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:From linear programming to statistics: Fast algori
thms for sampling based on interior point methods
- Martin Wainwright (UC Berkeley)
DTSTART;TZID=Europe/London:20171124T153000
DTEND;TZID=Europe/London:20171124T163000
UID:TALK96199AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/96199
DESCRIPTION:Sampling from distributions is a core challenge in
\nstatistics\, computer science and operations res
earch. An evolving\nbody of work is showing how a
lgorithms from optimization can be\nmodified so as
to sample from distributions. In this talk\, we
describe and analyze some novel algorithms\, based
on modifications of\ninterior point methods used
in linear programming\, for sampling points\nunifo
rmly from polytopes. Such sampling algorithms are
useful for\nvolume computation\, contigency table
analysis\, post selection\ninference\, and the ha
rd disk problem in statistical physics\, among\not
her applications. We propose and analyze the mixi
ng times of two new Markov chain methods\, referre
d as the Vaidya and John walks\, both of which\nyi
eld substantial improvements over the state-of-the
-art Dikin walk.\n\nBased on joint work with: Yua
nsi Chen\, Raaz Dwivedi\, and Bin Yu\nPre-print:
https://arxiv.org/abs/1710.08165
LOCATION:MR12
CONTACT:Quentin Berthet
END:VEVENT
END:VCALENDAR