Federated learning (FL) is a new machine learning paradigm to overcome the challenge of data silos and has garnered significant attention. However, through our observations, a globally effective trained model may performance disparities in different clients. This implies that the jointly trained models by clients may lead to unfair outcomes. On the other hand, relevant studies indicate that the transmission of gradients or models in federated learning can also give rise to privacy leakage issues, such as membership inference attacks. To address the first issue mentioned above, we propose a fairness-aware federated learning algorithm, termed FedFair. Building upon FedFair, we introduce privacy protection to form the FedFDP algorithm to address the second issue mentioned above. In FedFDP, we devise a fairness-aware clipping strategy to achieve differential privacy while adjusting fairness. Additionally, for the extra uploaded loss values, we present an adaptive clipping approach to maximize utility. Furthermore, we theoretically prove that our algorithm converges and ensures differential privacy. Lastly, extensive experimental results demonstrate that FedFair and FedFDP significantly outperform state-of-the-art solutions in terms of model performance and fairness. Code and data is accessible at https://anonymous.4open.science/r/FedFDP-5607.
翻译:暂无翻译