Workshop on Sequential Monte Carlo Methods: filtering and other applications.

St Anne's College, Oxford      3rd-5th July, 2006

Workshop on Sequential Monte Carlo Methods organised by

The local organiser is Prof. Terry Lyons.


The workshop is sponsored by

EPSRC                           London Mathematical Society

and is a satellite conference of the International Congress of Mathematicians, ICM2006.
Jump to [ Aims |Speakers and Abstracts| Programme |Student financial support| Registration | News | Travel Information and Accommodation | Relevant links]

St Anne's
Courtesy of St Anne's College.

Aims [back to index]

Sequential Monte Carlo methods (SMC) are very important statistical numerical techniques that have revolutionised statistical filtering, the art of sequential -in time- inference of dynamically evolving processes over the last 15 years and have had a profound impact in signal processing (e.g. target tracking) computer science (e.g. computer vision, robotics), but also financial statistics, econometrics and biostatistics. The methodology is now rather well developed and a theory of such processes has emerged that partly supports the methodology. However many open questions still remain, both at a methodological and theoretical level: for example, a satisfactory technique to address the problem of static parameter estimation in its full generality is yet to be found, while at a theoretical level, accuracy and stability of the method can only be proved under rather restrictive technical conditions which are only rarely met in practice.  More recently these techniques, whose main applicational emphasis has been in the area of sequential inference ("statistical filtering"), have attracted much interest in a different context where one is concerned with inference for "static" problems Doucet:Delmoral:Jasra:2006, Cappe:Robert. Traditionally Markov chain Monte Carlo (MCMC) methods have been routinely used in order to tackle these problems, but it has been recently suggested to use SMC in order to circumvent some of the MCMC limitations (e.g. sensitivity to multimodality). These methods are referred to as Sequential Monte Carlo samplers, or population Monte Carlo algorithms in the literature. This is a new and very promising area, where many questions are still unanswered e.g. how does one best use and share information provided by a swarm of samples, how does one implement the theoretical developments in Doucet:Delmoral:Jasra:2006?

We will publish Workshop proceedings within the European Series in Applied and Industrial Mathematics (ESAIM), under the auspices of Societe de Mathematiques Appliquees et Industrielles (SMAI). The workshop has been approved as a satellite conference by the Executive Committee of the International Congress of Mathematicians 2006.
On the grass
Courtesy of St Anne's College.


Invited speakers and abstracts  [back to index]

[Click on name for abstract]



Christophe Andrieu (Bristol)
 
Saadia Ghazali (Imperial)
 
Anastasia Papavasiliou (Warwick)
 
Olivier Cappe (ENST, Paris)
 
Simon Godsill (Cambridge)
 
Michael Pitt (Warwick)
 
Nicolas Chopin (Bristol)
 
Fredrik Gustafsson (Linköping)
 
Mathias Rousset (Nice)
 
Martin Clark (Imperial)
 
Arnaud Guyader (IRISA, Rennes)
 
Tobias Ryden (Lund)
 
Peter Clifford (Oxford)
 
Kari Heine (Tampere)
 
David Salmond (Dstl)
 
Dan Crisan (Imperial)
 
Ajay Jasra (Cambridge)
 
Vladislav Tadic (Sheffield)
 
Nando de Freitas (UBC)
 
Francois LeGland (INRIA/IRISA, Rennes)
 
Jie Xiong (Tennessee)
 
Pierre del Moral (Nice)
 
Terry Lyons (Oxford)
 
Arnaud Doucet (UBC)
 
Eric Moulines (ENST, Paris)
 
Andreas Eberle (Bonn)
 
Omiros Papaspiliopoulos (Warwick)




Tentative Programme [back to index]

 The meeting is scheduled to start on Monday 3rd July 2006 lunchtime (12pm-2pm lunch and registration) and finish on Wednesday 5th July 2006 at 1pm. The talks will be 30 minute long (including question time) and will be divided into four sessions as follows:

