Quentin Berthet
Research Scientist at Google DeepMind
Paris, Ile-de-France
Summary
🤩
Rockstar🎓
Top SchoolQuentin Berthet is a research scientist based in Paris with six years of industry experience bridging academic rigor and production ML at Google and DeepMind. Trained at Princeton (PhD) and École Polytechnique (BSc in Applied Mathematics), he has held roles from postdoc at Caltech to lecturer at Cambridge, blending theory with applied research. At Google/DeepMind he focuses on differentiable optimization and algorithmic building blocks for ML, contributing to notable open-source projects such as JAXopt where he implemented perturbation-based differentiable argmax/max routines. His work reflects a strong mathematical foundation applied to practical, hardware-accelerated tooling that aids end-to-end differentiable systems. Colleagues would point to his ability to translate subtle probabilistic ideas (Gumbel/Normal perturbations) into robust, test-covered code that scales in JAX. He combines deep theoretical insight with hands-on engineering to move advanced optimization methods from papers into widely used libraries.
6 years of coding experience
4 years of employment as a software developer
Bachelor of Science (BSc), Applied Mathematics, Bachelor of Science (BSc), Applied Mathematics at Ecole polytechniquePh.D., Ph.D. at Princeton University
Github Skills (8)
machine-learning10
differentiable-programming10
jax10
python10
optimization10
numpy9
testing8
deep-learning7
Programming languages (2)
Github contributions (5)
google/jaxopt
May 2021 - Jan 2023
Hardware accelerated, batchable and differentiable optimizers in JAX.
Role in this project:
ML Engineer
ML EngineerContributions:15 reviews, 5 commits, 1 PR in 1 year 8 months
Contributions summary:Quentin contributed to the development of differentiable optimizers within the JAXopt library. Their work focused on implementing and testing perturbations of argmax and max functions, as demonstrated by the code changes in `perturbations_test.py` and the `jaxopt/perturbations.py` file. These changes involved the use of Gumbel and Normal noise distributions, suggesting a focus on making optimization processes differentiable for use in machine learning applications. This involved defining new functions and adapting existing ones to be compatible with perturbations and differentiable operations.
pytorchdifferentiabledifferentiable-programmingautomatic-differentiationdeep-learning
q-berthet/q-berthet.github.io
Nov 2019 - Mar 2025
Contributions:73 pushes, 1 branch in 5 years 4 months
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.