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
-
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 theCIFRank
folder
python simulation_b.py "run_exp" "100" "False" "True"
-
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
andres10_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 folderOccupations_dataset_images_cropped
contains cropped versions of images from the Occupations dataset; we include the code used to crop the images insimulation_a.ipynb
. -
census_2010.csv
is taken from US Census Bureau website; it is available here.
@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}
}