Unlabeled Client

Unlabeled client participation in federated learning (FL) poses significant challenges for model accuracy and robustness, particularly when dealing with data heterogeneity and unseen clients. Current research focuses on developing methods that leverage unlabeled data effectively, such as through contrastive learning, semi-supervised techniques, and topology-aware approaches that account for client relationships. These advancements aim to improve the generalization capabilities of FL models and enable efficient and privacy-preserving inference for new or infrequent participants, impacting the scalability and real-world applicability of FL systems.