Attainability and Optimality: The Equalized Odds Fairness Revisited

TL;DR

This paper investigates the attainability and optimality of the Equalized Odds fairness criterion in machine learning, providing conditions for its achievement and comparing training-time fairness enforcement to post-processing methods.

Contribution

It offers theoretical conditions for achieving Equalized Odds with deterministic and stochastic predictors, and demonstrates benefits of training-time fairness over post-processing.

Findings

Conditions for Equalized Odds attainability with deterministic predictors

Existence of fair stochastic predictors under mild assumptions

Training-time fairness can outperform post-processing in prediction accuracy

Abstract

Fairness of machine learning algorithms has been of increasing interest. In order to suppress or eliminate discrimination in prediction, various notions as well as approaches have been proposed to impose fairness. Given a notion of fairness, an essential problem is then whether or not it can always be attained, even if with an unlimited amount of data. This issue is, however, not well addressed yet. In this paper, focusing on the Equalized Odds notion of fairness, we consider the attainability of this criterion and, furthermore, if it is attainable, the optimality of the prediction performance under various settings. In particular, for prediction performed by a deterministic function of input features, we give conditions under which Equalized Odds can hold true; if the stochastic prediction is acceptable, we show that under mild assumptions, fair predictors can always be derived. For classification, we further prove that compared to enforcing fairness by post-processing, one can always benefit from exploiting all available features during training and get potentially better prediction performance while remaining fair. Moreover, while stochastic prediction can attain Equalized Odds with theoretical guarantees, we also discuss its limitation and potential negative social impacts.