Snowball framework defends against backdoor attacks in federated learning through bidirectional elections from an individual perspective.