FedDST adapts sparse model connectivity during federated learning to reduce device computation and communication.