BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Reflection methods for user-friendly submodular optimization - Fra
 ncis Bach\, ENS Paris\, France 
DTSTART:20131015T120000Z
DTEND:20131015T130000Z
UID:TALK47827@talks.cam.ac.uk
CONTACT:Microsoft Research Cambridge Talks Admins
DESCRIPTION:Recently\, it has become evident that submodularity naturally 
 captures widely occurring concepts in machine learning\, signal processing
  and computer vision. In consequence\, there is need for efficient optimiz
 ation procedures for submodular functions\, in particular for minimization
  problems. While general submodular minimization is challenging\, we propo
 se a new approach that exploits existing decomposability of submodular fun
 ctions. In contrast to previous approaches\, our method is neither approxi
 mate\, nor impractical\, nor does it need any cumbersome parameter tuning.
  Moreover\, it is easy to implement and parallelize. A key component of ou
 r approach is a formulation of the discrete submodular minimization proble
 m as a continuous best approximation problem. It is solved through a seque
 nce of reflections and its solution can be automatically thresholded to ob
 tain an optimal discrete solution. Our method solves both the continuous a
 nd discrete formulations of the problem\, and therefore has applications i
 n learning\, inference\, and reconstruction. In our experiments\, we show 
 the benefits of our new algorithms for two image segmentation tasks (joint
  work with Stefanie Jegelka and Suvrit Sra).
LOCATION:Auditorium\, Microsoft Research Ltd\, 21 Station Road\, Cambridge
 \, CB1 2FB
END:VEVENT
END:VCALENDAR