Julemøde 1/12-2016 kl 16:45. Natural Language Processing and Multi-Task Learning: Need for Statistics?

Årets julemøde i Dansk Selskab for Teoretisk Statistik afholdes torsdag den 1. december 2016 kl 16.45

Mødet foregår på Københavns Universitets Center for Sundhed og Samfund CSS (det gamle kommunehospital, Øster Farimagsgade 5, 1014 K) i lokale CSS 5.2.46. (Biostatistiks bibliotek). Bemærk at vi starter lidt tidligere end normalt.

Foredragsholder: Anders Søgaard, Datalogisk Institut/Center for Sprogteknologi, KU.

Natural Language Processing and Multi-Task Learning: Need for Statistics?

Natural language processing is empirical by heart, part engineering and part a science of language and psychology. I will try to present the field in five minutes, for starters. Two things important: a) Labeled data (observations) are sparse. b) Many tasks are intuitively closely related. These observations have led to an interest in multi-task learning lately, and our research group is (among other things) at the forefront of this development. Story so far is mixed. Sometimes it works, sometimes it doesn’t. We are becoming increasingly interested in getting a better theoretical understanding of the method space, to be able to predict when it works, but before getting too deep into that I thought I would take advantage of this chance to hear whether we are, in fact, planning to reinvent the wheel?

Efter mødet vil der være middag på en restaurant i nærheden. Tilmelding til middagen bedes sendt til chair@dsts.dk søndag den 27/11-2016.

Download vedhæftet dokument: