Code for "Mitigating Bias in Set Selection with Noisy Protected Attributes"

This repository contains the code for reproducing the simulations in our paper

Mitigating Bias in Set Selection with Noisy Protected Attributes
Anay Mehrotra and L. Elisa Celis
Paper: https://arxiv.org/abs/2011.04219

Running Simulations

• simulation_a.ipynb reproduces the simulations from Section 4.2, Section 4.4, and Section 4.5. It also contains an implementation of our algorithm.

• CIFRank/simulation_b.py reproduces the simulations from Section 4.3. This code builds (heavily) on the repository CIFRank. To execute this code, run the following command inside the CIFRank folder

python simulation_b.py "run_exp" "100" "False" "True"

Acknowledgements

• The code for the simulation in Section 4.3 builds upon the repository CIFRank. We include the necessary code in the folder CIFRank.

• deploy.prototxt and res10_300x300_ssd_iter_140000.caffemodel are taken from this tutorial; they are also available in the opencv repository here.

• occupations_labels.csv is taken from the Occupations dataset. The folder Occupations_dataset_images_cropped contains cropped versions of images from the Occupations dataset; we include the code used to crop the images in simulation_a.ipynb.

• census_2010.csv is taken from US Census Bureau website; it is available here.

Citation

  @misc{mehrotra2020mitigating,
        title={Mitigating Bias in Set Selection with Noisy Protected Attributes},
        author={Anay Mehrotra and L. Elisa Celis},
        year={2020},
        eprint={2011.04219},
        archivePrefix={arXiv},
        primaryClass={cs.CY}
  }