Below you find a non-exhaustive list of possible projects to work on for theses. Whether it is a bachelor or master (or even PhD) thesis, feel free to reach out (jan.kieseler@kit.edu) if you are interested in any of these (or other tangential) topics! If you are looking for a BSc thesis and are intruiged by one of the MSc topics, please also reach out - I am happy to discuss potential adaptations.
Point cloud compression techniques for physics data.
(BSc or MSc)
In this thesis, you will develop (MSc) or test (BSc) advanced point cloud compression techniques for reconstructing particles. You will implement and evaluate different compression algorithms on real-world datasets from particle physics experiments.
Contact(s): Prof. Dr. Jan Kieseler
Improving electron identification for 4 top-quark production.
(MSc)
The identification of electrons is crucial for the analysis of 4 top-quark production. In this thesis, you will develop and evaluate new machine learning-based (transformer) electron identification algorithms to improve the efficiency and purity of electron selection in the context of 4 top-quark production.
Contact(s): Prof. Dr. Jan Kieseler
Differential measurement of 4 top-quark production.
(MSc)
The production of 4 top quarks is a rare process that can provide insights into the properties of the Higgs boson and the top quark. In this thesis, you will work towards a differential measurement of 4 top-quark production using data from the CMS experiment. The extent of the analysis will depend on the current state of the project, as well as your background and interests, but it can include data preparation, event selection, and statistical analysis.
Contact(s): Prof. Dr. Jan Kieseler
Robustness of particle reconstruction in high granular detectors with graph neural networks.
(BSc)
You will explore the robustness of our state-of-the art GNN based algorithm under changing hardware conditions to improve our understanding of the algorithms limitations in certain scenarios and to improve the reconstruction quality up to the final physics performance.
Contact(s): Prof. Dr. Jan Kieseler
Development of robust particle reconstruction in high granular detectors with graph neural networks.
(MSc)
The GNN will be trained on simulated data for a variety of existing and future detector designs. You will improve the robustness of our state-of-the art algorithm under changing hardware conditions to create a more robust particle reconstruction framework.
Contact(s): Prof. Dr. Jan Kieseler
CMS Particle Flow reconstruction with machine learning.
(MSc)
The Particle Flow algorithm is a key component of the CMS experiment's data processing chain and basis for about 100 analyses per year. In this thesis, you will apply our state-of-the art ML-based Particle Flow algorithm that can handle the high event densities expected at the HL-LHC to CMS data into the CMS computing model and will evaluate its performance on real data as well as simulations. This development can be used to improve the physics performance of the whole experiment.
Contact(s): Prof. Dr. Jan Kieseler
Development of a flexible simulation for a particle-flow detector
(MSc)
The simulation of particle detectors is a crucial step in the design and optimisation process. In this thesis, you will develop a flexible simulation framework for a simplified particle-flow detector that can be used to evaluate different detector designs and configurations. You will use state-of-the-art high fidelity simulation tools (Geant4).
Contact(s): Prof. Dr. Jan Kieseler
Optimisation of a particle flow detector design with generative AI.
(MSc)
The design of particle detectors is a complex task that requires a deep understanding of the physics involved. In this thesis, you will use machine learning to systematically optimise the design of a particle flow detector (a simple tracker and calorimeter) in an end-to-end fashion using differentiable surrogates - something never done before systematically. You will work with, and extend our existing software framework AIDO for the task of optimising tracking and calorimetry simultaneously.
Contact(s): Prof. Dr. Jan Kieseler
AI driven fast simulation for muon interactions with matter
(BSc or MSc)
Muon tomography is a technique that uses cosmic muons to image the internal structure of large objects, such as buildings or geological formations. In this thesis, you will develop a fast and differentiable simulation of muon interactions with matter using generative AI. The generative model is trained on state-of-the-art Geant4 simulations. (MSc only:) You will integrate the model into the public TomOpt muon tomography detector optiomisation framework.
Contact(s): Prof. Dr. Jan Kieseler
Reconstruction of muon tomography data using graph neural networks.
(MSc)
In this thesis, you will develop a graph neural network (GNN) model for the reconstruction of muon tomography data. The GNN will be trained on simulated data and will be evaluated on real-world datasets. You will explore different GNN architectures and training strategies to improve the reconstruction quality.
Contact(s): Prof. Dr. Jan Kieseler
Towards an implementation of modern Inference-as-a-Service for HEP at KIT
(MSc)
Inference-as-a-Service (IaaS) is a modern approach to provide machine learning inference capabilities as a service. In this thesis, you will work towards an implementation of IaaS for high-energy physics (HEP) at KIT. You will explore the SuperSONIC/Triton package on our local cluster and will work towards implemeting a prototype system that can be used for large scale inference in HEP.
Contact(s): Prof. Dr. Jan Kieseler
Evaluation of modern Inference-as-a-Service for HEP at KIT
(BSc)
Inference-as-a-Service (IaaS) is a modern approach to provide machine learning inference capabilities as a service. In this thesis, you will evaluate existing IaaS solutions for high-energy physics (HEP) at KIT. You will explore the SuperSONIC/Triton package on our local cluster and will work towards evaluating its performance and scalability for large scale inference in HEP.
Contact(s): Prof. Dr. Jan Kieseler