Presentation
IP Protection in Tiny ML
SessionUnleashing Privacy for AI
DescriptionMachine learning is a central component of emerging smart microcontrollers. Along with the advances brought by on-device learning, security risks such as model privacy leakage rise significant concerns. However, model protection mechanism deployment can introduce non-negligible overhead, which can undermine the real-time property and accuracy of machine learning tasks. In this work, a secure and timely learning framework for microcontrollers is proposed, which enables optimizing the inference procedures to meet performance and timing requirements while ensuring model privacy with ARM TrustZone. The effectiveness of our framework is demonstrated by the extensive evaluation of heterogeneous machine-learning tasks in microcontrollers.
Event Type
Research Manuscript
TimeTuesday, July 11th1:55pm - 2:10pm PDT
Location3004, 3rd Floor
AI
AI/ML Security/Privacy