Abstract: FL is vulnerable to model poisoning attacks due to the invisibility of local data and the decentralized nature of FL training. The adversary attempts to maliciously manipulate local model ...