Debiasing representations by removing unwanted variation due to protected attributes

TL;DR

This paper introduces a regression-based method to remove biases related to protected attributes in representations, improving fairness in tasks like recidivism risk scoring.

Contribution

It presents a statistically efficient approach for debiasing representations that satisfies conditional parity and demonstrates effectiveness in reducing racial bias.

Findings

Achieves better bias removal than existing methods

Satisfies a first order approximation of conditional parity

Reduces racial bias in recidivism risk scores

Abstract

We propose a regression-based approach to removing implicit biases in representations. On tasks where the protected attribute is observed, the method is statistically more efficient than known approaches. Further, we show that this approach leads to debiased representations that satisfy a first order approximation of conditional parity. Finally, we demonstrate the efficacy of the proposed approach by reducing racial bias in recidivism risk scores.