Fair Attribute Classification via Distance Covariance

Ruifan Huang, Haixia Liu

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: Fair classification, distance covariance, Lagrange dual optimization, convergence in probability

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

Abstract: With the increasing prevalence of machine learning, concerns about fairness have emerged. Mitigating potential discrimination risks and preventing machine learning algorithms from making unfair predictions are essential goals in fairness machine learning. We tackle this challenge from a statistical perspective, utilizing distance covariance—a powerful statistical method for measuring both linear and non-linear correlations—as a measure to assess the independence between predictions and sensitive attributes. To enhance fairness in classification, we integrate the sample distance covariance as a manageable penalty term into the machine learning process to promote independence. Additionally, we optimize this constrained problem using the Lagrangian dual method, offering a better trade-off between accuracy and fairness. Theoretically, we provide a proof for the convergence between sample and population distance covariance, establishing necessary guarantees for batch computations. Through experiments conducted on a range of real-world datasets, we demonstrate that our approach can seamlessly extend to existing machine learning models and deliver competitive results.

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: 4736