How unfair is private learning ?
This addresses the challenge of ensuring equitable outcomes in privacy-sensitive applications, such as critical decision-making, but is incremental in exploring trade-offs rather than proposing a new solution.
The paper tackles the problem of achieving both privacy and fairness in machine learning, particularly on long-tailed data, showing that strict privacy requirements can harm accuracy for minority subpopulations, but relaxing overall accuracy can improve fairness, with experimental validation on synthetic, vision, and tabular datasets.
As machine learning algorithms are deployed on sensitive data in critical decision making processes, it is becoming increasingly important that they are also private and fair. In this paper, we show that, when the data has a long-tailed structure, it is not possible to build accurate learning algorithms that are both private and results in higher accuracy on minority subpopulations. We further show that relaxing overall accuracy can lead to good fairness even with strict privacy requirements. To corroborate our theoretical results in practice, we provide an extensive set of experimental results using a variety of synthetic, vision (CIFAR10 and CelebA), and tabular (Law School) datasets and learning algorithms.