Thesis ALGORITMO DE MACHINE LEARNING PARA LA IDENTIFICACIÓN DEL BOOSTED DI-HIGGS EN EL EXPERIMENTO ATLAS
Loading...
Date
2019-11
Journal Title
Journal ISSN
Volume Title
Program
DEPARTAMENTO DE INFORMÁTICA. INGENIERÍA CIVIL INFORMÁTICA
Campus
Casa Central Valparaíso
Abstract
Este trabajo describe el desarrollo de algoritmos de machine learning para la
clasificación de eventos del LHC, los cuales pueden pertenecer a una de las dos clases: señal
o background, utilizando un conjunto de datos simulados del experimento ATLAS de CERN.
Específicamente, el objetivo es identificar la señal “generación de un boosted di-Higgs
decayendo en dos quarks bottom y dos leptones tau” de los eventos del tipo background. Se
utilizaron cuatro configuraciones de señal para construir los clasificadores binarios, usando
boosted decision trees y redes neuronales profundas. Para la optimización de parámetros
se utilizó una metodología grid search para la búsqueda del mejor valor de ROC AUC. Los
experimentos indicaron que el clasificador con el mejor desempeño fue un boosted decision
tree que alcanza un 96 % de ROC AUC y 85 % de F1 score.
This paper describes the development of machine learning-based algorithms for classifying ATLAS experiment events, in two classes: signals and background, on a simulated dataset. More precisely, this work aims to identify the generation of a boosted di-Higgs decaying in two bottom quarks and two tau leptons. Four main configurations were designed for building the binary classifiers, using boosted decision trees and deep neural networks approaches. The grid search technique was performed for optimal parameter selection. Finally, the best models were selected based on the ROC AUC metric for each approach and configuration. The experiments showed that the best performance was achieved using the boosted decision tree approach, reaching 96 % of ROC AUC and 85 % of the F1 score. Keywords— ATLAS Experiment, lhc events classification, boosted di-Higgs, machine lear- ning, boosted decision trees, deep learning.
This paper describes the development of machine learning-based algorithms for classifying ATLAS experiment events, in two classes: signals and background, on a simulated dataset. More precisely, this work aims to identify the generation of a boosted di-Higgs decaying in two bottom quarks and two tau leptons. Four main configurations were designed for building the binary classifiers, using boosted decision trees and deep neural networks approaches. The grid search technique was performed for optimal parameter selection. Finally, the best models were selected based on the ROC AUC metric for each approach and configuration. The experiments showed that the best performance was achieved using the boosted decision tree approach, reaching 96 % of ROC AUC and 85 % of the F1 score. Keywords— ATLAS Experiment, lhc events classification, boosted di-Higgs, machine lear- ning, boosted decision trees, deep learning.
Description
Keywords
EXPERIMENTO ATLAS, CLASIFICACIÓN DE EVENTOS DEL LHC, BOOSTED DI-HIGGS, MACHINE LEARNING, BOOSTED DECISION TREES, DEEP LEARNING