Fairness-Aware learning is a bias mitigation algorithm for protected groups and intersectional subgroups using input masking constraints.
We showcase how we can mitigate the bias not only in a single protected group such as race or gender but also in intersectional subgroups formed by race and gender.
In order to demonstrate our approach we built a two-layer Neural Network using Law School Admissions Council (LSAC) dataset which is avaialble in the jupyter notebook attached to this repository.
Before running the notebook make sure to install pytorch, pandas, numpy and captum model interpretability library for pytorch.