Charm tagging: An artificial neural network approach

by Stefaan Teetaert

Monday 26 Jun 2017, 11:00 → 12:00 Europe/Brussels

Large seminar root (Universe)

The CMS experiment at the LHC has an extended program in studying the proton-proton collisions that take place at this state-of-the-art particle collider. This includes searching for new physics phenomena as well as studying the known processes as described by the Standard Model of particle physics. In many of these analyses it has proven to be extremely useful to identify those jets that originate from bottom quarks (b tagging). Many ongoing studies could however also benefit from an algorithm to identify jets from charm quarks (c tagging). In order to observe charm jets effectively, good charm tagging algorithms are needed. Currently, a combination of two boosted decision trees (BDTs) is used for charm jet identification within the CMS experiment. Another machine learning algorithm, artificial neural networks (ANN), has shown its merits as a classifier in different domains of research and in daily life. In this paper we investigate if ANNs can be used as an alternative charm tagging algorithm. Performance of binary and multiclass feedforward ANN as a charm tagger have been compared to that of BDTs. We conclude that the current charm tagging performance, as achieved by using BDTs, can be surpassed by using ANNs as a classification algorithm.