Paul Garnier

  • 2024 - Back at CEMEF to start a PhD on ML+CFD

  • 2020/21 - 2024 Flaneer: built a startup providing computer in the cloud for the creative industry (as the CTO). We raised $1.4M, grew to team to 8 people and served up to 10k users daily.

  • 2019-2021 MCs at MINES Paristech. I spent some time doing internships with Amazon (formerly the A9 team) in 2019 in SF working on the first deep learning models for Learning to Rank, Artefact in China, CNRS and Microsoft. Spent my last year at MINES Paristech mostly working on a startup project: Flaneer.

  • 2017 - 2019 BSc at MINES Paristech, where I started working on Deep Reinforcement Learning. Worked with Elie Hachem to apply DRL to Fluid Mechanics.

  • 2015 - 2017 Prep. school at Ginette, doing mostly math and physics.

Resume available here.

Blog posts

I wrote several articles why being the CTO of Flaneer. You can find them below:

Narya: track soccer players and evaluate them

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.

Semi-Supervised Learning for Bilingual Lexicon Induction

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 migh help to produce better alignments in a Semi-Supervised fashion.

Football Data Analysis - Liverpool FC attacking system

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 : Pitch control for opponent analysis & Deep Reinforcement Learning (DRL) for accessing action value.

The PI-Mobile, a self-driving Lego car powered by a Raspberry Pi

One Raspberry Pi, some Lego and a bunch of Deep Learning? Challenge accepted.

Deep Reinforcement Learning and Fluid Mechanics?

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) still had several questions/goals on my mind :

  • How was DRL applied to Fluid Mechanics?
  • To improve in both subject, I wanted to try a test case on my own;
  • How could we code something as general as possible?

Legends of Code & Magic - a Codingame contest

Create AI for games? Ok why not. For an Hearthstone-like game? Sounds ok. Look at your AI playing against others? Now this sounds like a plan. This contest was host by Codingame in Aug. 2018., and I had to use several exciting concepts to (or try to at least) solve it: Genetic algorithms, MC and MCTS, etc.

Other projects

  • Narya, allows you to track soccer player from camera inputs, and evaluate them with an Expected Discounted Goal (EDG) Agent trained on a FIFA-like environment.
  • Deepfluid, a library for Deep Reinforcement Learning applied to fluid mechanics.
  • MeshGradientPy lets you compute particular field gradient on a mesh, and makes them compatible with Machine Learning tensors.
  • TwitchAI let's you scrap a twitch chat, and then apply NLP models on it.
  • Small project wondering how to make parking place prices evolve in Paris.
  • I also spent a lot of time building bots for computer games competition.

Publications

or here on Google Scholar.