Theses
Masters and Bachelors theses and research internships
If you are interested in a Masters or Bachelor thesis project or a research internship (Forschungspraxis) in our group we are happy to propose a concrete problem related to our current interests. Our group focuses on machine learning and optimization, deep learning for inverse problems, and DNA data storage and DNA information technologies. To get an idea about our current research, please check out our recent papers at google scholar. Projects usually involve a mixture of theory and applied work and require strong interest documented by excellent grades in relevant subjects such as linear algebra, probability and statistics, machine learning, signal processing, optimization, or related courses.
Below is a list of open topics. If you are interested in one of the projects, please send an email directly to the superviser and cc reinhard.heckel(at)tum.de. Include your transcript of records from TUM and your curriculum vitae and the planned start and end dates.
The list of topics is sometimes incomplete, and we are also happy to propose other topics if there is a good fit. If you are interested in a topic related to our current interests that is not listed below, please reach out to reinhard.heckel(at)tum.de and again include your transcript of records and CV.
External projects: If you are planning to carry our our project externally, for example at a company or another university, and you want us to supervise the project, please send an email to reinhard.heckel(at)tum.de including the name and contact of the exernal supervisor, your transcipt of records and CV, and a project description and an explanation how the project is related to our expertise. We can only supervise external projects if they are related to our expertise and current research interests.
Thesis and project administration and logistics: Here are some guidelines on the thesis and research project administration and on the grading.
Open projects
Automated Independent Component Selection for Fetal Magnetocardiography (fMCG)
Conventional signal reconstruction in Fetal Magnetocardiography (fMCG) requires labor-intensive, manual selection of Independent Component Analysis (ICA) components to separate fetal, maternal, and noise sources. The objective of this project is to explore learning-based methods (e.g. time‑series clustering and representation learning) to automate ICA component selection, thereby enabling automated reconstruction of fetal cardiac activity. Further extensions or related topics might be available upon inquiry. Own ideas are also welcome.
Supervisors: Jonas Emrich (jonas.emrich(at)tum.de)
Prerequisites: Strong coding skills. Background in machine learning or signal processing.
Type of project: Research Internship (Forschungspraxis) or Bachelor’s thesis (optionally Master’s thesis)
3D PET Self-Supervised Image Reconstruction
Self-supervised reconstruction is beneficial in PET imaging, where ground truth images are hard to obtain. We have developed a 2D self-supervised reconstruction method that has been thoroughly tested on simulated data. We are looking for a Master’s student to extend this method to 3D and evaluate it on real data.
Supervisors: Reinhard Heckel (reinhard.heckel(at)tum.de), Georg Schramm from KU Leuven
Prerequisites: Strong coding skills, especially in PyTorch. A solid understanding of deep learning frameworks and image reconstruction.
Type of project: Master’s thesis
Entropy-guided test-time inference for LLMs
We are looking for a Master's student to explore methods for enhancing the reasoning ability of Large Language Models (LLMs) through entropy-guided test-time inference. In this project, you will investigate methods to boost performance without training by analyzing token-level entropy distributions across reasoning tasks and developing test-time inference strategies that leverage entropy to refine model outputs.
Supervisor: Kun Wang (kun2000.wang(at)tum.de)
Prerequisites: Strong coding skills in Python and ML libraries (e.g., PyTorch), background in machine learning and foundation model.
Type of project: Master's thesis/Research internship
Training Spatiotemporal Diffusion Priors on Corrupted Cardiac MRI Datasets
We are looking for a motivated Master’s student to work on a project that aims to train spatiotemporal (video) diffusion models for cardiac MRI. In particular, you will explore methods that address samples that may be corrupted by motion artefacts - such as those caused by respiratory motion during acquisition. By addressing these issues you can contribute to simplifying data acquisition and improving robustness of large-scale MRI models.
Supervisor: Anselm Krainovic (anselm.krainovic(at)tum.de)
Prerequisites: Strong coding skills and a good understanding of deep learning for inverse problems.
