A Framework for Evaluating Privacy-Utility Trade-off in Vertical Federated Learning