Schefer, Meinrad (2025). Investigating and Leveraging Machine Learning Techniques for High Energy Particle Identification and Reconstruction. (Thesis). Universität Bern, Bern
|
Text
25schefer_m.pdf - Thesis Available under License Creative Commons: Attribution (CC-BY 4.0). Download (43MB) | Preview |
Abstract
Recent advances in artificial intelligence have revolutionised data analysis across scientific disciplines, and high energy physics is no exception. This thesis explores advancements in experimental particle physics through the integration of machine learning techniques in high energy physics experiments and simulations. The Standard Model of particle physics and the general theory of relativity are representing our current understanding of physics, however there are observations left unexplained. Phenomena, such as the matter-antimatter asymmetry, neutrino oscillations and the presence of dark matter, motivate the search for physics beyond the Standard Model. These searches are often heavily relying on advanced simulation tools and novel AI technologies, providing and classifying key signatures. The ATLAS experiment at the Large Hadron Collider plays a crucial role in searching for physics beyond the Standard Model, such as Supersymmetry, by applying sophisticated data acquisition and analysis techniques. A major contribution of this work is the application of the NeuralRinger algorithm for forward electron identification in ATLAS. By extending this machine learning based approach to regions of high pseudorapidity, electron reconstruction can be significantly improved, enhancing event selection for physics analyses. To further refine its performance, the NeuralRinger was integrated into the Lorenzetti Showers framework, a novel and highly flexible calorimetry simulation tool. Facilitating the NeuralRinger for forward regions in that framework allowed for further developments of the simulation tool and for direct comparison with the corresponding ATLAS studies. Finally, this thesis presents a search for pair production of the supersymmetric top squark in all-hadronic final states, utilising signatures identified by a novel graph neural network tool, recently developed within ATLAS. The promising results of the studies presented in this thesis, which are collectively relying on machine learning techniques, demonstrate the growing role of AI-driven methodologies in experimental particle physics. They o!er improved detection capabilities and enhance the search for new physics beyond the Standard Model.
| Item Type: | Thesis |
|---|---|
| Dissertation Type: | Cumulative |
| Date of Defense: | 1 October 2025 |
| Subjects: | 500 Science > 530 Physics |
| Institute / Center: | 08 Faculty of Science > Physics Institute 10 Strategic Research Centers > Albert Einstein Center for Fundamental Physics (AEC) |
| Depositing User: | Hammer Igor |
| Date Deposited: | 05 Dec 2025 09:39 |
| Last Modified: | 09 Dec 2025 09:55 |
| URI: | https://boristheses.unibe.ch/id/eprint/6939 |
Actions (login required)
 |
View Item |