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

Quantum information transfer is an processing technology with high speed and entanglement the help of quantum mechanics principles. To solve problem getting easily lost during transmission, we choose topological error correction codes as best candidate to improve fidelity information. The stability brings great convenience correction. represented by surface have produced very good effects in mechanism. In order strong spatial correlation optimal decoding codes, introduced a reinforcement learning decoder that can effectively characterize codes. At same time, use double-layer convolutional neural network model confrontation find better chain, generation approach model, ensure discriminant corrects more nontrivial errors. efficiency correction, double-Q algorithm ResNet increase success rate training code. Compared previous MWPM 0.005 threshold, has slightly improved, which reach up 0.0068 threshold. By using residual architecture, saved one-third time increased accuracy about 96.6%. Using successfully threshold from 0.0085, depolarized noise being used does not require priori basic noise, so entire improved. Finally, improved 0.2423 0.7423 protection schemes.