FedLWS: Federated Learning with Adaptive Layerwise Weight Shrinking