Anomaly Detection

Domain-specific edge devices such as autonomous vehicles often have strict realtime deadlines to meet. To meet high computational needs in an energy efficient manner, systems-on-a-chip (SoCs) are being used to operate such devices. Unfortunately, like most connected systems, SoCs for edge devices are inherently vulnerable to security attacks. Attacks targeting availability of on-chip resources can effectively cripple these systems, preventing the device from meeting realtime needs. I have developed an anomaly detection tool that can efficiently and accurately detect on-chip anomalous activity in realtime, enabling speedy recovery from attacks for these devices. The methodology leverages machine learning techniques to not only detect known resource availability attacks, but also novel future attacks. This work is currently in submission.

This work is part of the DECADES project.