BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.cam.ac.uk//v3//EN
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:19700329T010000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:19701025T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
CATEGORIES:Statistics
SUMMARY:Comparison of Markov chains via weak PoincarĂ© ineq
ualities\, with application to pseudo-marginal MCM
C - Sam Power (University of Bristol)
DTSTART;TZID=Europe/London:20221014T140000
DTEND;TZID=Europe/London:20221014T150000
UID:TALK182717AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/182717
DESCRIPTION:We investigate the use of a certain class of funct
ional inequalities known as weak PoincarĂ© inequali
ties to bound the convergence of Markov chains to
equilibrium. We show that this enables the straigh
tforward and transparent derivation of subgeometri
c convergence bounds for several 'pseudo-marginal'
sampling algorithms which are popular for carryin
g out Bayesian inference in the setting of intract
able likelihoods\, which are necessarily subgeomet
ric in many practical settings. These results rely
on novel quantitative comparison theorems between
Markov chains. Associated proofs are simpler than
those relying on drift / minorization conditions\
, and the tools developed allow us to recover and
further extend known results as particular cases.
As a consequence of our results\, we are then able
to provide new insights into the practical use of
pseudo-marginal algorithms\, analysing the effect
of averaging in Approximate Bayesian Computation
(ABC)\, the use of products of independent average
s\, and the complexity trade-offs which arise for
particle marginal Metropolis-Hastings (PMMH).\n\n(
https://arxiv.org/abs/2112.05605\, joint work with
Christophe Andrieu\, Anthony Lee\, and Andi Q. Wa
ng)
LOCATION:MR12\, Centre for Mathematical Sciences
CONTACT:Qingyuan Zhao
END:VEVENT
END:VCALENDAR