sibsutisВестник СибГУТИThe Herald of the Siberian State University of Telecommunications and Information Science1998-6920СибГУТИ10.55648/1998-6920-2026-20-2-3-15sibsutis-1058Research ArticleСтатьиПриватно-сохраняющее обнаружение вторжений в корпоративных сетях на основе федеративного обучения и дифференциально-приватного глубокого обученияPrivacy-preserving intrusion detection in corporate networks via federated and differentially private deep learninghttps://orcid.org/0009-0002-6226-0054КайюмАбдулQayyumAbdul

Postgraduate student, ITMO University (Saint Petersburg National Research University of Information Technologies, Mechanics and Optics) (197101, Russia, Saint Petersburg) e-mail: qayyum068@gmail.com  ORCID: 0009-0002-6226-0054.

PhD student, Faculty of Information Security Technologies, ITMO University (Saint Petersburg National Research University of Information Technologies, Mechanics and Optics), St. Petersburg, Russia. ORCID: 0009-0002-6226-0054.

qayyum068@gmail.comМуссаХамид ИдрисMussaHamid Idris

Master's degree (ITMO University), ITMO University (St. Petersburg National Research University of Information Technologies, Mechanics and Optics) (197101, Russia, St. Petersburg) e-mail: haidturkey@outlook.com

M.Sc. (ITMO University), ITMO University (Saint Petersburg National Research University of Information Technologies, Mechanics and Optics) (197101, Russia, St. Petersburg)

haidturkey@outlook.comhttps://orcid.org/0009-0000-4133-2923ИбрагимХалилIbrahimKhalil

Postgraduate student, ITMO University (Saint Petersburg National Research University of Information Technologies, Mechanics and Optics) (197101, Russia, Saint Petersburg) e-mail: khalil.ibrahim129@yahoo.com  ORCID: 0009-0000-4133-2923.

PhD student, ITMO University (Saint Petersburg National Research University of Information Technologies, Mechanics and Optics) (197101, Russia, St. Petersburg)

khalil.ibrahim129@yahoo.comhttps://orcid.org/0000-0002-0924-6221БеззатеевСергей ВалентиновичBezzateevSergey Valentinovich

Doctor of Engineering Sciences, Associate Professor, Associate Professor, ITMO University (Saint Petersburg National Research University of Information Technologies, Mechanics and Optics) (197101, Russia, Saint Petersburg) e-mail: bsv@guap.ru ORCID: 0000-0002-0924-6221.

D.Sc., Associate Professor, Associate Professor, ITMO University (Saint Petersburg National Research University of Information Technologies, Mechanics and Optics) (197101, Russia, St. Petersburg) e-mail: bsv@guap.ru  ORCID: 0000-0002-0924-6221.

bsv@guap.ruСанкт-Петербургский национальный исследовательский университет информационных технологий, механики и оптикиРоссияSaint Petersburg National Research University of Information Technologies, Mechanics and OpticsRussian FederationСанкт-Петербургский государственный университет аэрокосмического приборостроенияРоссияSaint-Petersburg State University of Aerospace InstrumentationRussian Federation202622062026202Вестник СибГУТИ315Copyright © Кайюм А., Мусса Х.И., Ибрагим Х., Беззатеев С.В., 20262026Кайюм А., Мусса Х.И., Ибрагим Х., Беззатеев С.В.Qayyum A., Mussa H.I., Ibrahim K., Bezzateev S.V.Данная работа распространяется под лицензией Creative Commons Attribution 4.0.This work is licensed under a Creative Commons Attribution 4.0 License.https://vestnik.sibsutis.ru/jour/article/view/1058

