Welcome Desk Opens
Our team will be ready to assist with your registration, provide conference materials, and answer any queries.
Our team will be ready to assist with your registration, provide conference materials, and answer any queries.
A distinguished member of the research community will set the tone with a brief overview of the conference's themes, objectives, and a warm welcome to all participants.
Explorez les dernières avancées en IA et analyse de données massives. Découvrez comment ces technologies transforment la recherche scientifique contemporaine et ouvrent de nouvelles perspectives.
Take a breather and mingle with peers during our coffee break. It's the perfect opportunity to discuss the morning's insights, refresh, and energize yourself for the upcoming sessions.
Isabeau Prémont-Schwarz , professeur agrégé, Dép. d'informatique et de génie logiciel, Université Laval Malgré ses succès spectaculaires, l'intelligence artificielle conserve au moins deux grandes faiblesses par rapport à l'intelligence humaine. D'abord, son appétit pour les données : là où un humain apprend une nouvelle tâche à partir de quelques exemples, les systèmes actuels en exigent des quantités considérables. Ensuite, la fragilité de ses réussites : quand une IA performe, c'est le plus souvent dans le domaine étroit où elle a été entraînée, et ses performances s'effondrent dès qu'on s'en éloigne. Cette présentation s'intéresse à la généralisation systématique - cette capacité, omniprésente chez l'humain, à transposer ce qu'on sait à des situations jamais rencontrées. Nous examinerons deux formes de généralisation qui nous sont naturelles mais qui mettent les IA en échec : d'une part, l'adaptation à une nouvelle configuration de l'environnement ; d'autre part, la capacité à traiter une combinaison inédite de facteurs de variation connus. Nous verrons comment ces deux situations résistent aux approches fondées sur l'apprentissage statistique classique, et nous explorerons quelques pistes de solutions pour concevoir des systèmes capables d'une telle généralisation.
Join us for a midday break where we will serve a variety of delicious, locally-sourced meals. This is a great chance to network, reflect on the morning's proceedings, and prepare for the afternoon's sessions.
Débattez des synergies entre disciplines scientifiques. Apprenez comment renforcer les partenariats de recherche pour résoudre les défis complexes de demain.
Explore a dynamic display of research findings during our poster session. Presenters will be on-hand to explain their work, answer questions, and engage in enlightening discussions. Perfect for broadening your perspective on current scientific advancements.
Unwind and savor a delightful dinner while we host a prestigious speaker from the scientific community. This evening event offers an informal platform for learning, networking, and inspiration. The perfect blend of academia, food, and conversation.
Our closing ceremony will offer a time for reflection on the conference's key learnings, with final remarks from the event organizers. We'll also be revealing student presentation prizes and details for next year's conference.
Rafael Pinot, Professeur junior, département de mathmatique, Sorbonne Université Major scientific discoveries increasingly rely on machine learning to extract knowledge from large datasets collected from multiple sources. This is particularly true in healthcare, where ambitious studies often require access to multicentric patient cohorts. Yet bringing these sensitive data together faces major privacy, legal, and organizational barriers. Federated learning (FL) offers a compelling alternative: learning a global model collectively while keeping data where they are. However, keeping data local is not enough to make collaboration trustworthy. Faulty or malicious participants may send arbitrarily corrupted updates and compromise the learning process. Designing learning algorithms that remain reliable in the presence of such participants is therefore particularly important in critical domains such as healthcare. In this talk, I will introduce the field of Byzantine-robust federated learning and the main challenges it seeks to address. I will present the fundamental ideas behind robust distributed optimization, discuss recent advances in the field, and highlight some of the key open problems that remain. Rafael Pinot is a Junior Professor in the Department of Mathematics at Sorbonne University. He holds a chair in the mathematical foundations of computer and data science within the LPSM research unit and is a founding member of the Responsible AI initiative at the Sorbonne Cluster for Artificial Intelligence (SCAI). From 2021 to 2023, he was a postdoctoral researcher at École Polytechnique Fédérale de Lausanne (EPFL). He completed his Ph.D. in Computer Science at PSL University (Paris Dauphine) and Paris-Saclay University (CEA LIST) from 2017 to 2020. His research centers on statistical machine learning, with a particular focus on the robustness and privacy of learning algorithms.
Profitez d'un moment pour réseauter et découvrir les projets d'étudiantes et d'étudiants en IA
Cocktail et remise de prix pour les affiches