This master thesis focuses on optimizing the inference phase of graph processing models, addressing the unique challenges posed by graph-structured data. The research aims to improve model performance for enhanced user experience while exploring hardware-specific optimizations to achieve maximum efficiency in graph-based computation.
- Performance Enhancement: Develop techniques to reduce inference latency and improve throughput in graph processing models without compromising accuracy or graph analysis capabilities. - Graph Algorithm Optimization: Investigate and implement specialized methods for optimizing graph algorithms, including graph traversal, message passing, neighborhood sampling, and aggregation operations. - Hardware Acceleration: Explore hardware-specific optimizations for various platforms (GPUs, TPUs, CPUs, edge devices) with a focus on irregular memory access patterns, sparse computation, and parallel execution strategies particular to graph processing. - User Experience Metrics: Define and measure user experience metrics related to graph model inference speed and establish benchmarks for real-world graph-based applications.
- Knowledge of mathematical machine learning fundamentals. - Proficiency in at least one graph processing framework (PyG, DGL, GraphScope). - Experience with neural network architectures. - Basic CUDA or triton programming skills are beneficial. - Understanding of memory hierarchies, bandwidth constraints regarding Linux systems. - Programming in Python, C++ or Zig.
Sebastian Baum
This master thesis explores the integration of physical laws and constraints into graph-based geometric reconstruction processes. The research focuses on developing optimization frameworks that not only reconstruct geometric structures represented as graphs but also ensure the resulting structures adhere to fundamental physical principles. The aim is to achieve a more stable and reliable optimization. By incorporating physics-based constraints such structural stability and physical feasibility, the reconstruction process produces results that are both geometrically accurate and physically plausible.
You will develop a mathematical framework for physics-constrained graph optimization that incorporates relevant physical properties into the reconstruction process. The framework will be evaluated on a FEM case studies.
- Knowledge of mathematical machine learning fundamentals. - Proficiency in at least one graph processing framework (PyG, DGL, GraphScope). - Experience with neural network architectures. - Programming in Python, C++ or Zig.
Sebastian Baum
This master thesis explores methods for decomposing geometric structures represented as graphs into independent topological units. By identifying natural boundaries within geometric graph structures, these units can be processed separately while maintaining the overall topological integrity.
You will develop algorithms to identify and extract topologically coherent subunits from geometric graphs. These algorithms must preserve critical structural information while allowing independent processing of each unit. One major problem could the stitching of these independent units. You will implement and evaluate reconstruction methods that can generalize from these topological units to more complex structures. The thesis will include theoretical analysis of the approach's mathematical foundations as well as practical demonstrations in application domains such as 3D simulation processing.
- Knowledge of mathematical machine learning fundamentals. - Proficiency in at least one graph processing framework (PyG, DGL, GraphScope). - Experience with neural network architectures. - Programming in Python, C++ or Zig.
Sebastian Baum
Testing of AI-based systems such as autonomous vehicles is challenging due to many situations and scenarios. Brute force is expensive and has gaps, as we see in practice. We thus use synthetic data for an AI-driven testing. This data covers real-world scenarios to train autonomous systems in a simulation-based environment. The training success is evaluated in a data loop and enhanced to close blind spots and unknown knowns. This thesis targets to integrate a requirements and test engine to an automated test system.
The goal of the thesis is to integrate existing parts of the system. A fully running system shall be implemented. The integration comprises verification and validation checks for the existing parts. Professional tools such as DOORS shall be used for industry-scale AI-based testing of autonomous systems.
Knowledge in Python Industry-scale software engineering and tools Work in a self-independent way Passionate about clean and good quality code Capable of integrating your work with other parts of the system
Christof Ebert