Speaker
Description
Quantum devices in the current Noisy Intermediate-Scale Quantum (NISQ)
era are inherently affected by noise [1], which degrades the
predictive performance of quantum machine learning models [2]. This
motivates the development of theoretical tools for analysing robustness
under realistic noise. In this work, motivated by the established links
between robustness and generalisation theory [3], we introduce the
cross-kernel margin, a novel margin-based measure of robustness
for Quantum Kernel-Assisted Support Vector Machines (QSVMs). The
cross-kernel margin quantifies the alignment between the learned
solution of a classifier under noise and the solution obtained in the
ideal noiseless setting. We derive analytical upper and lower bounds for
the cross-kernel margin, thereby
characterising noise-induced margin degradation. These bounds are
validated through numerical simulations across multiple datasets and
experiments performed on real quantum hardware. We further present
empirical results linking margin-based quantities with the
generalisation performance of QSVMs, providing additional motivation for
our margin-based robustness analysis. Finally, we empirically compare
the degradation of the test accuracy obtained under the local
depolarising noise model with the widely used global depolarising noise
model. Together, these results provide a theoretical framework for
analysing the robustness of quantum kernel methods in the NISQ era.
References:
[1] John Preskill. Quantum, 2:79, 2018
[2] Bhavna Bose and Muhammad Faryad. arXiv preprint
arXiv:2604.11541, 2026.
[3] Kenji Kawaguchi, Zhun Deng, Kyle Luh, and Jiaoyang Huang. In
International Conference on Machine Learning, pages 10866--10894. PMLR,
2022.
| Topic | Quantum kernel methods and their applications |
|---|---|
| Additional Topic | Quantum-enhanced robustness in machine learning models |