Signal-background classification is a central problem in high-energy physics, playing a major role for the discovery of new fundamental particles. The recent parametric neural network (pNN) leverages multiple signal mass hypotheses as an additional input feature to effectively replace a whole set of individual classifiers, each providing (in principle) the best response for the corresponding mass. PNNs have the potential to overcome the burden of training a multitude of independent classifiers, enabling interpolation as well as some degree of extrapolation, but also achieving better generalization and data-efficiency as a single model is trained on all the data at once. In our recent work (doi.org/10.1088/2632-2153/ac917c) we proposed both guidelines and improvements to the original pNN formulation (see: doi.org/10.1140/epjc/s10052-016-4099-4), to enable them for real-world physics analyses.
