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:Microsoft Research Cambridge\, public talks
SUMMARY:A Case for using Trend Filtering over Splines - A
aditya Ramdas\, Carnegie Mellon University
DTSTART;TZID=Europe/London:20140926T110000
DTEND;TZID=Europe/London:20140926T120000
UID:TALK54835AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/54835
DESCRIPTION:This talk will be about fast optimization algorith
ms (specifically specialized ADMM) for a common pr
actical problem - estimating piecewise constant/li
near/quadratic fits to time series data. I will fi
rst introduce Trend Filtering\, a recently propose
d tool for this problem by Kim\, Koh\, Boyd and Go
rinevsky (2009)\, and compare it to the popular sm
oothing splines and locally adaptive regression sp
lines. Tibshirani (2014) showed that trend filteri
ng estimates converge at the minimax optimal if th
e true underlying function (or its derivatives) ha
s bounded total variation. Hence\, the only roadbl
ock to using it in practice is having robust and e
fficient algorithms. We take a major step in overc
oming this problem\, by providing a more efficient
and robust solution than the current interior poi
nt methods in use. Furthermore\, the proposed ADMM
implementation is very simple\, and importantly\,
it is flexible enough to extend to many interesti
ng related problems\, such as sparse trend filteri
ng and isotonic trend filtering. Software for our
method will be made freely available\, written in
C++\, and also in R (see the {\\tt trendfilter} fu
nction in the R package {\\tt genlasso}).
LOCATION:Auditorium\, Microsoft Research Ltd\, 21 Station R
oad\, Cambridge\, CB1 2FB
CONTACT:Microsoft Research Cambridge Talks Admins
END:VEVENT
END:VCALENDAR