Type of project: Master thesis
Dataset design for self-supervised learning methods for accelerated MRI
Supervised learning is difficult for many accelerated MRI applications because ground-truth data is often unavailable. Self-supervised learning methods for imaging, which train without ground-truth, are a promising alternative. We are looking for a student to investigate how the performance of self-supervised learning methods can be improved by optimizing the training dataset, for example, by filtering data samples. We have research results for a similar question for supervised learning, and a key goal of this thesis is to explore data curation strategies for self-supervised learning methods.
Supervisor: Kang Lin (ka.lin(at)tum.de)
Prerequisites: Strong coding skills, especially in PyTorch. A solid understanding of deep learning frameworks and image reconstruction.
Type of project: Master thesis
Evolving Representations for the Abstraction and Reasoning Corpus Benchmark
The Abstraction and Reasoning Corpus (ARC) tests whether AI can generalize from just a few examples, a task that current deep learning models, including large language models, still struggle with. Each task requires identifying the hidden transformation between input and output and applying it to a new test input. The goal of this thesis is to apply the AlphaEvolve framework to the ARC benchmark by exploring different representations of input-output pairs, such as graph-based or object-centric formulations, and by developing new approaches to evaluate how good a proposed transformation is.
Supervisor: Franziska Weindel (franziska.weindel(at)tum.de)
Prerequisites: Strong coding skills in Python and ML libraries (e.g., PyTorch), understanding of large language models.
Type of project: Master’s thesis/Research internship
Reconstruction of Cardiac MRI using Diffusion Model with Optical-Flow Guidance
In cardiac cine MRI the task is the reconstruction of a video of the beating heart from undersampled dynamic measurements. The almost-periodic nature of the cardiac motion gives additional structural information that can help in the challenging reconstruction. The goal of this project is to train a strong motion prior using optical-flow, and apply it as guidance signal in reconstruction of cardiac videos.
Supervisor: Florian Fürnrohr (florian.fuernrohr(at)tum.de)
Prerequisites: Strong coding skills and a background in deep learning for inverse problems.
Type of project: Research Internship (Forschungspraxis)
Current and past theses in the group
Wenlong Li, ``Robust fine-tuning for accelerated MRI’’, Project/Forschungspraxis, ongoing
Hristo Stefanov, ``Perception-Distortion-Tradeoffs for Diffusion-Model based Reconstruction’’, Bachelor's thesis, 2025
Jakub Dvorak, ``Diffusion Prior for Reconstruction in Cryo-ET’’, Master’s thesis, 2025
Diyor Khayrutdinov, ``Denoising at Test Time for Better Membrane Segmentation of Cryo-Electron Tomograms’’, Master’s thesis, 2025
Xiufeng Yang, ``Retrieval and generation based test-time-training for imaging’’, Master’s thesis, 2025
Kai Eberl, ``Foundation Models prior for Image Restoration’’, Master’s thesis, 2025
Jonas Emrich, ``Establishing a Forward Model for Improved Signal Reconstruction in Fetal Magnetocardiography’’, Master’s thesis, 2025
Diyor Khayrutdinov, ``Denoising at Test Time for Better Membrane Segmentation of Cryo-Electron Tomograms’‘, Project/Forschungspraxis, 2025
Julian Streit, ``Addressing Synchronization Uncertainty in Foundational Error Correction Models’‘, Master’s thesis, 2025
Zeineb Ben Chaben, ``Self-Supervised PET Image Reconstruction from Subsampled Data with Variational Networks’‘, Master’s thesis, 2025
Moritz Bauman, ``Multiple Sequence Base Calling for DNA data storage’’, Project/Forschungspraxis, 2025
Oliver Kovacs, ``Adapting SAM for Data-Efficient Particle Picking in Cryo-ET’‘, Master’s thesis, 2024
Oliver Kovacs, ``Developing a VarNet for DeepDeWedge’’, Project/Forschungspraxis, 2024
