Data Science et Intelligence Artificielle

L'Ecole peut compter sur l'activité de 7 laboratoires impliquant 50 enseignants-chercheurs appuyés par 100 doctorants dans le domaine de l'IA et Data. Au niveau national, CentraleSupélec est un partenaire clé de l'Institut DATAIA.

Ces personnes concentrent leurs efforts sur 4 thèmes de recherche principaux ci-dessous :

Explainable AI: Computer Vision, Natural Language Processing, Data & Signal Understanding. Laboratoires impliqués : CVN - L2S - LRI - MICS Thèmes : Computer Vision & Graphics Medical Image Understanding Machine Learning & Optimization Social Network Analysis
Neurosciences & Cognition avec l'équipe Facial Analysis Synthesis and Tracking (FAST) : Expressions & emotion analysis Cognition-driven Interpretation System
"From Photonics to AI" avec le laboratoire LMOPS : Translating concepts / algorithms from machine learning on physical hardware with a focus on photonics and optoelectronic architectures
Decision Making & Support avec le laboratoire du Génie industriel. Decision aiding systems Recommender Systems, Deliberative decision Preference elicitation & learning

1 - Environnement

SAR imaging: detect and manage flooded areas by Pr. Frédéric Pascal from L2S

Context: Use of heterogeneous satellites images of same areas to detect natural disasters
Random models, manifold learning MRF, change detection.

2 - Finance quantitative

How to model the operators behavior on financial markets by Pr. Frédéric Abergel from MICS.
Context: quantitative finance - from data, understand financial markets at several levels (micro, operators behavior, products, warrants…)
 

Random models, numerical simulations, massive and unstructured data analysis.
 

3 - Human-Machine interaction

Modeling and reproducing human emotions by Pr Renaud Séguier (FAST team, Rennes campus)
Context: Medical applications such as psychological injuries (traumas), nervous breakdowns.

E-learning: application of Human-Machine interaction M. Geist (UMI, Metz)
Context: reinforcing learning – where software agents learn to control dynamical systems based on interactions

Machine learning, sequential decisions, optimization

4 - Personalized Digital Health

Improving the diagnosis tools Nikos Paragios from CVN.
Context: New in vivo acquisition devices, 3D and 4D massive data. Artiticial vision to assist the eyes of the medical staff.

Computer Vision Machine Learning Discrete Optimization

Integrating knowledge for a better interpretation by Pr. Céline Hudelot from MICS.
Context: Interaction between data and knowledge for a better interpretation  of massive and complex data.

Artificial Intelligence (KR) Machine Learning Graphs Computer Vision

5 - Telecommunication

Ultrafast data processing for optical telecom by Damien Rontani fom LMOPS.
Context: Translating concepts / algorithms from machine learning on physical hardware with a focus on photonics and optoelectronic architectures

Ultra-large scale photonics neural networks - Hardware friendly training methods - Simulation of optical neuromorphic chips

Information Theory and Deep learning beyond Communications Pablo Piantanida (L2S)
Context: capitalize on powerful and fertile concepts from information theory  in order to advance the state-of-the-art in deep learning

Information theory – Deep Learning - Deep neural networks

• MICS/LOGIMAS (Céline Hudelot): Télécom ParisTech, Lamsade, Verone University, ICST Trento, Univ Cardiff, TU Berlin, TU Dresden, Neurospin, CNAM, Science Po
• L2S (Pablo Piantanida) : LIA - CNRS (International Associated Laboratory CNRS) on ‘Information, Learning and Control’: interplay between information theory, deep learning and control in Smart Grids (CS, Université Paris Sud, CNRS and several Canadian institutions: Université de Montréal, INRS, MacGill University, Concordia University, ET)
• L2S (Pablo Piantanida) : H2020-MSCA-IF-2016 : Information Theory beyond Communications: Distributed Representations and Deep Learning, (with Prof. Y. Bengio (MILA - Université de Montréal and IVADO)
• LMOPS (Damien Rontani): ULB, U Ghent, KU Leuven (Belgium), IHP (Germany), FEMTO-ST (France), Georgia Tech (USA), Saitama U (Japan)

L'Ecole noue des partenariats avec de nombreux grands groupes. Une chaire data a été par exemple récemment signée avec Givaudan. Et également avec des startups prometteuses telles que Dynamixyz, 3D Sounds Labs ou Immersive Therapy.