Bayesian joint state and parameter tracking in autoregressive models

Ismail Senoz; Albert Podusenko; Wouter M. Kouw; A. (Bert) de Vries

Bayesian joint state and parameter tracking in autoregressive models

Ismail Senoz, Albert Podusenko, Wouter M. Kouw, A. (Bert) de Vries

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

Abstract

We address the problem of online Bayesian state and parameter tracking in autoregressive (AR) models with time-varying process noise variance. The involved marginalization and expectation integrals cannot be analytically solved. Moreover, the online tracking constraint makes sampling and batch learning methods unsuitable for this problem. We propose a hybrid variational message passing algorithm that robustly tracks the time-varying dynamics of the latent states, AR coefficients and process noise variance. Since message passing in a factor graph is a highly modular inference approach, the proposed methods easily extend to other non-stationary dynamic modeling problems.

Original language	English
Title of host publication	Conference on Learning for Dynamics and Control
Number of pages	10
Publication status	Published - Jun 2020
Event	2nd Annual Conference on Learning for DynamIcs & Control, L4DC 2020 - UC Berkeley, Berkeley, United States Duration: 11 Jun 2020 → 12 Jun 2020 Conference number: 2 https://l4dc.org

Publication series

Name	Proceedings of Machine Learning Research

Conference

Conference	2nd Annual Conference on Learning for DynamIcs & Control, L4DC 2020
Abbreviated title	L4DC
Country/Territory	United States
City	Berkeley
Period	11/06/20 → 12/06/20
Internet address	https://l4dc.org

Keywords

Autoregressive models
hierarchical Gaussian filter
factor graphs
online learning
variational message passing

Access to Document

https://openreview.net/pdf?id=uaYWeSAp5a5

Cite this

@inproceedings{1718e8b3a20f47b4a159aad526aba3d8,

title = "Bayesian joint state and parameter tracking in autoregressive models",

abstract = "We address the problem of online Bayesian state and parameter tracking in autoregressive (AR) models with time-varying process noise variance. The involved marginalization and expectation integrals cannot be analytically solved. Moreover, the online tracking constraint makes sampling and batch learning methods unsuitable for this problem. We propose a hybrid variational message passing algorithm that robustly tracks the time-varying dynamics of the latent states, AR coefficients and process noise variance. Since message passing in a factor graph is a highly modular inference approach, the proposed methods easily extend to other non-stationary dynamic modeling problems.",

keywords = "Autoregressive models, hierarchical Gaussian filter, factor graphs, online learning, variational message passing",

author = "Ismail Senoz and Albert Podusenko and Kouw, {Wouter M.} and {de Vries}, {A. (Bert)}",

year = "2020",

month = jun,

language = "English",

series = "Proceedings of Machine Learning Research",

booktitle = "Conference on Learning for Dynamics and Control",

note = "2nd Annual Conference on Learning for DynamIcs & Control, L4DC 2020, L4DC ; Conference date: 11-06-2020 Through 12-06-2020",

url = "https://l4dc.org",

}

Senoz, I , Podusenko, A , Kouw, WM & de Vries, A 2020, Bayesian joint state and parameter tracking in autoregressive models. in Conference on Learning for Dynamics and Control., 120, Proceedings of Machine Learning Research, 2nd Annual Conference on Learning for DynamIcs & Control, L4DC 2020, Berkeley, California, United States, 11/06/20. <https://openreview.net/pdf?id=uaYWeSAp5a5>

TY - GEN

T1 - Bayesian joint state and parameter tracking in autoregressive models

AU - Senoz, Ismail

AU - Podusenko, Albert

AU - Kouw, Wouter M.

AU - de Vries, A. (Bert)

N1 - Conference code: 2

PY - 2020/6

Y1 - 2020/6

N2 - We address the problem of online Bayesian state and parameter tracking in autoregressive (AR) models with time-varying process noise variance. The involved marginalization and expectation integrals cannot be analytically solved. Moreover, the online tracking constraint makes sampling and batch learning methods unsuitable for this problem. We propose a hybrid variational message passing algorithm that robustly tracks the time-varying dynamics of the latent states, AR coefficients and process noise variance. Since message passing in a factor graph is a highly modular inference approach, the proposed methods easily extend to other non-stationary dynamic modeling problems.

AB - We address the problem of online Bayesian state and parameter tracking in autoregressive (AR) models with time-varying process noise variance. The involved marginalization and expectation integrals cannot be analytically solved. Moreover, the online tracking constraint makes sampling and batch learning methods unsuitable for this problem. We propose a hybrid variational message passing algorithm that robustly tracks the time-varying dynamics of the latent states, AR coefficients and process noise variance. Since message passing in a factor graph is a highly modular inference approach, the proposed methods easily extend to other non-stationary dynamic modeling problems.

KW - Autoregressive models

KW - hierarchical Gaussian filter

KW - factor graphs

KW - online learning

KW - variational message passing

UR - http://www.scopus.com/inward/record.url?scp=85115057919&partnerID=8YFLogxK

M3 - Conference contribution

T3 - Proceedings of Machine Learning Research

BT - Conference on Learning for Dynamics and Control

T2 - 2nd Annual Conference on Learning for DynamIcs & Control, L4DC 2020

Y2 - 11 June 2020 through 12 June 2020

ER -

Bayesian joint state and parameter tracking in autoregressive models

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this