Marrying Fairness and Explainability in Supervised Learning
This addresses fairness and explainability issues in supervised learning for decision-making systems, with incremental improvements over existing methods.
The paper tackled the problem of discrimination in machine learning algorithms by formalizing direct and induced discrimination as causal effects, and found that existing fair learning methods can still induce discrimination. They proposed post-processing methods to nullify the influence of protected attributes, achieving relatively high accuracy and reducing disparity measures like demographic disparity.
Machine learning algorithms that aid human decision-making may inadvertently discriminate against certain protected groups. We formalize direct discrimination as a direct causal effect of the protected attributes on the decisions, while induced discrimination as a change in the causal influence of non-protected features associated with the protected attributes. The measurements of marginal direct effect (MDE) and SHapley Additive exPlanations (SHAP) reveal that state-of-the-art fair learning methods can induce discrimination via association or reverse discrimination in synthetic and real-world datasets. To inhibit discrimination in algorithmic systems, we propose to nullify the influence of the protected attribute on the output of the system, while preserving the influence of remaining features. We introduce and study post-processing methods achieving such objectives, finding that they yield relatively high model accuracy, prevent direct discrimination, and diminishes various disparity measures, e.g., demographic disparity.