A degressive quantum convolutional neural network for quantum state classification and code recognition

Qingshan Wu; Wenjie Liu; Yong Huang; Haoyang Liu; Hao Xiao; Zixian Li

doi:10.1016/j.isci.2024.109394

A degressive quantum convolutional neural network for quantum state classification and code recognition

iScience. 2024 Mar 4;27(4):109394. doi: 10.1016/j.isci.2024.109394. eCollection 2024 Apr 19.

Authors

Qingshan Wu¹, Wenjie Liu^{1

2

3}, Yong Huang⁴, Haoyang Liu¹, Hao Xiao⁵, Zixian Li¹

Affiliations

¹ School of Software, Nanjing University of Information Science and Technology, No. 219, Ning Liu Road, Nanjing, Jiangsu 210044, China.
² Jiangsu Collaborative Innovation Center of Atmospheric Environment and Equipment Technology (CICAEET), Nanjing University of Information Science and Technology, No. 219, Ning Liu Road, Nanjing, Jiangsu 210044, China.
³ Jiangsu Province Engineering Research Center of Advanced Computing and Intelligent Services, Nanjing University of Information Science and Technology, No. 219, Ning Liu Road, Nanjing, Jiangsu 210044, China.
⁴ Anhui Province Key Laboratory of Atmospheric Sciences and Satellite Remote Sensing, Anhui Institute of Meteorological Sciences, No.16, Shi He Road, Hefei, Anhui 230002, China.
⁵ School of Information Engineering, Huzhou University, No.759, Second Ring East Road, Huzhou, Zhejiang 313002, China.

Abstract

With the rapid development of quantum computing, a variety of quantum convolutional neural networks (QCNNs) are proposed. However, only $1 / 2^{\frac{n}{2}}$ features of an n-qubits input are transferred to the next layer in a quantum pooling layer, which results in the accuracy reduction. To solve this problem, a QCNN with a degressive circuit is proposed. In order to enhance the ability of extracting global features, we remove the parameters sharing strategy in the quantum convolutional layer and design a quantum convolutional kernel with global eyesight. In addition, to prevent a sharp feature reduction, a degressive parameterized quantum circuit is adopted to construct the pooling layer. Then the Z-basis measurement is only performed on the first qubit to control the operations on other qubits. Compared with the state-of-the-art QCNN, i.e., hybrid quantum-classical convolutional neural network, the accuracy of our model increased by 0.9%, 1%, and 3%, respectively, in three tasks: quantum state classification, binary code recognition, and quaternary code recognition.

Keywords: Physics; Quantum measurement.