Deep Learning-Based Intrusion Detection for Automotive Ethernet: Evaluating & Optimizing Fast Inference Techniques for Deployment on Low-Cost Platform

TL;DR

This paper evaluates fast neural network inference techniques like distilling and pruning to enable real-time deep learning-based intrusion detection on low-cost automotive Ethernet platforms, achieving high accuracy and low latency.

Contribution

It demonstrates the effectiveness of distilling and pruning methods for deploying real-time IDS models on low-cost hardware like Raspberry Pi 4.

Findings

Detection time of 727 microseconds on Raspberry Pi 4

AUCROC of 0.9890 indicating high detection accuracy

Fast inference techniques enable real-time intrusion detection on low-cost platforms

Abstract

Modern vehicles are increasingly connected, and in this context, automotive Ethernet is one of the technologies that promise to provide the necessary infrastructure for intra-vehicle communication. However, these systems are subject to attacks that can compromise safety, including flow injection attacks. Deep Learning-based Intrusion Detection Systems (IDS) are often designed to combat this problem, but they require expensive hardware to run in real time. In this work, we propose to evaluate and apply fast neural network inference techniques like Distilling and Prunning for deploying IDS models on low-cost platforms in real time. The results show that these techniques can achieve intrusion detection times of up to 727 {\mu}s using a Raspberry Pi 4, with AUCROC values of 0.9890.