Banca de DEFESA: JUAN LIEBER MARIN

Uma banca de DEFESA de DOUTORADO foi cadastrada pelo programa.
STUDENT : JUAN LIEBER MARIN
DATE: 08/02/2023
TIME: 09:00
LOCAL: https://conferenciaweb.rnp.br/webconf/laboratorio-de-sistemas-digitais-ufba
TITLE:

ONLINE PHOTONS DETECTION IN ATLAS EXPERIMENT USING MACHINE LEARNING

KEY WORDS:

Instrumentation, machine learning.

PAGES: 90
BIG AREA: Engenharias
AREA: Engenharia Elétrica
SUMMARY:

The ATLAS experiment, one of the largest of the LHC, operating at CERN, aims to expand the
knowledge of the structure of matter, as well to explore the properties which are already known.
The LHC collides proton each 25 ns at 13TeV as energy of center of mass, producing approximately
70 TB/s of information, which presents a need of a event selection system to investigation and
search of new particles. In special, photons have great interest of the ATLAS collaboration, once
rare particles, like Higgs boson, decays into photons. Therefore, the correct identification of
these particles is crucial. The NeuralRinger algorithm, which operates since 2017 on electron
identification, is used by the ATLAS trigger system. This method aims to capture the energy
development information in the ATLAS calorimeter system and, by using a neural classifier, aim
to identify electrons that interacts with the calorimeter. From the calorimetry point of view,
electrons and photons have similar behavior and, therefore, the use of NeuralRinger algorithm to
photon identification is promising, since the photon identification process in highly contaminated
with background noise. Thus, the main goal of this work is to adapt the NeuralRinger algorithm
for photon identification on the ATLAS trigger system. By adding this method to photon
identification chains, it is expected to maintain the efficiency of the trigger system, reducing the
acceptance of fake photons, saving computational resources during the collisions. Additionally,
as the NeuralRinger uses the information of the energy deposited by the particle to generate
the ringed signatures, the energy estimation methods will also be investigated in a high signal
stacking scenario, where the performance of traditional methods deteriorates. As a result, it is
shown that the use of the NeuralRinger algorithm can reduce at least 50% of the acceptance
of hadronic jets when compared to the current rate of the ATLAS experiment, this percentage
being higher when using the pre-processing by PCA, reaching a reduction of about 58% and 67%
when using a convolutional network. As a proof of concept, it is also shown that changing the
energy estimation method in the hadronic layers improves the discriminative capacity of the
signatures used by NeuralRinger, as well as increases this capacity in the current ATLAS photon
selection technique.

COMMITTEE MEMBERS:
Interno - 2506534 - EDUARDO FURTADO DE SIMAS FILHO
Interno - 1554822 - EDSON PINTO SANTANA
Externo à Instituição - LUCIANO MANHÃES DE ANDRADE FILHO - UFJF
Externo à Instituição - DANTON DIEGO FERREIRA
Externo à Instituição - DENIS OLIVEIRA DAMAZIO