BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Deep Neural Networks: A Nonparametric Bayesian Approach with Local
  Competition - Konstantinos P. Panousis
DTSTART:20190620T100000Z
DTEND:20190620T110000Z
UID:TALK125806@talks.cam.ac.uk
CONTACT:Robert Peharz
DESCRIPTION:The aim of this work is to enable inference of deep networks t
 hat retain high accuracy for the least possible model complexity\, with th
 e latter deduced from the data during inference. To this end\, we revisit 
 deep networks that comprise competing linear units\, as opposed to nonline
 ar units that do not entail any form of (local) competition. In this conte
 xt\, our main technical innovation consists in an inferential setup that l
 everages solid arguments from Bayesian nonparametrics. We infer both the n
 eeded set of connections or locally competing sets of units\, as well as t
 he required floating point precision for storing the network parameters. S
 pecifically\, we introduce auxiliary discrete latent variables representin
 g which initial network components are actually needed for modeling the da
 ta at hand\, and perform Bayesian inference over them by imposing appropri
 ate stick-breaking priors. As we experimentally show using benchmark\ndata
 sets\, our approach yields networks with less computational footprint than
  the state-of-the-art\, and with no compromises in predictive accuracy.
LOCATION:Engineering Department\, CBL Room BE-438.
END:VEVENT
END:VCALENDAR