BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:On the convergence of inexact projection primal first order method
 s for convex minimization
DTSTART:20170303T101500
DTEND:20170303T111500
DTSTAMP:20260924T163500Z
UID:b6693c8cc5d904c4b405578c43307935d8a5ddb43d0986a2fe2e695a
CATEGORIES:Conferences - Seminars
DESCRIPTION:Ion Necoara\, University Politehnica Bucharest\, Romania\nIt i
 s well-known that primal first order algorithms achieve sublinear (linear)
  convergence for smooth convex (smooth strongly convex) constrained minimi
 zation. However\, these methods encounter numerical difficulties when the 
 primal feasible set is complicated\, since they require exact projection o
 nto this set. Algorithmic alternatives to convex problems with complicated
  feasible set are the dual first order methods. Dual methods are able to h
 andle easily complicated constraints\, but they have difficulties in conve
 rgence when the norm of the optimal Lagrange multiplier is large\, since t
 his norm appears linearly in the convergence estimates of these methods. M
 oreover\, they have typically sublinear convergence rate in an average pri
 mal sequence\, even when the primal problem has smooth and strongly convex
  objective function.\n\nMotivated by these issues\, in this talk we analyz
 e the convergence of primal first order methods with inexact projections f
 or solving constrained convex problems with smooth and then strongly conve
 x objective function. In particular\, we consider the inexact variants of 
 Projected Gradient and Projected Fast Gradient methods\, where instead of 
 computing an exact projection onto the complicated primal feasible set\, a
 n approximate projection\, not necessarily feasible\, is used. We show tha
 t we can still achieve similar convergence rates for these inexact project
 ion first order algorithms with those given in the exact projection settin
 gs\, provided that the approximate projection is sufficiently accurate. Ou
 r convergence analysis allows to derive explicitly the accuracy of the ine
 xact projection and the number of iterations we need to perform in order t
 o obtain an approximate solution for our convex problem.   \n\nBio : Ion
  Necoara received the MSc degree in optimization and statistics from Unive
 rsity of Bucharest\, Romania in 2002 and the PhD degree in applied science
 s (cum laude) from Delft University of Technology\, The Netherlands in 200
 6. For the period 2007-2009\, he completed a postdoctoral fellowship at th
 e Katholieke Universiteit Leuven\, Belgium. Currently\, he is a full profe
 ssor at the Department of Automatic Control and Systems Engineering\, Univ
 ersity Politehnica Bucharest\, Romania. His research interests cover vario
 us topics in developing new optimization algorithms with a focus on struct
 ure exploiting and applying optimization techniques for developing new adv
 anced controller design algorithms for complex systems.
LOCATION:ME C2 405 http://plan.epfl.ch/?request_locale=fr&room=MEC2405&dom
 ain=places
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR