16

Kernel-based Equalized Odds: A Quantification of Accuracy-Fairness Trade-off in Fair Representation Learning

Main:10 Pages
Bibliography:2 Pages
1 Tables
Appendix:11 Pages
Abstract

This paper introduces a novel kernel-based formulation of the Equalized Odds (EO) criterion, denoted as EOkEO_k, for fair representation learning (FRL) in supervised settings. The central goal of FRL is to mitigate discrimination regarding a sensitive attribute SS while preserving prediction accuracy for the target variable YY. Our proposed criterion enables a rigorous and interpretable quantification of three core fairness objectives: independence (prediction Y^\hat{Y} is independent of SS), separation (also known as equalized odds; prediction Y^\hat{Y} is independent with SS conditioned on target attribute YY), and calibration (YY is independent of SS conditioned on the prediction Y^\hat{Y}). Under both unbiased (YY is independent of SS) and biased (YY depends on SS) conditions, we show that EOkEO_k satisfies both independence and separation in the former, and uniquely preserves predictive accuracy while lower bounding independence and calibration in the latter, thereby offering a unified analytical characterization of the tradeoffs among these fairness criteria. We further define the empirical counterpart, EO^k\hat{EO}_k, a kernel-based statistic that can be computed in quadratic time, with linear-time approximations also available. A concentration inequality for EO^k\hat{EO}_k is derived, providing performance guarantees and error bounds, which serve as practical certificates of fairness compliance. While our focus is on theoretical development, the results lay essential groundwork for principled and provably fair algorithmic design in future empirical studies.

View on arXiv
Comments on this paper