Projects

Visit at CERN

HUMANIA: We are involved actively in the democratization of IA. To that end, we received funding from ANR for a 4-year chair. With the current rapid growth of AI research and applications, there are both unprecedented opportunities and legitimate worries about its potential mis-uses. In this context, we are committed to help making AI easier to access and use by a large population segment. Making AI more accessible to all should both be an important factor of economical growth and help strengthen democracy.

CODALAB: We are community lead of the Codalab competitions project. Codalab is an open source platform to organize challenges in data science and machine learning. We host at Paris-Saclay a public instance and an instance dedicated to teaching. Codalab hosts hundreds of competitions world-wide and has tens of thousands of users. We gave recently a brief presentation of our activities as challenge organizer at the French academy of sciences.

Automated Machine Learning: Since 2014, we have co-organized dozens of events (workshops and challenges) around the theme of automated machine learning, namely creating predictive models without human intervention whatsoever. We are currently running several new challenges: AutoDL (to automate Deep Learning), AutoCV (to automate Computer Vision) and AutoNLP (to automate Natural Language Processing), with workshops at IJCNN2019, WAIC2019, ECML PKDD 2019, and NeurIPS 2019. Lisheng Sun finished her PhD on the subject of meta-learning as a Markov decision process in December 2019. Zhengying Liu works hard to finish his thesis! He participated to the AutoDL grand challenge project using the Jean Zay supercomputer.

Learning to run a Power Network: Since 2016, we are collaborating with RTE to help predicting the flows in the French power network grid. We co-organized the 2 million Euro See.4C challenge to predict power flows in the RTE grid (see details). The winners have been announced. Our students Benjamin Donnot (who defended his thesis in January 2019) and Balthazar Donon have produced their own methods that were recently presented at the US national academy of sciences. The first edition of our challenge Learning to run a Power Network just ended. This was built upon Marvin Lerousseau's nice open-source package Pypownet. A new challenge accepted as a WCCI 2020 competition is in preparation. It is based on the Grid2Op framework developed by Benjamin Donnot.

Causality: Our interest in causality started in 2007, with the Causality Workbench project. In 2013, we organize a challenge in cause-effect pairs. We published a BOOK on the results of the challenge with some nice tutorial chapters. Diviyan Kalainathan finished his PhD on Causal Generative Neural Networks in December 2019. His software package was published in JMLR.

High energy physics: We co-organized with CERN two challenges in high energy physics: The High Boson challenge and the Track ML challenge. Victor Estrade is a PhD student working on improving the estimation and correction of systematic errors in the detectors of the Large Hadron Collider.

MediChal: With RPI New York (prof Kristin Bennet), we have a project to design generative models of fake medical data. This can be used to train students or create data science challenges without exposing confidential data.