Researchers propose a new algorithm called FedCC to address the issue of label distribution skew in distillation-based federated learning. In this approach, clients are allowed to tag ambiguous samples as 'unknown' instead of classifying them, which helps balance confidence in majority classes against uncertainty in under-represented ones. The authors claim that FedCC significantly outperforms existing methods, especially in scenarios with severe label skew. --- Why it matters: This matters because it provides a solution to the problem of label distribution skew in federated learning, which can lead to inaccurate predictions and biased models. By allowing clients to tag ambiguous samples as 'unknown', FedCC helps mitigate client misclassification and improves overall performance. Source: https://arxiv.org/abs/2608.23031