BEGIN:VCALENDAR
VERSION:2.0
PRODID:IEEE vTools.Events//EN
CALSCALE:GREGORIAN
BEGIN:VTIMEZONE
TZID:Canada/Eastern
BEGIN:DAYLIGHT
DTSTART:20180311T030000
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
RRULE:FREQ=YEARLY;BYDAY=2SU;BYMONTH=3
TZNAME:EDT
END:DAYLIGHT
BEGIN:STANDARD
DTSTART:20181104T010000
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=11
TZNAME:EST
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20180609T150826Z
UID:A039A3E2-2DC1-44CD-8004-E6319604A783
DTSTART;TZID=Canada/Eastern:20180615T140000
DTEND;TZID=Canada/Eastern:20180615T150000
DESCRIPTION:This talk first introduces a standard generative model in ranki
 ng (or recommender) systems\, namely\, the Bradley-Terry-Luce (BTL) model 
 in which the chance of item i beating item j is proportional to the relati
 ve score of item i to item j. We consider two different problems related t
 o this model. First\, we study the top-K ranking problem where the goal is
  to recover the set of top-K ranked items out of a large collection of ite
 ms based on partially revealed preferences. We consider an adversarial cro
 wdsourced setting where there are two population sets. Pairwise comparison
  samples drawn from one of the populations follow the standard BTL model\,
  while in the other population\, the corresponding chance is inversely pro
 portional to the relative score. We derive information-theoretic limits fo
 r the recovery of the top-K set and efficient algorithms for doing so. Sec
 ond\, for the Bayesian BTL model\, we derive information-theoretic lower b
 ounds on the Bayes risk of estimators for norm-based distortion functions.
  We draw parallels between pairwise comparisons in the BTL model and inter
 -player games represented as edges in a graph and analyze the effect of va
 rious graph structures on the lower bounds.\n\nThis talk is based on joint
  work with Changho Suh (KAIST)\, Renbo Zhao\, Mine Alsan and Ranjitha Pras
 ad (NUS)\n\nSpeaker(s): Prof. Vincent Tan\, \n\nRoom: A113\, Bldg: ITB \, 
 1280 Main Street West\, McMaster University\, Hamilton\, Ontario\, Canada\
 , L8S 4K1
LOCATION:Room: A113\, Bldg: ITB \, 1280 Main Street West\, McMaster Univers
 ity\, Hamilton\, Ontario\, Canada\, L8S 4K1
ORGANIZER:junchen@ece.mcmaster.ca
SEQUENCE:10
SUMMARY:[Hamilton] Recent Advances in Ranking: Adversarial Respondents and 
 Lower Bounds on the Bayes Risk
URL;VALUE=URI:https://events.vtools.ieee.org/m/173818
X-ALT-DESC:Description: <br /><p>&nbsp\;</p>\n<p>&nbsp\;</p>\n<p>&nbsp\;</p
 >\n<p>&nbsp\;</p>\n<p>&nbsp\;</p>\n<p>&nbsp\;</p>\n<p>&nbsp\;</p>\n<p>&nbs
 p\;</p>\n<p>This talk first introduces a standard generative model in rank
 ing (or recommender) systems\, namely\, the Bradley-Terry-Luce (BTL) model
  in which&nbsp\;the chance of item i beating item j is proportional to the
  relative score of item i to item j. We consider two different problems re
 lated to this model. First\, we study the top-K ranking problem where the 
 goal is to recover the set of top-K ranked items out of a large collection
  of items based on partially revealed preferences. We consider an adversar
 ial crowdsourced setting where there are two population sets.&nbsp\;Pairwi
 se comparison samples drawn from one of the populations follow the standar
 d BTL model\, while in the other population\, the corresponding chance is 
 inversely proportional to the relative score. We derive information-theore
 tic limits for the recovery of the top-K set and efficient algorithms for 
 doing so. Second\, for the Bayesian BTL model\, we derive information-theo
 retic lower bounds on the Bayes risk of estimators for norm-based distorti
 on functions.&nbsp\;We draw parallels between pairwise comparisons in the 
 BTL model and inter-player games represented as edges in a graph and analy
 ze the effect of various graph structures on the lower bounds.&nbsp\;&nbsp
 \;</p>\n<p>This talk is based on joint work with Changho Suh (KAIST)\, Ren
 bo Zhao\, Mine Alsan and Ranjitha Prasad (NUS)</p>
END:VEVENT
END:VCALENDAR