MRIM - Modélisation et Recherche d’Information Multimédia

Between 2019 and 2024, the MRIM team focused on 3 key areas:

• Explainable multimedia models,
• IR Models: formal models of logical IR, models based on language models, etc.
• The evaluation of information retrieval systems.

In line with the scientific objectives of the MRIM team, we collaborate with the Getalp, APTIKAL and SLIDE teams of the SIDCH axis of the LIG: these collaborations have made it possible to explore the involvement of Large Language Models (LLM) in new IR models.

Within the laboratory, collaborations have made it possible to focus on applications related to our core business: WildCOunt project with Erods, Eye tracking for information retrieval with M-PSI, Transparency of search engines on the Web with Vasco. These interactions are essential to promote the exploration of innovative proposals.

The MRIM team is heavily involved in its research environment. This is reflected by active participation in the MIAI institute with participation in the Explainable and Responsible AI, MyWayToHealth, Artificial Intelligence & Language chairs, which led to the co-supervision of 3 theses. The Artelia chair of the Grenoble INP foundation is also funding a thesis. The Grenoble Labex Persyval has also enabled the funding of a thesis. The director of the Carnot institute is a member of MRIM. MRIM also participated in the creation of a start-up in the field of medical images by the SATT Linksium: Peektoria, created in 2022. Finally, MRIM regularly obtains Emergence funding from the LIG.

Our work is the subject of demonstrations (for example at SIGIR 2023), and some of our developments are also distributed, such as ML-WIKIR in 2020 (python library to build multilingual test collections for information retrieval, online on GitHub), or CaBRNet in 2024 (library for the development and evaluation of case-based reasoning models).

MRIM actively participates in the construction of test collections (LongEval 2023 and 2024). This work helps promote common frameworks in the field of information retrieval: the data (millions of documents, thousands of queries, tens of thousands of relevance measures) are accessible and have been downloaded more than 300 times in the last 2 years.