Fair Representations by Compression
Abstract
Organizations that collect and sell data face increasing scrutiny for the discriminatory use of data. We propose a novel unsupervised approach to map data into a compressed binary representation independent of sensitive attributes. We show that in an information bottleneck framework, a parsimonious representation should filter out information related to sensitive attributes if they are provided directly to the decoder. Empirical results show that the method achieves state-of-the-art accuracy-fairness trade-off and that explicit control of the entropy of the representation bit stream allows the user to move smoothly and simultaneously along both rate-distortion and rate-fairness curves.
Cite
Text
Gitiaux and Rangwala. "Fair Representations by Compression." AAAI Conference on Artificial Intelligence, 2021. doi:10.1609/AAAI.V35I13.17370Markdown
[Gitiaux and Rangwala. "Fair Representations by Compression." AAAI Conference on Artificial Intelligence, 2021.](https://mlanthology.org/aaai/2021/gitiaux2021aaai-fair/) doi:10.1609/AAAI.V35I13.17370BibTeX
@inproceedings{gitiaux2021aaai-fair,
title = {{Fair Representations by Compression}},
author = {Gitiaux, Xavier and Rangwala, Huzefa},
booktitle = {AAAI Conference on Artificial Intelligence},
year = {2021},
pages = {11506-11515},
doi = {10.1609/AAAI.V35I13.17370},
url = {https://mlanthology.org/aaai/2021/gitiaux2021aaai-fair/}
}