AI and the World's Hard Problems · Quantum physics & quantum computing · arXiv · 2026-10-19
Risk-aware neural pulse compilation of continuously parameterized single-qubit gates with optimal-control benchmarking and NMR hardware validation
The paper presents a physics-informed neural network that learns a reusable map from continuously parameterized single-qubit gate targets to control pulses, trained end-to-end without optimal-pulse labels and refined with risk-aware fine-tuning for robustness. On a three-qubit liquid-state NMR model, it reached mean nominal fidelity above 0.99 across Haar-distributed targets and generated new pulses in submillisecond inference time, while robust GRAPE achieved higher per-target fidelity. Risk-aware training widened tolerance to RF-amplitude errors, and an NMR hardware experiment reproduced this trend, though absolute correlations were lower than in simulation.