FairReweighing: density estimation-based reweighing framework for improving separation in fair regression

XIAOYIN XI, Zhe Yu

Primary Area: societal considerations including fairness, safety, privacy

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: fairness, separation, reweighing, machine learning

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

TL;DR: We designed a generalized reweighing technique that can improve the ratio estimation of separation on both regression and classification problem

Abstract: There has been a prevalence of implementing machine learning technologies in both high-stakes public-sector and industrial contexts. However, the lack of transparency in these algorithmic solutions has raised concerns over whether these data-informed decisions secure fairness against people from all racial, gender, or age groups. Despite the extensive research and work that emerged on fairness-aware machine learning, up till now, most efforts on solving this issue have been dedicated to binary classification tasks. In this work, we propose a density estimation-based pre-processing algorithm to train regression models satisfying the separation criterion $\hat{Y} \perp A \mid Y$. Evaluated by the ratio estimation of separation via probabilistic classification on both synthetic and real world data, we show that the proposed algorithm outperforms existing state-of-the-art regression fairness solutions in terms of maintaining high predicting accuracy while improving separation in fair regression.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 2987