BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Learning the Underlying Structure of NLP Tasks
DTSTART:20190709T100000
DTEND:20190709T120000
DTSTAMP:20260408T050644Z
UID:4b1b37d1ba0a1832ed877bf71e71a41ae5459c48b31b78013d4b70d0
CATEGORIES:Conferences - Seminars
DESCRIPTION:Jean-Baptiste Cordonnier\nEDIC candidacy exam\nExam president:
  Dr. François Fleuret\nThesis advisor: Prof. Martin Jaggi\nCo-examiner: P
 rof. Robert West\n\nAbstract\nProgress in Natural Language Processing has 
 been\ndriven by the quest of architectures capturing the structure of\ntex
 t and more recently by novel pre-training tasks on large\ntext corpora. In
  this report\, we discuss two important papers\nthat shifted the NLP resea
 rchers’ attention toward better semisupervised\ntasks and downstream tra
 ining using Multi-Task\nLearning. We then step back and study how the Comp
 uter Vision\ncommunity studies relationships between visual tasks and whic
 h\nlessons apply to NLP. I finally outline my research proposal for\nthe r
 est of the doctoral studies.\n\nBackground papers\nBERT: Pre-training of D
 eep Bidirectional Transformers for Language Understanding\, by Devlin\,J.
 \, et al.\nMulti-Task Deep Neural Networks for Natural Language Understand
 ing\, by Liu\, X.\, et al.\nTaskonomy: Disentangling Task Transfer Learnin
 g\, by  Zamir\, A.\, et al.
LOCATION:BC 010 https://plan.epfl.ch/?room==BC%20010
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR