BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Gradient-based hyperparameter optimization through reversible lear
 ning - Dr David Duvenaud (Harvard)
DTSTART:20150720T100000Z
DTEND:20150720T110000Z
UID:TALK60203@talks.cam.ac.uk
CONTACT:Zoubin Ghahramani
DESCRIPTION:Tuning hyperparameters of learning algorithms is hard because 
 gradients are usually unavailable. We compute exact gradients of cross-val
 idation performance with respect to all hyperparameters by chaining deriva
 tives backwards through the entire training procedure. This lets us optimi
 ze thousands of hyperparameters\, including step-size and momentum schedul
 es\, weight initialization distributions\, richly parameterized regulariza
 tion schemes\, and neural net architectures. We compute hyperparameter gra
 dients by exactly reversing the dynamics of stochastic gradient descent wi
 th momentum.
LOCATION:Engineering Department\, CBL Room BE-438
END:VEVENT
END:VCALENDAR