Q-ShiftDP: A differentially private parameter-shift rule for quantum machine learning

The researchers propose the differentially private parameter-shift rule Q-ShiftDP, which the authors call the first privacy mechanism customized for quantum machine learning (QML). The method exploits the bounded and random nature of quantum gradients under the parameter-shift rule to achieve tighter sensitivity analysis and reduce noise requirements, combines calibrated Gaussian noise with intrinsic quantum noise, and provides formal privacy and utility guarantees. Experiments on benchmark datasets show that Q-ShiftDP consistently outperforms classical differential privacy methods such as DP-SGD in QML.