Uncertainty analysis of Learning-from-Examples algorithms

Learning-from-Examples (LfE) algorithms are becoming popular building blocks for some type of measurement systems, like smart sensors. They enhance and extend the measurement capabilities of sensors allowing the use of sophisticated algorithms for sensor compensation, or for the automatic classification of physical phenomena. Machine learning systems differ quite a bit from components more commonly found in measurement systems, both in the way they introduce uncertainty and in the way that uncertainty is usually estimated. In this work we provide an analysis of uncertainty of such kind of systems, focusing on the peculiarities resulting from the presence of the LfE module in the measurement chain. The analysis is at a theoretical level, with support of numerical simulations.