Discipline: Physics
Subcategory: Physics (not Nanoscience)
Session: 3
Room: Exhibit Hall A
Sergi Castells - University of Illinois at Urbana-Champaign
Searches for cc ̅ from Higgs/Z decays have been done exclusively for the ground J/ψ state for leptonic decay modes while we aim to tag excited cc ̅ states via hadronic decay modes. The study of cc ̅ is relevant to Higgs coupling with the charm quark. Excited energy states such as ψ(2S) and χ_(c_0,c_1,c_2 ) are of interest as we can follow their decays into J/ψ γ. The production of excited states of cc ̅ is via the standard Higgs/Z production chain gg →H which produces cc ̅ via the H→cc ̅ γ process. The purpose of creating this tagging algorithm is to apply it to ATLAS data. The tagging is done using machine learning. Training data for the machine learning algorithm comes from Monte Carlo simulations of particle decays and simulations of interactions in ATLAS. Other Monte Carlo simulations are being tested to verify the stability of the algorithm. The accuracy for the fully-connected neural network trained on J/ψ, ψ(2S), χ_(c_0,c_1,c_2 ), and quark/gluon background is 93%. This novel approach to cc ̅ tagging resulted in a production-ready tagger. Further study is being done into using a convolution neural network for cc ̅ tagging.
Funder Acknowledgement(s): Duke University Physics REU Program through NSF Grant No. NSF-PHY-1757783.
Faculty Advisor: Nicolo de Groot, N.deGroot@hef.ru.nl
Role: I did the machine learning part of the project. The simulation part was done by my mentor.