Event-Driven Intrusion Detection Systems using Spiking Neural Networks for Edge and IoT Security

The rapid growth of Internet of Things (IoT) devices and the increasing difficulty of cyber-attacks highlight the need for novel, energy-efficient, and real-time Intrusion Detection Systems (IDS) capable of running on resource constrained edge devices. In this work, we present a Spiking Neural Networks (SNN) -based event-driven intrusion detection framework inspired by neuromorphic computing principles. Our event-driven detection framework is unique compared to traditional continuous databased IDS because it uses sparse, temporal event-based encoding to detect malicious activity with high precision and low power. We implement a biologically plausible, unsupervised SNN architecture that learns the spatio-temporal patterns of both normal and anomalous behavior using Spike-Timing Dependent Plasticity (STDP) under the concept of continuous learning to keep up with the changing threat landscape. In addition, we provide an innovative technique for spike-based feature encoding that preserves critical temporal elements while relieving some communication overhead. This work presents the first step towards implementing a reliable, low-latency, and energy-efficient neuromorphic security solution at the edge that integrates brain-inspired AI and IoT security applications. We evaluate our proposed framework on benchmark IoT intrusion datasets using a customized neuromorphic simulator that is designed to be deployed at the edge. Our results illustrate that this system is capable of achieving superior detection accuracy (as high as 97.8%), while consuming up to 55% less energy than traditional deep learning-based IDS. Furthermore, the SNN's event-based processing lends itself to real-time detection with latency on the order of milliseconds, making it ideal for low-latency in Cyber-Physical Systems (CPS) and smart edge ecosystems.