1 paper · 1 filter
Zeming Liu, Hang Lyu, Jingtao Zhang +1
Post-hoc probability calibration is usually evaluated under an optimistic assumption: the held-out calibration labels are clean. In many AI deployment settings, however, labels com…