Trustworthy AI at NPL

The Data Science and AI Department at NPL develops the tools needed to ensure the traceability of measurement information through AI systems, by addressing the challenges of data provenance and quality, uncertainty quantification, interpretability, software quality and evaluation metrics. We led the writing of the AI section of the MATHMET (European Metrology Network (EMN) for Mathematics and Statistics) Strategic Research Agenda (SRA) which laid out a 10-year roadmap for research. 

NPL is one of the three lead partners of the AI Standards Hub, which was established as part of the UK’s National AI Strategy and which is dedicated to building a vibrant and diverse community around AI standardisation. Our approach to developing AI systems is about ensuring that the entire life cycle, from design, through to development and deployment, is informed by an understanding of risks in context which are monitored through scientifically rigorous metrics. We have recently applied our approach to wearable devices which clinicians can use to screen patients suspected of having atrial fibrillation (AF).

As an impartial arm's length body advisor, we work alongside end users to implement trustworthy AI solutions. We participate in the following European-wide research collaborations:

• Uncertainty quantification methods for ML-based analysis of photoplethysmography (PPG) signals from wearable devices for assessment of cardiovascular health (QUMPHY project).
• Contribution to a new standard on automatic pollen and fungal spore monitoring through the provision of guidance on methods for quantifying accuracy and uncertainty of the ML algorithms involved (BioAirMet project). 
• A standardised and impartial framework for the assessment and validation of image-based AI tools for disease detection (MAIBAI project).

We also work across NPL and the international metrology community to incorporate machine learning into measurement science. We have developed a method for automating and quantifying the uncertainty of the NPL-pioneered self-validating thermocouple (licensed to CCPI Europe under the trade name INSEVA).

Products and services – consultancy

If you have a research collaboration where our expertise can help, please can contact the team and discuss your challenge.

Contact us