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:Learning and inference in probabilistic submodular
models - Andreas Krause (ETH Zürich)
DTSTART;TZID=Europe/London:20180116T114500
DTEND;TZID=Europe/London:20180116T123000
UID:TALK97636AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/97636
DESCRIPTION:I will present our work on inference and learning
in discrete probabilistic models defined through s
ubmodular functions. These generalize pairwise gr
aphical models and determinantal point processes\,
express natural notions such as attractiveness an
d repulsion and allow to capture richly parameteri
zed\, long-range\, high-order dependencies. The ke
y idea is to use sub- and supergradients of submod
ular functions\, and exploit their combinatorial s
tructure to efficiently optimize variational upper
and lower bounds on the partition function. This
approach allows to perform efficient approximate
inference in any probabilistic model that factoriz
es into log-submodular and log-supermodular potent
ials of arbitrary order. Our approximation is exa
ct at the mode for log-supermodular distributions\
, and we provide bounds on the approximation quali
ty of the log-partition function with respect to t
he curvature of the function. I will also discuss
how to learn log-supermodular distributions via b
i-level optimisation. In particular\, we show how
to compute gradients of the variational posterior\
, which allows integrating the models into modern
deep architectures. This talk is primarily based
on joint work with Josip Djolonga
LOCATION:Seminar Room 1\, Newton Institute
CONTACT:INI IT
END:VEVENT
END:VCALENDAR