通過額外湯川木耦合搜尋奇特希格斯粒子、利用向量注意機制標記粒子激漿流、以及粒子在高精密度量熱儀透過多任務學習模型及視覺變換器之重建

Abstract

本論文包含三個獨立的研究專題:第一個專題致力於搜尋新型希格斯粒子。我們利用新的湯川耦合ρtc和 ρtu，探索廣義雙希格斯粒子偶模型所預測的兩個新型希格斯粒子:CP偶稱的H重希格斯粒子和CP奇稱的A 重希格斯粒子。我們的研究主要聚焦於同電荷輕子與噴射流的信號特徵，並運用DeepJet演算法提供關 鍵的噴射流風味信息。實驗數據來自 CERN 大型強子對撞機的 CMS 實驗，涵蓋了完整的 Run 2 數據集。這項研究不僅探索了標準模型的可能擴展，還為未來新 型希格斯粒子的搜尋提供了重要的物理洞見。第二個專題旨在解決大型強子對撞機面臨的一個既具挑戰性又極為重要的物理問題:區分輕夸克和膠子風味噴流。 鑒於許多標準模型和新物理模型預測的事件都富含輕夸克噴流，我們開發了一個創新的深度學習模型。該模型整合了向量注意力機制和 Lund 變數，後者能提供噴流內部結構的關鍵信息。在公開的輕夸克和膠子噴流數據集上，我們的模型展現出優於當前最先進的 Particle Transformer 模型的性能。第三個專題引入了兩項創新:一是專為多任務學習設計的 AdaTask 梯度優化算法，二是將計算機視覺領域的 Vision Transformer應用於粒子重建。我們的模型在性能優化方面取得的進展，充分展示了這些先進技術在粒子物理學應用中的巨大潛力。這三個專題共同推進了粒子物理學和機器學習的前沿，為未來的研究開闢了新的方向。

This thesis incorporates three distinct research projects in particle physics and machine learning: The first project investigates the potential existence of heavy Higgs bosons predicted by the generalized two-Higgs doublet model. It focuses on two neutral Higgs bosons, CP- even H and CP-odd A, probed through new Yukawa couplings ρtc and ρtu. The analysis employs a signature of two same-sign leptons associated with jets, utilizing the DeepJet algorithm for crucial jet flavor information. Using the full Run 2 dataset collected by the CMS experiment at CERN’s Large Hadron Collider, this research explores potential extensions to the Standard Model of particle physics and provides physics insight for future heavy Higgs bosons searches. The second project addresses the challenge of separating light quark-initiated jets from gluon-heavy backgrounds at the LHC, crucial for both Standard Model and Beyond Standard Model processes. We develop deep learning models incorporating vector attention mechanisms and Lund variables to capture jet internal structure information. Our model demonstrates superior performance over the current state-of-the-art Particle Transformer when trained and evaluated on the Quark-Gluon public dataset. The third project introduces the AdaTask algorithm, an optimization algorithm dedicated to multi-task learning, and applies the Vision Transformer, a prominent deep learning model in computer vision, to the particle reconstruction in the next-generation detector, High Granularity Calorimeter. The improved performance of our model showcases the potential of these advanced techniques in particle physics applications.