BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Safe Learning: How to Modify Bayesian Inference when All Models ar
 e Wrong - Peter Grünwald\, Centrum voor Wiskunde en Informatica\, Amsterd
 am
DTSTART:20120210T160000Z
DTEND:20120210T170000Z
UID:TALK35223@talks.cam.ac.uk
CONTACT:Richard Samworth
DESCRIPTION:Standard Bayesian inference can behave suboptimally if the mod
 el under\nconsideration is wrong:  in some simple settings\, the posterior
  may fail to\nconcentrate even in the limit of infinite sample size.  We i
 ntroduce a test\nthat can tell from the data whether we are in such a situ
 ation. If we are\,\nwe can adjust the learning rate (equivalently: make th
 e prior\nlighter-tailed) in a data-dependent way. The resulting "safe" est
 imator\ncontinues to achieve good rates with wrong models. When applied to
 \nclassification problems\, the safe estimator achieves the optimal rates 
 for\nthe Tsybakov exponent of the underlying distribution\, thereby establ
 ishing a\nconnection between Bayesian inference and statistical learning t
 heory.
LOCATION:MR12\, CMS\, Wilberforce Road\, Cambridge\, CB3 0WB
END:VEVENT
END:VCALENDAR