Projects

Modèles de langue multimodaux et inclusifs pour le français général et clinique Le projet Pantagruel (ANR 23-IAS1-0001) ambitionne de développer et évaluer des modèles linguistiques multimodaux (écrit, oral, pictogrammes) inclusifs pour le français. Il mobilise des chercheurs de diverses disciplines telles que l’informatique, le traitement du signal, la sociologie et la linguistique pour assurer des résultats fiables et variés. Liste des partenaires : Responsable Scientifique pour le LIA : Yannick Estève (Equipe SLG) Date Début : 2023-11-20 Date Fin : 2026-11-20 En Savoir Plus

Grands modèles de langue adaptables et souverains pour le domaine médical français The recent arrival of Large Language Models (LLMs) and their associated tools for the general public reveals major challenges for society. Among the many fields that are, or will be, impacted by these generative models, the biomedical field is one of those that currently attract the attention of industrialists, researchers, but also the general public. Indeed, the need for tools and potential applications seems immense, whether, for example, at the level of the processing of textual documents, medical imaging, or even voice interaction. Due to the sensitive nature of the personal data handled and the fears of society associated with decision support tools, work in natural language processing (NLP) must innovate by addressing the issues inherent in this field. As part of the MALADES project, we presented innovative approaches for the integration of LLM in health centers. The aim is to equip these centers with NLP tools derived from LLMs and adapted for the biomedical field while maintaining sovereignty of the models and complete control of their health data. The work we carry out focuses on four areas of research: 1) the study of the legal and ethical Plus d'infos

Validity of a task of acoustic-phonetic decoding on anatomic deficits in paramedical assessment of speech disorders for patients treated for oral or oropharyngeal cancer Plus d'infos

Tracking and Detecting Fake News and Deepfakes in Arabic Social Networks The 4th Generation War (4GW) is known as an information war involving non-military populations. It is conducted by national or transnational groups following ideologies based on cultural beliefs, religion, economic or political interests, aiming to sow chaos in a targeted region globally. In 1989, authors discussing the 4th generation war, some of whom were military experts, explained that it would be widespread and challenging to define in the decades to come. With the emergence of social networks, the previously vague battlefield found a platform for 4GW. One of its penetration points is the extensive use of social networks to manipulate opinions, aiming to shape the targeted region’s perspective to accept a certain state of affairs and render it socially and politically acceptable. Much like the 4th generation war, cognitive warfare aims to blur comprehension mechanisms regarding politics, economy, religion, etc. Its consequence is destabilizing and weakening the adversary. This cognitive war targets what is assumed to be the enemy’s brain, altering reality by flooding the adversary’s population with misleading information, rumors, or manipulated videos. Furthermore, the proliferation of social bots today enables automated dissemination of disinformation on social networks. Plus d'infos

Efficient Self-Supervised Learning for Inclusive and Innovative Speech Technologies Self-Supervised Learning (SSL) has recently emerged as an incredibly promising artificial intelligence (AI) method. Through this method, massive amounts of unlabeled data that are accessible can be utilized by AI systems to surpass known performances. Particularly, the field of Automatic Speech Processing (ASP) is swiftly being transformed by the arrival of SSL, thanks in part to massive industrial investments and the explosion of data, both provided by a handful of companies. The performance gains are impressive, but the complexity of SSL models requires researchers and industry professionals in the field to have extraordinary computational capacity, drastically limiting access to fundamental research in this area and its deployment in everyday products. For instance, a significant portion of work using an SSL model for ASP relies on a system maintained and provided by a single company (wav2vec 2.0). The entire lifecycle of the technology, from its theoretical foundations to its practical deployment and societal analysis, therefore depends solely on institutions with the physical and financial means to support the intensity of this technique’s development. The E-SSL project aims to restore to the scientific community and ASP industry the necessary control over self-supervised learning Plus d'infos