Computer Engineering and Applications ›› 2018, Vol. 54 ›› Issue (20): 54-61.DOI: 10.3778/j.issn.1002-8331.1709-0423

Previous Articles     Next Articles

Design and implementation of convolution neural network based on quantum gate group

XU Xingyang, LIU Hongzhi   

  1. School of Computer and Information Engineering, Beijing Technology and Business University, Beijing 100048, China
  • Online:2018-10-15 Published:2018-10-19



  1. 北京工商大学 计算机信息与工程学院,北京 100048

Abstract: In order to improve the training speed of convolution neural network and reduce the training cost, combined with the latest research progress of convolution neural network and quantum computation, a model of Quantum Gate Convolutional Neural Network(QGCNN) is proposed. In order to construct QGCNN network structure, firstly, according to the characteristics of traditional CNN structure, the definition of convolution arithmetic circuit is given. Secondly, the relationship between the weight coefficients of ConvAC is explained by tensor decomposition, which provides the theoretical basis for constructing QGCNN. Then, QGCNN is divided into input presentation layer, hidden layer and output layer, and on this basis, the data is quantized, and a series of operations such as data initialization and network parameter updating are completed by using quantum gate group. Finally, QGCNN is applied to digital handwriting recognition. The experimental results show that the method has a good effect on the accuracy and convergence speed of handwriting recognition.

Key words: quantum computation, quantum neural networks, convolutional neural networks, quantum gates

摘要: 为进一步提高卷积神经网络的训练速度,减少训练成本,建立了量子门组卷积神经网络模型(Quantum Gate Convolutional Neural Network,QGCNN)。为了构建QGCNN网络结构,依据传统CNN结构的特点,给出卷积算术线路(Convolutional Arithmetic Circuit,ConvAC)的定义。用张量分解来说明ConvAC的权值系数之间的关系,为构建QGCNN提供理论依据。将QGCNN分为输入表示层、隐藏层和输出层,在此基础上实现对数据进行量子编码,利用量子门组完成数据初始化,网络参数更新等操作。将QGCNN应用到数字手写体识别中,实验结果表明,该方法在手写体识别的准确率和收敛速度上有不错的效果。

关键词: 量子计算, 量子神经网络, 卷积神经网络, 量子门