Convex version of multivariate adaptive regression splines for optimization

Thomas D. Shih, Victoria C P Chen, Seoung Bum Kim

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

1 Citation (Scopus)

Abstract

Multivariate Adpative Regression Splines (MARS) provide a flexible statistical modeling method that employs forward and backward search algorithms to identify the combination of basis functions that best fits the data. In optimization, MARS has been used successfully to estimate the value function in stochastic dynamic programming, and MARS could be potentially useful in many real world optimization problems where objective (or other) functions need to be estimated from data, such as in simulation optimization. Many optimization methods depend on convexity, but a nonconvex MARS approximation is inherently possible because interaction terms are products of univariate terms. In this paper, we propose a convex version of MARS. In order to ensure MARS convexity, two major modifications are made: (1) coefficients are constrained such that pairs of basis functions are guaranteed to jointly form convex functions; (2) The form of interaction terms is appropriately changed. Finally, MARS convexity can be achieved by the fact that the sum of convex functions is convex.

Original languageEnglish
Title of host publication2006 IIE Annual Conference and Exhibition
Publication statusPublished - 2006 Dec 1
Externally publishedYes
Event2006 IIE Annual Conference and Exposition - Orlando, FL, United States
Duration: 2006 May 202006 May 24

Other

Other2006 IIE Annual Conference and Exposition
CountryUnited States
CityOrlando, FL
Period06/5/2006/5/24

Fingerprint

Splines
Dynamic programming

Keywords

  • Convexity
  • Function approximation
  • Optimization
  • Regression splines

ASJC Scopus subject areas

  • Industrial and Manufacturing Engineering

Cite this

Shih, T. D., Chen, V. C. P., & Kim, S. B. (2006). Convex version of multivariate adaptive regression splines for optimization. In 2006 IIE Annual Conference and Exhibition

Convex version of multivariate adaptive regression splines for optimization. / Shih, Thomas D.; Chen, Victoria C P; Kim, Seoung Bum.

2006 IIE Annual Conference and Exhibition. 2006.

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

Shih, TD, Chen, VCP & Kim, SB 2006, Convex version of multivariate adaptive regression splines for optimization. in 2006 IIE Annual Conference and Exhibition. 2006 IIE Annual Conference and Exposition, Orlando, FL, United States, 06/5/20.
Shih TD, Chen VCP, Kim SB. Convex version of multivariate adaptive regression splines for optimization. In 2006 IIE Annual Conference and Exhibition. 2006
Shih, Thomas D. ; Chen, Victoria C P ; Kim, Seoung Bum. / Convex version of multivariate adaptive regression splines for optimization. 2006 IIE Annual Conference and Exhibition. 2006.
@inproceedings{b840ff1eee0d46ccad1a07489d48551d,
title = "Convex version of multivariate adaptive regression splines for optimization",
abstract = "Multivariate Adpative Regression Splines (MARS) provide a flexible statistical modeling method that employs forward and backward search algorithms to identify the combination of basis functions that best fits the data. In optimization, MARS has been used successfully to estimate the value function in stochastic dynamic programming, and MARS could be potentially useful in many real world optimization problems where objective (or other) functions need to be estimated from data, such as in simulation optimization. Many optimization methods depend on convexity, but a nonconvex MARS approximation is inherently possible because interaction terms are products of univariate terms. In this paper, we propose a convex version of MARS. In order to ensure MARS convexity, two major modifications are made: (1) coefficients are constrained such that pairs of basis functions are guaranteed to jointly form convex functions; (2) The form of interaction terms is appropriately changed. Finally, MARS convexity can be achieved by the fact that the sum of convex functions is convex.",
keywords = "Convexity, Function approximation, Optimization, Regression splines",
author = "Shih, {Thomas D.} and Chen, {Victoria C P} and Kim, {Seoung Bum}",
year = "2006",
month = "12",
day = "1",
language = "English",
booktitle = "2006 IIE Annual Conference and Exhibition",

}

TY - GEN

T1 - Convex version of multivariate adaptive regression splines for optimization

AU - Shih, Thomas D.

AU - Chen, Victoria C P

AU - Kim, Seoung Bum

PY - 2006/12/1

Y1 - 2006/12/1

N2 - Multivariate Adpative Regression Splines (MARS) provide a flexible statistical modeling method that employs forward and backward search algorithms to identify the combination of basis functions that best fits the data. In optimization, MARS has been used successfully to estimate the value function in stochastic dynamic programming, and MARS could be potentially useful in many real world optimization problems where objective (or other) functions need to be estimated from data, such as in simulation optimization. Many optimization methods depend on convexity, but a nonconvex MARS approximation is inherently possible because interaction terms are products of univariate terms. In this paper, we propose a convex version of MARS. In order to ensure MARS convexity, two major modifications are made: (1) coefficients are constrained such that pairs of basis functions are guaranteed to jointly form convex functions; (2) The form of interaction terms is appropriately changed. Finally, MARS convexity can be achieved by the fact that the sum of convex functions is convex.

AB - Multivariate Adpative Regression Splines (MARS) provide a flexible statistical modeling method that employs forward and backward search algorithms to identify the combination of basis functions that best fits the data. In optimization, MARS has been used successfully to estimate the value function in stochastic dynamic programming, and MARS could be potentially useful in many real world optimization problems where objective (or other) functions need to be estimated from data, such as in simulation optimization. Many optimization methods depend on convexity, but a nonconvex MARS approximation is inherently possible because interaction terms are products of univariate terms. In this paper, we propose a convex version of MARS. In order to ensure MARS convexity, two major modifications are made: (1) coefficients are constrained such that pairs of basis functions are guaranteed to jointly form convex functions; (2) The form of interaction terms is appropriately changed. Finally, MARS convexity can be achieved by the fact that the sum of convex functions is convex.

KW - Convexity

KW - Function approximation

KW - Optimization

KW - Regression splines

UR - http://www.scopus.com/inward/record.url?scp=36448991219&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=36448991219&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:36448991219

BT - 2006 IIE Annual Conference and Exhibition

ER -