CV

PhD

Mila - Université de Montréal, Montréal, Canada

• Supervised by Prof. Yoshua Bengio
• Deep Learning Applications to Climate Change Mitigation
• Visualizing the consequences of climate change (2019-2022)
  • Problematic: GANs were invented at Mila. Can we use them for good?
  • Solution: Use GANs to trigger empathy through personalized visualizations of climate change-related extreme events: floods, wildfires and smog.
  • Website: thisclimatedoesnotexist.com where you can search for any address on Google Street View and see what that place would look like if it were impacted by one of the aforementioned extreme events. The ultimate test set!
  • Paper: ClimateGAN: Raising Climate Change Awareness by Generating Images of Floods (ICLR 2022)
• Measuring the carbon impact of AI (2019-2022)
  • Problematic: AI is a promising tool to fight climate change, but it also has a carbon footprint. How can we measure it?
  • Solution: create tools for the community to measure the carbon impact of their AI models and experiments.
  • Online calculator: Quantifying the Carbon Emissions of Machine Learning (Climate Change AI workshop, NeurIPS 2019) -> mlco2.github.io/impact
  • Python package: codecarbon
• Materials discovery (2022-2024)
  • Problematic: Materials discovery is a slow and expensive process. Can we use AI to accelerate it?
  • Solution: Use GFlowNets to explore the space of electro-catalysts in order to discover materials able to improve the energy efficiency of electro-chemical reactions such as the electrolysis of water for energy storage.
  • Paper: PhAST: Physics-Aware, Scalable, and Task-specific GNNs for Accelerated Catalyst Design (JMLR 2024)
  • Paper: FAENet: Frame Averaging Equivariant GNN for Materials Modeling (ICML 2023)
  • Paper: Crystal-GFN: sampling crystals with desirable properties and constraints (AI4Mat @ Neurips 2023)

MSc. Machine Learning

University College London, London, UK

• Distinction award
• Supervised learning, Graphical models, RL, NLP, Computer Vision.
• NLP Research Internship with Prof. Sebastian Riedel (4 months, UCL Machine Reading Group): Deep RL & NLP to answer questions using Wikipedia.

Engineering degree

École polytechnique, Palaiseau, France

• Entrepreneurship and Data Science majors
• Courses: Quantum Physics, Computer Science, Operations Research, Entrepreneurship, Data Science

NLP Data Scientist

Ministry for the Economy and Finance, Paris, France

• Public Interest Entrepreneur
  • "During 10 months, a cohort of about 30 Public Interest Entrepreneurs work within public administrations in teams of 2 or 3 on challenges linked to data and digital transformation." (Etalab.gouv)
• Document classification and sentence highlighting
• Validation web app to mesure noise in the labels and improve understanding of the models.
• Project page (French)

Data Science Engineer

Rythm, Inc, Paris, France

• Deep learning infrastructure using Keras & Lasagne
• Signal quality prediction based on two seconds of electro-encephalogram signal with LSTM networks
• Backend server to store and compare predictors

Software Engineer Intern

Gemalto, Singapore

• Development of a mobile web app

PaperMemory

• A browser extension to automatically record the papers you read. Also finds code repositories and matches Arxiv papers to their published version. Never lose a paper again.
  • JavaScript (+Gulp, Mocha, Pupeeteer), HTML, CSS.
  • Started as a personal tool, now maintained for a community
• 450+ ⭐️

CodeCarbon

• A popular Python package to estimate the carbon footprint of your code.
  • Creator of the project
  • Project leader for 2 years
  • Coordinator of a team of 5-10 open source contributors
  • Major code contributor
  • Data For Good France Season 9 leader
• 1.1k+ ⭐️

GitmoPy

• A simple Python CLI to use the Gitmoji convention in your commits
  • Mostly a personal tool to learn how to build a Python CLI with typer and GitPython

• Druide
• Antidote fellowship winner

• University College London
• Distinction

• École polytechnique
• Outstanding Investment award