BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Strong Structural Priors for Neural Network Architectures - Tim Ro
 cktäschel ( UCL)
DTSTART:20160610T110000Z
DTEND:20160610T120000Z
UID:TALK65342@talks.cam.ac.uk
CONTACT:Kris Cao
DESCRIPTION:Many current state-of-the-art methods in natural language proc
 essing and information extraction rely on representation learning. Despite
  the success and wide adoption of neural networks in the field\, we still 
 face major challenges such as (i) efficiently estimating model parameters 
 for domains where annotation is costly and only few training examples are 
 available\, (ii) interpretable representations that allow inspection and d
 ebugging of deep neural networks\, as well as (iii) ways to incorporate co
 mmonsense knowledge and task-specific prior knowledge. To tackle these iss
 ues\, advanced neural network architectures such as differentiable memory\
 , attention\, data structures and even Turing machines\, program interpret
 ers and theorem provers have been proposed very recently. In this talk I w
 ill give an overview of our work on such strong structural priors for sequ
 ence modeling\, knowledge base completion and program induction. 
LOCATION:FW26\, Computer Laboratory
END:VEVENT
END:VCALENDAR