Orthogonal Gaussian process models

Plumlee, Matthew; Joseph, V. Roshan

doi:10.5705/ss.202015.0404

Title: Orthogonal Gaussian process models

Full Record
Other Related Research

Abstract

Gaussian processes models are widely adopted for nonparameteric/semi-parametric modeling. Identifiability issues occur when the mean model contains polynomials with unknown coefficients. Though resulting prediction is unaffected, this leads to poor estimation of the coefficients in the mean model, and thus the estimated mean model loses interpretability. This paper introduces a new Gaussian process model whose stochastic part is orthogonal to the mean part to address this issue. As a result, this paper also discusses applications to multi-fidelity simulations using data examples.

Authors:

Plumlee, Matthew ^[1]; Joseph, V. Roshan ^[2]

Univ. of Michigan, Ann Arbor, MI (United States)
Georgia Inst. of Technology, Atlanta, GA (United States)

Publication Date:: Sun Jan 01 00:00:00 EST 2017

Research Org.:: Georgia Institute of Technology, Atlanta, GA (United States)

Sponsoring Org.:: USDOE Office of Science (SC), Advanced Scientific Computing Research (ASCR)

OSTI Identifier:: 1405183

Report Number(s):: DOE-GT-0010548-8
Journal ID: ISSN 1017-0405; FG02-13ER26159

Grant/Contract Number:: SC0010548

Resource Type:: Accepted Manuscript

Journal Name:: Statistica Sinica

Additional Journal Information:: Journal Volume: 28; Journal ID: ISSN 1017-0405

Publisher:: Institute of Statistical Science, Academia Sinica - International Chinese Statistical Association

Country of Publication:: United States

Language:: English

Subject:: 97 MATHEMATICS AND COMPUTING; Computer experiments; Identifiability; Kriging; Multi-fidelity simulations; Universal kriging

Citation Formats


                    Plumlee, Matthew, and Joseph, V. Roshan. Orthogonal Gaussian process models.  United States: N. p., 2017. 
Web.  doi:10.5705/ss.202015.0404.

Copy to clipboard


                    Plumlee, Matthew, & Joseph, V. Roshan. Orthogonal Gaussian process models.  United States.  https://doi.org/10.5705/ss.202015.0404

Copy to clipboard


                    Plumlee, Matthew, and Joseph, V. Roshan. Sun .  
"Orthogonal Gaussian process models".  United States.  https://doi.org/10.5705/ss.202015.0404.  https://www.osti.gov/servlets/purl/1405183.

Copy to clipboard


                    
@article{osti_1405183,

  title        = {Orthogonal Gaussian process models},

  author       = {Plumlee, Matthew and Joseph, V. Roshan},

  abstractNote = {Gaussian processes models are widely adopted for nonparameteric/semi-parametric modeling. Identifiability issues occur when the mean model contains polynomials with unknown coefficients. Though resulting prediction is unaffected, this leads to poor estimation of the coefficients in the mean model, and thus the estimated mean model loses interpretability. This paper introduces a new Gaussian process model whose stochastic part is orthogonal to the mean part to address this issue. As a result, this paper also discusses applications to multi-fidelity simulations using data examples.},

  doi          = {10.5705/ss.202015.0404},

  journal      = {Statistica Sinica},

  number       = ,

  volume       = 28,

  place        = {United States},

  year         = {Sun Jan 01 00:00:00 EST 2017},

  month        = {Sun Jan 01 00:00:00 EST 2017}

}

Copy to clipboard

Journal Article:

Free Publicly Available Full Text

Accepted Manuscript (DOE)

Publisher's Version of Record

https://doi.org/10.5705/ss.202015.0404

Other availability

Search WorldCat to find libraries that may hold this journal

Citation Metrics:

Cited by: 8 works

Citation information provided by
Web of Science

Save / Share:

Export Metadata

Save to My Library

Similar Records in DOE PAGES and OSTI.GOV collections:

Bayesian learning of orthogonal embeddings for multi-fidelity Gaussian Processes

Journal Article Tsilifis, Panagiotis ; Pandita, Piyush ; Ghosh, Sayan ; ... - Computer Methods in Applied Mechanics and Engineering

