Research Article
Federated Learning-Based IDS Against Poisoning Attacks
@INPROCEEDINGS{10.1007/978-3-030-96791-8_25, author={Mengfan Xu and Xinghua Li}, title={Federated Learning-Based IDS Against Poisoning Attacks}, proceedings={Security and Privacy in New Computing Environments. 4th EAI International Conference, SPNCE 2021, Virtual Event, December 10-11, 2021, Proceedings}, proceedings_a={SPNCE}, year={2022}, month={3}, keywords={Federated learning Privacy computing Poisoning attacks Intrusion detection system Homomorphic encryption}, doi={10.1007/978-3-030-96791-8_25} }- Mengfan Xu
Xinghua Li
Year: 2022
Federated Learning-Based IDS Against Poisoning Attacks
SPNCE
Springer
DOI: 10.1007/978-3-030-96791-8_25
Abstract
With the implementation of the General Data Protection Regulation (GDPR), the federated learning scheme has become a hot topic in the field of private computing. However, existing federated learning scheme can only encrypt the models to ensure the privacy of the data, but can not guarantee the correctness of the uploaded models, which will lead to a significant decrease in the detection performance of the global model. In this paper, we propose a federated learning-based intrusion detection scheme (IDS) against poisoning attacks. Specifically, we first design an anti-poisoning attacks algorithm based on the encryption model. Then we define the anti-attack strategy and objective function. To achieve high detection performance for the availability and concealment of attack, we introduce the poisoning rate into the objective function. The privacy preservation for local data sources also be provided while the IDS model based on knowledge sharing among islands is constructed. We leverage the Paillier public key cryptosystem to prevent data leakage for each entity. The results of security analysis show that our scheme can meet the security requirements of local data sources. In addition, the experiment results demonstrate that the proposed scheme can significantly improve the robustness of the detection model, and its accuracy rate can reach 83.11% even after being poisoned, which means the detection performance has not significantly decreased compared with non-poisoning attacks scheme.
