DocumentCode :
3716222
Title :
Adaptive approximate filtering of state-space models
Author :
Kamil Dedecius
Author_Institution :
Institute of Information Theory and Automation, Czech Academy of Sciences, Pod Vodarenskou veZi 1143/4, 182 08 Prague, Czech Republic
fYear :
2015
Firstpage :
2191
Lastpage :
2195
Abstract :
Approximate Bayesian computation (ABC) filtration of state-space models replaces popular particle filters in cases where the observation models (i.e. likelihoods) are either computationally too demanding or completely intractable, but it is still possible to simulate from them. These sequential Monte Carlo methods evaluate importance weights based on the distance between the true observation and the simulated pseudoobservations. The paper proposes a new adaptive method consisting of probability kernel-based evaluation of importance weights with online determination of kernel scale. It is shown that the resulting algorithm achieves performance close to particle filters in the case of well-specified models, and outperforms generic particle filters and state-of-art ABC filters under heavy-tailed noise and model misspecification.
Keywords :
"Kernel","Computational modeling","Approximation methods","Biological system modeling","Adaptation models","Bayes methods","State-space methods"
Publisher :
ieee
Conference_Titel :
Signal Processing Conference (EUSIPCO), 2015 23rd European
Electronic_ISBN :
2076-1465
Type :
conf
DOI :
10.1109/EUSIPCO.2015.7362773
Filename :
7362773
Link To Document :
بازگشت