Vehicular Intrusion Detection System Based on Hybrid Quantum Neural Networks

Vehicular Intrusion Detection System Based on Hybrid Quantum Neural Networks

연구 분야: Cryptography

학회: GLOBECOM 2024 - 2024 IEEE Global Communications Conference

Traditional Deep Learning (DL) method is increasingly used in vehicular Intrusion Detection Systems (IDSs). However, there are still some limitations. It combines various models, resulting in a massive model size and numerous parameters, requiring more computational resources. Furthermore, real-time performance is crucial in vehicular IDS. Large models typically require more time for detection, failing to meet the timeliness requirements. Quantum computing offers parallel computing capabilities, prompting our proposal of a Hybrid Quantum Neural Network (HQNN)-based IDS to improve the timeliness of intrusion detection. This IDS employs HQNN to enhance feature extraction and speed up traditional convolution. It retains most of the structure of classical Convolutional Neural Network (CNN), comprising attention layers, batch normalization layers, and quantum circuits. Through quantum superposition and entanglement, certain complex non-linear functions are compressed. Furthermore, residual connections enable seamless gradient flow during back propagation. This accelerates computation and expedites convergence. Comparative experiments on the Car-hacking dataset show improved detection time and convergence speed, with exceptionally high accuracy in practical applications.