More about HKUST
Privacy-preserving machine learning based on homomorphic encryption
PhD Qualifying Examination Title: "Privacy-preserving machine learning based on homomorphic encryption" by Mr. Yiteng PENG Abstract: The rapid deployment of machine learning (ML) systems in real-world scenarios, particularly in sensitive domains such as healthcare and finance, has raised significant privacy concerns regarding data exposure and privacy leakage during both model training and inference processes. Homomorphic encryption (HE), a kind of cryptographic scheme that allows computations to be performed directly on encrypted data, is considered as a promising solution to address these privacy challenges. However, the practical adoption of HE in ML faces significant obstacles, due to the limited range of supported operations in current HE schemes and the substantial computational overhead in encrypted computations. To overcome these hurdles, recent research focuses on developing efficient adaptations and optimizations techniques. For further facilitating HE integration into ML workflows with these innovative techniques, several HE-ML frameworks have also been developed to enable data scientists to leverage HE's benefits without extensive cryptographic expertise. In this survey, we present a comprehensive review of privacy-preserving ML based on HE. We begin with the fundamental concepts of HE and its applications across diverse ML models. We then examine current adaptation and optimization techniques that enhance the compatibility and efficiency of HE within ML contexts. Finally, we discuss emerging frameworks that abstract the complexities of HE-ML. By synthesizing current knowledge, this survey aims to provide valuable insights for applying HE in ML systems and to stimulate future research in this field. Date: Friday, 25 July 2025 Time: 4:00pm - 6:00pm Venue: Room 3494 Lifts 25/26 Committee Members: Dr. Shuai Wang (Supervisor) Dr. Xiaomin Ouyang (Chairperson) Dr. Binhang Yuan