Direct versus prediction-based particle filter algorithms

François Desbouvries; Boujemaa Ait-El-Fquih

doi:10.1109/MLSP.2008.4685497

Direct versus prediction-based particle filter algorithms

François Desbouvries^*, Boujemaa Ait-El-Fquih

^*Corresponding author for this work

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

3 Scopus citations

Abstract

Particle Filtering (PF) algorithms propagate in time a Monte Carlo (MC) approximation of the a posteriori filtering measure in a Hidden Markov Chain (HMC) model. In this paper we first shed some new light on two classical PF algorithms, which can be considered as natural MC implementations of two two-step direct recursive formulas for computing the filtering distribution. We next address the Particle Prediction (PP) problem, which happens to be simpler than the PF problem because the optimal prediction conditional importance distribution (CID) is much easier to sample from. Motivated by this result we finally develop two PP-based PF algorithms, and we compare our algorithms via simulations.

Original language	English (US)
Title of host publication	Proceedings of the 2008 IEEE Workshop on Machine Learning for Signal Processing, MLSP 2008
Pages	303-308
Number of pages	6
DOIs	https://doi.org/10.1109/MLSP.2008.4685497
State	Published - 2008
Externally published	Yes
Event	2008 IEEE Workshop on Machine Learning for Signal Processing, MLSP 2008 - Cancun, Mexico Duration: Oct 16 2008 → Oct 19 2008

Other

Other	2008 IEEE Workshop on Machine Learning for Signal Processing, MLSP 2008
Country/Territory	Mexico
City	Cancun
Period	10/16/08 → 10/19/08

Keywords

Hidden Markov chains
Optimal importance function
Particle filtering
Sampling importance resampling
Sequential importance sampling

ASJC Scopus subject areas

Artificial Intelligence
Software
Electrical and Electronic Engineering

Access to Document

10.1109/MLSP.2008.4685497

Cite this

@inproceedings{9f4d1edfd8114206af0e6a906feb0345,

title = "Direct versus prediction-based particle filter algorithms",

abstract = "Particle Filtering (PF) algorithms propagate in time a Monte Carlo (MC) approximation of the a posteriori filtering measure in a Hidden Markov Chain (HMC) model. In this paper we first shed some new light on two classical PF algorithms, which can be considered as natural MC implementations of two two-step direct recursive formulas for computing the filtering distribution. We next address the Particle Prediction (PP) problem, which happens to be simpler than the PF problem because the optimal prediction conditional importance distribution (CID) is much easier to sample from. Motivated by this result we finally develop two PP-based PF algorithms, and we compare our algorithms via simulations.",

keywords = "Hidden Markov chains, Optimal importance function, Particle filtering, Sampling importance resampling, Sequential importance sampling",

author = "Fran{\c c}ois Desbouvries and Boujemaa Ait-El-Fquih",

year = "2008",

doi = "10.1109/MLSP.2008.4685497",

language = "English (US)",

isbn = "9781424423767",

pages = "303--308",

booktitle = "Proceedings of the 2008 IEEE Workshop on Machine Learning for Signal Processing, MLSP 2008",

note = "2008 IEEE Workshop on Machine Learning for Signal Processing, MLSP 2008 ; Conference date: 16-10-2008 Through 19-10-2008",

}

TY - GEN

T1 - Direct versus prediction-based particle filter algorithms

AU - Desbouvries, François

AU - Ait-El-Fquih, Boujemaa

PY - 2008

Y1 - 2008

N2 - Particle Filtering (PF) algorithms propagate in time a Monte Carlo (MC) approximation of the a posteriori filtering measure in a Hidden Markov Chain (HMC) model. In this paper we first shed some new light on two classical PF algorithms, which can be considered as natural MC implementations of two two-step direct recursive formulas for computing the filtering distribution. We next address the Particle Prediction (PP) problem, which happens to be simpler than the PF problem because the optimal prediction conditional importance distribution (CID) is much easier to sample from. Motivated by this result we finally develop two PP-based PF algorithms, and we compare our algorithms via simulations.

AB - Particle Filtering (PF) algorithms propagate in time a Monte Carlo (MC) approximation of the a posteriori filtering measure in a Hidden Markov Chain (HMC) model. In this paper we first shed some new light on two classical PF algorithms, which can be considered as natural MC implementations of two two-step direct recursive formulas for computing the filtering distribution. We next address the Particle Prediction (PP) problem, which happens to be simpler than the PF problem because the optimal prediction conditional importance distribution (CID) is much easier to sample from. Motivated by this result we finally develop two PP-based PF algorithms, and we compare our algorithms via simulations.

KW - Hidden Markov chains

KW - Optimal importance function

KW - Particle filtering

KW - Sampling importance resampling

KW - Sequential importance sampling

UR - http://www.scopus.com/inward/record.url?scp=58049184929&partnerID=8YFLogxK

U2 - 10.1109/MLSP.2008.4685497

DO - 10.1109/MLSP.2008.4685497

M3 - Conference contribution

AN - SCOPUS:58049184929

SN - 9781424423767

SP - 303

EP - 308

BT - Proceedings of the 2008 IEEE Workshop on Machine Learning for Signal Processing, MLSP 2008

T2 - 2008 IEEE Workshop on Machine Learning for Signal Processing, MLSP 2008

Y2 - 16 October 2008 through 19 October 2008

ER -

Direct versus prediction-based particle filter algorithms

Abstract

Other

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this