Amplitude transformed quantum convolutional neural network

With the rapid development of quantum neural networks (QNN), several quantum simulations of convolutional neural networks (CNN) have been proposed. Among them, Google has proposed three quantum convolutional neural network (QCNN) models, but its purely QCNN model suffers from slow convergence and low training efficiency. In this work, we design low-depth parameterized quantum circuits with only two quantum bits interacting and construct a QCNN framework with lower depths, fewer parameters and global correlation. Based on this, we propose an Amplitude Transformed Quantum Convolutional Neural Network (ATQCNN). Experiments show that our model achieves 100% and 97.92% accuracy and faster convergence on the quantum cluster state and CICMalDroid2020 datasets compared to the purely QCNN proposed by Google. In particular, the required parameters and depth of ATQCNN are in reduced by about 27% for the same scale of qubits. It will be more suitable for current noisy intermediate-scale quantum (NISQ) devices.