A label-free method identifies spurious concepts in vision models by ranking NMF-derived concept vectors using gradient interactions on misclassified examples, enabling suppression that improves worst-group accuracy on Waterbirds and CelebA.
arXiv preprint arXiv:2206.03680 (2023)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Bias Leaves a Gradient Trail: Label-Free Bias Identification via Gradient Probes on Concept Decompositions
A label-free method identifies spurious concepts in vision models by ranking NMF-derived concept vectors using gradient interactions on misclassified examples, enabling suppression that improves worst-group accuracy on Waterbirds and CelebA.