Parallel Block Particle Filtering
| Autor: | Christelle Garnier, Rui Min, John Klein, François Septier |
|---|---|
| Přispěvatelé: | Centre de Recherche en Informatique, Signal et Automatique de Lille - UMR 9189 (CRIStAL), Centrale Lille-Université de Lille-Centre National de la Recherche Scientifique (CNRS), Ecole nationale supérieure Mines-Télécom Lille Douai (IMT Lille Douai), Institut Mines-Télécom [Paris] (IMT), Université Lille Nord (France), Laboratoire de Mathématiques de Bretagne Atlantique (LMBA), Université de Bretagne Sud (UBS)-Université de Brest (UBO)-Centre National de la Recherche Scientifique (CNRS), Université de Bretagne Sud (UBS), ANR-11-LABX-0020,LEBESGUE,Centre de Mathématiques Henri Lebesgue : fondements, interactions, applications et Formation(2011), Septier, François |
| Rok vydání: | 2021 |
| Předmět: |
010302 applied physics
[STAT.AP]Statistics [stat]/Applications [stat.AP] Computer science Gaussian Posterior probability 02 engineering and technology 021001 nanoscience & nanotechnology [STAT.CO] Statistics [stat]/Computation [stat.CO] 01 natural sciences symbols.namesake [STAT.AP] Statistics [stat]/Applications [stat.AP] Resampling 0103 physical sciences symbols State space [STAT.CO]Statistics [stat]/Computation [stat.CO] 0210 nano-technology Particle filter [SPI.SIGNAL]Engineering Sciences [physics]/Signal and Image processing Algorithm ComputingMilieux_MISCELLANEOUS Subspace topology [SPI.SIGNAL] Engineering Sciences [physics]/Signal and Image processing Curse of dimensionality Block (data storage) |
| Zdroj: | SSP IEEE Statistical Signal Processing Workshop (SSP) IEEE Statistical Signal Processing Workshop (SSP), Jul 2021, Rio de Janeiro, Brazil |
| Popis: | Particle filtering (PF) is a powerful method to estimate the posterior distribution in nonlinear/ non Gaussian state space models. To overcome the curse of dimensionality of PF, the block PF (BPF) partitions the state space and runs correction and resampling steps separately on each subspace. Using a blocking step can significantly reduce the variance of the filtering distribution estimate, but it breaks correlation across subspaces.In this paper, we introduce a parallelisation scheme in the block PF. The scheme consists in dispatching the set of particles into M parallel BPFs. We show that the usual benefit of parallelisation in terms of bias-variance trade-off remains valid and, most importantly, that assigning different partitions to the parallel filters leads to far better performance than naive parallelisation using only one partition. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|