Can a neural network learn your quantum computer's quirks from scratch?

Quantum processors like transmon-based chips suffer from subtle, hardware-specific errors that are hard to characterize without thousands of measurements. This work shows that a neural network can learn a compact model of those errors from as few as 12 tomography measurements — and that plugging this model into a quantum optimization algorithm reduces its error by roughly 20× for two qubits. Scaling to three qubits, simple linear regression performs nearly as well, suggesting practical error-aware calibration doesn't require deep hardware access.