Application of Machine Learning in Top+b jet analyses at the CMS Experiment
- 주제(키워드) Machine Learning , High Energy Physics , CMS experiments , Top quark , bottom quark
- 발행기관 한양대학교 대학원
- 지도교수 Tae Jeong Kim
- 발행년도 2026
- 학위수여년월 2026. 2
- 학위명 박사
- 학과 및 전공 대학원 물리학과
- 세부분야 해당없음
- 실제URI http://www.dcollection.net/handler/hanyang/200000943824
- UCI I804:11062-200000943824
- 본문언어 영어
- 저작권 한양대학교 논문은 저작권에 의해 보호받습니다.
초록/요약
기계학습(Machine Learning)은 현대 고에너지물리 분석에서 필수적인 기법으로 자리 잡고 있다. 본 학위논문은 CMS 실험에서의 여러 기계학습 응용을 탐구하며, 검출기의 한계 보완과 물리 분석 성능 향상을 동시에 목표로 한다. 첫째, Run 3 및 향후 HL-LHC 환경에서의 b-jet 태깅 성능 향상을 위해 CMS High-Level Trigger에서의 heavy-flavour 식별을 위한 새로운 알고리즘을 개발하고 검증하였다. 둘째, ttbb 사건에서의 추가 젯 식별에 다변량 방법을 적용하여 heavy-flavour 생성과 관련된 측정의 정밀도를 향상시켰다. 마지막으로, 많은 젯을 동반한 hadronic 최종상태에서 vector-like T quark이 top quark과 Higgs boson으로 붕괴하는 과정(T →tH)을 탐색하기 위해 신경망 기반 접근법을 적용하여 기존의 컷 기반 (cut-based) 방법보다 향상된 민감도를 달성하였다. 이러한 발전은 CMS 실험에서 기계학습이 검출기 성능과 물리 분석 모두에 어떤 영향을 미치는지를 잘 보여준다.
more초록/요약
Machine learning techniques have become an essential tool in modern high-energy physics analyses. This thesis explores several applications of machine learning within the CMS experiment, targeting both detector-level limitations and physics analyses. First, new algorithms for heavy-flavour identification at the CMS High-Level Trigger are developed and validated, enabling improved b-jet tagging performance under Run 3 and future HL-LHC conditions. Second, multivariate methods are employed to identify additional jets in ttbb events, enhancing the precision of measurements involving heavy-flavour production. Finally, a neural network based approach is used in the search for singly produced vector-like T quarks decaying into a top quark and a Higgs boson in fully hadronic final states, yielding improved sensitivity compared to traditional cut-based selections. Together, these developments illustrate the impact of machine learning on both detector performance and physics analyses at CMS.
more목차
ABSTRACT xxx
1 Introduction 1
1.1 The Standard Model 3
1.2 Vector Like Quarks 5
2 Detectors 9
2.1 Large Hadron Collider 9
2.2 The Compact Muon Solenoid experiment 10
2.2.1 Inner tracking system 11
2.2.2 Electromagnetic calorimeter 11
2.2.3 Hadron calorimeter 12
2.2.4 Muon system 13
2.2.5 The CMS trigger system 15
2.3 Muon detector complementarity in Phase II CMS Level 1 Trigger: iRPC+CSC 19
2.3.1 Motivation 20
2.3.2 Trigger Primitives 20
2.3.3 Ghost hits in CSC 21
2.3.4 Matching algorithm 21
2.3.5 Conclusion 24
2.4 Identification of Heavy-flavour jet in HLT 24
2.4.1 Performance of Online Flavour Taggers 27
2.4.2 Monitoring and Validation of Trigger 31
2.4.3 Impact on T (→ tH) in Run 3 33
2.4.4 Impact in physics program and perspective 36
2.5 Jet Energy Scale in γ + jets Channel 40
2.5.1 Implementation of a dedicated low pT Photon HLT Path 40
2.5.2 Impact in JES in relation to ∆R between γ and jet 42
2.5.3 Preliminary JES in γ + jets in Run3 44
3 Physics analysis using Neural Networks 54
3.1 Machine Learning 54
3.2 Identification of additional jets in the ttbb events by using deep neural network 56
3.2.1 Introduction 56
3.2.2 Samples and Selections 56
3.2.3 Strategy to select two additional b-jets 57
3.2.4 Results 53
3.2.5 Perspective in ttbb inclusive and differential cross section measurement 65
3.3 Search for resolved Vectorlike T quark in hadronic final state using Neural Network 67
3.3.1 Introduction 67
3.3.2 Samples 68
3.3.3 Physics objects and corrections 69
3.3.4 Event selection 75
3.3.5 Region of interests 75
3.3.6 χ2 Reconstruction of W, Higgs boson and top quark 76
3.3.7 Neural Network 77
3.3.8 Transfer Function 87
3.3.9 Optimization of cut values 89
3.3.10 Signal modeling and Background Estimation 90
3.3.11 Systematic uncertainties 92
3.3.12 Result 104
3.4 Summary and Perspective 105
3.4.1 Perspective in HL-LHC 106