A constrained semi-parametric gaussian process using bayesian-entropy regression

Yuhao Wang, Yi Gao, Yongming Liu, Sayan Ghosh, Liping Wang

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

Abstract

A Gaussian Process makes prediction based on the existing observed data. But in many cases, information is not limited to observations. Extra information, such as physical constraints and empirical knowledge, exists in many engineering problems. This paper presents a Bayesian-Entropy method to encode constraints into a Semiparametric Gaussian Process. The Bayesian-Entropy method can encode various types of constraints by adding an additional term to the Bayesian equation. The Bayesian-Entropy regression method can incorporate values and derivative information into the classical Bayesian regression as constraint. By adjusting the mean function in Semiparametric Gaussian Process according to the Bayesian-Entropy regression principle, extra information, such as the expected value and/or the derivative at a specific point, can be encoded into the regression function. Comparing with the traditional method, the constrained Semiparametric Gaussian Process benefits from the available extra information and can make better prediction outside the range of training data.

Original languageEnglish (US)
Title of host publicationAIAA Scitech 2021 Forum
PublisherAmerican Institute of Aeronautics and Astronautics Inc, AIAA
Pages1-9
Number of pages9
ISBN (Print)9781624106095
StatePublished - 2021
EventAIAA Science and Technology Forum and Exposition, AIAA SciTech Forum 2021 - Virtual, Online
Duration: Jan 11 2021Jan 15 2021

Publication series

NameAIAA Scitech 2021 Forum

Conference

ConferenceAIAA Science and Technology Forum and Exposition, AIAA SciTech Forum 2021
CityVirtual, Online
Period1/11/211/15/21

ASJC Scopus subject areas

  • Aerospace Engineering

Fingerprint

Dive into the research topics of 'A constrained semi-parametric gaussian process using bayesian-entropy regression'. Together they form a unique fingerprint.

Cite this