Establish the reliability of label-free proxies for anomaly-detection performance

Establish how accurately label-free quality proxies, particularly the normalized pseudo-discrepancy (NPD), track detection performance for unsupervised quantum-kernel anomaly detectors.

Background

The paper uses the normalized pseudo-discrepancy (NPD) to select quantum-kernel hyperparameters without labeled validation data. The authors find that NPD identifies a promising region of the hyperparameter space, but its relationship with test AUC is weak within each kernel family and does not provide a consistently fine-grained ranking of configurations.

Because labels are unavailable in the targeted operational setting, determining the extent to which NPD and related label-free criteria reliably predict anomaly-detection quality is an unresolved methodological issue. The problem is relevant beyond the malware-beaconing dataset because the proposed selection protocol is intended for broader unsupervised quantum-kernel applications.

References

Establishing how far label-free proxies track detection performance is an open question that our results suggest deserves dedicated study.

— Pareto-optimal quantum kernel selection for unsupervised anomaly detection on real malware beaconing data  (2610.09717 - Micah et al., 7 Oct 2026) in Section IV, subsection “Fidelity and projected quantum kernels: NPD, GD, and Pareto optimization”