A Survey on Privacy-preserving machine learning with Confidential Computing

PhD Qualifying Examination


Title: "A Survey on Privacy-preserving machine learning with Confidential 
Computing"

by

Mr. Dong CHEN


Abstract:

The rapid growth of data collection and processing has heightened privacy and 
security concerns in machine learning (ML), with frequent attacks exposing 
critical vulnerabilities. Confidential Computing, using Trusted Execution 
Environments (TEE), addresses these challenges by isolating sensitive data and 
computations within a hardware-protected enclave, ensuring confidentiality and 
integrity even with untrusted components. Unlike software-based methods like 
Homomorphic Encryption or Multi-Party Computation, Confidential Computing 
offers enhanced performance while maintaining security, making it increasingly 
popular in both academia and industry for securing ML workloads from training 
to deployment.

This survey explores the evolving role of CC in privacy-preserving machine 
learning. We begin by outlining the potential threats to ML systems and the key 
hardware features of CC technologies. We then review current research on secure 
ML training and inference, highlighting how CC can be effectively used to 
protect both data and models. Our goal is to offer insights into designing 
secure ML systems and to foster future research in this important field.


Date:                   Friday, 30 August 2024

Time:                   1:00pm - 3:00pm

Venue:                  Room 4475
                        Lifts 25/26

Committee Members:      Dr. Wei Wang (Supervisor)
                        Dr. Shuai Wang (Supervisor)
                        Prof. Bo Li (Chairperson)
                        Prof. Kai Chen
                        Prof. Song Guo