ATTACK COPING METHOD AND FEDERATED LEARNING DEVICE

摘要

An attack coping method and a federated learning device. The method comprises: according to a preconfigured dynamic monitoring mechanism, monitoring model parameters fed back by various participants of the current round of training of an alliance (S101); according to the model parameters and a preconfigured attack probability model, determining the attack probabilities of the various participants attacking a federated learning model, and identifying a malicious attacker from the various participants according to the attack probabilities (S102); according to the attack probability of the malicious attacker in the previous round and a preconfigured monitoring intensity model, determining a target monitoring intensity of the previous round, and according to the target monitoring intensity of the previous round and a preconfigured punishment intensity model, determining a punishment intensity of the alliance in the current round for the malicious attacker (S103); and according to the punishment intensity, determining a punishment loss for the malicious attacker and sending the punishment loss to the malicious attacker to be used for deterring the malicious attacker so as to achieve the purpose of preventing the malicious attacker from attacking the federated learning model, thereby effectively reducing the success rate of the malicious attacker in attacking the federated learning model (S104).