FAIR-FATE: Fair Federated Learning with Momentum

While fairness-aware machine learning algorithms have been receiving increasing attention, the focus has on centralized learning, leaving decentralized methods underexplored. Federated Learning is a form of where clients train local models with server aggregating them to obtain shared global model. Data heterogeneity amongst common characteristic Learning, which may induce or exacerbate discrimination unprivileged groups defined by sensitive attributes such as race gender. In this work we propose FAIR-FATE: novel FAIR FederATEd algorithm that aims achieve group fairness while maintaining high utility via aggregation method computes model taking into account clients. To that, update computed estimating fair using Momentum term helps overcome oscillations non-fair gradients. best our knowledge, first approach in estimate. Experimental results real-world datasets demonstrate FAIR-FATE outperforms state-of-the-art under different levels data heterogeneity.