A Review of Homomorphic Encryption Applications in Deep Learning

Junhao Wu

doi:10.54097/257h8077

Authors

Junhao Wu

DOI:

https://doi.org/10.54097/257h8077

Keywords:

Deep Learning, Homomorphic Encryption, Privacy Preservation

Abstract

In the era of artificial intelligence, deep learning has achieved success in fields such as image recognition and natural language processing, but the training process, which relies on sensitive data, raises privacy risks. Traditional privacy protection techniques, such as differential privacy and secure multi-party computation, either result in accuracy loss or high computational overhead. Homomorphic encryption (HE) can perform operations directly on ciphertext without decrypting the data, providing a new approach for deep learning to protect data privacy. Microsoft's CryptoNets system in 2016 first verified the feasibility of performing neural network inference on encrypted data, leading to a research boom combining Homomorphic encryption with deep learning. This paper systematically reviews the latest progress in combining homomorphic encryption with deep learning, introducing typical works from three aspects: encrypted inference, encrypted training, and hybrid methods, analyzing performance bottlenecks, and looking forward to future development directions.

Downloads

Download data is not yet available.

References

[1] Gentry, C. (2009). Fully homomorphic encryption using ideal lattices. Proceedings of the 41st Annual ACM Symposium on Theory of Computing (STOC), 9, 169–178.

[2] Fan, J., & Vercauteren, F. (2012). Somewhat practical fully homomorphic encryption. IACR Cryptology ePrint Archive, 2012, Article 144.

[3] Cheon, J.H., Kim, A., Kim, M., & Song, Y. (2017). Homomorphic encryption for arithmetic of approximate numbers. Advances in Cryptology – ASIACRYPT 2017, 10624, 409–437.

[4] Chillotti, I., Gama, N., Georgieva, M., & Izabachène, M. (2020). TFHE: Fast fully homomorphic encryption over the torus. Journal of Cryptology, 33(1), 34–91.

[5] Brakerski, Z., & Vaikuntanathan, V. (2021). Fully homomorphic encryption from learning with errors. SIAM Journal on Computing, 50(2), 628–657.

[6] Gilad-Bachrach, R., Dowlin, N., Laine, K., Lauter, K., Naehrig, M., & Wernsing, J. (2016). CryptoNets: Applying neural networks to encrypted data with high throughput and accuracy. Proceedings of the 33rd International Conference on Machine Learning (ICML), 48, 201–210.

[7] Juvekar, C., Vaikuntanathan, V., & Chandrakasan, A. (2018). Gazelle: A low latency framework for secure neural network inference. Proceedings of the 27th USENIX Security Symposium, 2018, 1651–1669.

[8] Boemer, F., Lao, Y., Wierzynski, C., & Naehrig, M. (2019). nGraph-HE: A graph compiler for deep learning on homomorphically encrypted data. Proceedings on Privacy Enhancing Technologies, 2019(1), 3–24.

[9] Dowlin, N., Gilad-Bachrach, R., Laine, K., Lauter, K., Naehrig, M., & Wernsing, J. (2019). Manual and automated homomorphic evaluation of neural networks. Proceedings of the 2019 ACM SIGSAC Conference on Computer and Communications Security, 2019, 135–152.

[10] Kim, M., Song, Y., Wang, S., Xia, Y., & Jiang, X. (2021). Secure logistic regression based on homomorphic encryption: Design and evaluation. IEEE Access, 9, 109021–109035.

[11] Bourse, F., Minelli, M., Minihold, M., & Paillier, P. (2018). Fast homomorphic evaluation of deep discretized neural networks. Advances in Cryptology – CRYPTO 2018, 10993, 483–512.

[12] Brutzkus, A., & Gilad-Bachrach, R. (2020). Low latency privacy-preserving inference. Proceedings of the 37th International Conference on Machine Learning Workshops, 2020, 812–821.

[13] He, Z., Zhang, Y., & Lee, J. (2021). Efficient deep learning inference on encrypted data. IEEE Transactions on Parallel and Distributed Systems, 32(6), 1325–1338.

[14] Lou, Q., Jiang, Y., & Zhou, X. (2022). Efficient CNN inference over homomorphically encrypted data with optimized bootstrapping. Neurocomputing, 481, 112–125.

[15] Li, J., & Wang, H. (2023). Hardware acceleration of homomorphic encryption for privacy-preserving artificial intelligence. IEEE Transactions on Circuits and Systems I: Regular Papers, 70(4), 1502–1514.

[16] Wei, X., Zhang, Q., & Xu, L. (2022). Hybrid privacy-preserving deep learning with homomorphic encryption and secure multi-party computation. Computers & Security, 114, 102602.

[17] Chen, X., Wang, Y., & Liu, Z. (2023). ReBoot: Efficient fully homomorphic encryption for neural network training. IEEE Transactions on Information Forensics and Security, 18, 4210–4224.

[18] Wu, T., & Zhang, C. (2024). Privacy-preserving federated learning with homomorphic encryption and differential privacy. IEEE Internet of Things Journal, 11(3), 4125–4138.

[19] Xu, Y., & Liu, H. (2024). Homomorphic encryption-based secure medical image classification. BMC Medical Informatics and Decision Making, 24(1), 56.

[20] Zhao, S., & Li, K. (2025). Towards practical privacy-preserving deep learning: A survey of homomorphic encryption and secure multi-party computation integration. ACM Computing Surveys, 57(1), 1–38.