Neural Networks taking systematic uncertainties in the input-feature space into account under real-life conditions

Many people can seperate a signal from a background process with the help of an neural network. But only few people can do this taking systematic uncertainties in features of the neural network input space into account. With a brandnew paper ETP has shown, that our neural networks can cope even with 86 partially non-trivially correlated uncertainties facilitating measurements with minimal combined uncertainty, consisting of statistical uncertainties and systematic variations. This is the first time that a neural network training, which takes systematic uncertainties into account, has been demonstrated to work with such a complex data model, as used for real physics measurements at the LHC. A preprint of the paper can be found following this link.