Detecting measurement correlations with graphical models

As quantum information with superconducting circuits scales up, detecting and quantifying correlated errors becomes crucial, as fault tolerance and error correction schemes usually assume uncorrelated errors. Measurement errors, in particular, are likely candidates for correlation, especially in superconducting circuit systems as the readout of multiple qubits is often performed through shared measurement chains to reduce resource requirements (such as number of amplifiers and other cryogenic components). A naive approach to quantify these correlations by reconstructing the distribution of outcomes conditioned on the systems' state scales exponentially, and is therefore impractical. We present an approach that avoids this pitfall by reconstructing a sparse undirected graphical model to approximate the distribution of errors, and demonstrate its use superconducting transmon qubits.