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:Isaac Newton Institute Seminar Series
SUMMARY:Step size control for Newton type MCMC samplers
Jonathan Goodman - Jonathan Goodman ()
DTSTART;TZID=Europe/London:20191120T150500
DTEND;TZID=Europe/London:20191120T153500
UID:TALK135031AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/135031
DESCRIPTION:ABSTRACT: MCMC sampling can use ideas from the opt
imization community.  \;Optimization via Newto
n&rsquo\;s method can fail without line search\, e
ven for smooth strictly convex problems.  \;Af
fine invariant Newton based MCMC sampling uses a G
aussian proposal based on a quadratic model of the
potential using the local gradient and Hessian. &
nbsp\;This can fail (conjecture: give a transient
Markov chain) even for smooth strictly convex pote
ntials.  \;We describe a criterion that allows
a sequence of proposal distributions from X_n wit
h decreasing &ldquo\;step sizes&rdquo\; until (wit
h probability 1) a proposal is accepted.  \;&l
dquo\;Very detailed balance&rdquo\; allows the who
le process to preserve the target distribution. &n
bsp\;The method works in experiments but the theor
y is missing.
LOCATION:Seminar Room 2\, Newton Institute
CONTACT:INI IT
END:VEVENT
END:VCALENDAR