BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Calibration properties of conformal predictive systems - Sam Allen
  (Karlsruhe Institute of Technology (KIT))
DTSTART:20250827T143000Z
DTEND:20250827T150000Z
UID:TALK234502@talks.cam.ac.uk
DESCRIPTION:Conformal prediction has gained considerable attention recentl
 y since it produces forecasts that have out-of-sample calibration guarante
 es by construction\, under the assumption of exchangeability. In this work
 \, we study the calibration properties of conformal predictive systems\, w
 hich issue sets of predictive distributions for real-valued outcomes. We d
 emonstrate that conformal predictive systems implicitly exploit prediction
  methods with in-sample calibration guarantees to construct prediction set
 s with out-of-sample calibration guarantees. While the calibration guarant
 ees are typically that prediction intervals derived from the predictive di
 stributions have the correct marginal coverage\, we show that this line of
  reasoning can be extended to stronger notions of calibration that are com
 mon in statistical forecasting theory. This allows us to take any predicti
 on method that is calibrated in-sample\, and conformalize it to obtain a c
 onformal predictive system with out-of-sample calibration guarantees. Usin
 g this\, we introduce two conformal predictive systems that exhibit strong
 er calibration properties than existing approaches. The first method corre
 sponds to a binning of the data\, while the second leverages isotonic dist
 ributional regression (IDR)\, a non-parametric distributional regression m
 ethod under order constraints. We study the theoretical properties of thes
 e new conformal predictive systems\, and compare their performance in a si
 mulation experiment and two case studies. Both approaches are found to out
 perform existing conformal predictive systems\, while conformal IDR additi
 onally provides a natural method for quantifying epistemic uncertainty of 
 the predictions.&nbsp\;
LOCATION:Seminar Room 1\, Newton Institute
END:VEVENT
END:VCALENDAR