Nicolas Vayatis (École Normale Supérieure Paris-Saclay) ran a Seminar@SystemX on June 28 from 2:00 pm to 3:30 pm in SystemX’s premises (Building 862, Amphitheater 34) on the following topic: “Nouvelles perspectives sur l’optimisation séquentielle et les plans d’expérience grâce au machine learning”.
Abstract
Dans la simulation numérique ou la conception de systèmes, on peut souvent formaliser les questions de plan d’expérience ou les problèmes inverses comme des problèmes de régression généralisés qui mettraient en correspondance espace de conception et espace de performance. Dans ces contextes, il est plutôt question de “small data” car les évaluations ou la réalisation de prototypes sont coûteux, il est pourtant tout aussi crucial de développer des approches dites prédictives permettant d’explorer ces espaces, parfois structurés, de façon ciblée. Au cours de l’exposé, un cadre méthodologique permettant de mettre en oeuvre des algorithmes d’apprentissage répondant aux enjeux de la prise de décision sera proposé. Les perspectives de la démarche proposée seront illustrées sur des exemples académiques ainsi que sur dans des applications dans le cadre de projets collaboratifs.
Biography
Nicolas Vayatis is Full Professor at the Department of Mathematics of ENS Paris-Saclay and is the Director of the Center for Mathematics and Their Applications (CMLA – CNRS & ENS Paris-Saclay). He also leads a research group on Machine Learning and Massive Data Analysis (MLDA) of about 20 people which is highly involved in interdisciplinary projects in the areas of network science, healthcare, digital marketing, and scientific computing.
His main research interests are machine learning theory and algorithms, predictive modeling, sequential optimization and inference problems arising from real graph data. Nicolas Vayatis has been the advisor of 13 PhD students (defended) and 9 postdoctoral researchers between 2009 and 2016 and has coauthored more than 70 publications in peer-reviewed international journals and conferences. He also serves as an Action Editor for the Journal of Machine Learning Research since 2007. He is regularly invited to participate to scientific committees for the main conferences of the field of machine learning (NIPS, COLT, ALT…).
Over the years, Nicolas has been intensively teaching applied mathematics, statistics and probability, data mining and statistical learning for engineers, math students, economists and psychologists, at various institutions (Université Pierre-et- Marie-Curie, ENS Paris-Saclay, Ecole Centrale Paris, ENSAE, Université Paris Nanterre, GeorgiaTech, Universitat Pompeu Fabra), and he currently is the main coordinator of the graduate master program MVA on Mathematics, Vision, and Learning which offers high level research training for more than 130 students per year. Nicolas Vayatis also provides expertise as a scientific advisor for the French Nuclear Agency (CEA).
