Fig. 2: Results of training the quantum end-to-end model.
From: Experimental quantum end-to-end learning on a superconducting processor
a, b The typical training process of the end-to-end model. For better clarity, all data points are averaged over the neighboring four points. c–f The classification performance of the trained model. The horizontal labels show the digits to be classified, while the vertical labels show the majority vote of the computational basis measurement results. The hollow bars (nearly invisible) in the experimental results (c, e) correspond to the standard deviation of multiple repeated measurements. c, d Experimental and simulation results for the 2-digit classification task, respectively. The averaged accuracies for the classification are 0.986 ± 0.001 and 0.982 in the experiment and the simulation, respectively. e, f The 4-digit classification of the QNN. The averaged accuracies are 0.894 ± 0.015 and 0.889 in the experiment and the simulation, respectively.
