Estimating parameters in stochastic systems: a variational Bayesian approach

Michail Vrettas, Dan Cornford, Manfred Opper

Research output: Contribution to journalArticle

Abstract

This work is concerned with approximate inference in dynamical systems, from a variational Bayesian perspective. When modelling real world dynamical systems, stochastic differential equations appear as a natural choice, mainly because of their ability to model the noise of the system by adding a variation of some stochastic process to the deterministic dynamics. Hence, inference in such processes has drawn much attention. Here a new extended framework is derived that is based on a local polynomial approximation of a recently proposed variational Bayesian algorithm. The paper begins by showing that the new extension of this variational algorithm can be used for state estimation (smoothing) and converges to the original algorithm. However, the main focus is on estimating the (hyper-) parameters of these systems (i.e. drift parameters and diffusion coefficients). The new approach is validated on a range of different systems which vary in dimensionality and non-linearity. These are the Ornstein–Uhlenbeck process, the exact likelihood of which can be computed analytically, the univariate and highly non-linear, stochastic double well and the multivariate chaotic stochastic Lorenz ’63 (3D model). As a special case the algorithm is also applied to the 40 dimensional stochastic Lorenz ’96 system. In our investigation we compare this new approach with a variety of other well known methods, such as the hybrid Monte Carlo, dual unscented Kalman filter, full weak-constraint 4D-Var algorithm and analyse empirically their asymptotic behaviour as a function of observation density or length of time window increases. In particular we show that we are able to estimate parameters in both the drift (deterministic) and the diffusion (stochastic) part of the model evolution equations using our new methods.
Original languageEnglish
Pages (from-to)1877–1900
Number of pages24
JournalPhysica D
Volume240
Issue number23
Early online date18 Sep 2011
DOIs
Publication statusPublished - 15 Nov 2011

Fingerprint

Stochastic systems
Bayesian Approach
Stochastic Systems
estimating
inference
dynamical systems
Dynamical systems
Dynamical system
Hybrid Monte Carlo
state estimation
Polynomial approximation
Local Polynomial
Hyperparameters
Local Approximation
Lorenz System
Time Windows
stochastic processes
Kalman filters
Polynomial Approximation
State Estimation

Bibliographical note

NOTICE: this is the author’s version of a work that was accepted for publication in Physica D: Nonlinear Phenomena. Changes resulting from the publishing process, such as peer review, editing, corrections, structural formatting, and other quality control mechanisms may not be reflected in this document. Changes may have been made to this work since it was submitted for publication. A definitive version was subsequently published in Vrettas, M, Cornford, D & Opper, M, 'Estimating parameters in stochastic systems: a variational Bayesian approach', PHYSICA D-NONLINEAR PHENOMENA, vol 240, no. 23, pp. 1877–1900., (2011) DOI http://dx.doi.org/10.1016/j.physd.2011.08.013

Keywords

  • Bayesian inference
  • variational techniques
  • dynamical systems
  • stochastic differential equations
  • parameter estimation

Cite this

Vrettas, Michail ; Cornford, Dan ; Opper, Manfred. / Estimating parameters in stochastic systems : a variational Bayesian approach. In: Physica D. 2011 ; Vol. 240, No. 23. pp. 1877–1900.
@article{24d8a595b4584a26965af06e4ab56f42,
title = "Estimating parameters in stochastic systems: a variational Bayesian approach",
abstract = "This work is concerned with approximate inference in dynamical systems, from a variational Bayesian perspective. When modelling real world dynamical systems, stochastic differential equations appear as a natural choice, mainly because of their ability to model the noise of the system by adding a variation of some stochastic process to the deterministic dynamics. Hence, inference in such processes has drawn much attention. Here a new extended framework is derived that is based on a local polynomial approximation of a recently proposed variational Bayesian algorithm. The paper begins by showing that the new extension of this variational algorithm can be used for state estimation (smoothing) and converges to the original algorithm. However, the main focus is on estimating the (hyper-) parameters of these systems (i.e. drift parameters and diffusion coefficients). The new approach is validated on a range of different systems which vary in dimensionality and non-linearity. These are the Ornstein–Uhlenbeck process, the exact likelihood of which can be computed analytically, the univariate and highly non-linear, stochastic double well and the multivariate chaotic stochastic Lorenz ’63 (3D model). As a special case the algorithm is also applied to the 40 dimensional stochastic Lorenz ’96 system. In our investigation we compare this new approach with a variety of other well known methods, such as the hybrid Monte Carlo, dual unscented Kalman filter, full weak-constraint 4D-Var algorithm and analyse empirically their asymptotic behaviour as a function of observation density or length of time window increases. In particular we show that we are able to estimate parameters in both the drift (deterministic) and the diffusion (stochastic) part of the model evolution equations using our new methods.",
keywords = "Bayesian inference, variational techniques, dynamical systems, stochastic differential equations, parameter estimation",
author = "Michail Vrettas and Dan Cornford and Manfred Opper",
note = "NOTICE: this is the author’s version of a work that was accepted for publication in Physica D: Nonlinear Phenomena. Changes resulting from the publishing process, such as peer review, editing, corrections, structural formatting, and other quality control mechanisms may not be reflected in this document. Changes may have been made to this work since it was submitted for publication. A definitive version was subsequently published in Vrettas, M, Cornford, D & Opper, M, 'Estimating parameters in stochastic systems: a variational Bayesian approach', PHYSICA D-NONLINEAR PHENOMENA, vol 240, no. 23, pp. 1877–1900., (2011) DOI http://dx.doi.org/10.1016/j.physd.2011.08.013",
year = "2011",
month = "11",
day = "15",
doi = "10.1016/j.physd.2011.08.013",
language = "English",
volume = "240",
pages = "1877–1900",
journal = "Physica D",
issn = "0167-2789",
publisher = "Elsevier",
number = "23",

