BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Active Sampling for Pairwise Comparisons via Approximate Message P
 assing and Information Gain Maximization - Aliaksei Mikhailiuk\, Rainbow G
 roup
DTSTART:20201126T150000Z
DTEND:20201126T160000Z
UID:TALK154435@talks.cam.ac.uk
CONTACT:Maryam Azimi
DESCRIPTION:Pairwise comparison data arise in many domains with subjective
  assessment experiments\, for example in image and video quality assessmen
 t. In these experiments observers are asked to express a preference betwee
 n two conditions. However\, many pairwise comparison protocols require a l
 arge number of comparisons to infer accurate scores\, which may be unfeasi
 ble when each comparison is time-consuming (e.g. videos) or expensive (e.g
 . medical imaging). This motivates the use of an active sampling algorithm
  that chooses only the most informative pairs for comparison. In this pape
 r we propose ASAP\, an active sampling algorithm based on approximate mess
 age passing and expected information gain maximization. Unlike most existi
 ng methods\, which rely on partial updates of the posterior distribution\,
  we are able to perform full updates and therefore much improve the accura
 cy of the inferred scores. The algorithm relies on three techniques for re
 ducing computational cost: inference based on approximate message passing\
 , selective evaluations of the information gain\, and selecting pairs in a
  batch that forms a minimum spanning tree of the inverse of information ga
 in. We demonstrate\, with real and synthetic data\, that ASAP offers the h
 ighest accuracy of inferred scores compared to the existing methods. We al
 so provide an open-source GPU implementation of ASAP for large-scale exper
 iments.
LOCATION:Virtual Zoom meeting
END:VEVENT
END:VCALENDAR