions sequentially generated by adopting an extended Expected Improvement criterion. This paper presents the algorithm and the preliminary numerical results.",
author = "Giulia Pedrielli and Ng, {Szu Hui}",
year = "2017",
month = "1",
day = "17",
doi = "10.1109/WSC.2016.7822143",
language = "English (US)",
pages = "803--814",
booktitle = "2016 Winter Simulation Conference: Simulating Complex Service Systems, WSC 2016",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
address = "United States",

}

TY - GEN

T1 - G-STAR

T2 - A new kriging-based trust region method for global optimization

AU - Pedrielli, Giulia

AU - Ng, Szu Hui

PY - 2017/1/17

Y1 - 2017/1/17

N2 - Trust region methods are an efficient technique to identify good solutions when the sampling effort needs to be controlled due to the cost of running simulation. Meta-model based applications of trust region methods have already been proposed and their convergence has been characterized. Nevertheless, these approaches keep the strongly local characteristic of the original trust region method. This is not desirable in that information generated at local level are "lost" as the search progresses. A first consequence is that the search technique cannot guarantee global convergence. We propose a global version of the trust region method, the Global Stochastic Trust Augmented Region (G-STAR). The trust region is used to focus the simulation effort and balance between exploration and exploitation. Such an algorithm focuses the sampling effort in trust regions sequentially generated by adopting an extended Expected Improvement criterion. This paper presents the algorithm and the preliminary numerical results.

AB - Trust region methods are an efficient technique to identify good solutions when the sampling effort needs to be controlled due to the cost of running simulation. Meta-model based applications of trust region methods have already been proposed and their convergence has been characterized. Nevertheless, these approaches keep the strongly local characteristic of the original trust region method. This is not desirable in that information generated at local level are "lost" as the search progresses. A first consequence is that the search technique cannot guarantee global convergence. We propose a global version of the trust region method, the Global Stochastic Trust Augmented Region (G-STAR). The trust region is used to focus the simulation effort and balance between exploration and exploitation. Such an algorithm focuses the sampling effort in trust regions sequentially generated by adopting an extended Expected Improvement criterion. This paper presents the algorithm and the preliminary numerical results.

UR - http://www.scopus.com/inward/record.url?scp=85014142840&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85014142840&partnerID=8YFLogxK

U2 - 10.1109/WSC.2016.7822143

DO - 10.1109/WSC.2016.7822143

M3 - Conference contribution

AN - SCOPUS:85014142840

SP - 803

EP - 814

BT - 2016 Winter Simulation Conference: Simulating Complex Service Systems, WSC 2016

PB - Institute of Electrical and Electronics Engineers Inc.

ER -