Overview of fair federated learning for fairness and privacy preservation

In the rapidly advancing field of machine learning, federated learning (FL) has facilitated a paradigm shift, enabling collaborative model development across multiple distributed entities while preserving data privacy. FL has recently gained considerable attention as a solution for collaborative machine learning that does not compromise data privacy. However, it faces significant challenges related to fairness, which result in performance disparities across subsets of data. This paper explores the complexities and challenges associated with implementing fairness in the FL framework. Our study substantially contributes to the field of fair FL, detailing how this innovation addresses critical issues such as bias, variance, and model performance degradation in heterogeneous data environments. We provide a comprehensive taxonomy of fairness in FL, categorizing solutions based on data partitioning strategies, privacy mechanisms, applicable machine learning models, communication architectures, and heterogeneity mitigation techniques. Additionally, we highlight ongoing challenges and propose future directions to enhance the integrity and effectiveness of FL systems in this domain.