Generalized method of moments approach for hyperparameter estimation for Gaussian Markov random fields
Open Access
Author:
Dong, Yi
Graduate Program:
Industrial Engineering
Degree:
Master of Science
Document Type:
Master Thesis
Date of Defense:
March 30, 2018
Committee Members:
Eunhye Song, Thesis Advisor/Co-Advisor
Keywords:
Generalized method of moments Gaussian Markov Random Field Discrete Optimization via Simulation Maximum Likelihood Estimation
Abstract:
Gaussian Markov random field (GMRF) has shown good performance as a metamodel to design an adaptive random search algorithm for a Discrete Optimization via Simulation (DOvS) problem. A popular method to estimate the hyperparameters of GMRF is maximum likelihood estimation (MLE). MLE has its limitations in computation time and numerical precision in estimating the hyperparameter of a GMRF. These limitations motivated us to develop new generalized method of moment (GMM) estimators of the hyperparameters of GMRF. A numerical experiment is presented in this thesis which shows that our proposed 1-step sequential GMM (SGMM) approach can significantly reduce the computation time and reduce the estimation error compared to MLE under the same simulation effort.
