2022, 39(2): 64-73.
Quantum neural network (QNN) is a research field that combines traditional artificial neural network (ANN) with quantum information and quantum computing. QNN has the advantages of both, which is of theoretical importance to improve the shortcomings of traditional ANN and to essentially improve the performance of neural network. In this paper, we propose an optimized variational quantum neural network (VQNN) model based on the ANN model by combining quantum theory and quantum mechanical concepts such as quantum parallel computing, quantum gate circuits, and variational quantum circuits, which is a quantum-classical hybrid computing model consisting of quantum circuits that can operate on Noisy Intermediate-Scale Quantum (NISQ) devices combined with machine learning (ML) strategies. The quantum circuits consist of two parts: the quantum state encoding circuits for encoding classical data into quantum state data; and the variational quantum circuits (VQC) which simulates the classical probability distribution of the learning target and encodes the information into the quantum circuits parameters. Finally, the classical probability output distribution is obtained by measuring the VQC quantum state output, and the parameters of the variational quantum circuits are optimized by using a classical computer. This structure makes it easy to integrate VQC with classical ML. Further, we explored the use of VQNN to build a classifier based on practical applications to apply it in the field of network attack detection. The experimental results show that for the KDD CUP 99 dataset, VQNN has relatively high detection performance, both of which are higher than other classical and the quantum gate circuits neural network model detection comparison models. In addition, this VQNN can be deployed in most of the recent NISQ devices. Also, our proposed VQNN is the first model that can be deployed in NISQ for network intrusion detection.
