UC Riverside

The increasing adoption of Internet-of-Things (IoT) devices and an explosion in sensor data are fueling frequent data communication between edge devices and intelligence moving towards the Internet edge. As a side effect, the number of potential threats and possible attacks against security and privacy among edge devices has grown drastically. In this dissertation, we focus on strengthening security at the Internet edge, from securing communication between edge devices to achieving trustworthiness of on-device classification. First, we propose PowerKey to secure communication between multiple plugged edge devices in an electrical domain. Concretely, PowerKey generates secret communicating keys for communications between devices pluggedinto nearby power outlets by exploiting electromagnetic interferences (EMI) spikes with randomly varying but consistent frequencies. Second, to achieve secure communications for unplugged edge devices, we propose another secret key generation method, called CompKey, which allows wireless edge devices in the proximity of a third-party computer to securely associate with each other by exploiting electromagnetic radiation (EMR) emitted from the computer. Next, for trustworthy on-device classification, we study the adversarial attacks on brain-inspired hyperdimensional computing (HDC) classifiers. Finally, we consider an ultra-efficient version of HDC classifiers --- low-dimensional computing (LDC) classifiers --- and propose an interval bound propagation (IBP) technique to achieve certified robustness against adversarial attacks subject to L-infinity norm-bounded perturbation.