Loop Improvement (LI) method enhances feature extraction in federated learning and multi-task learning without a central server.