Pairwise Fairness for Ranking and Regression

Improving fairness for ranking and regression models has less mature algorithmic tooling than classifiers. Here, we present pairwise formulations of fairness for ranking and regression models that can express analogues of statistical fairness notions like equal opportunity or equal accuracy, as well as statistical parity. The proposed framework supports both discrete protected groups, and continuous protected attributes. We show that the resulting training problems can be efficiently and effectively solved using constrained optimization or robust optimization algorithms. Experiments illustrate the broad applicability and trade-offs of these methods.