System Log Anomaly Detection Based on Spiking Neural Network Trained with Backpropagation

System log analysis is one of the most important ways to defend attacks and avoid system exceptions. However, rule-based log anomaly detection cannot meet the requirements of complex systems. Log anomaly detection methods based on machine learning and deep learning have gradually become a research hotspot. However, for some complex mobile devices, such as unmanned aerial vehicle (UAV) and intelligent vehicles, log monitoring systems based on complex deep learning or machine learning could be with high power consumption in local anomaly detection. It is an urgent problem to be solved for such energy limited devices. In recent years, brain-like intelligence based on spiking neural networks (SNN) has been proved to have the potential for accurate detection with low power consumption. In this study, we proposed a log anomaly detection method based on SNN, called Log-SNN. This method parsed and converted text logs into embedding vectors with template and parameter information. Then, an SNN model based on backpropagation (BP) training (SNN-BP) was used to detect the log information. We evaluated the proposed method on three commonly used log anomaly datasets and proved the proposed method has higher accuracy than machine learning and deep learning algorithms. The algorithm has high inference speed and low power consumption. The Log-SNN proposed in this study provides an accurate, high-speed, and low-power feasible solution for local log anomaly detection of mobile devices.