SCBD uses supervised contrastive losses to disentangle target-correlated features from environment-correlated features, improving out-of-distribution accuracy on Camelyon17 and batch correction on a 26M-cell optical pooled screen.
We additionally skip gradients with a norm above 1 × 1012, and clip gradients with a norm above 1 × 106, as done in Child (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
-
Supervised Contrastive Block Disentanglement
SCBD uses supervised contrastive losses to disentangle target-correlated features from environment-correlated features, improving out-of-distribution accuracy on Camelyon17 and batch correction on a 26M-cell optical pooled screen.