BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Efficient and Structured Uncertainty: Challenges and Opportunities
  - Andrey Malinin\, Yandex Research
DTSTART:20200722T100000Z
DTEND:20200722T110000Z
UID:TALK150025@talks.cam.ac.uk
CONTACT:Eric T Nalisnick
DESCRIPTION:Uncertainty estimation is important for ensuring safety and ro
 bustness of AI systems. Ensembles of models yield improvements in system p
 erformance as well as principled and interpretable uncertainty estimates. 
 However\, ensemble-based uncertainty estimation comes at a computational a
 nd memory cost which may be prohibitive for many applications. This has li
 mited both practical application and the scale of problems which are exami
 ned in research. In this talk we examine pushing the scale-limit of ensemb
 le-based uncertainty estimation in two ways. Firstly\, we introduce the ta
 sk of "Ensemble Distribution Distillation". Here\, the goal is to distill 
 an ensemble into a single model such that it emulates the ensemble\, retai
 ning both its improved predictive performance and interpretable uncertaint
 y estimates. Secondly\, we investigate principled ensemble-based uncertain
 ty estimation for autoregressive structured prediction tasks\, such as mac
 hine translation and speech recognition\, an area which has received limit
 ed attention so far. Through the lens of these two scale-limits we pose po
 ssible directions of future research.\n\nZoom meeting: https://yandex.zoom
 .us/j/93618382511?pwd=cHJ0MkhZWXhobzZteG9YTUVJV25iUT09
LOCATION:Virtual (see abstract for Zoom link)
END:VEVENT
END:VCALENDAR