Many problems, from CT scans to MRI, require reconstructing parameters we can't measure directly: an inverse problem. We study models with dynamical, controllable components, treating measurement design itself as optimal control. Data-driven methods have advanced reconstruction and measurement design, but mostly in isolation. This Special Research Area unites both into one framework, applied concretely to MRI. Based in Graz, we bring together optimization, inverse problems, and machine learning to improve clinical imaging and train the next generation of researchers.
Coordination project
The Coordination project manages administration, communication, outreach, and central research data stewardship of the SFB, ensuring seamless collaboration and FAIR-compliant dissemination of all research outputs.
Subproject Head
Team
Controlled parameter identification in time dependent PDEs
IdCONTROL establishes a mathematical framework for jointly solving reconstruction, parameter identification, and control design problems in MRI using time-dependent PDE models.
Subproject Head
Team
Optimal control for optimal operators
OptOP develops an integrated optimization framework for designing measurement operators in dynamical systems, aiming at optimal parameter reconstruction from noisy data with a focus on MRI applications.
Subproject Head
Team
Bilevel control of sequential parameter identification
BiCONTROL formulates optimal design in time dependent inverse PDE problems as a bilevel and model predictive control problem to compute design parameters such as MRI pulse sequences that minimize reconstruction error.
Subproject Head
Team
Bilevel learning for joint motion and image reconstruction
BiLEARN develops and analyzes bilevel learning schemes for image reconstruction, focusing on parameter optimization, motion estimation, texture-aware regularization, and stability, with applications to MRI.
Subproject Head
Team
Structured model learning
ModLEARN advances MRI acquisition models by integrating learned components into PDE based physical models to capture nonlinear effects and model imperfections, with a focus on structured model learning for the Bloch Torrey equation.
Subproject Head
Team
Prior learning
PriorLEARN develops Bayesian and data driven approaches for dynamic MRI reconstruction by learning robust priors, enabling uncertainty quantification, and advancing probabilistic methods for linear and nonlinear inverse problems.
Subproject Head
Team
Generalized pulse sequences for MRI
GenMRI develops a comprehensive numerical framework for designing and validating generalized MRI experiments that integrates mathematical advances from the SFB with practical hardware, pulse sequence, and reconstruction considerations.
Subproject Head
Team
Associated members
Associated members are researchers who contribute to the SFB's research activities. They may be involved in specific projects, collaborations, or provide expertise in certain areas relevant to the SFB's goals.
Contact us
If you would like to get in touch you can email us at mr-dynamo@uni-graz.at.
