Federal learning multi-attacker backdoor attack method based on multi-objective optimization

The invention belongs to the technical field of information security, and discloses a federal learning multi-attacker backdoor attack method based on multi-objective optimization. In order to solve the problem of model parameter conflict caused by pointing similar triggers to different target tags, a homomorphic encryption-based backdoor task optimization method is designed to help a plurality of attackers find the existence of conflicting backdoor tasks on the premise of protecting own backdoor task privacy, and triggers with low similarity are generated for the attackers. In order to relieve conflicts among different attacker backdoor model parameters and balance attack success rates of different attacker backdoor models, the invention provides a multi-target optimization-based multi-backdoor model coordination method, and a group of models beneficial to updating of each backdoor task is searched, so that the attack success rate of each backdoor is improved. 本发明属于信息安全技术领域，公开了一种基于多目标优化的联邦学习多攻击者后门攻击方法。为了解决由于将