Conversion of an Unsupervised Anomaly Detection System to Spiking Neural Network for Car Hacking Identification

Across industry, there is an increasing availability of streaming, time-varying data, where it is important to detect anomalous behavior. These data are found in an enormous number of sensor-based applications, in cybersecurity (where anomalous behavior could indicate an attack), and in finance. Spiking Neural Networks (SNNs) have come under the spotlight for machine learning applications due to the extreme energy efficiency of their implementation on neuromorphic processors like the Intel Loihi research chip. In this paper we explore the applicability of spiking neural networks for in vehicle cyberattack detection. We show exemplary results by converting an autoencoder model to spiking form. We present a learning model comparison that shows the proposed SNN autoencoder outperforms a One Class Support Vector Machine and an Isolation Forest. Furthermore, only a slight reduction in accuracy is observed when compared to a traditional autoencoder.