BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:In-sample calibration yields conformal calibration guarantees - Jo
 hanna Ziegel (ETH Zurich)
DTSTART:20250530T130000Z
DTEND:20250530T140000Z
UID:TALK231532@talks.cam.ac.uk
CONTACT:Qingyuan Zhao
DESCRIPTION:Conformal predictive systems allow forecasters to issue predic
 tive distributions for real-valued future outcomes that have out-of-sample
  calibration guarantees. On a more abstract level\, conformal prediction m
 akes use of in-sample calibration guarantees to construct bands of predict
 ions with out-of-sample guarantees under exchangeability. The calibration 
 guarantees are typically that prediction intervals derived from the predic
 tive distributions have the correct marginal coverage. We extend this line
  of reasoning to stronger notions of calibration that are common in statis
 tical forecasting theory.\n\nWe take two prediction methods that are calib
 rated in-sample\, and conformalize them to obtain conformal predictive sys
 tems with stronger out-of-sample calibration guarantees than existing appr
 oaches. The first method corresponds to a binning of the data\, while the 
 second leverages isotonic distributional regression (IDR)\, a non-parametr
 ic distributional regression method under order constraints. We study the 
 theoretical properties of these new conformal predictive systems\, and com
 pare their performance in a simulation experiment. They are then applied t
 o two case studies on European temperature forecasts and on predictions fo
 r the length of patient stay in Swiss intensive care units. Both approache
 s are found to outperform existing conformal predictive systems\, while co
 nformal IDR additionally provides a natural method for quantifying epistem
 ic uncertainty of the predictions.
LOCATION:MR12\, Centre for Mathematical Sciences
END:VEVENT
END:VCALENDAR