BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Efficient implementation of Markov chain Monte Carlo when using an
  unbiased likelihood estimator - Arnaud Doucet\, University of Oxford
DTSTART:20130215T143000Z
DTEND:20130215T153000Z
UID:TALK42411@talks.cam.ac.uk
CONTACT:Richard Samworth
DESCRIPTION:When an unbiased estimator of the likelihood is used within an
 \nMarkov chain Monte Carlo\n(MCMC) scheme\, it is necessary to tradeoff th
 e number of samples used\nagainst the computing\ntime. Many samples for th
 e estimator will result in a MCMC scheme which has\nsimilar properties\nto
  the case where the likelihood is exactly known but will be expensive. Few
 \nsamples for the\nconstruction of the estimator will result in faster est
 imation but at the\nexpense of slower mixing\nof the Markov chain. We expl
 ore the relationship between the number of\nsamples and the\nefficiency of
  the resulting MCMC estimates. Under specific assumptions about\nthe likel
 ihood\nestimator\, we are able to provide guidelines on the number of samp
 les to\nselect for a general\nMetropolis-Hastings proposal. We provide the
 ory which justifies the use of\nthese assumptions\nfor a large class of mo
 dels. On a number of examples\, we find that the\nassumptions on the\nlike
 lihood estimator are accurate.\n\nThis is joint work with Mike Pitt (Unive
 rsity of Warwick) and Robert Kohn\n(UNSW).\n
LOCATION:MR12\, CMS\, Wilberforce Road\, Cambridge\, CB3 0WB
END:VEVENT
END:VCALENDAR