Copy from a previous blog post, written in 12/2020.

Evolution of our Tracking and EDG model over time. The EDG model learns to capture actions with potential in a very general manner, and compute such potential with the player coordinates our Tracking model gathers from the live Camera.

This blog post is the markdown version of a list of Jupyter Notebooks you can find inside Narya’s repository. This post allows to have each Notebook at the same place. It will probably be replaced by a Jupyter Book whenever I find the time and the solution to integrate them into this blog.

This project is also an evolution from a previous blog post.

We tried to make everything easy to reuse, we hope anyone will be able to:

Copy from a previous blog post, written in 09/2020.

This blog post contains an introduction to Unsupervised Bilingual Alignment and Multilingual Alignment. We also go through the theoretical framework behind Learning to Rank and discuss how it might help produce better alignments in a Semi-Supervised fashion.

This post was written with Gauthier Guinet and was initially supposed to explore the impact of learning to rank to Unsupervised Translation. We both hope that this post will serve as a good introduction to anyone interested in this topic.


Example of the embedding of “Etudiant” in 6 dimensions.

Word vectors are conceived to synthesize and quantify semantic nuances, using a few hundred…

Copy from a previous blog post, written in 05/2020.

This post was made possible by @lastrowview and @Soccermatics which shared the tracking data of 19 goals scored by LFC during 2018–2019 and 2019–2020 seasons. Our code is available on this GitHub repository.

This article was co-written with Théophane, a classmate, while we worked on this project for a few days.

Sadio Mané, Mohamed Salah and Firmino celebrating.

This post illustrates how data analysis and machine learning can be applied to football players’ tracking data in order to reveal key insights. Our analysis will be articulated in two parts :

Copy from a previous blog post, written in 03/2020.

When we started talking about this project with Matthias and then Arthur, we knew that both building a Lego motorized car, and learning to drive with real-life Deep Reinforcement Learning was possible. However, we wanted to do both at the same time. We also wanted to try another approach: Imitation Learning. We recently acquired a Raspberry Pi 4b and knew that the power limitations set by the previous model vanished. However, we didn’t know exactly how to fit everything into a lego car, and how to train it properly. …

Copy from a previous blog post, written in 08/2019. The paper is now published.

When I started an Internship at the CEMEF, I’ve already worked with both Deep Reinforcement Learning (DRL) and Fluid Mechanics, but never used one with the other. I knew that several people already went down that lane, some where even currently working at the CEMEF when I arrived. However, I (and my tutor Elie Hachem) still had several questions/goals on my mind :

Copy from a previous blog post, written in 09/2018.


Codingame is a website where you can solve puzzles, work on optimization problems, and build bots for games. You have access to a vast range of codding languages, and I personally mainly used Java, Python & C++. What I enjoy the most is to build AI for multiplayer games: a card game, a quidditch match, a Tron battle, etc.

A quidditch game from Codingame where your AI controls 2 players. This is a great game that includes great AI and physics.

Paul Garnier

Building, robots, and trying to solve one equation at a time

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store