Privacy-Preserving AI: A Survey of Encrypted Techniques

Abstract

The increasing integration of artificial intelligence (AI) into various domains has underscored the critical need for preserving privacy in AI applications. This survey explores the landscape of privacy-preserving techniques in AI, focusing on encrypted methods that safeguard sensitive data while enabling robust machine learning models. We categorize and examine a range of techniques, including homomorphic encryption, secure multi-party computation, differential privacy, and federated learning, highlighting their principles, advantages, and limitations. The survey provides a comparative analysis of these methods in terms of computational efficiency, security guarantees, and applicability to different AI tasks. We also discuss current challenges and future directions, emphasizing the importance of balancing privacy with performance. This comprehensive review aims to guide researchers and practitioners in selecting appropriate privacy-preserving techniques for their AI applications, fostering the development of secure and trustworthy AI systems.