The fairness‐accuracy Pareto front

Algorithmic fairness seeks to identify and correct sources of bias in machine learning algorithms. Confoundingly, ensuring often comes at the cost accuracy. We provide formal tools this work for reconciling fundamental tension algorithm fairness. Specifically, we put use concept Pareto optimality from multiobjective optimization seek fairness-accuracy front a neural network classifier. demonstrate that many existing algorithmic methods are performing so-called linear scalarization scheme, which has severe limitations recovering optimal solutions. instead apply Chebyshev scheme is provably superior theoretically no more computationally burdensome solutions compared scheme.