Artificial Intelligence

Artificial Intelligence (AI) is today one of the fastest growing areas of research, impacting many fields from healthcare to environmental science to engineering. In metrology, artificial intelligence can be considered from two sides:  

  • How does the use of AI change metrology?  

  • How can metrology and standardisation help to improve the quality of AI? 

Artificial Intelligence is based on the development and application of algorithms, executed in a dynamic computing environment, that are able to learn from information. These algorithms, including the very popular neural networks, are parametric models with high degrees of freedom offering a general and flexible framework to model a variety of complex physical systems. For example, in classification of images these algorithms have significantly outperformed traditional interpretable models. So-called AI models are therefore increasingly adopted in a wide variety of metrology applications, such as healthcare, sensor networks, autonomous transport, biophysics and mechanical engineering. As an emerging research field, AI often does not meet the standards and requirements for conventional methods of data analysis. 

Producing information that can be perceived as trustworthy is a major challenge in AI, as there is a pressing need to incorporate AI models into quality standards and to use their outputs to make decisions and inform policy. Metrology has a major role to play in this context by: 

  • characterising training data sets (for example addressing uncertainty, mislabelling, labelling noise) 

  • assessing the robustness of AI models (for example through adversarial robustness) 

  • ensuring the traceability of the model outputs (such as providing details about the training process, the validation data, the test data, the model architecture, subsequent decisions made about the model) 

  •  developing a framework for the evaluation of uncertainty.  

Current EMPIR European Joint Research Projects (such as Metrology of automated data analysis for cardiac arrhythmia management 18HLT07 MedalCare) are already on the lookout for answers to these open questions.  



Do you have a measurement science related need in this area? Or would you like to collaborate with MATHMET on a research topic in this area? Please don’t hesitate and contact us!