BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Provable and Practical Methods for Nonconvex Optimazation
DTSTART:20190225T100000
DTEND:20190225T120000
DTSTAMP:20260407T065949Z
UID:fa57eecdd64f495724a03d295f9573bf66c32d128f38fe158e8e34da
CATEGORIES:Conferences - Seminars
DESCRIPTION:Fatih Sahin\nEDIC candidacy exam\nExam president: Prof. Martin
  Jaggi\nThesis advisor: Prof. Volcan Cevher\nCo-examiner: Prof. Daniel Kuh
 n\n\nAbstract\nDue to its capability of handling large scale problems\, no
 nlinear and nonconvex optimization has increased in popularity in the last
  decade. Modern applications in signal processing\, statistics and machine
  learning exhibits distinguished performance with the nonconvex models. Th
 ese models often offer reduced prediction times and can better capture the
  problem structure. Therefore\, it is of high importance to design practic
 al and efficient algorithms for this class of problems. The challenge in c
 ontemporary nonconvex optimization is to establish global convergence guar
 antees. In this proposal\, we are going to discuss three papers which focu
 ses on these challenges. First paper proposes a conditional gradient frame
 work which is a combination of homotopy and smoothing techniques for a com
 posite convex minimization template. Even though it is a convex method\, i
 t can efficiently deal large scale semidefinite programming(SDP) problems 
 in small space via sketching. Second paper theoretically analyzes the conv
 ergence properties of a powerful yet experimental splitting method for sol
 ving large scale SDP's. The last paper provides a primal dual Lagrangian-b
 ased method for a wider class of problems beyond SDP's. However\, the auth
 ors does not demonstrate any experimental results. Finally\, we conclude w
 ith our research proposal on how to address the aforementioned challenges 
 while maintaining some key aspects of the discussed papers.\n\nBackground 
 papers\nA Conditional Gradient Framework for Composite Convex Minimization
 with Applications to Semidefinite Programming\, by Yurtsever\, A.\, et al.
 \nLocal Minima and Convergencein Low-Rank Semidefinite Programming\, by Bu
 rer\, S.\, Monteiro\, R.\nNonconvex Lagrangian-Based Optimization: Monitor
 ing Schemes and Global Convergence by Bolte\, J.\, et al.\n\n 
LOCATION:ELD 120 https://plan.epfl.ch/?room=ELD120
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR