Fairness Warnings & Fair-MAML: Learning Fairly from Minimal Data

This is the code for our paper, "Fairness Warnings & Fair-MAML: Learning Fairly from Minimal Data."

Checkout the full paper.

Getting started

Setup virtual environment and install requirements:

conda create -n fairwarnmaml python=3.7
source activate fairwarnmaml
pip install -r requirements.txt

To run Fairness Warnings, you'll need a full version of CPLEX. For students and faculty members, install a free full version using the instructions from here.

Run compas_warning_example.py to generate a fairness warning for the COMPAS data set. Detailed read outs of the SLIM results will be given in a folder called ./SLIMLOGS that will be created on run.

Run fair_maml_cc_example.py to sweep over a range of gammas for Fair-MAML on the communities and crime task.

References

Please consider citing our paper if you found this work useful!

@article{Slack2019FairWarningsFairMAML},
	title={Fairness Warnings and Fair-MAML: Learning Fairly with Minimal Data},
	author={Dylan Slack and Sorelle Friedler and Emile Givental.},
	journal={Proceedings of the Conference on Fairness, Accountability, and Transparency (FAT*)},
	year={2020},
}

Citations

The fair-MAML code in fair_maml_cc_example.py is modified from a MAML implementation from github user Jakie Loong found in this repository. Please check out the original implementation as its quite elegant!