Machine learning offers a powerful avenue for improving noisy quantum devices. This work introduces a convolutional neural network that learns the disorder landscape of a disordered extended Hubbard model directly from charge-stability diagrams. By processing site-specific disorder in hopping terms, on-site potentials and Coulomb interactions simultaneously, the network predicts disorder parameters with very high accuracy (R² ≈ 0.994) and reduces the number of constraints compared with previous approaches. The method enables automated tuning of five or more quantum dots and mitigates crosstalk between device parameters.

---

← Back to publications