On conditional parity as a notion of non-discrimination in machine learning

TL;DR

This paper introduces conditional parity as a comprehensive framework for non-discrimination in machine learning, unifying various existing notions and providing statistical tools for its analysis and enforcement.

Contribution

It formalizes conditional parity as a general non-discrimination criterion and develops statistical methods, including kernel-based tests, to assess and achieve it.

Findings

Conditional parity encompasses several existing non-discrimination notions.

Randomization can be used to achieve conditional parity.

Kernel-based tests effectively evaluate conditional parity.

Abstract

We identify conditional parity as a general notion of non-discrimination in machine learning. In fact, several recently proposed notions of non-discrimination, including a few counterfactual notions, are instances of conditional parity. We show that conditional parity is amenable to statistical analysis by studying randomization as a general mechanism for achieving conditional parity and a kernel-based test of conditional parity.