Backward sequential Monte Carlo for marginal smoothing

Joel Kronander, Thomas B. Schön, Johan Dahlin
Proceedings of the IEEE Statistical Signal Processing Workshop (SSP), Gold Coast, Australia - July 2014
Download the publication : main.pdf [207Ko]   SMCFFBSIposterSSP14.pdf [652Ko]  
In this paper we propose a new type of particle smoother with linear computational complexity. The smoother is based on running a sequential Monte Carlo sampler backward in time after an initial forward filtering pass. While this introduces dependencies among the backward trajectories we show through simulation studies that the new smoother can outperform existing forward-backward particle smoothers when targeting the marginal smoothing densities.

Images and movies

 

BibTex references

@InProceedings\{KBD14,
  author       = "Kronander, Joel and B. Schön, Thomas and Dahlin, Johan",
  title        = "Backward sequential Monte Carlo for marginal smoothing",
  booktitle    = "Proceedings of the IEEE Statistical Signal Processing Workshop (SSP), Gold Coast, Australia ",
  month        = "July ",
  year         = "2014"
}

Author publication list