Monday afternoon

14:00-14:30 Ajay Jasra (Cambridge) Interacting Sequential Monte Carlo samplers for trans-dimensional simulation
14:30-15:00 Arnaud Guyader (IRISA, Rennes) Adaptive multilevel splitting for rare event analysis
15:00-15:30 Mathias Rousset (Nice) Some Population Monte Carlo issues in Molecular Dynamics.
15:30-16:00 Coffee break

16:00-16:30
 Pierre del Moral (Nice) Precise Propagations of Chaos Estimates for Feynman-Kac and Genealogical Particle Models
16:30-17:00
 Eric Moulines (ENST, Paris) Limit Theorems for Branching Selection Rules
17:00-17:30
 Andreas Eberle (Bonn) Convergence of sequential MCMC methods
17:30-18:00
 Kari Heine (Tampere) The stability of the discrete time filter in terms of the tails of the signal and observation noise distributions

Tuesday morning

09:00-09:30
 Dan Crisan (Imperial) An approximate McKean-Vlasov model for the stochastic filtering problem
09:30-10:00
 Jie Xiong (Tennessee) Numerical solutions for a class of SPDEs over bounded domains
10:00-10:30
 Saadia Ghazali (Imperial) Weak approximations of SDEs with applications to the filtering problem
10:30-11:00
 Christian Litterer (Oxford) Cubature on Wiener Space continued
11:00-11:30 Coffee break

11:30-12:00 Francois LeGland (INRIA/IRISA, Rennes) Particle approximations of Feynman-Kac flows depending on a parameter
12:00-12:30
 Anastasia Papavasiliou (Warwick) Particle filters for multiscale systems
12:30-13:00
 Omiros Papaspiliopoulos (Warwick) Particle Filters for Partially Observed Diffusions
Tuesday afternoon

14:00-14:30
 Nicolas Chopin (Bristol) Likelihood inference for continuous-time hidden Markov models.
14:30-15:00 Michael Pitt (Warwick) Bayesian estimation and particle filtering for diffusion driven state space models
15:00-15:30
 Olivier Cappe (ENST, Paris) On the use of sequential Monte Carlo methods for approximating smoothing functionals,
with application to fixed parameter estimation.
15:30-16:00
 Tobias Ryden (Lund) Particle filter-based approximate likelihood inference in state-space models
16:00-16:30 Coffee break

16:30-17:00
 Christophe Andrieu (Bristol) On the online parameter estimation in general state-spaces problem.
17:00-17:30
 Vladislav Tadic (Sheffield) TBA
17:30-18:00
 Arnaud Doucet (UBC) Sequential Monte Carlo for Optimal Filter Derivative - Application to Parameter Estimation and Control

Wednesday

9:00-9:30 Martin Clark (Imperial) The shifted Rayleigh filter: an efficient approach to bearings only tracking
9:30-10:00 Simon Godsill (Cambridge) Variable Rate Models and Algorithms and Applications to Tracking
10:00-10:30 Fredrik Gustafsson (Linköping) Marginalization issues for the particle filter
1030-11:00 Coffee break

10:00-11:00 David Salmond (Dstl) Target tracking exploiting supplementary information
11:30-12:00 Nando de Freitas (UBC) N-particle interactions: Toward efficient algorithms
Round table discussion.

 
Breakfast, lunch and dinner will be provided on-site at St Anne's College.

Diner

[Courtesy of St Anne's College.]



Support [back to index]

Thanks to the London Mathematical Society, limited financial support is available for a number of UK based PhD students working in the areas of sequential Monte Carlo methods. For more information and application please contact either Dr. C. Andrieu or Dr. D. Crisan.

Registration [back to index]

Conference registration is now open. As the number of places 
for the non-speaker participants is limited we strongly 
recommend that you register early. Prospective participants 
should send an e-mail to either Christophe Andrieu or Dan Crisan 
confirming their intention. We will inform you in due course 
of the registration and accommodation payments.

News [back to index]



 TBA

Travel information and accommodation [back to index]


Information about how to get to the workshop can be found on St Anne's College website.

Accommodation will be provided on-site at St Anne's College.

Diner

[Courtesy of St Anne's College.]

Relevant links [back to index]

Nonlinear Statistical Signal Processing Workshop 2006, Cambridge, 13-15 September 2006
Last update 20 June, 2006