データ同化研究開発センター

Research and Development Center for Data Assimilation

  • フォントサイズを大きくする
  • デフォルトフォントサイズ
  • フォントサイズを小さく
Home お知らせ セミナー Special seminar for Data Assimilation

Special seminar for Data Assimilation

Title: Understanding and countering the particle filter degeneracy phenomenon

Date : Mon. 25 March 2013, between 14:00 and 16:00

Place: Seminar room 2 (D304), The Institute of Statistical Mathematics, Tachikawa, Tokyo, Japan

Speaker: Dr. Edson H. Aoki (University of Twente, the Netherlands)

*No registration fee is required.

Abstract:

The particle filter is a numerical Bayesian estimation algorithm which is, in principle, applicable to highly general non-linear, non-Gaussian partially observed systems. Yet particle filters have also their limitations, and for some scenarios, their performance can be even worse than of filters based on linear approximations. This is because of the well-known SIR degeneracy phenomenon, which is, however, not always well-understood from an intuitive and/or practical point of view.

This talk aims to provide the audience a both mathematical and intuitive understanding of the particle filter SIR degeneracy phenomenon, to explain its practical consequences, and how particle filter implementations may be adapted to counter degeneracy. For the last purpose, we will focus on a recently developed technique, the Rao-Blackwelllized marginal particle filter (RBMPF), illustrated using practical examples.

 

Home お知らせ セミナー Special seminar for Data Assimilation