Federated fairness-aware classification under differential privacy

Researchers Gengyu Xue and Yi Yu have published a paper introducing FDP-Fair, a novel algorithm designed to address two critical challenges in modern machine learning simultaneously: privacy protection and algorithmic fairness. The method specifically targets classification tasks in federated settings where data is distributed across multiple servers and cannot be centralized. FDP-Fair operates under differential privacy constraints while enforcing demographic parity fairness requirements, providing a framework for building AI systems that are both private and equitable.

The paper makes significant theoretical contributions by disentangling the sources of performance loss in such systems. The authors identify four distinct components of what they term the 'private fairness-aware excess risk': the intrinsic cost of classification, the cost of private classification, the non-private cost of fairness, and the private cost of fairness. This breakdown provides valuable insights into the trade-offs involved when combining these two important requirements. For simpler scenarios with only one server, the researchers also propose CDP-Fair, a computationally efficient alternative algorithm.

Extensive experiments on both synthetic and real-world datasets demonstrate the practicality of the proposed approaches. The work represents an important step toward developing AI systems that can be deployed in sensitive domains like healthcare or finance, where protecting individual privacy and ensuring fair outcomes across demographic groups are both paramount concerns. By providing both algorithms and theoretical analysis, this research offers concrete tools for practitioners building responsible distributed AI.