MIFair defines fairness via mutual information independence between predictions and sensitive attributes, supplies a flexible metric template plus regularization-based mitigation, proves equivalences to standard notions, and handles intersectionality and multiclass settings in a single framework.
The dataset is highly imbalanced across demographic groups; for example, theFemale,Non−Chubbysubgroup has hundreds of times more samples than theFemale,Chubbysubgroup
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
MIFair: A Mutual-Information Framework for Intersectionality and Multiclass Fairness
MIFair defines fairness via mutual information independence between predictions and sensitive attributes, supplies a flexible metric template plus regularization-based mitigation, proves equivalences to standard notions, and handles intersectionality and multiclass settings in a single framework.