Filtragem online baseada em calorimetria de altas energias e com alta taxa de eventos

Abstract

In the acquisition of large amounts of data, cases are observed where the rate of (and rare) interest events is small. This occurs in the area of high energy physics, which involves a large amount of data and experiments with fine segmentation and very restrictive operations. The ATLAS experiment has in its calorimetry system (energy measurement) more than 200 thousand readout channels to record the colli- sions that occur every 25 ns in LHC. Due to this high collision rate, which produces almost 70 TB/s of data, an online filtering system is required. Electrons are present in large parts of the physics channels that ATLAS intends to observe and therefore this work is dedicated to filtering them online based on the calorimetry system. The online electron filtering system in ATLAS is, since 2017, a strategy proposed by COPPE/UFRJ, called NeuralRinger, which consists of an ensemble of neural net- works as a method for decision making only for transverse energy above 15 GeV, due to a statistical restriction. Bellow this threshold, another algorithm that uses cuts in highly discriminating calorimetry variables are used. This thesis will present an extension of the NeuralRinger algorithm for the region below 15 GeV, which shows fake rate reduction up to 3 times with respect to the current algorithm based on cuts.