机器学习基准

本研究为量子错误纠正（QEC）领域提供了首个全面的机器学习基准。该数据集通过将表面码表示为网格或图形结构，评估了包括卷积神经网络、图形神经网络和变换器在内的七种神经架构在QEC中的应用。数据集的创建旨在捕捉表面码中的隐式长程依赖关系，以提高QEC的准确性。通过广泛的实验，数据集展示了利用远距离辅助量子位信息可以显著提高QEC的性能，特别是在错误检测方面。此数据集的应用领域主要集中在量子计算中的错误纠正，旨在解决现有量子计算机中数据量子位不可靠的问题，确保量子计算系统的稳定性。

This study presents the first comprehensive machine learning benchmark for the field of quantum error correction (QEC). This dataset evaluates the performance of seven neural architectures, including Convolutional Neural Networks (CNNs), Graph Neural Networks (GNNs), and Transformers, for QEC by representing surface codes as grid or graph structures. The dataset was created to capture the implicit long-range dependencies in surface codes, with the goal of improving the accuracy of QEC. Through extensive experiments, this dataset demonstrates that leveraging information from distant ancillary qubits can significantly improve the performance of QEC, particularly in error detection. The application scope of this dataset is mainly focused on error correction in quantum computing, aiming to address the issue of unreliable data qubits in current quantum computers and ensure the stability of quantum computing systems.