Increasing Fairness via Combination with Learning Guarantees

2023-01-25

AI Summary

Machine learning models often struggle to simultaneously satisfy individual and group fairness, lacking rigorous theoretical guarantees. This paper introduces Discriminative Risk (DR), a unified fairness metric that organically integrates individual and group fairness for the first time. We establish the first learning-augmented fairness framework with provable first- and second-order oracle bounds—delivering the first theoretically grounded, learnable guarantee for fairness improvement. Our approach applies to both binary and multiclass classification. Through rigorous fairness-theoretic analysis, ensemble modeling, oracle-bound derivation, and a DR-driven pruning algorithm, we empirically validate our method on multiple benchmark datasets. Results demonstrate significant fairness gains without compromising accuracy, and the derived oracle bounds align closely with empirical observations.

Abstract

The concern about hidden discrimination in ML models is growing, as their widespread real-world application increasingly impacts human lives. Various techniques, including commonly used group fairness measures and several fairness-aware ensemble-based methods, have been developed to enhance fairness. However, existing fairness measures typically focus on only one aspect -- either group or individual fairness, and the hard compatibility among them indicates a possibility of remaining biases even when one of them is satisfied. Moreover, existing mechanisms to boost fairness usually present empirical results to show validity, yet few of them discuss whether fairness can be boosted with certain theoretical guarantees. To address these issues, we propose a fairness quality measure named 'discriminative risk (DR)' to reflect both individual and group fairness aspects. Furthermore, we investigate its properties and establish the first- and second-order oracle bounds to show that fairness can be boosted via ensemble combination with theoretical learning guarantees. The analysis is suitable for both binary and multi-class classification. A pruning method is also proposed to utilise our proposed measure and comprehensive experiments are conducted to evaluate the effectiveness of the proposed methods.

Problem

• Addressing hidden discrimination in ML models
• Combining individual and group fairness measures
• Ensuring fairness with theoretical learning guarantees

Innovation

• Combines individual and group fairness via discriminative risk
• Ensemble combination with theoretical learning guarantees
• Pruning method for fairness measure optimization