Централизованное обнаружение вторжений в распределённых корпоративных инфраструктурах (филиальные сети, удалённые офисы и промышленный/корпоративный IoT) создаёт ограничения по приватности и соблюдению требований (комплаенсу), поскольку требует централизации конфиденциальной телеметрии. В работе исследуется приватно-сохраняющий дизайн IDS на основе федеративного обучения (FL) с дифференциально-приватным локальным обучением (DP-SGD). Клиенты обучаются локально на признаках, извлечённых из сетевых потоков, и структурированной событийной телеметрии, а координатору для агрегации передаются только обновления модели. Мы оцениваем бюджет приватности (ε, δ) с использованием учёта приватности на основе RDP и качество обнаружения по метрикам Accuracy и Macro-F1. Эксперименты на CICIDS 2018 и наборе данных CERT Insider Threat v6.2 демонстрируют ожидаемый компромисс «приватность–полезность»: обучение с DP снижает полезность по сравнению с не-приватным централизованным обучением, а FL с DP обычно приводит к дополнительному снижению при гетерогенных (non-IID) разбиениях клиентов, оставаясь практичным при явных бюджетах приватности (основное сравнение: ε ≈ 1.8-2.0). Также приводятся эксплуатационные показатели, релевантные для SOC: время обучения, пиковое потребление памяти, задержка вывода (inference) и размер модели, а также сравнение с базовыми моделями Random Forest и XGBoost.

Centralized intrusion detection in distributed corporate infrastructures (branches, remote offices, and enterprise/industrial IoT) creates privacy and compliance constraints because it requires centralizing sensitive telemetry. We study a privacy-preserving IDS design based on federated learning (FL) with differentially private local training (DP-SGD). Clients train locally on flow-derived features and structured event telemetry, and only model updates are shared with a coordinator for aggregation. We report the privacy budget (ε, δ) using an RDP accountant and evaluate detection quality using Accuracy and Macro-F1. Experiments on CICIDS 2018 and the CERT Insider Threat v6.2 dataset show the expected privacy-utility tradeoff: DP training reduces utility compared to non-private centralized learning, and FL with DP typically incurs an additional decrease under heterogeneous (non-IID) client partitions, while remaining practical at explicit privacy budgets (main comparison: ε ≈ 1.8 -2.0). We also report SOC-relevant operational indicators: training time, peak memory usage, inference latency, and model size-and compare against Random Forest and XGBoost baselines.

