SILENT Uncertainty quantification in data fitting Neural and Hilbert Networks

In the framework of artificial intelligence, neural networks are one of the most popular tools. When considering the representation of data, namely data fitting, an alternative is furnished by Hilbert networks, which have a higher complexity, since they consider general Hilbert basis instead of affine functions for the transfer of information between the nodes of the network. In this work, we examine the effects of errors in the data on the quality of the approximation furnished by these networks. The techniques of uncertainty quantification are used to determine the effects of these errors on the final fitting and in the forecasting for points outside the training data.