Efficient quantum state tomography method based on convolutional neural networks

Various reconstruction algorithms of quantum state tomography are sorted out systematically． Combining with MATLAB numerical simulation, the reconstruction effects of linear reconstruction, linear regression estimation, maximum likelihood estimation and deep neural network-based quantum state tomography are compared and analyzed． For 1 to 3 qubits, convolutional neural network (CNN) based reconstruction algorithms achieves a fidelity of > 99.5% with a shorter period of time, which has significant advantages in algorithm complexity and fidelity compared to other classical reconstruction algorithms． Due to the strong fitting ability to complex models, CNN helps to solve the problem of negative eigenvalues in estimated density matrices, making all the estimated density matrices reconstructed with it physically meaningful．