Citation
Y. B. Erol, L. Li, B. Ramsundar, S. Russell. "The Extended Parameter Filter". 30th International Conference on Machine Learning (ICML), 2013.

Abstract
The parameters of temporal models, such as dynamic Bayesian networks, may be modelled in a Bayesian context as static or atemporal variables that in uence transition probabilities at every time step. Particle lters fail for models that include such variables, while methods that use Gibbs sampling of parameter variables may incur a per-sample cost that grows linearly with the length of the observation sequence. Storvik (2002) devised a method for incremental computation of exact sufficient statistics that, for some cases, reduces the per-sample cost to a constant. In this paper, we demonstrate a connection between Storvik's filter and a Kalman Filter in parameter space and establish more general conditions under which Storvik's filter works. Drawing on an analogy to the extended Kalman filter, we develop and analyze, both theoretically and experimentally, a Taylor approximation to the parameter posterior that allows Storvik's method to be applied to a broader class of models. Our experiments on both synthetic examples and real applications show improvement over existing methods.

Electronic downloads

• bugraerol13.pdf · application/pdf · 1044 kbytes

• HTML

  Y. B. Erol, L. Li, B. Ramsundar, S. Russell. <a
  href="http://robotics.eecs.berkeley.edu/pubs/8.html"
  >The Extended Parameter Filter</a>, 30th
  International Conference on Machine Learning (ICML), 2013.

• Plain text

  Y. B. Erol, L. Li, B. Ramsundar, S. Russell. "The
  Extended Parameter Filter". 30th International
  Conference on Machine Learning (ICML), 2013.

• BibTeX

  @inproceedings{ErolLiRamsundarRussell13_ExtendedParameterFilter,
      author = {Y. B. Erol and L. Li and B. Ramsundar and S.
                Russell},
      title = {The Extended Parameter Filter},
      booktitle = {30th International Conference on Machine Learning
                (ICML)},
      year = {2013},
      abstract = {The parameters of temporal models, such as dynamic
                Bayesian networks, may be modelled in a Bayesian
                context as static or atemporal variables that in
                uence transition probabilities at every time step.
                Particle lters fail for models that include such
                variables, while methods that use Gibbs sampling
                of parameter variables may incur a per-sample cost
                that grows linearly with the length of the
                observation sequence. Storvik (2002) devised a
                method for incremental computation of exact
                sufficient statistics that, for some cases,
                reduces the per-sample cost to a constant. In this
                paper, we demonstrate a connection between
                Storvik's filter and a Kalman Filter in parameter
                space and establish more general conditions under
                which Storvik's filter works. Drawing on an
                analogy to the extended Kalman filter, we develop
                and analyze, both theoretically and
                experimentally, a Taylor approximation to the
                parameter posterior that allows Storvik's method
                to be applied to a broader class of models. Our
                experiments on both synthetic examples and real
                applications show improvement over existing
                methods.},
      URL = {http://robotics.eecs.berkeley.edu/pubs/8.html}
  }
  

Posted by Ehsan Elhamifar on 16 May 2014.
Groups: ehumans
For additional information, see the Publications FAQ or contact webmaster at robotics eecs berkeley edu.