Quantum noise protects quantum classifiers against adversaries

Noise in quantum information processing is often viewed as a disruptive and difficult-to-avoid feature, especially near-term technologies. However, noise has played beneficial roles, from enhancing weak signals stochastic resonance to protecting the privacy of data differential privacy. It then natural ask, can we harness power that computing? An important current direction for computing its application machine learning, such classification problems. One outstanding problem learning sensitivity adversarial examples. These are small, undetectable perturbations original where perturbed completely misclassified otherwise extremely accurate classifiers. They also be considered `worst-case' by unknown sources. We show taking advantage depolarisation circuits classification, robustness bound against adversaries derived improves with increasing noise. This property intimately connected an security concept called which extended For protection data, this first protocol used most general adversaries. Furthermore, how classical case sensitive details model, but model absent, thus providing potential independent speedups. opens opportunity explore other ways our favour, well identifying algorithms helpful