BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Revisiting optimal scaling of Metropolis-Hastings methods - Jure V
 ogrinc
DTSTART:20210511T130000Z
DTEND:20210511T140000Z
UID:TALK160138@talks.cam.ac.uk
CONTACT:Perla Sousi
DESCRIPTION:Deriving useful bounds on the mixing of high dimensional Marko
 v chain Monte Carlo (MCMC) algorithms still seems impossibly difficult. In
  the analysis of their performance we therefore restrict ourselves to answ
 ering simpler proxy questions\, such as how to optimally\nchoose design pa
 rameters in an MCMC method. We will review some classical results on how t
 o optimally scale the proposal variance in the class of Metropolis-Hasting
 s algorithms. Then we will approach the same problem with a novel axiomati
 c framework enabling a more\ndetailed analysis. We will look at counterexa
 mples exhibiting anomalous optimal scaling rates and identifying exact smo
 othness assumptions required for classical scaling. We will discuss mathem
 atical background\, intuitive understanding and further applications of th
 e\naxiomatic framework. \n\nThe talk is based on joint work with Wilfrid K
 endall.
LOCATION:Zoom
END:VEVENT
END:VCALENDAR