The PhD will be conducted in two laboratories at Paris-Saclay: one group with expertise in artificial intelligence developed over many years, MIA-PS (INRAE), and another in the physics of matter – soft matter, biology – MMB-LLB (CEA/CNRS).
Small-Angle Scattering techniques (X-rays, neutrons, light) involve a constantly growing community, particularly active in France, especially in soft matter and biology. The transition of data from reciprocal space to real space is achieved via different models – in which the MMB group is an expert – whether concerning shape – sphere, rod, platelet, polymer chain – or interactions – attraction, aggregation, repulsion, arrangement. Furthermore, more complex structures, such as proteins or irregular aggregates, require computational or empirical approaches. In all cases, the results are not unequivocal. This is particularly challenging for research groups new to the technique.
In this thesis, thanks to MIA-PS's expertise in AI (machine learning, optimization, visualization), the focus will be on developing explainable AI methods. Part of the modeling involves explained mathematical and physical models, while another part relies on so-called 'black box' models, which will be progressively explained. The doctoral candidate will have access to data from three use cases provided by the LLB, and to their experts, to develop a generic methodology. A first step could be based on the globally shared software SasView, a treasure trove of explicit models. We have already received a positive response from the SasView developers, which could therefore serve as a dissemination tool. A valuable contribution will be the access to complementary DPA measurements via the LLB platforms and the SOLEIL and ESRF synchrotrons.
Subsequently, a component focusing on human-computer interaction—ensuring that users remain fully responsible for constructing a physico-chemical-biological explanation—can be implemented. MIA-PS is also an expert in advanced interactive visualization methods.

This project therefore combines highly advanced developments in computer science with a wealth of real-world systems chosen for their originality and, of course, their potential applications.