FedMDFG: Federated Learning with Multi-Gradient Descent and Fair Guidance

Fairness has been considered as a critical problem in federated learning (FL). In this work, we analyze two direct causes of unfairness FL - an unfair direction and improper step size when updating the model. To solve these issues, introduce effective way to measure fairness model through cosine similarity, then propose multiple gradient descent algorithm with fair guidance (FedMDFG) drive fairer. We first convert into multi-objective optimization (MOP) design advanced calculate by adding fair-driven objective MOP. A low-communication-cost line search strategy is designed find better for update. further show theoretical analysis on how it can enhance guarantee convergence. Finally, extensive experiments several scenarios verify that FedMDFG robust outperforms SOTA algorithms convergence fairness. The source code available at https://github.com/zibinpan/FedMDFG.