BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//RLASKEY//CALENDEROUS//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
BEGIN:VEVENT
DTSTAMP:20190824T205221Z
LAST-MODIFIED:20180301T140604Z
DTSTART:20180302T170000Z
DTEND:20180302T180000Z
UID:event1931@bu.edu
URL:http://physics.bu.edu/events/show/1931
SUMMARY:"Unbiased Markov chain Monte Carlo with couplings joint work with J
ohn O'Leary\, Yves F. AtchadÃ©"
DESCRIPTION:Featuring Pierre E. Jacob\, Harvard University\n\nPart of the C
ondensed Matter Theory Seminar Series.\n\nMarkov chain Monte Carlo (MCMC) m
ethods provide consistent approximations of integrals as the number of iter
ations goes to infinity. MCMC estimators are generally biased after any fix
ed number of iterations\, which complicates both parallel computation and t
he construction of confidence intervals. We propose to remove this bias by
using couplings of Markov chains together with a telescopic sum argument of
Glynn & Rhee (2014). The resulting unbiased estimators can be computed in
parallel\, with confidence intervals following directly from the Central Li
mit Theorem for i.i.d. variables. We discuss practical couplings for popula
r algorithms such as Metropolis-Hastings\, Gibbs samplers\, and Hamiltonian
Monte Carlo. We establish the theoretical validity of the proposed estimat
ors and study their efficiency relative to the underlying MCMC algorithms.
Finally\, we illustrate the performance and limitations of the method on va
rious examples\, involving discrete and continuous high dimensional state s
paces.
LOCATION:SCI 352\, 590 Commonwealth Avenue\, 02215
STATUS:CONFIRMED
CLASS:PUBLIC
END:VEVENT
END:VCALENDAR