Abstract
Differential privacy (DP) is a widely used technique for securing the privacy of data in deep learning (DL). However, despite its broad adoption, the effectiveness of DP in safeguarding privacy has not yet been challenged. This paper introduces a novel approach, a fast and robust DP optimizer for general DL tasks, which addresses the limitations of existing methods and offers a fresh perspective on the issue. We first examine the convergence condition of current Differential Privacy (DP) algorithms. Then, we propose a new general optimization algorithm called ACNSI to reduce the effect of noise. ACNSI comprises five steps: affine transformation, clipping, noising, smoothing, and inverse affine transformation. By using affine and inverse affine transforms, we can significantly reduce the added Gaussian noise, which not only improves existing differential private stochastic gradient descent methods but also holds promise for enhancing the practical application of DP in DL tasks. We provide a theoretical analysis for privacy-preserving properties and utility improvement with noise reduction and smoothing effects. Finally, we conduct an empirical study to demonstrate that ACNSI achieves the best accuracy with a tight privacy budget and economical computational cost compared to existing methods.
Supported by the National Key R&D Program of China (No. 2023YFB2703700).
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Similar content being viewed by others
References
Abadi, M., et al.: Deep learning with differential privacy. In: CCS, pp. 308–318 (2016)
Bassily, R., Feldman, V., Talwar, K., Thakurta, A.G.: Private stochastic convex optimization with optimal rates. In: NeurIPS, pp. 11282–11291 (2019)
Bassily, R., Smith, A., Thakurta, A.: Private empirical risk minimization: efficient algorithms and tight error bounds. In: FOCS, pp. 464–473. IEEE (2014)
Bertsekas, D.P., Tsitsiklis, J.N.: Parallel and Distributed Computation: Numerical Methods, vol. 23. Prentice hall Englewood Cliffs, NJ (1989)
Blocki, J., Blum, A., Datta, A., Sheffet, O.: The Johnson-lindenstrauss transform itself preserves differential privacy. In: 2012 IEEE 53rd Annual Symposium on Foundations of Computer Science, pp. 410–419. IEEE (2012)
Bu, Z., Dong, J., Long, Q., Su, W.J.: Deep learning with gaussian differential privacy. arXiv preprint arXiv:1911.11607 (2019)
Chaudhuri, K., Monteleoni, C., Sarwate, A.D.: Differentially private empirical risk minimization. J. Mach. Learn. Res. 12(3), 1069–1109 (2011)
Chen, L., Hu, X., Wu, H.: Randomized fast subspace descent methods. arXiv preprint arXiv:2006.06589 (2020)
Chen, Y., Qin, X., Wang, J., Yu, C., Gao, W.: FedHealth: a federated transfer learning framework for wearable healthcare. IEEE Intell. Syst. 35, 83–93 (2020)
Davody, A., Adelani, D.I., Kleinbauer, T., Klakow, D.: Robust differentially private training of deep neural networks. arXiv preprint arXiv:2006.10919 (2020)
Dong, J., Roth, A., Su, W.J.: Gaussian differential privacy. arXiv preprint arXiv:1905.02383 (2019)
Dwork, C., Roth, A., et al.: The algorithmic foundations of differential privacy. Found. Trends Theor. Comput. Sci. 9(3–4), 211–407 (2014)
Geyer, R.C., Klein, T., Nabi, M.: Differentially private federated learning: a client level perspective. arXiv preprint arXiv:1712.07557 (2017)
Haddadpour, F., Mahdavi, M.: On the convergence of local descent methods in federated learning. arXiv preprint arXiv:1910.14425 (2019)
Horn, R.A., Johnson, C.R.: Matrix Analysis. Cambridge University Press (2012)
Jin, C., Ge, R., Netrapalli, P., Kakade, S.M., Jordan, M.I.: How to escape saddle points efficiently. In: ICML, pp. 1724–1732. PMLR (2017)
Mironov, I.: Rényi differential privacy. In: 2017 IEEE 30th Computer Security Foundations Symposium (CSF), pp. 263–275. IEEE (2017)
Nesterov, Y.: Gradient methods for minimizing composite functions. Math. Program. 140(1), 125–161 (2013)
Nikolov, A., Talwar, K., Zhang, L.: The geometry of differential privacy: the sparse and approximate cases. In: Proceedings of the Forty-fifth Annual ACM Symposium on Theory of Computing, pp. 351–360 (2013)
Osher, S., et al.: Laplacian smoothing gradient descent. arXiv preprint arXiv:1806.06317 (2018)
Pichapati, V., Suresh, A.T., Yu, F.X., Reddi, S.J., Kumar, S.: AdaCliP: adaptive clipping for private SGD. arXiv preprint arXiv:1908.07643 (2019)
Saad, Y.: Iterative Methods for Sparse Linear Systems. SIAM (2003)
Shang, F., Zhang, Z., Xu, T., Liu, Y., Liu, H.: Principal component analysis in the stochastic differential privacy model. In: UAI, pp. 1110–1119. PMLR (2021)
Shen, D., Wu, G., Suk, H.I.: Deep learning in medical image analysis. Annu. Rev. Biomed. Eng. 19, 221–248 (2017)
Shokri, R., Stronati, M., Song, C., Shmatikov, V.: Membership inference attacks against machine learning models. In: 2017 IEEE Symposium on Security and Privacy (SP), pp. 3–18. IEEE (2017)
Upadhyay, J.: Randomness efficient fast-Johnson-lindenstrauss transform with applications in differential privacy and compressed sensing. arXiv preprint arXiv:1410.2470 (2014)
Wang, B., Gu, Q., Boedihardjo, M., Wang, L., Barekat, F., Osher, S.J.: DP-LSSGD: a stochastic optimization method to lift the utility in privacy-preserving ERM. In: Mathematical and Scientific Machine Learning, pp. 328–351. PMLR (2020)
Wang, D., Ye, M., Xu, J.: Differentially private empirical risk minimization revisited: faster and more general. arXiv preprint arXiv:1802.05251 (2018)
Wright, J., Ma, Y.: High-Dimensional Data Analysis with Low-Dimensional Models: Principles, Computation, and Applications. Cambridge University Press (2022)
Xiao, H., Rasul, K., Vollgraf, R.: Fashion-MNIST: a novel image dataset for benchmarking machine learning algorithms. arXiv preprint arXiv:1708.07747 (2017)
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2025 The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.
About this paper
Cite this paper
Wu, H. (2025). Fast and Robust Differential Private Stochastic Gradient Descent with Preconditioner. In: Wu, S., Su, X., Xu, X., Kang, B.H. (eds) Knowledge Management and Acquisition for Intelligent Systems. PKAW 2024. Lecture Notes in Computer Science(), vol 15372. Springer, Singapore. https://doi.org/10.1007/978-981-96-0026-7_15
Download citation
DOI: https://doi.org/10.1007/978-981-96-0026-7_15
Published:
Publisher Name: Springer, Singapore
Print ISBN: 978-981-96-0025-0
Online ISBN: 978-981-96-0026-7
eBook Packages: Computer ScienceComputer Science (R0)