The Evaluation of Uncertainty in Measurements Using Artificial Neural Network Techniques

One of the main challenges in applying Artificial Intelligence (AI) is to develop reliable systems that simultaneously give valid and reproducible results. The popularity of such techniques for classification and prediction has had a disruptive impact bringing numerous benefits and accelerating progress in several areas such as autonomous driving, cybersecurity, industrial process condition monitoring, and predictive maintenance. On the other hand, the lack of or little treated evaluation of the quality of these methodologies, is noteworthy, particularly in assessing the uncertainty of measurements obtained through such approaches. This paper highlights the current state of the uncertainty assessment in Artificial Intelligence-based measurement approaches with the primary objective of proposing a new methodology relying on the law of propagation of uncertainty based on the principles described in the ISO GUM (Guide to the Expression of Uncertainty in Measurement) standard. Thus, it is necessary to investigate the aspects related to the propagation of uncertainty to the model of an Artificial Neural Network, taking into account all the issues related to the black-box nature of the networks and the non-linearities that characterize them.