Fair Learning with Private Demographic Data

TL;DR

This paper proposes a method for privately releasing sensitive demographic data, enabling fair and non-discriminatory machine learning even when such attributes are limited or privatized, with theoretical performance guarantees.

Contribution

It introduces a scheme for private release of sensitive attributes and adapts fair learning algorithms to work with privatized data, providing theoretical guarantees.

Findings

Enables fair learning with privatized sensitive attributes.

Provides theoretical guarantees on the performance of fair classifiers.

Applicable when protected attributes are only partially available.

Abstract

Sensitive attributes such as race are rarely available to learners in real world settings as their collection is often restricted by laws and regulations. We give a scheme that allows individuals to release their sensitive information privately while still allowing any downstream entity to learn non-discriminatory predictors. We show how to adapt non-discriminatory learners to work with privatized protected attributes giving theoretical guarantees on performance. Finally, we highlight how the methodology could apply to learning fair predictors in settings where protected attributes are only available for a subset of the data.