BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Structured deep models: Deep learning on graphs and beyond - Thoma
 s Kipf\, University of Amsterdam
DTSTART:20180525T130000Z
DTEND:20180525T140000Z
UID:TALK106102@talks.cam.ac.uk
CONTACT:Petar Veličković
DESCRIPTION:In the recent years there has been an increasing number of suc
 cess stories in applying deep learning techniques to graph-structured data
 .\nThe main workhorse in this emerging field is the graph neural network: 
 a message passing algorithm parameterized by neural networks\, trained via
 \nbackpropagation. Variants of graph neural networks now define the state 
 of the art in many classical graph or network problems\, such as node clas
 sification\, graph classification\, and link prediction.\n\nIn this talk\,
  I will give an overview of structured deep models that employ graph neura
 l networks as a key component and discuss trade-offs for a few popular mod
 el variants such as graph convolutional networks (GCNs) [1] and graph atte
 ntion networks (GATs) [2]. \n\nI will further introduce three emerging res
 earch directions: learning deep generative models of graphs\, inference of
  latent graph structure\, and hierarchical\nconcept learning (“learning 
 to pool”). Structured deep models are ideal candidates for these areas a
 nd hold great promise for applications such as program induction\, chemica
 l synthesis\, causal inference\, and\ninteracting physical and multi-agent
  systems.\n\n[1] Kipf & Welling\, Semi-supervised classification with grap
 h\nconvolutional networks\, ICLR 2017\n[2] Veličković et al.\, Graph att
 ention networks\, ICLR 2018
LOCATION:Lecture Theatre 2\, Cambridge University Computer Laboratory\, J 
 J Thompson Avenue\, Madingley Road\, Cambridge
END:VEVENT
END:VCALENDAR