Edinburgh Napier University

  Digital Infrastructures are complex systems that are built upon computing and communication hardware, and whose resilience is undermined by the security of its hardware building blocks, which has not received much consideration in the past. A focus on security is the key to enhance the resilience of the digital infrastructure. The massive deployment of smart IoT devices (which are driven by on-device Deep Learning (DL) systems) in the global campaign for building digital infrastructures, however, has unveiled an extensive attack surface threatening the security, and resilience of digital infrastructures.

Side-channel information, such as observable power/electromagnetic emissions or timing, is generated by the IoT devices during data computation. Side-Channel Analysis (SCA) attacks can observe the emissions or timing and correlate them with the internal activities and data. This unintentional leak of information allows the divulging of important internal details (e.g., a DL model’s structure and parameters) and exposing private information (e.g., IP and sensitive input data). The leaked information also can help adversaries trick the on-device DL systems more easily toward misclassifications. Even though significant effort has been made to enable efficient DL implementations on resource-constrained IoT devices and secure data sharing, protecting DL implementations is less attended.

This project aims to enhance the resilience of digital infrastructures by setting out to develop a secure DL implementation scheme for smart IoT devices to defend against SCA attacks on the hardware and software components of digital infrastructures.