(Local) Differential Privacy has NO Disparate Impact on Fairness
This addresses fairness concerns in privacy-preserving machine learning for industries handling sensitive data, showing an incremental improvement over existing methods.
The paper tackles the problem of collecting multiple sensitive attributes under Local Differential Privacy (LDP) and its impact on fairness, finding that LDP slightly improves fairness without significantly affecting model performance, with a novel privacy budget allocation scheme outperforming state-of-the-art solutions.
In recent years, Local Differential Privacy (LDP), a robust privacy-preserving methodology, has gained widespread adoption in real-world applications. With LDP, users can perturb their data on their devices before sending it out for analysis. However, as the collection of multiple sensitive information becomes more prevalent across various industries, collecting a single sensitive attribute under LDP may not be sufficient. Correlated attributes in the data may still lead to inferences about the sensitive attribute. This paper empirically studies the impact of collecting multiple sensitive attributes under LDP on fairness. We propose a novel privacy budget allocation scheme that considers the varying domain size of sensitive attributes. This generally led to a better privacy-utility-fairness trade-off in our experiments than the state-of-art solution. Our results show that LDP leads to slightly improved fairness in learning problems without significantly affecting the performance of the models. We conduct extensive experiments evaluating three benchmark datasets using several group fairness metrics and seven state-of-the-art LDP protocols. Overall, this study challenges the common belief that differential privacy necessarily leads to worsened fairness in machine learning.