Open positions
Join us
We are always looking for talented, motivated researchers. If you are interested in working with us, check our open positions to apply to a specific opening, or send a spontaneous application by filling in the form.
All the research activities will be carried out at the University of Genova within a Machine Learning center across the Mathematics and Computer Science departments.
PhD students
Supervisors
Luca Calatroni - DIBRIS
luca.calatroni@unige.itAbout
In recent years, optimisation-based learning approaches have gained popularity in image reconstruction due to their interpretability, strong performance with limited training data, and applicability to problems involving well-defined (albeit approximate) physical models. A notable example is image super-resolution in fluorescence microscopy and, more broadly, the field of computational imaging. Typically, these approaches focus on learning specific algorithmic components—such as proximal operators, step sizes, or regularisation parameters—through neural network parameterisation. With appropriate network architectures, it is often possible to establish convergence and stability properties, which is crucial for producing reliable results in many applications.
This PhD project will explore this framework with a focus on models and algorithms suited to non-Gaussian noise and imperfect or potentially unknown forward models. Numerical validation will be conducted using real-world 2D/3D+t fluorescence microscopy data, in collaboration with G. Vicidomini’s research group at the Italian Institute of Technology, Genoa, Italy.
The position is estimated to start in March Prospective candidates should make an expression of interest before 05/12/2024.Perspective candidates should make an expression of interest before December 5, 2024.
Useful links
Supervisors
Luca Calatroni - DIBRIS
luca.calatroni@unige.itAbout
Model-aware learning approaches for image reconstruction effectively integrate the theoretical guarantees and interpretability of model-based regularisation methods for solving inverse problems with the expressive capabilities of data-driven techniques. These approaches are now widely used in imaging inverse problems and frequently achieve state-of-the-art results when combined with appropriate optimisation schemes. Notable examples include optimisation-based methods (such as unrolling and plug-and-play) and physics-inspired generative approaches (like GANs, VAEs, and diffusion models) where an expressive prior model is trained on large amounts of data.
The task of this PhD project is to design model-aware reconstruction procedures with provable reconstruction guarantees in the case where non-Gaussian noise statistics, imperfect space-variant and potentially time-dependent forward models (including stochasticity) are involved.
Numerical validations will be performed on real-world fluorescence 2D/3D+t microscopy data, in collaboration with the research group of G. Vicidomini (Italian Institute of Technology, Genova, Italy).
The position is estimated to start in March Prospective candidates should make an expression of interest before 05/12/2024.Perspective candidates should make an expression of interest before December 5, 2024.
Useful links