A soft-mask that ranks each network weight by its contribution to bias versus accuracy lets a pretrained medical classifier be debiased in a few fine-tuning epochs on a small balanced dataset, with accuracy maintained.
Leveling down in com- puter vision: Pareto inefficiencies in fair deep classifiers
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
-
SWiFT: Soft-Mask Weight Fine-tuning for Bias Mitigation
A soft-mask that ranks each network weight by its contribution to bias versus accuracy lets a pretrained medical classifier be debiased in a few fine-tuning epochs on a small balanced dataset, with accuracy maintained.