Q-SafeML: Safety Monitoring for Quantum Machine Learning
- Aug 2
- 2 min read
Artificial intelligence safety is still an active engineering challenge. As machine learning begins to use quantum computing, those assurance questions become even more important.
At the 9th International Symposium on Model-Based Safety and Assessment (IMBSA 2025) in Athens, Dr Koorosh Aslansefat presented new research from the Dependable Intelligent Systems (DEIS) group at the University of Hull: “Q-SafeML: A Quantum-Statistical Approach to Safety Monitoring in Quantum Machine Learning”.
The paper was co-authored by Oliver Dunn, Dr Koorosh Aslansefat and Professor Yiannis Papadopoulos. It extends ideas from SafeML to quantum machine-learning models.
Machine-learning systems can become unreliable when the data or behaviour encountered during operation differs from what was represented during development. Runtime monitoring helps engineers detect these changes while a system is operating. Q-SafeML adapts this principle to the statistical characteristics of quantum machine learning.
The approach monitors a quantum model at runtime, tracks distribution shifts and behavioural changes, quantifies uncertainty and aims to flag potentially unsafe states early. Rather than assuming that a model remains trustworthy after deployment, it provides continuing evidence about whether its operating conditions and behaviour are changing.
Q-SafeML builds on the wider SafeML research programme. The original SafeML work has been cited in DIN SPEC 92005 on the quantification of uncertainties in machine learning, reflecting the practical relevance of runtime statistical monitoring to trustworthy AI.
Quantum machine learning remains an emerging area, but safety cannot be postponed until the technology is mature. Monitoring and assurance methods need to develop alongside new computational capabilities. By addressing uncertainty and behavioural change now, researchers can establish stronger foundations for future quantum-AI applications.
The work was presented to the model-based safety community at IMBSA 2025, with participants from research and industry, including Airbus and Monohakobi. Discussion and feedback from this community will help guide the next stages of validation and development.
The DEIS team is grateful to the National Edge Artificial Intelligence Hub and colleagues who supported this work. The research contributes to an open and forward-looking programme on runtime safety, dependable AI and the assurance of emerging intelligent systems.
Original LinkedIn update:
Comments