BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Approximate Bayesian Computation for model selection - Christian R
 obert\, Universite Paris-Dauphine and IUF
DTSTART:20120127T160000Z
DTEND:20120127T170000Z
UID:TALK32950@talks.cam.ac.uk
CONTACT:Richard Samworth
DESCRIPTION:Approximate Bayesian computation (ABC)\, also known as likelih
 ood-free\nmethods\, has become a standard tool for the analysis of complex
  models\,\nprimarily in population genetics but also for complex financial
  models. The\ndevelopment of new ABC methodology is undergoing a rapid inc
 rease in the\npast years\, as shown by multiple publications\, conferences
  and even\nsoftwares. While one valid interpretation of ABC based estimati
 on is\nconnected with nonparametrics\, the setting is quite different for 
 model\nchoice issues. We examined in Grelaud et al. (2009) the use of ABC 
 for\nBayesian model choice in the specific of Gaussian random fields (GRF)
 \,\nrelying on a sufficient property to show that the approach was legitim
 ate.\nDespite having previously suggested the use of ABC for model choice 
 in a\nwider range of models in the DIY ABC software (Cornuet et al.\, 2008
 )\, we\npresent in Robert et al. (PNAS\, 2011) theoretical evidence that t
 he general\nuse of ABC for model choice is fraught with danger in the sens
 e that no\namount of computation\, however large\, can guarantee a proper 
 approximation\nof the posterior probabilities of the models under comparis
 on. In a more\nrecent work (Marin et al.\, 2011)\, we expand on this warni
 ng to derive\nnecessary and sufficient conditions on the choice of summary
  statistics for\nABC model choice to be asymptotically consistent.\n
LOCATION:MR12\, CMS\, Wilberforce Road\, Cambridge\, CB3 0WB
END:VEVENT
END:VCALENDAR