BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Department of Statistics and Operations Research - ECPv4.5.12.3//NONSGML v1.0//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
X-WR-CALNAME:Department of Statistics and Operations Research
X-ORIGINAL-URL:http://stat-or.unc.edu
X-WR-CALDESC:Events for Department of Statistics and Operations Research
BEGIN:VEVENT
DTSTART;TZID=UTC-4:20170828T153000
DTEND;TZID=UTC-4:20170828T163000
DTSTAMP:20170923T091725
CREATED:20170811T203540Z
LAST-MODIFIED:20170822T182947Z
UID:3086-1503934200-1503937800@stat-or.unc.edu
SUMMARY:STOR Colloquium: Ion Necoara\, Polytechnic University of Bucharest
DESCRIPTION:Ion Necoara \nPolytechnic University of Bucharest \n \nConditions for linear convergence of (stochastic) first order methods \n \nFor convex optimization problems deterministic first order methods have linear convergence when the objective function is smooth and strongly convex. Moreover\, under the same conditions – smoothness and strong convexity – sublinear convergence rates have been derived for stochastic first order methods. However\, in many applications (machine learning\, statistics\, control\, signal processing) the strong convexity condition does not hold\, but the objective function still has a special structure. In this talk we replace the smoothness/strong convexity conditions with several other conditions\, that are less conservative\, for which we are able to prove that several (stochastic) first order methods are converging linearly. We also provide necessary conditions for linear convergence of (stochastic) gradient method. Finally\, we discuss several applications of these results (Lasso problem\, linear systems\, linear programming\, convex feasibility\, etc). \n \n \n \nRefreshments will be served at 3:00pm in the 3rd floor lounge of Hanes Hall \n \n
URL:http://stat-or.unc.edu/event/stor-colloquium-ion-necoara
CATEGORIES:STOR Colloquium
END:VEVENT
END:VCALENDAR