Using Advanced Machine Learning Techniques to Study Poorly Modeled Processes in pp Collisions with the ATLAS Detector - Dissertation

Aug 5, 2026

Upload by Violet

Dissertation research on applying advanced machine learning to analyze poorly modeled physics processes in proton-proton (pp) collisions recorded with the ATLAS detector. The work positions the study within the Standard Model framework, describing the Large Hadron Collider and event reconstruction, then details data handling, Monte Carlo simulation, and event selection. Machine learning methods are developed across supervised, weakly supervised, and unsupervised regimes, including neural architectures, density estimation via normalizing flows, and autoencoders, to improve background modeling and investigate specific interference effects. Results culminate in a summary and conclusion with supporting appendices on the developed machine learning tools and fit procedures.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages167

Reading Time7 h 1 min

AccessFree to download

Preamble

Introduction

The Standard Model

Structure of a Proton

Physics at Particle Colliders

Top-Quark Physics

Tau Physics

The Large Hadron Collider and ATLAS

The Large Hadron Collider

ATLAS

Reconstruction and Identification of Objects

Data, Monte Carlo Simulation, and Event Selection

Datasets

Monte Carlo Simulation

Signal and Sources of Background

Machine Learning

Introduction

Artificial Neural Networks

Supervision

Weak Supervision

No Supervision

Imprecise Modeled Processes

Interference Between tW and tt

Poorly Modeled Backgrounds

Hadronic Leptons

Summary and Conclusion

Appendices

Using Advanced Machine Learning Techniques to Study Poorly Modeled Processes in pp Collisions with the ATLAS Detector Dissertation zur Erlangung des Doktorgrades (Dr. rer. nat.) der Mathematisch-Naturwissenschaftlichen Fakultät der Rheinischen Friedrich-Wilhelms-Universität Bonn von Federico Guillermo Diaz Capriles aus Ft. Lauderdale, FL, USA Bonn, 10.3.2023 Angefertigt mit Genehmigung der Mathematisch-Naturwissenschaftlichen Fakultät der Rheinischen Friedrich-Wilhelms-Universität Bonn 1. Gutachter: Prof. Dr. Ian C. Brock 2. Gutachter: Priv-Doz. Dr. Philip Bechtle Tag der Promotion: 15.05.2023 Erscheinungsjahr: 2023 Acknowledgements I would like to thank my family and friends for their continued support with the distinction of my mother, and best friend whose contributions are what made this possible. Prof. Dr. Ian Brock for the opportunity to take part in this ﬁeld, the patience, the support, and the freedom to pursue my ideas. Finally, the friends that helped during these tough times. Corona in the oﬃce was rough, but at least being able to vent to each other made it a little more bearable. Contents 1 Preamble 1 2 Introduction 3 2. 1 The Standard Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 2.1. 1 Fermions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 2.1.2 Bosons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 2.2 Structure of a Proton . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 2.3 Physics at Particle Colliders .............................. 14 2.4 Top-Quark Physics ................................... 17 2.5 Tau Physics ....................................... 22 3 The Large Hadron Collider and ATLAS 25 3.1 The Large Hadron Collider ............................... 25 3.2 ATLAS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 3.3 Reconstruction and Identiﬁcation of Objects ...................... 32 4 Data, Monte Carlo Simulation, and Event Selection 39 4. 1 Datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 4.2 Monte Carlo Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 4.3 Signal and Sources of Background ........................... 41 5 Machine Learning 49 5. 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 5.2 Artiﬁcial Neural Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 5.2.1 Architecture and Components ......................... 50 5.2.2 Data Preparation ................................ 53 5.2.3 Over-and Underﬁtting ............................. 54 5.2.4 Regularization ................................. 56 5.2.5 Batch Sizes and Hardware . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 5.3 Supervision ....................................... 60 5.4 Weak Supervision .................................... 60 5.4.1 Learning from Label Proportions ....................... 60 5.4.2 Classiﬁcation Without Labels . . . . . . . . . . . . . . . . . . . . . . . . . 61 5.5 No Supervision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61 5.5. 1 Autoencoders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 5.5.2 Masked Autoregressive Density Estimators (MADE) . . . . . . . . . . . . . 64 5.5.3 Normalizing Flows ............................... 66 5.5.4 Masked Autoregressive Flows ......................... 67 5.5.5 Anomaly Detection with Density Estimation ................. 68 6 Imprecise Modeled Processes 71 6. 1 Interference Between tW and tt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 7 Poorly Modeled Backgrounds 91 7.1 Hadronic 􀁧 Leptons ................................... 91 8 Summary and Conclusion 117 A Machine Learning Package 119 B Autoencoder Di

FAQs

What physics setting and detector are used in this dissertation?

The dissertation studies proton-proton (pp) collisions analyzed with the ATLAS detector at the Large Hadron Collider, using the Standard Model as the baseline theory for expected behavior.

How does the work structure its machine learning approaches?

It develops machine learning across supervised learning, weak supervision, and unsupervised learning, including neural networks, autoencoders, masked autoregressive density estimators, and normalizing flows for density estimation and anomaly detection.

What types of challenges related to modeling are targeted?

The research focuses on poorly modeled processes and backgrounds, including interference effects between tW and tt and specific hadronic-lepton background contributions that motivate improved modeling strategies.

Aug 5, 2026

Upload by Violet