New quantum noise fingerprinting method achieves 84% accuracy

Accurate noise classification is critical for operating near-term quantum processors but traditional methods like quantum process tomography scale exponentially with system size. In a new paper on arXiv, researchers Vridhi Jain and Lei Zhang introduce a scalable noise fingerprinting pipeline that combines structured classical shadow tomography with physics-informed feature engineering. They use 3-qubit probe circuits and construct 279-dimensional feature vectors from randomized Pauli measurements and derived observables to resolve physically similar noise channels that produce overlapping signatures under generic measurement sets.

Evaluated on a dataset of 14,000 labeled samples spanning 10 noise types, three classifiers were tested: random forest, extra trees, and a multilayer perceptron. The random forest achieved the best test accuracy of 84.26% with a macro F1 score of 0.8437. Confusion analysis shows many noise types classified with high reliability; remaining confusions occur between channels sharing similar physical decay mechanisms. This work motivates future research on richer probe states and noise parameter estimation, offering a practical path toward routine quantum processor calibration without exponential overhead.