A DNN for CMS track classification and selection

Abstract

The upgrade of the track classification and selection step of the CMS tracking to a Deep Neural Network is presented. The CMS tracking follows an iterative approach: tracks are reconstructed in multiple passes starting from the ones that are easiest to find and moving to the ones with more complex characteristics (lower transverse momentum, high displacement). The track classification comes into play at the end of each iteration. A classifier using a multivariate analysis is applied after each iteration and several selection criteria are defined. If a track meets the high purity requirement, its hits are removed from the hit collection, thus simplifying the later iterations, and making the track classification an integral part of the reconstruction process. Tracks passing loose selections are also saved for physics analysis usage. The CMS experiment improved the track classification starting from a parametric selection used in Run 1, moving to a Boosted Decision Tree in Run 2, and finally to a Deep Neural Network in Run 3. An overview of the Deep Neural Network training and current performance is shown.