Balancing Similarity and Complementarity for Federated Learning