A Modified Proximal-Perturbed Lagrangian for Non-Convex Non-Smooth Representatives of Fairness Constraints
Keywords: fairness constraints, non-convexity, non-smoothness, primal-dual method
Abstract: We study classification problems under fairness constraints and introduce an algorithmic framework designed to prevent discrimination against different groups. These problems are often reformulated as continuous constrained optimization problems and are typically solved using continuous relaxations (surrogates) of the fairness constraints. However, many current algorithms do not provide theoretical guarantees, which possibly is due to the resulting fairness constraints being both non-convex and non-smooth. We propose a novel primal-dual algorithm, based on a newly developed Lagrangian, that converges to a stationary solution of the reformulated problem. Our algorithm is not only efficient and robust, but it also enjoys strong performance guarantees on the fairness of its solutions. Furthermore, experimental results demonstrate that our algorithm is highly effective in terms of computational cost and fairness guarantees, outperforming related algorithms that use regularization (penalization) techniques and/or standard Lagrangian relaxation.
Primary Area: alignment, fairness, safety, privacy, and societal considerations
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.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
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: 2204
Loading