BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Bayesians turn to Experts for Advice! - Steven de Rooij (Cambridge
 )
DTSTART:20081119T170000Z
DTEND:20081119T180000Z
UID:TALK14680@talks.cam.ac.uk
CONTACT:HoD Secretary\, DPMMS
DESCRIPTION:<I>In Bayesian model selection and model averaging\, inference
  is normally based on a posterior distribution on the models\, usually\n i
 nterpreted as a measure of how likely we consider each of the models to\n 
 be "true"\, or at least in some sense close to true\, given on the\n obser
 vations.\n     Rather than with truth\, I will be concerned with the more 
 practical\n goal of finding a "useful" model\, in the sense that it predic
 ts future\n outcomes of the underlying process well. As it turns out\, the
  most\n useful model may well vary depending on the number of available\n 
 observations! For instance\, given ten samples from some continuous\n dens
 ity\, a seven-bin histogram model is more useful than a 1\,000-bin\n model
 \, even though the latter is arguably closer to being "true".\n     As it 
 turns out\, methods for tracking transient performance of\n prediction str
 ategies have already been developed in the learning theory\n literature un
 der the heading "prediction with expert advice". I will\n illustrate how t
 hese methods can improve model selection performance\n using results from 
 computer simulations on density estimation problems.</I>\n
LOCATION:CMS\, MR4
END:VEVENT
END:VCALENDAR