MPAF: Model Poisoning Attacks to Federated Learning based on Fake Clients | IEEE Conference Publication | IEEE Xplore