DMCA
Scotland, UK.
Cached
Download Links
BibTeX
@MISC{Solak_scotland,uk.,
author = {E. Solak and D. J. Leith and R. Murray-smith and W. E. Leithead and C. E. Rasmussen},
title = {Scotland, UK.},
year = {}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
Gaussian processes provide an approach to nonparametric modelling which allows a straightforward combination of function and derivative observations in an empirical model. This is of particular importance in identification of nonlinear dynamic systems from experimental data. 1) It allows us to combine derivative information, and associated uncertainty with normal function observations into the learning and inference process. This derivative information can be in the form of priors specified by an expert or identified from perturbation data close to equilibrium. 2) It allows a seamless fusion of multiple local linear models in a consistent manner, inferring consistent models and ensuring that integrability constraints are met. 3) It improves dramatically the computational efficiency of Gaussian process models for dynamic system identification, by summarising large quantities of near-equilibrium data by a handful of linearisations, reducing the training set size – traditionally a problem for Gaussian process models. 1
Keyphrases
derivative information gaussian process model particular importance computational efficiency consistent manner empirical model gaussian process seamless fusion experimental data straightforward combination integrability constraint dynamic system identification derivative observation perturbation data large quantity multiple local linear model near-equilibrium data consistent model nonlinear dynamic system normal function observation inference process
