Keith Kenemer | Searcharxiv

arXiv2025

Realistic quantum network simulation for experimental BBM92 key distribution

Quantum key distribution (QKD) can provide secure key material between two parties without relying on assumptions about the computational power of an eavesdropper. QKD is performed over quantum links and quantum networks, systems which are resource-intensive to deploy and maintain. To evaluate and optimize performance prior to, during, and after deployment, accurate simulations with attention to physical realism are necessary. Quantum network simulators can simulate a variety of quantum and classical protocols and can assist in quantum network design and optimization by offering realism and flexibility beyond mathematical models which rely on simplifying assumptions and can be intractable to solve as network complexity increases. We use a versatile discrete event quantum network simulator to simulate the entanglement-based QKD protocol BBM92 and compare it to our experimental implementation and to existing theory. We find the discrete event quantum network simulator can match experimental key rates and error rates with a lower mean squared error than analytical theory. Furthermore, we simulate secure key rates in a repeater key distribution scenario for which no experimental implementations exist and find agreement between simulation and analytical theory. Hence, we demonstrate discrete event simulators can meet needs in quantum network simulations which cannot be filled solely by experiment or theory: discrete event simulators can accurately simulate QKD protocols and match experiments in regimes where theoretical models may require more simplifying assumptions, and they can match theoretical models in the opposite scenario where experiments have not yet been performed but theoretical models exist.

quant-ph↗

arXiv2019

Deep Detector Health Management under Adversarial Campaigns

Machine learning models are vulnerable to adversarial inputs that induce seemingly unjustifiable errors. As automated classifiers are increasingly used in industrial control systems and machinery, these adversarial errors could grow to be a serious problem. Despite numerous studies over the past few years, the field of adversarial ML is still considered alchemy, with no practical unbroken defenses demonstrated to date, leaving PHM practitioners with few meaningful ways of addressing the problem. We introduce turbidity detection as a practical superset of the adversarial input detection problem, coping with adversarial campaigns rather than statistically invisible one-offs. This perspective is coupled with ROC-theoretic design guidance that prescribes an inexpensive domain adaptation layer at the output of a deep learning model during an attack campaign. The result aims to approximate the Bayes optimal mitigation that ameliorates the detection model's degraded health. A proactively reactive type of prognostics is achieved via Monte Carlo simulation of various adversarial campaign scenarios, by sampling from the model's own turbidity distribution to quickly deploy the correct mitigation during a real-world campaign.

cs.LG↗