BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Optimal and efficient learning with random features - Lorenzo Rosa
 sco (Massachusetts Institute of Technology\; Massachusetts Institute of Te
 chnology\; Istituto Italiano di Tecnologica (IIT))
DTSTART:20180117T094500Z
DTEND:20180117T103000Z
UID:TALK97720@talks.cam.ac.uk
CONTACT:INI IT
DESCRIPTION:Random features approaches correspond to one hidden layer neur
 al networks with random hidden units\, and can be seen as approximate kern
 el methods. We study the statistical and computational properties of rando
 m features within a ridge regression scheme. We prove for the first time t
 hat a number of random features much smaller than the number of data point
 s suffices for optimal statistical error\, with a corresponding huge compu
 tational gain. We further analyze faster rates under refined conditions an
 d the potential benefit of  random features chosen according to adaptive s
 ampling schemes.
LOCATION:Seminar Room 1\, Newton Institute
END:VEVENT
END:VCALENDAR