BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:A simple neural network module for relational reasoning - David Ba
 rrett\, DeepMind
DTSTART:20170706T100000Z
DTEND:20170706T110000Z
UID:TALK73097@talks.cam.ac.uk
CONTACT:Adrian Weller
DESCRIPTION:Relational reasoning is a central component of generally intel
 ligent behavior\, but has proven difficult for neural networks to learn. I
 n this paper we describe how to use Relation Networks (RNs) as a simple pl
 ug-and-play module to solve problems that fundamentally hinge on relationa
 l reasoning. We tested RN-augmented networks on three tasks: visual questi
 on answering using a challenging dataset called CLEVR\, on which we achiev
 e state-of-the-art\, super-human performance\; text-based question answeri
 ng using the bAbI suite of tasks\; and complex reasoning about dynamic phy
 sical systems. Then\, using a curated dataset called Sort-of-CLEVR we show
  that powerful convolutional networks do not have a general capacity to so
 lve relational questions\, but can gain this capacity when augmented with 
 RNs. Our work shows how a deep learning architecture equipped with an RN m
 odule can implicitly discover and learn to reason about entities and their
  relations.\n\nhttps://arxiv.org/abs/1706.01427 
LOCATION:CBL Room BE-438\, Department of Engineering
END:VEVENT
END:VCALENDAR