Federated Averaging

Federated Averaging (FedAvg) is a machine learning technique enabling collaborative model training across decentralized devices without directly sharing sensitive data. Current research focuses on addressing challenges like data heterogeneity (variations in data distributions across devices) and communication efficiency, often employing techniques such as variance reduction, adaptive learning rates, and novel aggregation methods beyond simple averaging, including generative approaches and model recombination. These advancements aim to improve the accuracy and scalability of FedAvg for diverse applications, particularly in privacy-sensitive domains like healthcare and mobile computing, where data decentralization is crucial.