The workshop “Italian Visual and Language models: Challenges and Activities” took place in Modena on February 5, 2024, organized by the Interdepartmental Center AI Research and Innovation (AIRI) and the AImageLab Laboratory of the “Enzo Ferrari” Department of Engineering at the University of Modena and Reggio Emilia, in collaboration with the FAIR Foundation.
The workshop presented the results of FAIR’s Transversal Project “Vision, Language, and Multimodal Challenges“, coordinated by Professors Rita Cucchiara (University of Modena and Reggio Emilia, CNR) and Roberto Navigli (Sapienza University of Rome), focusing on future developments of LMMs—foundational Artificial Intelligence models that integrate Large Language Models with image, video, audio, and multimodal data processing, representing the frontier of AI research.
A key highlight of the workshop was the presentation of results on Italian-language foundational models and the possibility of querying and retrieving visual data by interacting in Italian with multimodal archives. Among the projects presented were the first Italian language model, “LLaMAantino“, developed within FAIR by the University of Bari; the extensive effort in collecting multimodal and “egocentric” video data carried out by the University of Catania; and the “MORE” model – Multimodal mOdel and REtrieval, developed by UNIMORE for multilingual interaction between images and text and multimodal data retrieval. The MORE model was trained thanks to an ISCRA-B grant on the Leonardo supercomputer at CINECA, using public datasets, including those from LAION.
According to Lorenzo Baraldi, researcher at the “Enzo Ferrari” Department of Engineering, Deputy Director of the DHMORE Center and creator of MORE: “The MORE model will integrate dialogue capabilities, multimodal data analysis, retrieval techniques, and validation methods to create a system capable of interacting in Italian with visual content, retrieving accurate knowledge from external data sources, and justifying its responses by indicating their sources.”
The goal of the event was to initiate a discussion within the Italian community on the future challenges of designing, developing, and evaluating large language models, visual models and their integration, as well as their applications in sectors such as healthcare and manufacturing. During the final discussion session, participants also addressed the definition of the next FAIR project activities together with the Leonardo supercomputing center at CINECA, thanks to the agreement signed between CINECA and FAIR to support the entire Italian academic community in foundational AI research.