обнаружение вторженийфедеративное обучениедифференциальная приватностьDP-SGDбезопасная агрегацияинсайдерские угрозыинформационная безопасностькорпоративные сетицентр мониторинга и реагирования на инциденты (SOC)Intrusion DetectionFederated LearningDifferential PrivacyDP-SGDSecure AggregationInsider ThreatsInformation SecurityCorporate NetworksSecurity Operations Center (SOC)Работа выполнена в рамках государственного задания (проект FSER-2025-0003).The work was performed within the framework of the state assignment (project FSER-2025-0003)ReferencesBuczak A.L., Guven E. A survey of data mining and machine learning methods for cyber security intrusion detection. IEEE Communications Surveys & Tutorials, 2016, vol. 18, no. 2, pp. 1153–1176. doi:10.1109/COMST.2015.2494502.ReferencesScarfone K., Mell P. Guide to Intrusion Detection and Prevention Systems (IDPS). NIST Special Publication 800-94. National Institute of Standards and Technology, 2007. doi:10.6028/NIST.SP.800-94.Buczak A.L., Guven E. A survey of data mining and machine learning methods for cyber security intrusion detection. IEEE Communications Surveys & Tutorials, 2016, vol. 18, no. 2, pp. 1153–1176. doi:10.1109/COMST.2015.2494502.McMahan H.B., Moore E., Ramage D., Hampson S., Agüera y Arcas B. Communication-efficient learning of deep networks from decentralized data. In: Proceedings of AISTATS, 2017. arXiv:1602.05629. doi:10.48550/arXiv.1602.05629.Scarfone K., Mell P. Guide to Intrusion Detection and Prevention Systems (IDPS). NIST Special Publication 800-94. National Institute of Standards and Technology, 2007. doi:10.6028/NIST.SP.800-94.Kairouz P., McMahan H.B., Avent B., et al. Advances and open problems in federated learning. Foundations and Trends in Machine Learning, 2021, vol. 14, no. 1–2, pp. 1–210. doi:10.1561/2200000083.McMahan H.B., Moore E., Ramage D., Hampson S., Agüera y Arcas B. Communication-efficient learning of deep networks from decentralized data. In: Proceedings of AISTATS, 2017. arXiv:1602.05629. doi:10.48550/arXiv.1602.05629.Lazzarini R., Tianfield H., Charissis V. Federated learning for IoT intrusion detection. AI, 2023, vol. 4, no. 3, pp. 509–530. doi:10.3390/ai4030028.Kairouz P., McMahan H.B., Avent B., et al. Advances and open problems in federated learning. Foundations and Trends in Machine Learning, 2021, vol. 14, no. 1–2, pp. 1–210. doi:10.1561/2200000083.Chen J., Yan H., Liu Z., Zhang M., Xiong H., Yu S. When federated learning meets privacy-preserving computation. ACM Computing Surveys, 2024, vol. 56, no. 12, pp. 1–36. doi:10.1145/3679013.Lazzarini R., Tianfield H., Charissis V. Federated learning for IoT intrusion detection. AI, 2023, vol. 4, no. 3, pp. 509–530. doi:10.3390/ai4030028.Zhu L., Liu Z., Han S. Deep leakage from gradients. In: Advances in Neural Information Processing Systems (NeurIPS 2019), 33rd Conference on Neural Information Processing Systems, Vancouver, Canada, 2019, pp. 14747–14756. Available at: https://proceedings.neurips.cc/paper_files/paper/2019/file/60a6c4002cc7b29142def8871531281a-Paper.pdfChen J., Yan H., Liu Z., Zhang M., Xiong H., Yu S. When federated learning meets privacy-preserving computation. ACM Computing Surveys, 2024, vol. 56, no. 12, pp. 1–36. doi:10.1145/3679013.Geiping J., Bauermeister H., Dröge H., Moeller M. Inverting gradients – how easy is it to break privacy in federated learning? arXiv:2007.05657, 2020. doi:10.48550/arXiv.2007.05657.Zhu L., Liu Z., Han S. Deep leakage from gradients. In: Advances in Neural Information Processing Systems (NeurIPS 2019), 33rd Conference on Neural Information Processing Systems, Vancouver, Canada, 2019, pp. 14747–14756. Available at: https://proceedings.neurips.cc/paper_files/paper/2019/file/60a6c4002cc7b29142def8871531281a-Paper.pdfShokri R., Stronati M., Song C., Shmatikov V. Membership inference attacks against machine learning models. 2017 IEEE Symposium on Security and Privacy (SP), 2017, pp. 3–18. doi:10.1109/SP.2017.41.Geiping J., Bauermeister H., Dröge H., Moeller M. Inverting gradients – how easy is it to break privacy in federated learning? arXiv:2007.05657, 2020. doi:10.48550/arXiv.2007.05657.Nasr M., Shokri R., Houmansadr A. Comprehensive privacy analysis of deep learning: passive and active white-box inference attacks against centralized and federated learning. 2019 IEEE Symposium on Security and Privacy (SP), 2019. doi:10.1109/SP.2019.00065.Shokri R., Stronati M., Song C., Shmatikov V. Membership inference attacks against machine learning models. 