TU Graz Computer Scientist Awarded €2.5 Million ERC Grant for Generative AI Research
Thomas Pock, head of the Institute of Visual Computing at Graz University of Technology (TU Graz), has been awarded a 2.5 million euro Advanced Grant by the European Research Council (ERC). The five-year project, titled EAGLE (Efficient Algorithms for Generative Learning), aims to develop novel mathematical methods for generative AI, specifically focusing on improving the reliability of image reconstruction in fields like medical imaging.
What is the EAGLE project?
The EAGLE project seeks to address limitations in current AI systems regarding image synthesis and data interpretation. According to TU Graz, the research will focus on “Bayesian inverse problems.” Unlike standard AI models that often provide a single, deterministic solution, Pock’s team aims to create algorithms capable of generating multiple, equally plausible solutions from incomplete measurement data. This approach is intended to make uncertainties visible, allowing researchers to see which image details are supported by data and which are speculative.
Why is this research significant for medical imaging?
Modern medical diagnostics, particularly magnetic resonance imaging (MRI), rely heavily on reconstructing high-resolution images from limited measurement data. While AI is already used for this, a common challenge is the potential for “hallucinations” or the loss of critical diagnostic details. By developing generative models that quantify uncertainty, Pock’s research aims to provide clinicians with more reliable data. The project emphasizes that these models will not just analyze existing data but will be capable of synthesizing realistic examples, bridging the gap between data interpretation and generation.
How does the EAGLE project differ from current AI trends?
The project distinguishes itself by prioritizing mathematical efficiency over the current industry trend of building increasingly massive, computationally expensive models. Thomas Pock intends to demonstrate that significant scientific progress is achievable with moderate data requirements and lower computing power. By anchoring the research in sound mathematical theory rather than brute-force computing, the project aims to create more sustainable and interpretable AI systems.
ERC Advanced Grant Selection Process
The European Research Council awards Advanced Grants to established, leading researchers to pursue high-risk, high-reward projects. Competition for these grants is rigorous; in the most recent funding round, the ERC selected 319 projects from a pool of 3,329 applications across Europe. This marks the first time a researcher at TU Graz has received an ERC Advanced Grant, a milestone noted by Andrea Höglinger, the university’s Vice Rector for Research, as a testament to the institution’s growing influence in the fields of computer vision and artificial intelligence.
Key Project Details
- Principal Investigator: Thomas Pock, Institute of Visual Computing, TU Graz
- Funding Amount: 2.5 million euros
- Duration: 5 years
- Primary Objective: Develop efficient algorithms for generative learning and Bayesian inverse problems
- Application Area: Medical imaging and broader scientific data synthesis