WALDSPURGER Irène

Mathematics

waldspurger [at] ceremade.dauphine.fr

Irène Waldspurger

Short bio

I prepared a PhD at École Normale Supérieure de Paris, on phase retrieval and scattering transforms, under the supervision of Stéphane Mallat. I defended in 2015, then spent one year as a post-doc at MIT, mentored by Philippe Rigollet. Since then, I am a CNRS researcher.

Topics of interest

Non-convex optimization, inverse problems, scattering transform

Project in Prairie

It is known that simple non-convex algorithms can sometimes efficiently solve a priori difficult problems, like phase retrieval. This phenomenon has been rigorously explained under strong statistical assumptions only. I would like to understand better under which circumstances it happens. Another direction of research I would like to pursue is to study the links between the low layers of learned convolutional networks and the scattering transform.

ROBERT Christian

Computational statistics

xian [at] ceremade.dauphine.fr / Blog: xianblog.wordpress.com

Christian Robert

Short bio

Professor at Université Paris Dauphine since 2000, part-time professor at University of Warwick (Fall 2013- ), fellow of the ASA (2012) and the IMS (1996), former editor of the Journal of the Royal Statistical Society (2006-2010) and deputy editor of Biometrika (2018-), senior member of Institut Universitaire de France (2010-2021)

Topics of interest

Foundations of Bayesian analysis, Bayesian decision theory, Markov chain simulation methods, approximate Bayesian inference

Project in Prairie

To assess and improve approximate inference methods that handle complex and big data models, in particular developing novel ABC and MCMC technology. Contribution to the PSL maths graduate school by teaching and administrating the MASH program. Animation of international conferences and summer schools in Bayesian computational statistics.

Quote

Christian Robert travaille depuis une quinzaine d’années sur les méthodes d’inférence bayésienne approximatives, induites par la complexité ou la taille des données. Ses résultats valident des méthodes de Monte Carlo sur des modèles génératifs et aident à la construction de techniques de réduction de dimension efficaces.

PEYRÉ Gabriel

Applied mathematics

gabriel.peyre [at] ens.fr

Gabriel Peyre

Short bio

CNRS research director and professor at Ecole Normale Supérieure. Director of the data sciences center of the ENS. Blaise Pascal Prize 2017 of Académie des sciences, Magenes prize 2019 from the UMI. ERC starting grant 2012, ERC consolidator grant 2017.

Topics of interest

Optimal transport, imaging sciences, machine learning

Project in Prairie

The goal of my research project is to scale Optimal Transport methods both computationally and statistically to handle high dimensional machine learning problems. As deputy scientific director of PRAIRIE, I help to coordinate the research and teaching effort of the project. I am also be involved through my chair in fundamental and collaborative researches, as well as in teaching and dissemination of research.

Quote

Optimal transport (OT) is a fundamental mathematical theory at the interface between optimization, partial differential equations and probability. It has recently emerged as an important tool to tackle a surprisingly large range of problems in data sciences, such as shape registration in medical imaging, structured prediction problems in supervised learning and training deep generative networks.

GAIFFAS Stéphane

Machine learning

stephane.gaiffas [at] lpsm.paris

Stephane Gaiffas

Short bio

Professor at University Paris Diderot (since 2017) and part time professor at Ecole normale supérieure (since 2019). Previously associate professor at Ecole polytechnique (2012-2017).

Topics of interest

Statistical learning theory, online learning, optimization for machine learning and applications of machine learning in healthcare

Project in Prairie

« PERHAPS: deeP learning for ElectRonic HeAlth records and applications in ProStatic pathologies ». The aim is to exploit a set of electronic health records of patients with prostatic problems in order to improve heath care of such medical conditions. An example is deep learning models that predict longterm complications of surgery based on the full health care pathways of patients prior to it.

Quote

Recent successes in machine learning and deep learning applied to health are mostly concerned with computer vision (medical imaging) and biological signals (such as ECG data). A challenge is with huge electronic health records (such as accounting databases that contain codes for diagnoses, drug prescription and medical acts) and a combination of such databases with clinical data. The use of deep learning models for such huge databases that contains indirect clinical signals is an important challenge, that this chair will try to address.