Embedded Files
Preserving Privacy in the Age of AI with Homomorphic Encryption
Preserving Privacy in the Age of AI with Homomorphic Encryption
Artificial intelligence and machine learning are unlocking incredible insights from massive datasets, but this comes with real privacy risks. As AI models are trained on credit card records, medical histories, online behavior and more, how can sensitive personal data be kept secure while still enabling the power of AI?
Homomorphic encryption (HE) offers a solution to privacy concerns in AI by enabling computations on encrypted data without decryption [1]. This technology allows AI companies to analyze sensitive information while maintaining data privacy, analogous to performing calculations on data inside a locked box without opening it [1].
HE utilizes advanced mathematics to preserve relationships within encrypted data, enabling useful computations [2]. Various HE approaches exist, ranging from partial schemes for limited operations to fully homomorphic encryption allowing any computation [1], [3].
The applications for privacy-preserving AI using HE are extensive. It can facilitate clinical research and personalized medicine through analysis of multiple encrypted health databases [1]. HE also enables cloud services to process encrypted financial data for fraud detection without accessing sensitive information [2]. Additionally, companies can collaboratively train AI models while protecting proprietary data [2].
However, HE is still developing, with challenges in computational complexity, speed, and usability [3]. Ongoing research is making significant progress in addressing these issues, and the technology is becoming increasingly practical [1], [4]. Nevertheless, HE presents an opportunity to harness the benefits of AI while enhancing data security and privacy protections.
References
References
[1] A. Wood, K. Najarian, and D. Kahrobaei, “Homomorphic Encryption for Machine Learning in Medicine and Bioinformatics,” ACM Computing Surveys, vol. 53, no. 4, pp. 1–35, Aug. 2020, doi: https://doi.org/10.1145/3394658.
[2] M. Brand and G. Pradel, “Practical Privacy-Preserving Machine Learning using Fully Homomorphic Encryption,” Cryptology ePrint Archive (eprint.iacr.org), 2023. Available: https://ia.cr/2023/1320.
[3] B. Liu, M. Ding, S. Shaham, W. Rahayu, F. Farokhi, and Z. Lin, “When Machine Learning Meets Privacy,” ACM Computing Surveys, vol. 54, no. 2, pp. 1–36, Apr. 2021, doi: https://doi.org/10.1145/3436755.
[4] B. Paul, T. K. Yadav, B. Singh, S. Krishnaswamy, and G. Trivedi, “A Resource Efficient Software-Hardware Co-Design of Lattice-Based Homomorphic Encryption Scheme on the FPGA,” IEEE Trans. Comput., vol. 72, no. 5, pp. 1247–1260, May 2023, doi: https://doi.org/10.1109/TC.2022.3198628.
Preserving Privacy in the Age of AI with Differential Privacy
Preserving Privacy in the Age of AI with Differential Privacy
In today's data-driven world, artificial intelligence (AI) has the potential to revolutionise industries and enhance our lives. However, as AI systems become more advanced, concerns about data privacy and security have grown.
Differential privacy is a mathematical framework that protects individual privacy during data analysis while allowing valuable insights to be gained. It works by adding carefully calibrated noise to data, ensuring that the presence or absence of any single individual does not significantly affect the analysis results [1].
This approach offers several key advantages: It is resistant to attacks based on an attacker's background knowledge, as the privacy guarantees hold regardless of what information an attacker may possess. It can handle multiple analyses on the same dataset through the composition theorem, which allows for quantification and control of privacy loss [1].
Differential privacy has applications in various sectors, including healthcare, organizations, and finance [3]. For companies seeking to balance data analytics with customer privacy protection, differential privacy provides a robust solution.
References
References
[1] T. Zhu, D. Ye, W. Wang, W. Zhou, and P. Yu, “More Than Privacy: Applying Differential Privacy in Key Areas of Artificial Intelligence,” IEEE Transactions on Knowledge and Data Engineering, vol. 34, no. 6, pp. 1–1, 2021, doi: https://doi.org/10.1109/tkde.2020.3014246.
[2] M. Gong, Y. Xie, K. Pan, K. Feng, and A. K. Qin, “A Survey on Differentially Private Machine Learning [Review Article],” IEEE Computational Intelligence Magazine, vol. 15, no. 2, pp. 49–64, May 2020, doi: https://doi.org/10.1109/mci.2020.2976185.
[3] K. Pan, Y.-S. Ong, M. Gong, H. Li, A. K. Qin, and Y. Gao, “Differential privacy in deep learning: A literature survey,” Neurocomputing, pp. 127663–127663, Apr. 2024, doi: https://doi.org/10.1016/j.neucom.2024.127663.
Page updated
Report abuse