Quantum Information Protection Scheme Based on Reinforcement Learning for Periodic Surface Codes

Figure 5

We used two decoders at d = 3 (marked with pink squares), d = 5 (marked with blue circles), and d = 7 for the ResNet network (marked with yellow triangles) as training models. (a) The variation of the logical error rate of the MWPM decoder for different d values, and the zoomed-in plot shows the variation of the threshold at around 0.005. The logical error rate increases gradually with the increase of the physical error rate. (b) The variation of the CNN decoder logical error rates for different d values with a threshold value of 0.0068.

(a)

(b)