Minimax Pareto Fairness: A Multi Objective Perspective

TL;DR

This paper introduces a multi-objective fairness framework for classifiers that ensures minimax risk and Pareto efficiency across groups, optimizing fairness without needing sensitive attributes at test time.

Contribution

It formulates group fairness as a multi-objective optimization problem and proposes a simple deep learning-compatible algorithm to achieve minimax Pareto fairness.

Findings

Outperforms existing fairness methods in real-world case studies.

Effectively reduces worst-case classification errors in unbalanced datasets.

Does not require sensitive attribute access during testing.

Abstract

In this work we formulate and formally characterize group fairness as a multi-objective optimization problem, where each sensitive group risk is a separate objective. We propose a fairness criterion where a classifier achieves minimax risk and is Pareto-efficient w.r.t. all groups, avoiding unnecessary harm, and can lead to the best zero-gap model if policy dictates so. We provide a simple optimization algorithm compatible with deep neural networks to satisfy these constraints. Since our method does not require test-time access to sensitive attributes, it can be applied to reduce worst-case classification errors between outcomes in unbalanced classification problems. We test the proposed methodology on real case-studies of predicting income, ICU patient mortality, skin lesions classification, and assessing credit risk, demonstrating how our framework compares favorably to other approaches.