New methods CM-ATC and CM-DoC estimate a classifier's confusion matrix on unlabeled medical images by applying class-specific confidence thresholds and offsets, outperforming the prior baseline on real-world shifts but failing under prevalence shifts.
In: NeurIPS Workshop on Distribution Shifts: Connecting Methods and Applications (2021)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Label-free estimation of clinically relevant performance metrics under distribution shifts
New methods CM-ATC and CM-DoC estimate a classifier's confusion matrix on unlabeled medical images by applying class-specific confidence thresholds and offsets, outperforming the prior baseline on real-world shifts but failing under prevalence shifts.