BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Convex Factorization Machines - Mathieu Blondel (NTT Communication
  Science Laboratorie)
DTSTART:20150914T100000Z
DTEND:20150914T110000Z
UID:TALK60262@talks.cam.ac.uk
CONTACT:Dr Jes Frellsen
DESCRIPTION:Factorization machines are a generic framework which allows to
  mimic many factorization models simply by feature engineering. In this wa
 y\, they combine the high predictive accuracy of factorization models with
  the flexibility of feature engineering.  Unfortunately\, factorization ma
 chines involve a non-convex optimization problem and are thus subject to b
 ad local minima. In this paper\, we propose a convex formulation of factor
 ization machines based on the nuclear norm. Our formulation imposes fewer 
 restrictions on the learned model and is\nthus more general than the origi
 nal formulation.  To solve the corresponding optimization problem\, we pre
 sent an efficient globally-convergent two-block coordinate descent algorit
 hm.  Empirically\, we demonstrate that our approach achieves comparable or
  better predictive accuracy than the original factorization machines on 4 
 recommendation tasks and scales to datasets with 10\nmillion samples.
LOCATION:Engineering Department\, CBL Room BE-438
END:VEVENT
END:VCALENDAR