Overview
The AI for Science Group at TH Köln offers courses and thesis projects covering AI methods for scientific applications. Our teaching emphasizes both theoretical foundations and practical implementations with real-world data.
Lectures
Bachelor Level
MLOps (MLO)
Machine Learning Operations for production-ready ML systems, covering data versioning, experiment tracking, pipeline orchestration, and deployment strategies for scientific computing applications.
User name and password can be found in ILU.
Distributed Systems (BVS2)
Fundamentals of distributed computing architectures, parallel processing, and scalable systems design.
User name and password can be found in ILU.
Master Level
Quantum Computing (QC)
Introduction to quantum computing principles, quantum algorithms, and quantum hardware platforms with focus on semiconductor spin qubits and ML applications for quantum device control.
Advanced AI Topics (AAI)
Advanced topics in artificial intelligence including physics-informed machine learning, generative models, object-centric learning, and domain-specific AI applications in science and industry.
Thesis Topics
We offer Bachelor’s and Master’s thesis projects across all our research areas. Below is a list of currently available topics. Interested? Contact us to discuss details.
Metrology and Object-Centric Learning
AI-Based Analysis, Optimization, and Generation of Integrated Circuits
This thesis explores the use of AI methods for the automated design and optimization of electronic circuits. A key focus is the calculation of derivatives of target metrics (e.g., cutoff frequencies) with respect to component parameters, enabling efficient gradient-based local optimization towards predefined specifications. In addition, a transformer model is developed that generates complete circuits with continuous component values from partial netlists and target parameters. The resulting workflow combines classical simulation techniques with modern AI approaches, significantly accelerating the circuit design process.
Project with ASML: AI for Semiconductor Technology
This project explores using advanced large language models (LLMs) to automatically generate physical priors for machine learning models in semiconductor metrology. The goal is to enhance model accuracy by incorporating additional terms into cost functions that embed physical prior knowledge, traditionally designed by domain experts. The project involves developing an automated pipeline leveraging LLMs to generate such priors, improving the accuracy of measured parameters like misalignment between layers or feature sizes in semiconductor devices.
Domain Transfer with AI
Make data from device A resemble data from device B (e.g., oscilloscopes). Also applicable to simulated data.
Measurement Interpretation with AI
Extract physical parameters (e.g., impedance) from data using neural networks. Applicable to simulated data.
System Characterization with AI
Learn linear or nonlinear transfer functions of systems (e.g., step response). Applicable to simulated data.
Quantum Computing + ML
Digital Twin Modeling of Charge Sensors in Quantum Dot Devices
This project aims to build a digital twin of a single-electron transistor (SET) charge sensor using machine learning. You will train a neural network on recorded sensor scans to predict the charge sensor’s conductivity as a function of gate voltages. This model will simulate the SET response at various tuning points and assist in real-time tuning, eliminating the need for frequent manual re-tuning. The work contributes towards autonomous operation of charge sensors as a step towards scalable quantum dot architectures.
In collaboration with RWTH Aachen University. Details
“Shuttling” VAEs for Quantum Hardware Modeling
Use of VAEs to reconstruct and optimize the transport of single electrons in quantum processors.
Review, Evaluation, and Test Development for QML
Systematic analysis of QML approaches with implementation, benchmarking, and test suite design.
Biology + ML
Biomass Modeling with VAEs and Contrastive Learning
Training a VAE to estimate biomass from sensor data, including image-based inputs.
Integrating Domain Knowledge into VAEs via LLMs
Using large language models to provide domain-specific priors for VAEs (e.g., biological models for plant weight).
ML Engineering and MLOps
MLOps for Physics Simulations and/or Real Experiments
Development of data and model tracking systems for physical simulations and experiments based on existing MLOps tools.
Development of a Low-Code/No-Code AI Builder
Tool for ML applications in domains like biomass or quantum data.
Energy and Electrical Grids
Analysis of Electrical Grids with ML Methods
Prediction of load states and fault detection using autoencoders with local priors.
Simulation von Stromnetzen in KI-Frameworks (Master)
This thesis aims to implement power grid equations in differentiable ML frameworks (Keras, JAX) instead of traditional simulation environments. By formulating physical equations (Kirchhoff’s rules, component models) in autodifferentiable frameworks, the approach enables GPU-accelerated simulation, direct integration with ML pipelines, efficient gradient computation, and improved handling of missing parameters. The work involves transferring equations to an ML framework, simulating example networks, and evaluating the algorithmic approach.
Synthetische Generierung von Stromnetzen (Master)
This thesis focuses on generating synthetic power grids using machine learning to address the limited availability of real power grid datasets due to privacy requirements. The work involves creating ML models (particularly Variational Autoencoders and Graph Diffusion Models) to generate both realistic graph topologies and associated parameters. Generated networks will be verified through statistical analysis and load flow calculations to ensure they exhibit realistic characteristics across different consumer structures.
Industry Collaborations
Project with Industry Partner: AI and Computer Vision
Industry collaboration focused on AI and computer vision; topic details on request.
Contact
For questions about courses, thesis opportunities, or teaching collaborations, please contact us.