Updated on 2024/02/28


XIE, Tianying
Faculty of Science and Engineering, School of Fundamental Science and Engineering
Job title
Research Associate

Internal Special Research Projects

  • Available but Invisible: Privacy-Preserving Techniques for Federated Learning in Differential Privacy


     View Summary

    The primary objective of this research is to develop and evaluate privacy-preserving techniques for federated learning that adhere to the principles of differential privacy. The aim is to allow sensitive data to remain available for deep learning models, especially generative models, while ensuring the privacy of the training dataset’s information is maintained, effectively making the data invisible to unauthorized entities, which is also called privacy computing. The study employs a two-fold approach: (1) designing novel privacy-preserving algorithms for federated learning that meet the requirements of Gaussian mechanism-differential privacy, and (2) analyzing and quantifying the trade-offs between metrics (FID, IS, etc.) and privacy budget (Epsilon) in the proposed methods. The research begins by conducting a thorough review of existing literature on differential privacy and federated learning. Based on this review, the study proposes a set of privacy-preserving algorithms tailored for federated learning scenarios. These algorithms are rigorously analyzed, both theoretically and empirically, to assess their performance and privacy guarantees. Furthermore, the research conducts experiments on real-world datasets to evaluate the practical implications of the proposed techniques. The results are compared with existing state-of-the-art methods to demonstrate the effectiveness and efficiency of the developed privacy-preserving techniques for federated learning in differential privacy.We submitted the results to the Computer Security Symposium 2023 and ACM AsiaCCS 2024.