Defense against local model poisoning attacks to byzantine-robust federated learning

摘要

1 Introduction As a new mode of distributed learning,Federated Learning(FL)helps multiple organizations or clients to jointly train an artificial intelligence model without sharing their own datasets.Compared with the model trained by each client alone,a high-accuracy federated model can be obtained after multiple communication rounds in FL.Due to the characteristics of privacy protection and distributed learning,FL has been applied in many fields,such as the prognosis of pandemicdiseases,smartmanufacturing systems,etc.