Personalized Federated Learning With Server-Side Information

Personalized Federated Learning (FL) is an emerging research field in FL that learns easily adaptable global model the presence of data heterogeneity among clients. However, one main challenges for personalized heavy reliance on clients’ computing resources to calculate higher-order gradients since client segregated from server ensure privacy. To resolve this, we focus a problem setting where may possess independent – prevalent various applications, yet relatively unexplored existing literature. Specifically, propose FedSIM, new method actively utilizes such server data improve meta-gradient calculation increased personalization performance. Experimentally, demonstrate through benchmarks and ablations FedSIM superior methods terms accuracy, more computationally efficient by calculating full meta-gradients server, converges up 34.2% faster.