June 2024 – Laboratoire Informatique d’Avignon

ANR DELIGHT Project

4 June 2024

Faire progresser l’apprentissage fédéré tout en réduisant l’empreinte carbone Les technologies de l’IA sont aujourd’hui trop gourmandes en énergie pour être compatibles avec nos objectifs de développement durable. Si des travaux récents ont permis d’évaluer l’empreinte carbone des méthodes d’apprentissage traditionnelles, celle d’une approche émergente comme l’apprentissage fédéré est mal connue. Le projet DELIGHT vise à évaluer finement et à réduire la consommation énergétique de l’apprentissage fédéré en utilisant différents leviers (compression de gradient, data summarization, speed-scaling, etc.). Étant donné l’hétérogénéité des données, un autre objectif sera l’étude du processus de négociation et de formation d’une coalition entre les nœuds pour comprendre dans quelle mesure un nœud a intérêt à collaborer avec d’autres. Les techniques développées seront validées empiriquement sur des tâches de vision par ordinateur et de TALN en utilisant le toolkit Flower Liste des partenaires : Responsable Scientifique pour le LIA : Rachid El Azouzi Date Début : 2023-03-01 Date Fin : 2026-09-01 En Savoir Plus

Cornet Seminar – Piet Van Mieghem – 10/06/2024

4 June 2024

Speaker: Prof. Piet Van Mieghem, Delft University of Technology (https://www.nas.ewi.tudelft.nl/people/Piet) Time: 10/06/2024 – 12pm Title: Linear Processes on Networks Abstract: From a network science point of view, we will discuss linear processes on a graph, which are the easiest, but also the most elegant processes. We start with the Laplacian matrix of a graph and concentrate on a flow problem in networks. From the spectral decomposition of the Laplacian, we will introduce the simplex of an undirected, possibly weighted graph. The simplex geometry of a graph is, besides the topology domain and the spectral domain, the third equivalent description of a graph. Continuous-time Markov processes are described by the Chapman-Kolmogorov (linear) equations, in which the appearing infinitesimal generator is, in fact, a weighted Laplacian of the Markov graph. We will discuss Markovian epidemics on a fixed graph, show limitations of the linear theory on graphs and introduce a mean-field approximation, a powerful method from physics, that results into non-linear governing equations. We will end with recent work on non-Markovian epidemics on a fixed graph. Bio available here: https://www.nas.ewi.tudelft.nl/people/Piet/

ANR MALADES Project

4 June 2024

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