The neural network Z vertex trigger for the Belle II detector

A novel approach for track triggering is currently studied for the Belle II detector: neural networks are used to predict the event vertex in z direction, using only information from the central drift chamber. The lack in accuracy of classical online vertex reconstruction motivates new studies for the z vertex trigger. Since neural networks are general function approximators, they are well suited for problems where the model is not known a priori. Several methods were investigated, but our studies for single tracks in geometrically restricted areas of the detector have proven the multi layer perceptron to produce the most accurate results, even in the presence of background. This encourages the use of a set of multi layer perceptrons to cover the entire detector. Additionally, the methods presented may lead to online event reconstruction, for Belle II as well as for other running or future detectors.