Edouard Grave
Research Scientist at Kyutai
Paris, Ile-de-France
Summary
🤩
Rockstar🎓
Top SchoolEdouard Grave is a research scientist in Paris with nine years of experience building scalable, efficient machine learning systems for natural language understanding. His work blends theoretical advances in low-supervision NLP with practical engineering—contributions include fast, production-oriented code fixes and refactors to the widely used fastText library and architecture and decoder improvements to the jiant NLP toolkit. He has transitioned from strong academic roots (PhD and postdocs at Inria, UC Berkeley, Columbia) to research roles at Facebook and Apple and now Kyutai, focusing on algorithms that scale to billions of words while remaining computationally efficient. Notably, his background combines inventing unsupervised word-representation methods with hands-on C++ and model-level changes that improve robustness and training efficiency.
9 years of coding experience
9 years of employment as a software developer
Master's Degree Applied Mathematics, Master's Degree Applied Mathematics at École normale supérieure, CachanBachelor's Degree Computer Science Applied Mathematics, Bachelor's Degree Computer Science Applied Mathematics at École PolytechniqueDoctor of Philosophy (Ph.D.) Computer Science, Doctor of Philosophy (Ph.D.) Computer Science at Pierre and Marie Curie University
French, English
Stackoverflow
1,503reputation
60kreached
12answers
3questions
Github Skills (24)
transformers10
machine-translation10
pytorch10
c-language10
refactor10
sequence-to-sequence10
nlp10
refactoring10
cprogramming-language10
data-structure9
algorithm9
algorithms9
bert9
dropout9
data-structures9
Programming languages (4)
Github contributions (5)
facebookresearch/fastText
Jul 2016 - Apr 2019
Library for fast text representation and classification.
Role in this project:
Back-end Developer
Back-end DeveloperContributions:93 commits, 8 PRs, 15 pushes in 2 years 8 months
Contributions summary:Edouard primarily focused on bug fixes and code refactoring within the `facebookresearch/fasttext` repository. They addressed a rounding bug within the `utils::log` function and removed unused code in `Utils.cpp` and `Utils.h`. The user also refactored code related to model training, including changes to the argument order in matrix operations and changes to the `binaryLogistic` function. This involved changes across multiple C++ files, indicating a focus on improving the codebase's functionality and efficiency.
classifierbertword-embeddingsword2vectopic-modeling
nyu-mll/jiant
Jun 2018 - Aug 2018
jiant is an nlp toolkit
Role in this project:ML Engineer
ML EngineerContributions:11 commits in 1 month
Contributions summary:Edouard's contributions center around the development and modification of a sequence-to-sequence decoder module. They implemented a machine translation pretraining task, added dropout regularization to the decoder, and refactored the code related to initialization. These changes, along with modifications to the model's architecture, indicate a focus on improving the performance and functionality of NLP models, particularly for sequence generation tasks.
nlptransformersmultitask-learningsentence-representationbert
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.