Meta-learning and Personalization Layer in Federated Learning

Federated learning systems are confronted with two challenges: systemic and statistical. Non-IID data is acknowledged to be a primary component in causing statistical challenges. To address the federated learning system's substantial performance loss on non-IID data, we offer the FedMeta-Per algorithm (which combines meta-learning methods and personalization layer approaches into a federated learning system). In terms of performance and personalization, FedMeta-Per has been shown in experiments to outperform typical federated learning algorithm, algorithms using personalization layer techniques and algorithms using meta-learning in system optimization.