Supervised-model-based OOD detection methods are fundamentally misspecified: they conflate label uncertainty and feature distance with true distribution shift, producing irreducible detection errors.
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
-
Out-of-Distribution Detection Methods Answer the Wrong Questions
Supervised-model-based OOD detection methods are fundamentally misspecified: they conflate label uncertainty and feature distance with true distribution shift, producing irreducible detection errors.