EFFL: Egalitarian Fairness in Federated Learning for Mitigating Matthew Effect