Uncertainty measures derived from proper scoring rules should be chosen to match the task loss; total uncertainty suits selective prediction, zero-one loss suits active learning.
Title resolution pending
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
-
Uncertainty Quantification with Proper Scoring Rules: Adjusting Measures to Prediction Tasks
Uncertainty measures derived from proper scoring rules should be chosen to match the task loss; total uncertainty suits selective prediction, zero-one loss suits active learning.