Fair Regression via Post-Processing

A differentially private post-processing algorithm for fair regression. Supports statistical parity under the attribute-aware setting.

To reproduce our results, see the notebooks law.ipynb and communities.ipynb.

LP solvers. Our algorithm involves solving linear programs, and they are set up in our code using the cvxpy package. For large-scale problems, we recommend the Gurobi optimizer for speed.

Citation

@inproceedings{xian2024DifferentiallyPrivatePost,
  title     = {{Differentially Private Post-Processing for Fair Regression}},
  booktitle = {{Proceedings of the 41st International Conference on Machine Learning}},
  author    = {Xian, Ruicheng and Li, Qiaobo and Kamath, Gautam and Zhao, Han},
  year      = {2024}
}