From 2–5 September 2025, the University of the Balearic Islands (UIB) hosted the Workshop UIB–UBS: New Trends in Inverse Problems Applied to Images, a joint scientific meeting between the Mathematical Imaging and Learning (MIA) group at UIB–IAC3 and the Laboratoire de Mathématiques de Bretagne Atlantique (LMBA) at Université Bretagne-Sud (UBS).
This event builds upon years of collaboration between the two groups, which began with a jointly supervised PhD thesis, defended back in 2021, and has since developed into several joint publications and research visits. The workshop aimed to exchange expertise on inverse problems in imaging, an area where mathematics and artificial intelligence meet to recover high-quality images from degraded or incomplete data.
Over four days, researchers from both institutions shared their current work and explored new avenues for collaboration. The program included presentations from:
- Franck Vermet (UBS) – Mathematical foundations and advances in diffusion-based generative models.
- Corentin Vazia (UBS) – Solving inverse problems with diffusion models: a general framework with application to spectral CT reconstruction.
- Jacques Froment (UBS) – Explainability in diffusion models: techniques and applications to hallucination detection.
- Nazim Kerkech (UBS) – A survey of conditional generative models for satellite image fusion and future research directions.
- Claire Launay (UBS) – Beyond denoising diffusion: an overview of recent generative models.
From the UIB side, several members of the MIA group presented their contributions:
- Francesc Alcover (UIB) – Nonlocal BV and nonlocal Sobolev spaces induced by nonfractional weight functions.
- Cristian Comellas (UIB) – Deep learning methods for joint denoising and demosaicking.
- Ivan Pereira (UIB) – Model-based deep learning approaches for multi-modal image super-resolution.
- Marco Sanchez (UIB) – Patch-based hybrid method for image denoising.
- Dani Torres (UIB) – A Retinex-based variational model with a nonlocal gradient-type constraint for low-light image enhancement.
Beyond the talks, the afternoons were dedicated to working sessions, where participants discussed the presented research, identified common interests, and drafted ideas for future collaborative projects.
The topics addressed during the workshop reflect the complementary expertise of the two groups: while the UBS team has recently advanced the use of generative diffusion models for inverse problems, the UIB team develops hybrid approaches that combine variational models, nonlocal regularizers, and deep learning techniques. Applications ranged from medical imaging (such as spectral CT) to satellite image fusion and low-light image enhancement.
The workshop fostered a stimulating exchange of ideas and set the stage for future cooperation between UIB and UBS in the fields of mathematical imaging, computer vision, and artificial intelligence.
