BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Uncertainty Estimation in Structured Prediction - Andrey Malinin\,
  Yandex
DTSTART:20200317T120000Z
DTEND:20200317T130000Z
UID:TALK140281@talks.cam.ac.uk
CONTACT:Konstantinos Kyriakopoulos
DESCRIPTION:Uncertainty estimation is important for ensuring safety and ro
 bustness of AI systems\, especially for high-risk applications. While much
  progress has recently been made in this area\, most research has focused 
 on un-structured prediction\, such as image classification and regression 
 tasks. However\, while task-specific forms of confidence score estimation 
 have been investigated by the speech and machine translation communities\,
  limited work has investigated general uncertainty estimation approaches f
 or structured prediction. Thus\, this work aims to investigate uncertainty
  estimation for structured prediction tasks within a single unified and in
 terpretable probabilistic ensemble-based framework. We consider uncertaint
 y estimation for sequence data at the token-level and complete sequence-le
 vel\, provide interpretations for\, and applications of\, various measures
  of uncertainty and discuss the challenges associated with obtaining them.
  This work also explores the practical challenges associated with obtainin
 g uncertainty estimates for structured predictions tasks and provides base
 lines for token-level error detection\, sequence-level prediction rejectio
 n\, and sequence-level out-of-domain input detection using ensembles of au
 to-regressive transformer models trained on the WMT'14 English-French and 
 WMT'17 English-German translation and LibriSpeech speech recognition datas
 ets.
LOCATION: LT6\, First floor Baker building\, Dept of engineering\, Cambrid
 ge university\, Trumpington street\, CB2 1PZ.
END:VEVENT
END:VCALENDAR