Quantum computing relies on developing quantum devices that are robust against small and uncontrolled parameter variations in the Hamiltonian. We focus on real-time closed-loop feedback protocols for fast estimation of stochastic fluctuations of superconducting qubit Hamiltonian and decoherence parameters [1, 2].

We present adaptive Bayesian schemes for efficiently tracking frequency [3] and relaxation times [4] fluctuations in superconducting qubits. In real time, we implement the Bayesian algorithm to estimate low-frequency magnetic flux noise in a flux-tunable transmon qubit, achieving exponential scaling in calibration precision with the number of measurements [3], up to the limit imposed by decoherence. The algorithm is validated by improved coherence and single-qubit gate fidelity through feed-forward of the updated qubit frequency.

We also perform fast estimation of relaxation times averaging 0.17 ms and occasionally exceeding 0.5 ms in only a few milliseconds [4], more than two orders of magnitude faster than previous nonadaptive methods. We observe telegraphic relaxation time fluctuations up to 10 Hz, four orders of magnitude faster than previously measured.

These works emphasize the need for online Hamiltonian learning to enhance the performance and stability of quantum devices affected by quasistatic noise, and to quickly identify the lowest-performing qubit outliers in quantum processing units.