An Exhaustive Survey on Privacy Preserving Machine Learning using Homomorphic Encryption and Secure Multiparty Computation Techniques

Abstract

The advent of privacy-preserving machine learning techniques, rooted in Secure Multiparty Computation (SMC) and Homomorphic Encryption (HE), has ushered in a new era of data security and collaborative analytics. This survey paper provides an exhaustive examination of the state-of-the-art in privacy-preserving machine learning, focusing on the innovative applications and advancements brought forth by SMC and HE. Data privacy is paramount in today's data-centric world, and the inherent conflict between sharing data for machine learning and maintaining privacy has spurred the development of privacy-preserving techniques. SMC, a cryptographic approach that enables parties to jointly compute a function over their inputs while keeping those inputs private, has been at the forefront of this endeavour. HE, on the other hand, allows for computations on encrypted data, providing a novel solution for privacy preservation. In this survey, we comprehensively explore the principles of SMC and HE, elucidating their underlying mechanisms and key attributes. We delve into the theoretical foundations and practical implementations of these techniques, offering insights into their strengths and limitations. The survey uncovers a myriad of applications where privacy-preserving machine learning, driven by SMC and HE, has made a significant impact. From healthcare to finance, and from secure data sharing to personalized recommendations, the domains benefiting from these techniques are diverse and expansive. We present case studies and real-world applications that showcase the transformative power of SMC and HE in preserving data privacy while reaping the benefits of machine learning. Furthermore, this paper offers a detailed comparative analysis between SMC and HE in terms of security guarantees, computational overhead, applicability to different deep learning architectures, and scalability. By providing a nuanced understanding of when and how to use these techniques, we empower practitioners and researchers to make informed decisions in selecting the right approach for their specific use cases. In conclusion, the survey paper paints a comprehensive portrait of the dynamic landscape of privacy-preserving machine learning. It underscores the pivotal role of SMC and HE in ensuring data privacy and highlights their potential to revolutionize the way organizations handle sensitive information. As the world becomes increasingly data-driven, these techniques offer a promising path forward, where privacy and innovation coexist harmoniously.