BEGIN:VCALENDAR
PRODID:-//eluceo/ical//2.0/EN
VERSION:2.0
CALSCALE:GREGORIAN
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1793
DTSTAMP:20261028T051951Z
SUMMARY:Great Expectations: Calibration and Its Consequences
DESCRIPTION:Speaker: Sagnik Chatterjee (TIFR)\n\nAbstract: \nA probabilisti
 c predictor $f:X\\to[0\,1]$ is calibrated if $\\mathbb{E}[Y\\mid f(X)]=f(X
 )$ almost surely. This talk surveys two developments concerning the effici
 ent construction of approximately calibrated predictors and the learning g
 uarantees calibration enables. We first consider adversarial online predic
 tion\, where we present a remarkably simple\, efficient deterministic pred
 ictor by Arunachaleswaran et al. (SODA 2025) that achieves $O(\\sqrt{T})$ 
 distance to a perfectly calibrated prediction sequence over $T$ rounds.\nG
 lobal calibration\, however\, is not a panacea and can conceal systematic 
 errors within subgroups. On the other hand\, multicalibration requires cal
 ibration simultaneously on every subgroup in a specified family. Multiaccu
 racy is a computationally cheaper relaxation of multicalibration\, requiri
 ng predictions to be accurate on average within each subgroup without cond
 itioning further on individual prediction values. We finally discuss a sur
 prising result of Casacuberta et al. (FOCS 2025) who show that adding glob
 al calibration to multiaccuracy recovers much of the learning-theoretic an
 d complexity-theoretic power of multicalibration\, at a computational cost
  comparable to multiaccuracy alone.\n
URL:https://www.tcs.tifr.res.in/web/events/1793
DTSTART;TZID=Asia/Kolkata:20261030T160000
DTEND;TZID=Asia/Kolkata:20261030T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
END:VCALENDAR