An Intrusion Detection System Using Vision Transformer for Representation Learning

Intrusion Detection System (IDS) is important in safeguarding cybersecurity by identifying and responding to malicious activities. Traditional IDSs filter the abnormal traffic through rules or learn the behaviors of normal and abnormal network data. Nevertheless, these methods utilize the manually designed feature set that introduces limitations in this field. Machine learning shows advantages in the traffic classification domain but still faces challenges of computing resource consumption and a high false positive rate. This paper presents an innovative approach to lightweight IDS using vision transformer techniques for feature representation learning in the context of network intrusion detection. Specifically, our IDS uses a self-attention mechanism to process network traffic, which flattens the splitted network flow to images for training the model. It utilizes Natural Language Processing techniques to capture temporal-spatial information from network traffic. We conduct several experiments to show the effectiveness of our proposed method. The results show that our approach can achieve high accuracy in intrusion detection tasks and keep the false positive rate very low at the same time. The findings highlight the potential of vision transformers in IDS and contribute to the development of robust network security solutions for critical domains like civil aviation.