A distance-based softmax loss trained with scaled logits, unscaled at inference, plus an entropy score, improves out-of-distribution detection without extra data, tuning, or accuracy loss.
Probability models for open set recognition,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2019 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Entropic Out-of-Distribution Detection
A distance-based softmax loss trained with scaled logits, unscaled at inference, plus an entropy score, improves out-of-distribution detection without extra data, tuning, or accuracy loss.