BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Markov chain Monte Carlo on Riemann manifolds - Mark Girolami  (UC
 L)
DTSTART:20110225T160000Z
DTEND:20110225T170000Z
UID:TALK28563@talks.cam.ac.uk
CONTACT:Richard Nickl
DESCRIPTION:Markov chain Monte Carlo (MCMC) provides the dominant methodol
 ogy for inference over\nstatistical models with non-conjugate priors. Desp
 ite a wealth of theoretical characterisation of\nmixing times\, geometric 
 ergodicity\, and asymptotic step-sizes\, the design and implementation of 
 MCMC\nmethods remains something of an engineering art-form. An attempt to 
 address this issue in a\nsystematic manner leads one to consider the geome
 try of probability distributions\, as has been the\ncase previously in the
  study of e.g. higher-order efficiency in statistical estimators. By\ncons
 idering the natural Riemannian geometry of probability distributions MCMC 
 proposal mechanisms\nbased on Langevin diffusions that are characterised b
 y the metric tensor and associated manifold\nconnections are proposed and 
 studied. Furthermore\, optimal proposals that follow the geodesic paths\nr
 elated to the metric are defined via the Hamilton-Jacobi approach and thes
 e are empirically\nevaluated on some challenging modern-day inference task
 s.\n\nThis talk is based on work that was presented as a Discussion Paper 
 to the Royal Statistical Society\nand a dedicated website with Matlab code
 s is available at\nhttp://www.ucl.ac.uk/statistics/research/rmhmc\n\n\nhtt
 p://videolectures.net/mark_girolami/
LOCATION:MR12\, CMS\, Wilberforce Road\, Cambridge\, CB3 0WB
END:VEVENT
END:VCALENDAR