BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Graph Convolutional Networks for Natural Language Processing and R
 elational Modeling - Ivan Titov (Edinburgh)
DTSTART:20180208T110000Z
DTEND:20180208T120000Z
UID:TALK100522@talks.cam.ac.uk
CONTACT:Dimitri Kartsaklis
DESCRIPTION:Graph Convolutional Networks (GCNs) is an effective tool for m
 odeling graph structured data. We investigate their applicability in the c
 ontext of natural language processing (machine translation and semantic ro
 le labelling) and modeling relational data (link prediction). For natural 
 language processing\,  we introduce a version of GCNs suited to modeling s
 yntactic and/or semantic dependency graphs and use them to construct lingu
 istically-informed sentence encoders. We demonstrate that using them resul
 ts in a substantial boost in machine translation performance and state-of-
 the-art results on semantic role labeling of English and Chinese. We also 
 experiment with GCNs over latent graphs (i.e. use structure of a sentence 
 as a latent variable). For link prediction\, we propose Relational GCNs (R
 GCNs)\, GCNs developed specifically to deal with highly multi-relational d
 ata\, characteristic of realistic knowledge bases. By explicitly modeling 
 neighbourhoods of entities\, RGCNs accumulate evidence over multiple infer
 ence steps in relational graphs and yield competitive results on standard 
 link prediction benchmarks. \n\nJoint work with Diego Marcheggiani\, Micha
 el Schlichtkrull\, Joost Bastings\, Thomas Kipf\, Wilker Aziz\, Max Wellin
 g\, Khalil Sima’an\, Rianna van den Berg and Peter Bloem. 
LOCATION:Boardroom\, Faculty of English\, West Road
END:VEVENT
END:VCALENDAR