Jakub Dvorak, ``Simultaneous Self Supervised Image Denoising and Deconvolution’’, Project/Forschungspraxis, 2024
Isabel Schorr, ``DPO’’, Master's thesis, 2024
Andreas Faika, ``Optimizing data mixtures ’’, Master's thesis, 2024
Tim Lindenau, ``Optimization of mixing proportions with zero-shot optimization’’, Project/Forschungspraxis, 2024
Serden Sait Eranil, ``Signal processing for fetal magnetocardiography’’, Project/Forschungspraxis, 2024
Kun Wang, ``Motion reconstrution for MRI’’, Master's thesis, 2024
Claudio Kaserer, ``Improving Mathematical Reasoning of Language Models Using Supervision Data’’, Master’s thesis, 2024
Serdar Caglar, ``Test time training for denoising distribution shifts’’, Project/Forschungspraxis, 2024
Andreas Faika, ``Comparison of Tokenizers for LLMs’’, Project/Forschungspraxis, 2024
Raimundo Parra, ``Unintentional Bilingualism in Large Language Models’’, Project/Forschungspraxis, 2024
Mamdouh Aljoud, ``Filtering techniques in next generation multimodal datasets’’, Master's thesis, 2023
Cheng Yan, ``Meta-Learning for mulit-task MRI reconstruction’’, Master's thesis, 2023
Francesco Bollero, ``Data pruning for image reconstruction’’, Master's thesis, 2023
Faidra Patsatzi, ``Randomized smoothing for inverse problems’’, Bachelor's thesis, 2023
Xiaodong Lei, ``Adversarial robustness of deblurring methods’’, Project/Forschungspraxis, 2023
Dogukan Atik, ``Scaling laws for self-supervised image denoising’’, Master's thesis, 2023
Mamdouh Aljoud, ``Deep networks for nanopore basecalling’’, Project/Forschungspraxis, 2023
Rafael Vorländer, ``Reimplementing CryoGAN’’, Project/Forschungspraxis, 2023
Juan Cao, ``Uncertainty quantification methods for compressed sensing’’, Project/Forschungspraxis, 2023
Litao Li, ``Datawork for accelerated MRI’’, Master's thesis, 2023
Guang Chai, ``Evaluating deep-learning based imaging systems’’, Master's thesis, 2023
Johannes Kunz, ``Generative models for Cardiac Magnetic Resonance Imaging’’, Master's thesis, 2023
Ali Can, ``Understanding the Contribution of Training Samples on a Prediction of a Single Test Image in a Denoising Task Using Attention Mechanism’’, Project/Forschungspraxis, 2022
Deniz Uysal, ``Spectral Computed Tomography Image Reconstruction’’, Master's thesis, 2022
Xuyang Zhong, ``Self-Supervised Learning for Image Denoising’’, Master's thesis, 2022
Johannes Kunz, ``Dynamic MRI reconstruction’’, Project/Forschungspraxis, 2022
Weixing Wang, ``Graph neural networks for clustering and aligning DNA sequences for DNA storage'', Project/Forschungspraxis, 2022
Yundi Zhang, ``Coordinate-based image priors'', Master's thesis, 2022.
Samuel Eadie, ``Rate-Distortion Stochastic Autoencoding for Robust Representation Learning and Out-of-Distribution Detection'' (carried out at Bosch Research), Master's thesis, 2022
Kang Lin, ``Transformers for image recovery'', Master's thesis, 2021.
Frederik Fraaz, ``Image recovery with invertible neural networks'', Master's thesis, 2021.
Youssef Mansour, ``Neural network architectures for image recovery and denoising'', Master's thesis, 2021.
Benedikt Böck, ``Multiplicative filter networks for image processing applications'', Project/Forschungspraxis, 2021
Mohamed Ketata, ``Data standardisation, multi-domain learning, and artifact robustness for improved MRI'', Bachelor's thesis, 2021.
Deniz Uysal, ``A simple encoder and decoder for DNA data storage with Polar codes'', Project/Forschungspraxis, 2021.
Yundi Zhang, ``Deep matrix decoder for collaborative filtering'', Project/Forschungspraxis, 2021.
Youssef Mansour, ``Ensembles of image reconstruction method for MRI'', Project/Forschungspraxis, 2021.
Jacob Geussen, ``Diffusion MRI denoising with neural networks'', Bachelor's thesis, 2020.
Lena Heidemann, ``FastMRI with untrained neural networks'', Master's thesis, 2020.
Tobit Klug, ``Image separation with untrained neural networks'', Master's thesis, 2020.
Oleksii Khakhlyuk, ``Convolutional neural networks with fixed kernels'', Bachelor's thesis, 2019.
Zi Yang, ``Probabilistic matching networks for few-shot learning'', Master's thesis, 2019.