Uncertainty propagation in complex engineering systems often poses significant computational challenges related to modeling and quantifying probability distributions of model outputs, as those emerge as the result of various sources of uncertainty that are inherent in the system under investigation. Gaussian Processes regression (GPs) is a robust meta-modeling technique that allows for fast model prediction and exploration of response surfaces. Multi-fidelity variations of GPs further leverage information from cheap and low fidelity model simulations in order to improve their predictive performance on the high fidelity model. In order to cope with the high volume of data required to train GPsmore »« less
https://doi.org/10.1016/j.cma.2021.114147

Full Text Available
Gaussian processes with built-in dimensionality reduction: Applications to high-dimensional uncertainty propagation

Journal Article Tripathy, Rohit ; Bilionis, Ilias ; Gonzalez, Marcial - Journal of Computational Physics

Uncertainty quantification (UQ) tasks, such as model calibration, uncertainty propagation, and optimization under uncertainty, typically require several thousand evaluations of the underlying computer codes. To cope with the cost of simulations, one replaces the real response surface with a cheap surrogate based, e.g., on polynomial chaos expansions, neural networks, support vector machines, or Gaussian processes (GP). However, the number of simulations required to learn a generic multivariate response grows exponentially as the input dimension increases. This curse of dimensionality can only be addressed, if the response exhibits some special structure that can be discovered and exploited. A wide range ofmore »« less
https://doi.org/10.1016/J.JCP.2016.05.039
PERIOD ESTIMATION FOR SPARSELY SAMPLED QUASI-PERIODIC LIGHT CURVES APPLIED TO MIRAS

Journal Article He, Shiyuan ; Huang, Jianhua Z. ; Long, James ; ... - Astronomical Journal (Online)

We develop a nonlinear semi-parametric Gaussian process model to estimate periods of Miras with sparsely sampled light curves. The model uses a sinusoidal basis for the periodic variation and a Gaussian process for the stochastic changes. We use maximum likelihood to estimate the period and the parameters of the Gaussian process, while integrating out the effects of other nuisance parameters in the model with respect to a suitable prior distribution obtained from earlier studies. Since the likelihood is highly multimodal for period, we implement a hybrid method that applies the quasi-Newton algorithm for Gaussian process parameters and search the period/frequencymore »« less
https://doi.org/10.3847/0004-6256/152/6/164
Nested polynomial trends for the improvement of Gaussian process-based predictors

Journal Article Perrin, G., E-mail: guillaume.perrin2@cea.fr ; Soize, C. ; Marque-Pucheu, S. ; ... - Journal of Computational Physics

The role of simulation keeps increasing for the sensitivity analysis and the uncertainty quantification of complex systems. Such numerical procedures are generally based on the processing of a huge amount of code evaluations. When the computational cost associated with one particular evaluation of the code is high, such direct approaches based on the computer code only, are not affordable. Surrogate models have therefore to be introduced to interpolate the information given by a fixed set of code evaluations to the whole input space. When confronted to deterministic mappings, the Gaussian process regression (GPR), or kriging, presents a good compromise betweenmore »« less
https://doi.org/10.1016/J.JCP.2017.05.051
An Adaptive Bayesian Parameter Estimation of a Synchronous Generator Under Gross Errors

Journal Article Xu, Yijun ; Mili, Lamine ; Korkali, Mert ; ... - IEEE Transactions on Industrial Informatics

Polynomial-chaos-expansion-based surrogate models have recently been advocated in the literature for power system dynamic parameter estimation. Regarding the estimation of the uncertain generator parameters, a Bayesian inference framework has been proposed based on a polynomial-based reduced-order representation of the synchronous machines using assumed parameter values. Then, the non-Gaussian posterior probability distribution functions (pdfs) of these parameters are recovered through the stochastic sampling approach efficiently. However, facing very large parameter errors, the reliability of the surrogate model decreases, yielding biased estimation results. To overcome this problem, here we develop hierarchical Bayesian inference framework that processes the measurements provided by phasor measurementmore »« less
https://doi.org/10.1109/tii.2019.2950238

Full Text Available

Similar Records