Fig. 1: Overview of our method for fairness and privacy.
From: Achieving flexible fairness metrics in federated medical imaging
a Overview of the proposed FlexFair and its comparison with both the centralized learning and the vanilla FL method, FedAvg. FlexFair effectively mitigates prediction disparities from task models through a weighted penalty mechanism while prioritizing data privacy by integrating a federated framework. b Detailed design of FlexFair. FlexFair addresses fairness and privacy challenges in federated environments by incorporating multiple sensitive attributes, e.g., age, gender, and site, into its framework. It evaluates fairness using metrics like EA, DP, and EO, and integrates these attributes into a weighted regularized loss to ensure the training process promotes fairness across all groups.
