Defending Quantum Machine Learning Models from Adversarial Attacks

Develop practical and effective defense mechanisms for quantum machine learning models, including quantum variational circuits and hybrid quantum–classical architectures, and rigorously establish robustness under standard white-box and black-box adversarial threat models.

Background

The paper surveys recent findings showing quantum classifiers’ susceptibility to adversarial examples and cross-model transferability of attacks, paralleling vulnerabilities observed in classical systems. It emphasizes the importance of robustness for deployment in sensitive domains.

Despite some promising indications of inherent resilience in certain quantum models, the authors note that practical implementations of defenses are limited and that securing QML against adversarial manipulation remains a key open research challenge.