BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Advocating for a Combined Use of OOD Detection and Conformal Predi
 ction
DTSTART:20250930T091500Z
DTEND:20250930T101500Z
DTSTAMP:20260907T120100Z
UID:indico-event-14449@indico.math.cnrs.fr
DESCRIPTION:Speakers: Paul Novello (IRT Saint Exupéry\, INSA Toulouse)\n\
 nResearch on Out-Of-Distribution (OOD) detection focuses mainly on buildin
 g scores that efficiently distinguish OOD data from In Distribution (ID) d
 ata. On the other hand\, Conformal Prediction (CP) uses non-conformity sco
 res to construct prediction sets with probabilistic coverage guarantees. I
 n other words\, the former designs scores\, while the latter designs proba
 bilistic guarantees based on scores. This position paper argues that these
  two fields exhibit some potentially impactful synergies. We defend this p
 osition by formalizing this link and emphasizing the benefits of consideri
 ng this link in both OOD detection and the CP fields. First\, for OOD det
 ection\, we show that in standard OOD benchmark settings\, evaluation metr
 ics can be affected by the validation dataset's finite sample size. Extend
 ing the work of Bates et al. 2022\, we define new conformal AUROC and c
 onformal FPR@TPR95 metrics\, which are corrections that provide probabili
 stic guarantees on the variability of the FPR involved in these metrics wi
 th respect to the validation datasets. We show the effect of these correct
 ions on two reference OOD and anomaly detection benchmarks\, OpenOOD (Yang
  et al. 2022)\, and ADBench (Han et al. 2022). Second\, for CP\, we explo
 re using OOD scores as non-conformity scores and show that they can improv
 e the efficiency of the prediction sets obtained with CP.\n\nhttps://indic
 o.math.cnrs.fr/event/14449/
LOCATION:Salle K. Johnson (1R3\, 1er étage)
URL:https://indico.math.cnrs.fr/event/14449/
END:VEVENT
END:VCALENDAR