BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Drupal iCal API//EN
X-WR-CALNAME:Events items teaser
X-WR-TIMEZONE:America/Toronto
BEGIN:VTIMEZONE
TZID:America/Toronto
X-LIC-LOCATION:America/Toronto
BEGIN:DAYLIGHT
TZNAME:EDT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
DTSTART:20240310T070000
END:DAYLIGHT
BEGIN:STANDARD
TZNAME:EST
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
DTSTART:20241103T060000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
UID:69d158d081d3c
DTSTART;TZID=America/Toronto:20241122T153000
SEQUENCE:0
TRANSP:TRANSPARENT
DTEND;TZID=America/Toronto:20241122T163000
URL:https://uwaterloo.ca/combinatorics-and-optimization/events/tutte-colloq
 uium-kate-larson
SUMMARY:Tutte colloquium-Kate Larson
CLASS:PUBLIC
DESCRIPTION:TITLE: Soft Condorcet Optimization\n\nSPEAKER:\n Kate Larson\n\
 nAFFILIATION:\n University of Waterloo\n\nLOCATION:\n MC 5501\n\nABSTRACT:
 \n\nA common way to drive the progress of AI models and agents is to\ncomp
 are their performance on standardized benchmarks. This often\ninvolves agg
 regating individual performances across a potentially wide\nvariety of tas
 ks and benchmarks and many of the leaderboards that draw\ngreatest attenti
 on are Elo-based. \n\n \n\nIn this paper\, we describe a novel ranking s
 cheme inspired by social\nchoice frameworks\, called Soft Condorcet Optimi
 zation (SCO)\, to\ncompute the optimal ranking of agents: the one that mak
 es the fewest\nmistakes in predicting the agent comparisons in the evaluat
 ion data.\nThis optimal ranking is the maximum likelihood estimate when\ne
 valuation data (which we view as votes) are interpreted as noisy\nsamples 
 from a ground truth ranking\, a solution to Condorcet's\noriginal voting s
 ystem criteria and inherits desirable social-choice\ninspired properties s
 ince SCO ratings are maximal for Condorcet\nwinners when they exist\, whic
 h we show is not necessarily true for the\nclassical rating system Elo.\n\
 n \n\nWe propose three optimization algorithms to compute SCO ratings and
 \nevaluate their empirical performance across a variety of synthetic and\n
 real-world datasets\, to illustrate different properties.\n\n \n\nWith Ma
 rc Lanctot\, Ian Gemp\, Quentin Berthet\, Yoram Bachrach\, Manfred\nDiaz\,
  Roberto-Rafael Maura-Rivero\,  Anna Koop\, and Doina Precup\n\n \n\n 
DTSTAMP:20260404T183040Z
END:VEVENT
END:VCALENDAR