BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Seminar by Georgina Hall\, Princeton University
DTSTART:20161006T150000
DTEND:20161006T163000
DTSTAMP:20260427T200832Z
UID:ea81d13ab9faf69d985adb010fbbca25a99da3c0ba93c3b8d4e58bb5
CATEGORIES:Conferences - Seminars
DESCRIPTION:Georgina Hall\, Princeton University\n"Sum of squares optimiza
 tion: scalability improvements and applications to difference of convex pr
 ogramming" \n\nAbstract\nOver the past decade\, sum of squares techniques 
 have impacted many areas of computational mathematics\, including optimiza
 tion\, controls\, statistics\, and operations research. It is well known h
 owever that sum of squares (sos) problems are limited by the large semidef
 inite programs (SDPs) that they generate\, which are slow to solve with cu
 rrent technology. In the first part of the talk\, we describe methods to a
 pproximate such SDPs by a series of linear or second order cone programs i
 n order to obtain more scalable algorithms. In the second part of this tal
 k\, we show how sos techniques can be used to tackle the problem of optimi
 zing over convex polynomials. We focus on applications of this problem to 
 computer vision and to nonconvex polynomial optimization. In particular\, 
 we prove that any nonconvex polynomial can be decomposed efficiently (via 
 semidefinite programming) as the difference of two convex polynomials and 
 present the impacts that such a result has on a subclass of optimization p
 roblems called difference of convex programming.
LOCATION:EPFL\, PH H3 31 http://plan.epfl.ch/?request_locale=en&room=ph+h3
 +31&domain=places
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR