The computer scientist is receiving a grant of 2.5 million euros to develop innovative algorithms in computer vision. These innovative AI models are designed to be able to both understand and realistically generate images, thereby improving medical imaging, for example.
Thomas Pock, head of the Institute of Visual Computing at Graz University of Technology (TU Graz), has been awarded an Advanced Grant from the European Research Council (ERC) for his project EAGLE – Efficient Algorithms for Generative Learning. In his research, the computer scientist aims to develop novel generative learning methods and algorithms. He will receive funding of 2.5 million euros for this, with the project running for five years.
First ERC Advanced Grant at TU Graz
“This ERC Advanced Grant is a first for our university and a well-deserved honour for Thomas Pock, who has been setting new standards for over a decade with his research at the interface of computer science, mathematics and medical imaging,” says Andrea Höglinger, Vice Rector for Research at TU Graz. “I offer my warmest congratulations on this grant, which is testament to the outstanding quality and relevance of our research focus on Computer Vision and Artificial Intelligence.”
AI systems now play an important role in reconstructing images and other data from incomplete measurements. A well-known example from the field of medicine is magnetic resonance imaging (MRI), in which high-resolution images are generated from a small amount of measurement data. However, there is a risk that important details may be lost or that uncertainties may remain hidden.
More than just a solution: making uncertainties visible
As part of the ERC project EAGLE, Thomas Pock is developing new mathematical methods for what are known as Bayesian inverse problems. Rather than calculating just a single solution from the measurement data, the system is designed to generate many different, but equally plausible, solutions using generative AI and novel, efficient sampling algorithms. This makes it clear which information – in the case of MRI, for example, the image details – is actually supported by the measurement data, and where uncertainties lie.
Understanding data through data synthesis
A key objective of the project is to establish a close link between data analysis and data generation. The models developed should not only be able to interpret data, but also to generate realistic examples of the same kind. This results in an approach in which an understanding of data is specifically supported by the ability to synthesise information. This enables more reliable predictions and improved quantification of uncertainty – in medical imaging as well as in numerous other scientific and technical fields of application.
The approach taken in the EAGLE project deliberately sets itself apart from the global trend towards ever-larger and more computationally intensive AI models. “The aim is to show that significant progress can be made even with a moderate use of data and computing power – based on sound mathematical theory,” says Thomas Pock.
Only 319 Advanced Grants awarded by the ERC
Through its Advanced Grants, the European Research Council offers established leading researchers the opportunity to undertake highly ambitious projects that promise scientific breakthroughs. Of the 3,329 project applications submitted across Europe in this funding round, 319 were selected for funding. In Austria, twelve projects were awarded funding.