Data Classification Based On Quantum-Classical Hybrid Convolutional Neural Networks

With the upgrading of computer hardware and the intellectualization of mobile terminals,network technology has become increasingly critical in people’s lives,and then network security is an important research field.Machine learning technology has been widely utilized in the field of network intrusion detection and image processing.At the same time,quantum computing provides a new model to solve complicated computing problems.Consequently,how to enhance the traditional machine learning algorithms by adopting quantum computing has also become a research frontier.Based on the conventional dimension reduction technology and the quantum neural network,a novel model for network intrusion detection was firstly presented in this dissertation to distinguish normal access and abnormal attack.To balance classification accuracy and robustness in the image classification task with a traditional convolutional neural network,a quantum-classical hybrid convolutional neural network model was also designed from the perspective of machine learning model security.With the advent of cloud computing and big data,the network intrusion methods are also steadily diversified.Hence,efficient and generalized intrusion detection technology is necessary to ensure cyber security.In this dissertation,a new intrusion detection model based on principal component analysis and quantum neural networks was explored.There is a certain correlation between the features of the benchmark dataset for network intrusion detection.The redundant components in the dataset were removed and the dimension of the data was reduced with the principal component analysis.Data dimensionality reduction can not only extract useful components from raw data,but also facilitate the training of quantum neural networks.The quantum neural network was equivalent to a classifier,which was mainly utilized to identify normal access and abnormal attacks in the network traffic data.After iterative optimization via the classical optimizer,the quantum neural network can detect the abnormal access in network traffic data.The experimental results verify the feasibility and the validity of the proposed intrusion detection model.As the machine learning model becomes more and more ubiquitous in the actual business systems,the accuracy and the security of the models become increasingly vital.To enhance the performance of classical convolutional neural networks in image classification,a hybrid quantum-classical convolutional neural network model was designed.The quantum convolution layer of the hybrid model was composed of quantum convolution filters,which were exploited to enhance the feature extraction ability of the conventional convolution layer and accelerate the convolution process.It is shown that quantum convolution kernels may not only heighten the accuracy of image classification but also speed up the convergence of the model.In addition,the impact of quantum computing on the robustness of the hybrid quantum-classical convolutional neural network model was investigated as well.It is indicated that quantum computing can enhance the robustness of the hybrid model in the classical adversarial scenarios.