The main objectives of this proposal are threefold: (1) to study the impact of data poisoning in sensed vehicular data and neural network training architectures; (2) to design a reconfigurable accelerator based on an adaptive framework for secure design, implementation, and evaluation of the Internet of Vehicles (IoV); and (3) to drive the reconfigurability of (2) using findings of (1). This project aims to make revolutionary progress to close the gap between the existing security mechanisms (e.g. multi-factor authentication), current decentralized vehicular security solutions (e.g. defense in depth), and the security needs of the IoV data. The project’s closely intertwined research activities include: (1) designing a modular framework for secure implementation of emerging autonomous and connected vehicles, covering deterrent, preventive, detective, corrective, and recovery controls; (2) developing and tuning Deep Learning architectures to classify malicious behaviors and target agents using sensed data as the input; and (3) designing the accelerator hardware to translate the security findings into actionable criteria.
California State University, Fresno