Securing Execution of Neural Network Models on Edge Devices
Neural network model deployment in the cloud may not be feasible or effective in many cases. When the application cannot tolerate the round-trip latency associated with calls to a remote cloud server, edge computation is often the only viable solution. The models are often trained using private datasets that are very expensive to collect, or highly sensitive. They are commonly exposed either through online APIs, or used in hardware devices deployed in the field and available to the end users. Such access provides malicious parties with opportunities to steal these ML models as a proxy for gathering the underlying datasets. While API-based model exfiltration has been studied before, the theft and protection of machine learning models on hardware devices have not been examined. In this work, we develop a hardware module named Trusted Inference Engine (TIE), and an anonymous authentication model distribution protocol that allows designers to securely distribute their models without the risk of exfiltration. The engine protects non-volatile memory against probing attacks and prevents API-based extraction by ensuring rate-limiting operations. With its associated anonymous authentication protocol, it fulfills the desired functionality for authentication and privacy while providing strong security guarantees for edge deployments.
Michael A. Kinsey
Michel A. Kinsy is an Assistant Professor in the Department of Electrical and Computer Engineering at Boston University (BU), where he directs the Adaptive and Secure Computing Systems (ASCS) Laboratory. He focuses his research on computer architecture, hardware-level security, and neural network accelerator designs. Dr. Kinsy is an MIT Presidential Fellow, the 2018 IEEE MWSCAS Myril B. Reed Best Paper Award Recipient, DFT'17 Best Paper Award Finalist, and FPL'11 Tools and Open-Source Community Service Award Recipient. Dr. Kinsy earned his PhD in Electrical Engineering and Computer Science in 2013 from the Massachusetts Institute of Technology. His doctoral work in algorithms to emulate and control large-scale power systems at the microsecond resolution inspired further research by the MIT spin-off Typhoon HIL, Inc. Before joining the BU faculty, Dr. Kinsy was an assistant professor in the Department of Computer and Information Systems at the University of Oregon, where he directed the Computer Architecture and Embedded Systems (CAES) Laboratory. From 2013 to 2014, he was a Member of the Technical Staff at the MIT Lincoln Laboratory.
Host: Professor Patrick Schaumont