Differentiable Particle Filtering via Entropy-Regularized Optimal Transport

Adrien Corenflos; James Thornton; George Deligiannidis; Arnaud Doucet

Differentiable Particle Filtering via Entropy-Regularized Optimal Transport

Adrien Corenflos, James Thornton, George Deligiannidis, Arnaud Doucet

University of Oxford

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

Abstract

Particle Filtering (PF) methods are an established class of procedures for performing inference in non-linear state-space models. Resampling is a key ingredient of PF, necessary to obtain low variance likelihood and states estimates. However, traditional resampling methods result in PF-based loss functions being non-differentiable with respect to model and PF parameters. In a variational inference context, resampling also yields high variance gradient estimates of the PF-based evidence lower bound. By leveraging optimal transport ideas, we introduce a principled differentiable particle filter and provide convergence results. We demonstrate this novel method on a variety of applications.

Original language	English
Title of host publication	Proceedings of Machine Learning Research
Editors	M Meila, T Zhang
Publisher	JMLR
Number of pages	12
Publication status	Published - 2021
MoE publication type	A4 Conference publication
Event	International Conference on Machine Learning - Virtual, Online Duration: 18 Jul 2021 → 24 Jul 2021 Conference number: 38

Publication series

Name	Proceedings of Machine Learning Research
Volume	139
ISSN (Electronic)	2640-3498

Conference

Conference	International Conference on Machine Learning
Abbreviated title	ICML
City	Virtual, Online
Period	18/07/2021 → 24/07/2021

Keywords

LIKELIHOOD EVALUATION

Access to Document

OpenUrl availability

Full text

1 Finished
1 Active

ADAFUME: Advanced data fusion methods for environmental modeling
Särkkä, S., Corenflos, A., Raitoharju, M., Gao, R., Merkatas, C., Sarmavuori, J., Yaghoobi, F., Ma, X. & Hassan, S. S.
01/01/2020 → 31/12/2023
Project: Academy of Finland: Other research funding
-: Parallel and distributed computing for Bayesian graphical models
Särkkä, S., Merkatas, C., Yamin, A., Corenflos, A., Ma, X., Emzir, M., Yaghoobi, F. & Hassan, S. S.
04/09/2019 → 31/12/2022
Project: Academy of Finland: Other research funding

Cite this

@inproceedings{2bddf7519e4d4272b08bf0e7fd182202,

title = "Differentiable Particle Filtering via Entropy-Regularized Optimal Transport",

abstract = "Particle Filtering (PF) methods are an established class of procedures for performing inference in non-linear state-space models. Resampling is a key ingredient of PF, necessary to obtain low variance likelihood and states estimates. However, traditional resampling methods result in PF-based loss functions being non-differentiable with respect to model and PF parameters. In a variational inference context, resampling also yields high variance gradient estimates of the PF-based evidence lower bound. By leveraging optimal transport ideas, we introduce a principled differentiable particle filter and provide convergence results. We demonstrate this novel method on a variety of applications.",

keywords = "LIKELIHOOD EVALUATION",

author = "Adrien Corenflos and James Thornton and George Deligiannidis and Arnaud Doucet",

year = "2021",

language = "English",

series = "Proceedings of Machine Learning Research",

publisher = "JMLR",

editor = "M Meila and T Zhang",

booktitle = "Proceedings of Machine Learning Research",

address = "United States",

note = "International Conference on Machine Learning, ICML ; Conference date: 18-07-2021 Through 24-07-2021",

}

Corenflos, A, Thornton, J, Deligiannidis, G & Doucet, A 2021, Differentiable Particle Filtering via Entropy-Regularized Optimal Transport. in M Meila & T Zhang (eds), Proceedings of Machine Learning Research. Proceedings of Machine Learning Research, vol. 139, JMLR, International Conference on Machine Learning, Virtual, Online, 18/07/2021. <https://arxiv.org/abs/2102.07850>

Differentiable Particle Filtering via Entropy-Regularized Optimal Transport. / Corenflos, Adrien; Thornton, James; Deligiannidis, George et al.
Proceedings of Machine Learning Research. ed. / M Meila; T Zhang. JMLR, 2021. (Proceedings of Machine Learning Research; Vol. 139).

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

TY - GEN

T1 - Differentiable Particle Filtering via Entropy-Regularized Optimal Transport

AU - Corenflos, Adrien

AU - Thornton, James

AU - Deligiannidis, George

AU - Doucet, Arnaud

N1 - Conference code: 38

PY - 2021

Y1 - 2021

N2 - Particle Filtering (PF) methods are an established class of procedures for performing inference in non-linear state-space models. Resampling is a key ingredient of PF, necessary to obtain low variance likelihood and states estimates. However, traditional resampling methods result in PF-based loss functions being non-differentiable with respect to model and PF parameters. In a variational inference context, resampling also yields high variance gradient estimates of the PF-based evidence lower bound. By leveraging optimal transport ideas, we introduce a principled differentiable particle filter and provide convergence results. We demonstrate this novel method on a variety of applications.

AB - Particle Filtering (PF) methods are an established class of procedures for performing inference in non-linear state-space models. Resampling is a key ingredient of PF, necessary to obtain low variance likelihood and states estimates. However, traditional resampling methods result in PF-based loss functions being non-differentiable with respect to model and PF parameters. In a variational inference context, resampling also yields high variance gradient estimates of the PF-based evidence lower bound. By leveraging optimal transport ideas, we introduce a principled differentiable particle filter and provide convergence results. We demonstrate this novel method on a variety of applications.

KW - LIKELIHOOD EVALUATION

M3 - Conference contribution

T3 - Proceedings of Machine Learning Research

BT - Proceedings of Machine Learning Research

A2 - Meila, M

A2 - Zhang, T

PB - JMLR

T2 - International Conference on Machine Learning

Y2 - 18 July 2021 through 24 July 2021

ER -

Differentiable Particle Filtering via Entropy-Regularized Optimal Transport

Abstract

Publication series

Conference

Keywords

Access to Document

OpenUrl availability

Fingerprint

Projects

ADAFUME: Advanced data fusion methods for environmental modeling

-: Parallel and distributed computing for Bayesian graphical models

Cite this