}

Estimating parameters in stochastic systems : a variational Bayesian approach. / Vrettas, Michail; Cornford, Dan; Opper, Manfred.

In: Physica D, Vol. 240, No. 23, 15.11.2011, p. 1877–1900.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Estimating parameters in stochastic systems

T2 - a variational Bayesian approach

AU - Vrettas, Michail

AU - Cornford, Dan

AU - Opper, Manfred

N1 - NOTICE: this is the author’s version of a work that was accepted for publication in Physica D: Nonlinear Phenomena. Changes resulting from the publishing process, such as peer review, editing, corrections, structural formatting, and other quality control mechanisms may not be reflected in this document. Changes may have been made to this work since it was submitted for publication. A definitive version was subsequently published in Vrettas, M, Cornford, D & Opper, M, 'Estimating parameters in stochastic systems: a variational Bayesian approach', PHYSICA D-NONLINEAR PHENOMENA, vol 240, no. 23, pp. 1877–1900., (2011) DOI http://dx.doi.org/10.1016/j.physd.2011.08.013

PY - 2011/11/15

Y1 - 2011/11/15

N2 - This work is concerned with approximate inference in dynamical systems, from a variational Bayesian perspective. When modelling real world dynamical systems, stochastic differential equations appear as a natural choice, mainly because of their ability to model the noise of the system by adding a variation of some stochastic process to the deterministic dynamics. Hence, inference in such processes has drawn much attention. Here a new extended framework is derived that is based on a local polynomial approximation of a recently proposed variational Bayesian algorithm. The paper begins by showing that the new extension of this variational algorithm can be used for state estimation (smoothing) and converges to the original algorithm. However, the main focus is on estimating the (hyper-) parameters of these systems (i.e. drift parameters and diffusion coefficients). The new approach is validated on a range of different systems which vary in dimensionality and non-linearity. These are the Ornstein–Uhlenbeck process, the exact likelihood of which can be computed analytically, the univariate and highly non-linear, stochastic double well and the multivariate chaotic stochastic Lorenz ’63 (3D model). As a special case the algorithm is also applied to the 40 dimensional stochastic Lorenz ’96 system. In our investigation we compare this new approach with a variety of other well known methods, such as the hybrid Monte Carlo, dual unscented Kalman filter, full weak-constraint 4D-Var algorithm and analyse empirically their asymptotic behaviour as a function of observation density or length of time window increases. In particular we show that we are able to estimate parameters in both the drift (deterministic) and the diffusion (stochastic) part of the model evolution equations using our new methods.

AB - This work is concerned with approximate inference in dynamical systems, from a variational Bayesian perspective. When modelling real world dynamical systems, stochastic differential equations appear as a natural choice, mainly because of their ability to model the noise of the system by adding a variation of some stochastic process to the deterministic dynamics. Hence, inference in such processes has drawn much attention. Here a new extended framework is derived that is based on a local polynomial approximation of a recently proposed variational Bayesian algorithm. The paper begins by showing that the new extension of this variational algorithm can be used for state estimation (smoothing) and converges to the original algorithm. However, the main focus is on estimating the (hyper-) parameters of these systems (i.e. drift parameters and diffusion coefficients). The new approach is validated on a range of different systems which vary in dimensionality and non-linearity. These are the Ornstein–Uhlenbeck process, the exact likelihood of which can be computed analytically, the univariate and highly non-linear, stochastic double well and the multivariate chaotic stochastic Lorenz ’63 (3D model). As a special case the algorithm is also applied to the 40 dimensional stochastic Lorenz ’96 system. In our investigation we compare this new approach with a variety of other well known methods, such as the hybrid Monte Carlo, dual unscented Kalman filter, full weak-constraint 4D-Var algorithm and analyse empirically their asymptotic behaviour as a function of observation density or length of time window increases. In particular we show that we are able to estimate parameters in both the drift (deterministic) and the diffusion (stochastic) part of the model evolution equations using our new methods.

KW - Bayesian inference

KW - variational techniques

KW - dynamical systems

KW - stochastic differential equations

KW - parameter estimation

UR - http://www.scopus.com/inward/record.url?scp=80955130770&partnerID=8YFLogxK

U2 - 10.1016/j.physd.2011.08.013

DO - 10.1016/j.physd.2011.08.013

M3 - Article

VL - 240

SP - 1877

EP - 1900

JO - Physica D

JF - Physica D

SN - 0167-2789

IS - 23

ER -