SemiFL: Semi-Supervised Federated Learning for Unlabeled Clients with Alternate Training

We propose SemiFL to address the problem of combining communication-efficient FL such as FedAvg with Semi-Supervised Learning (SSL).