2017 IEEE Symposium on Security and Privacy (SP), 2017, pp. 3–18. doi:10.1109/SP.2017.41.Dwork C., Roth A. The algorithmic foundations of differential privacy. Foundations and Trends in Theoretical Computer Science, 2014, vol. 9, no. 3–4, pp. 211–407. doi:10.1561/0400000042.Nasr M., Shokri R., Houmansadr A. Comprehensive privacy analysis of deep learning: passive and active white-box inference attacks against centralized and federated learning. 2019 IEEE Symposium on Security and Privacy (SP), 2019. doi:10.1109/SP.2019.00065.Abadi M., Chu A., Goodfellow I., et al. Deep learning with differential privacy. Proceedings of the 2016 ACM SIGSAC Conference on Computer and Communications Security (CCS), 2016, pp. 308–318. doi:10.1145/2976749.2978318.Dwork C., Roth A. The algorithmic foundations of differential privacy. Foundations and Trends in Theoretical Computer Science, 2014, vol. 9, no. 3–4, pp. 211–407. doi:10.1561/0400000042.Mironov I. Rényi differential privacy. 2017 IEEE 30th Computer Security Foundations Symposium (CSF), 2017, pp. 263–275. doi:10.1109/CSF.2017.11.Abadi M., Chu A., Goodfellow I., et al. Deep learning with differential privacy. Proceedings of the 2016 ACM SIGSAC Conference on Computer and Communications Security (CCS), 2016, pp. 308–318. doi:10.1145/2976749.2978318.Wang Y.-X., Balle B., Kasiviswanathan S.P. Subsampled Rényi differential privacy and analytical moments accountant. In: Proceedings of the 22nd International Conference on Artificial Intelligence and Statistics (AISTATS 2019). PMLR, vol. 89, 2019. Available at: https://proceedings.mlr.press/v89/wang19a.htmlMironov I. Rényi differential privacy. 2017 IEEE 30th Computer Security Foundations Symposium (CSF), 2017, pp. 263–275. doi:10.1109/CSF.2017.11.Geyer R.C., Klein T., Nabi M. Differentially private federated learning: a client level perspective. arXiv:1712.07557, 2017. doi:10.48550/arXiv.1712.07557.Wang Y.-X., Balle B., Kasiviswanathan S.P. Subsampled Rényi differential privacy and analytical moments accountant. In: Proceedings of the 22nd International Conference on Artificial Intelligence and Statistics (AISTATS 2019). PMLR, vol. 89, 2019. Available at: https://proceedings.mlr.press/v89/wang19a.htmlBonawitz K., Ivanov V., Kreuter B., et al. Practical secure aggregation for privacy-preserving machine learning. Proceedings of the 2017 ACM SIGSAC Conference on Computer and Communications Security (CCS), 2017, pp. 1175–1191. doi:10.1145/3133956.3133982.Geyer R.C., Klein T., Nabi M. Differentially private federated learning: a client level perspective. arXiv:1712.07557, 2017. doi:10.48550/arXiv.1712.07557.Bagdasaryan E., Veit A., Hua Y., Estrin D., Shmatikov V. How to backdoor federated learning. In: Proceedings of AISTATS, 2020. arXiv:1807.00459. doi:10.48550/arXiv.1807.00459.Bonawitz K., Ivanov V., Kreuter B., et al. Practical secure aggregation for privacy-preserving machine learning. Proceedings of the 2017 ACM SIGSAC Conference on Computer and Communications Security (CCS), 2017, pp. 1175–1191. doi:10.1145/3133956.3133982.Sharafaldin I., Lashkari A.H., Ghorbani A.A. Toward generating a new intrusion detection dataset and intrusion traffic characterization. Proceedings of ICISSP, 2018, pp. 108–116. doi:10.5220/0006639801080116.Bagdasaryan E., Veit A., Hua Y., Estrin D., Shmatikov V. How to backdoor federated learning. In: Proceedings of AISTATS, 2020. arXiv:1807.00459. doi:10.48550/arXiv.1807.00459.Canadian Institute for Cybersecurity, University of New Brunswick. CSE-CIC-IDS2018 Dataset. Electronic resource. Available at: https://www.unb.ca/cic/datasets/ids-2018.htmlSharafaldin I., Lashkari A.H., Ghorbani A.A. Toward generating a new intrusion detection dataset and intrusion traffic characterization. Proceedings of ICISSP, 2018, pp. 108–116. doi:10.5220/0006639801080116.CERT Division, Software Engineering Institute, Carnegie Mellon University. Insider Threat Test Dataset (CERT), version 6.2. Electronic resource. Available at: https://resources.sei.cmu.edu/library/asset-view.cfm?assetid=508099Canadian Institute for Cybersecurity, University of New Brunswick. CSE-CIC-IDS2018 Dataset. Electronic resource. Available at: https://www.unb.ca/cic/datasets/ids-2018.htmlCERT Division, Software Engineering Institute, Carnegie Mellon University. Insider Threat Test Dataset (CERT), version 6.2. Electronic resource. Available at: https://resources.sei.cmu.edu/library/asset-view.cfm?assetid=508099CERT Division, Software Engineering Institute, Carnegie Mellon University. Insider Threat Test Dataset (CERT), version 6.2. Electronic resource. Available at: https://resources.sei.cmu.edu/library/asset-view.cfm?assetid=508099

The authors declare that there are no conflicts of interest present.