A Survey on Uncertainty Assessment in ANN-Based Measurements

Abstract

The concept of artificial intelligence (AI) is becoming progressively less abstract and much more concrete in the public imagination. This is due to the remarkable spread of methodologies that are part of this family of techniques in many fields, both research and industrial. The driving force behind this diffusion and the strong interest created in AI, particularly in Deep Learning algorithms, is mainly due to their ability to learn from the enormous amounts of data produced nowadays. The different disciplines affected by this real revolution include natural language processing, bioinformatics, robotics, control, signal processing, computer vision, cybersecurity and early earthquake warnings. AI techniques have also been employed in mission-critical areas, where the reliability of the produced data must be the highest. In this sense, metrology defines measurement uncertainty as a reference standard for providing information about the greenness of measurements produced by an instrument or algorithm. The state of the art, however, has not investigated these aspects adequately enough or not by the requirements of existing metrology standards, such as ISO GUM. For these reasons, this work aims to propose a methodology for direct estimation, without the use of numerical methods, of the uncertainty of measurements produced by artificial neural networks, the most widely used deep learning tool to date.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 16 citations exist upstream.