MHT: Dual-Layered Meta-Driven Hypernetworks for Unknown Attack Detection

With the rapid proliferation of the Internet of Things (IoT), a multitude of vulnerabilities have emerged, which attackers exploit to carry out attacks. Most existing intrusion detection systems struggle to identify unknown attacks, such as zero-day attacks. Recently proposed open-set recognition methods have shown promise in detecting unknown attacks but often misclassify some normal traffic as unknown network flows. To address these issues, we propose Meta-HyperTwin Intrusion Detection System (MHT), a model based on meta-learning and hypernetworks for feature extraction and recognition. Meta-learning effectively adapts to new data samples, while hypernetworks generate neural network parameters, significantly reducing training time and mitigating the risk of overfitting. In our experiments, the proposed approach attains an accuracy of 0.99 and an F1 Score of 0.99 in detecting multiple unknown attacks, and both values exceed 0.95 when identifying most individual unknown attacks.