FPGA-based Machine Learning
Field-programmable gate array (FPGA) technology is eminently applicable to a large class of computing problems in general and machine learning in particular. Compared with other technologies including central processing units, digital signal processors, graphics processing units and application specific integrated circuits, FPGAs offer unique Enhancement, Parallelism, Integration and Customisation (EPIC) opportunities, leading to improved latency, Size, Weight, Power and Cost (SWaP-C). This white paper describes the benefits of FPGAs for machine learning at the edge and is intended for non-specialists.
Cybersecurity Analysis Neural Engine
A fundamental challenge with cyber security systems is the associated requirement to perform sophisticated data analysis at high speed. While machine learning (ML) is effective at addressing many Cyber problems, its computational complexity often makes its implementation infeasible at line rates. Signature-based intrusion detection systems (IDSs) identify known attacks and fall into the misuse detection class. Machine Learning approaches that learn the behaviour of the traffic flow fall into the class of anomaly detectors. We propose CruxML CANE, an FPGA-based IDS which achieves line rate speeds and combines signature (existing capability) and anomaly based (proposed capability) detectors. Compared with software implementations on processors and GPUs, our hybrid IDS (HIDS) system is more secure, accurate and performant. Moreover, it has a greatly reduced attack surface as the insertion of malicious code, injection attacks and viruses do not have an FPGA counterpart.