BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Efficiency and Transferability of Neural Networks - Amos Storkey\,
  School of Informatics\, University of Edinburgh
DTSTART:20190712T100000Z
DTEND:20190712T110000Z
UID:TALK127447@talks.cam.ac.uk
CONTACT:Robert Peharz
DESCRIPTION:Neural Networks are expensive tools for doing very specific ta
 sks. The computational costs of neural network use hinder deployment on em
 bedded devices for time-sensitive tasks\, and the costs of learning are su
 bstantial in all settings. At the same time\, neural network learning is o
 verly dependent on large labelled training dataset that directly matches t
 he test time task\, and by default transfers poorly to slight perturbation
 s of task.\n\nIn this talk I will explore a number of investigations in th
 e efficiency and transferability of neural networks\; we demonstrate that 
 pruning is typically ineffective\, but structured replacement can work wel
 l. We show that full pipeline optimization to hardware can provide substan
 tial benefit. At the same time we explore how meta learning approaches can
  be adapted for transductive learning\, by learning to learn unsupervised 
 loss functions.
LOCATION:Engineering Department\, CBL Room BE-438.
END:VEVENT
END:VCALENDAR