Data-driven optimal control of quantum gates

Improving the fidelity of quantum gates is essential for advancing quantum computing applications, though it is often hindered by experimental burden and error interference in closed-loop optimization. In this work, we implement a quantum-classical hybrid optimization process, combining a classical simulator with experimental hardware to iteratively improve gate fidelity by refining the control-pulse envelope. We further employ machine-learning-assisted protocols to reconstruct quantum processes and mitigate state-preparation-and-measurement errors, requiring significantly fewer measurements compared to standard tomography methods. Using a gradient-based optimizer, we achieve high-fidelity two-qubit gates for superconducting qubits. Our results demonstrate a substantial reduction in measurement cost and data overhead, providing an efficient framework for enhancing gate performance in practical quantum computing applications.