The paper shows that encoder dropout variance dominates under covariate shift in twin networks and better predicts errors on out-of-distribution samples than head variance.
What uncertainties do we need in bayesian deep learning for computer vision? InAdvances in Neural Information Processing Systems, pages 5574–5584
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Localising Dropout Variance in Twin Networks
The paper shows that encoder dropout variance dominates under covariate shift in twin networks and better predicts errors on out-of-distribution samples than head variance.