Speeding up Discoveries in the Era of Machine Learning - Accurate and precise density estimation for the LHC - Dissertation

Aug 5, 2026

Upload by Asher

A new era of measurements and tests of the Standard Model at the Large Hadron Collider is enabled by novel simulation and data-analysis methodologies. Machine learning drives this shift by developing tools that exploit correlations in high-dimensional phase spaces in a statistically principled way. For collider applications, the work improves both searches for physics beyond the Standard Model and advanced simulation chains. Two critical LHC challenges are addressed: fast, accurate, precise surrogate models for detector response and assumption-agnostic tools for new-physics searches, unified through precise density estimation using physics-informed representations such as symmetries.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages148

Reading Time6 h 13 min

AccessFree to download

Introduction

Physics at the LHC

Standard model of particle physics

Simulations ex machina

Discovering New Physics

Machine learning

Deep learning basics

The Mexican standoff

Architectures

Fast detector simulations

High-dimensional calorimeters

CaloINN

CaloDREAM

Understanding generative networks

Speeding up Discoveries in the Era of Machine Learning Accurate and precise density estimation for the LHC Dissertation Luigi Favaro Dissertation submitted to the Combined Faculty of Mathematics, Engineering and Natural Sciences of Heidelberg University, Germany for the degree of Doctor of Natural Sciences Put forward by Luigi Favaro born in Castrovillari, Italy Oral examination: 24.10.2024 Speeding up Discoveries in the Era of Machine Learning Accurate and precise density estimation for the LHC Referees: Prof. Dr. Tilman Plehn Prof. Dr. Fred Hamprecht Abstract A new era of measurements and tests of the Standard Model of particle physics at the Large Hadron Collider has been shaped by novel methodologies applied to simulations and data analysis. Machine learning is leading this revolution with constant and progressive development of new tools for understanding our data. We have access to techniques that exploit correlations in high-dimensional phase spaces in a statistically principled way. In the context of colliders, these techniques not only boost searches for physics beyond the Standard Model but also improve the already advanced simulation chain. We consider two obstacles that will be critical for the LHC. We propose fast, accurate, and precise surrogate models for simulating the detector response, answering the increasing demand for simulations in the future runs of the LHC. Second, we develop tools for searches of new physics that do not rely on assumptions of the specific new physics signature These tools aim to complement the current paradigm of direct tests of extensions of the SM, which can carry limiting assumptions. We unify these two applications under the lens of precise density estimation using modern machine learning tools, and we demonstrate the importance of using powerful representations that leverage our physics knowledge, e.g. symmetries. Zusammenfassung Die Einführung neuer Methoden und Techniken für Simulationen und Datenanalyse hat eine neue Ära von Tests und Messungen des Standardmodels am Large Hadron Collider geprägt. Diese Revolution wird dabei vor allem durch das Machine Learning und dessen ständigen Fortschritt bei der Entwicklung neuer Werkzeuge zum Verständnis unserer Daten angetrieben. Diese erlauben es uns nun Korrelationen im hochdimensionalen Phasenraum auf statistisch fundierte Weise auszunutzen. Im Kontext von Teilchenbeschleunigern hilft dies nicht nur bei der Suche nach Physik jenseits des Standardmodels, sondern verbessert auch die so schon weit fortgeschrittene Simulationskette. Wir betrachten hier zwei kritische Hürden für den LHC. Zum einen schlagen wir schnelle, präzise und akkurate, sogenannte Surrogate Models für die Simulation von Detektoreffekten vor, um dem steigenden Bedarf an Simulationen am LHC nachzukommen. Zumanderen entwickeln wir Werkzeuge für die Suche nach neuer Physik, welche unabhängig von spezifischen Annahmen über dessen Signatur ist. Diese sollen komplementär zum momentanen Paradigma direkter Suchen nach neuer Physik sein, welche auf einschränkenden Annahmen beruhen können. Beide diese Anwendungen vereinen wir unter der Nutzung von Precise Density Estimation mithilfe moderner Machine Learning tools. Hierbei demonstrieren wir die Stärke von Repräsentationen durch die wir unser physikalisches Wissen, wie z.B. Symmetrien, effektiv nutzen können. Contents List of Abbreviations iii Preface v 1 Introduction 1 2 Physics at the LHC 5 2.1 Standard model of particle physics ...................... 5 2.2 Simulations ex machina . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 2.2.1 Hard scattering ............................. 7 2.2.2 QCD effects ............................... 9 2.2.3 Interaction with matter . . . . . . . . . . . . . . . . . . . . . . . . 10 2.2.4 Calorimeter showers . . . . . . . . . . . . . . . . . . . . . . . . . . 12 2.3 Discovering New Physics . . . . . . . . . . . . . . . . . . .

FAQs

What is the main goal of this dissertation for LHC physics?

To accelerate discoveries at the LHC by using machine learning to perform accurate and precise density estimation, improving both simulation and new-physics searches.

How does the work contribute to detector simulation efficiency?

It proposes fast, accurate, and precise surrogate models that emulate detector response, addressing the increasing demand for simulations in future LHC runs.

What second challenge does the dissertation address in addition to simulations?

It develops tools for searching new physics that do not rely on assumptions about a specific new-physics signature.

Aug 5, 2026

Upload by Asher