Light-weight AI acceleration
Dataflow deep learning accelerators at mixed (reduced) precision for in-network line rate intrusion detection
The project explores the case for line rate intrusion detection using deep learning models for in-vehicle networks such as Controller Area Networks and Ethernet. The key contributions of this research are – Developing highly quantised DL models with state of the art detection performance – Dataflow mapping optimisations and development of custom layers – Tight integration into the Network controller’s datapath without affecting protocol functionality or incurring additional latency in message movement
The project is supported by the Department of EEE through a PhD scholarship.