BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//University of California\, Berkeley//UCB Events Calendar//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
BEGIN:VTIMEZONE
TZID:America/Los_Angeles
BEGIN:STANDARD
TZOFFSETFROM:-0700
TZOFFSETTO:-0800
DTSTART:19701029T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:19700402T020000
TZOFFSETFROM:-0800
TZOFFSETTO:-0700
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20170306T105302Z
DTSTART;TZID=America/Los_Angeles:20170308T110000
DTEND;TZID=America/Los_Angeles:20170308T120000
TRANSP:OPAQUE
SUMMARY:Matrix Computations and Scientific Computing Seminar: Sparse inverse covariance matrix estimation for a million variables
UID:107795-ucb-events-calendar@berkeley.edu
ORGANIZER;CN="UC Berkeley Calendar Network":
LOCATION:380 Soda Hall
DESCRIPTION:Inderjit Dhillon\, University of Texas at Austin\n\nThe L1-regularized Gaussian maximum likelihood estimator has been shown to have strong statistical guarantees in recovering a sparse inverse covariance matrix even under high-dimensional settings. However\, it requires solving a difficult non-smooth log-determinant program with number of parameters that scale quadratically with the number of Gaussian variables. Earlier methods thus do not scale to problems with more than 20\,000 variables. In this talk\, I will describe a quadratic approximation method that can solve 1-million dimensional L1-regularized log determinant problems (which would thus have a trillion parameters) on a single computer. In order to do so\, we carefully exploit the underlying structure of the problem. Our innovations include (i) a second-order Newton-like method\, (ii) division of the variables into free and fixed sets\, (iii) a block co-ordinate descent method\, and (iv) a memory efficient scheme that approximates key quantities within the algorithm. Even with the latter approximations\, the proposed BIGQUIC algorithm can achieve a quadratic convergence rate. Experimental results using synthetic and real application data demonstrate the improvements in performance over previous state-of-the-art methods.\n\nThis is joint work with Cho-Jui Hsieh\, Matyas Sustik and Pradeep Ravikumar.
URL:http://events.berkeley.edu/index.php/calendar/sn/pubaff.html?event_ID=107795&view=preview
SEQUENCE:0
CLASS:PUBLIC
CREATED:20170306T105302Z
LAST-MODIFIED:20170306T105302Z
X-MICROSOFT-CDO-BUSYSTATUS:BUSY
X-MICROSOFT-CDO-INSTTYPE:0
X-MICROSOFT-CDO-IMPORTANCE:1
X-MICROSOFT-CDO-OWNERAPPTID:-1
END:VEVENT
END:VCALENDAR