Adaptive incentive for cross-silo federated learning: A multi-agent reinforcement learning approach