May 26, 2025
Precise knowledge of the error sources of quantum devices can lead to improved design of quantum control protocols and allows for less resource-intensive QEC codes tailored to the device’s noise profile. Learning the noise via tomography is computationally and experimentally intensive for large system sizes. From the perspective of QEC, however, the only necessary information is the one used for decoding. This information is encoded in a detector error model (DEM), a weighted (hyper-)graph, whose weights are the Pauli error rates. We propose a top-down approach for learning the noise of QEC experiments using only the syndrome information. We estimate Pauli noise directly on decoding graphs or hypergraphs of memory QEC experiments and apply our methods on circuit-level simulations of repetition, surface, and color codes. Our methods precisely reconstruct the error rates of a DEM and the expected logical error compared to noise-aware models. We also find an increase in logical error suppression when the decoder exploits information about inhomogeneous qubit and gate error rates. The results are posted on the arXiv: https://arxiv.org/html/2504.20212v1
