Joint State and Parameter Estimation in Temporal Models

Yusuf Erol, Stuart J. Russell and Laurent El Ghaoui

EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2018-28
May 7, 2018

http://www2.eecs.berkeley.edu/Pubs/TechRpts/2018/EECS-2018-28.pdf

Many problems in science and engineering involve the modeling of dynamic processes using state-space models (SSMs). Online parameter and state estimation---computing the posterior probability for both static parameters and dynamic states, incrementally over time–--is crucial for many applications. Many sequential Monte Carlo algorithms have been proposed for this problem; some apply only to restricted model classes, while others are computationally expensive. We propose two new algorithms, namely, the extended parameter filter and the assumed parameter filter, that try to close the gap between computational efficiency and generality. We compare our new algorithms with several state-of-the-art solutions on many benchmark problems. Finally, we discuss our work on joint state and parameter estimation for physiological models in intensive-care medicine.

Advisor: Stuart J. Russell


BibTeX citation:

@mastersthesis{Erol:EECS-2018-28,
    Author = {Erol, Yusuf and Russell, Stuart J. and El Ghaoui, Laurent},
    Title = {Joint State and Parameter Estimation in Temporal Models},
    School = {EECS Department, University of California, Berkeley},
    Year = {2018},
    Month = {May},
    URL = {http://www2.eecs.berkeley.edu/Pubs/TechRpts/2018/EECS-2018-28.html},
    Number = {UCB/EECS-2018-28},
    Abstract = {Many problems in science and engineering involve the modeling of
dynamic processes using state-space models (SSMs). Online parameter
and state estimation---computing the posterior probability for both
static parameters and dynamic states, incrementally over time–--is
crucial for many applications. Many sequential Monte Carlo algorithms
have been proposed for this problem; some apply only to restricted
model classes, while others are computationally expensive. We propose
two new algorithms, namely, the extended parameter filter and the
assumed parameter filter, that try to close the gap between
computational efficiency and generality.  We compare our new
algorithms with several state-of-the-art solutions on many benchmark
problems. Finally, we discuss our work on joint state and
parameter estimation for physiological models in intensive-care
medicine.}
}

EndNote citation:

%0 Thesis
%A Erol, Yusuf
%A Russell, Stuart J.
%A El Ghaoui, Laurent
%T Joint State and Parameter Estimation in Temporal Models
%I EECS Department, University of California, Berkeley
%D 2018
%8 May 7
%@ UCB/EECS-2018-28
%U http://www2.eecs.berkeley.edu/Pubs/TechRpts/2018/EECS-2018-28.html
%F Erol:EECS-2018-28