Nicolas Dufour
Postdoctoral Researcher at Kyutai
Paris, Ile-de-France
Summary
🤩
Rockstar🎓
Top SchoolNicolas Dufour is a Paris-based postdoctoral researcher with eight years of experience specializing in generative models, deep learning and reinforcement learning, combining strong academic training (PhD-level work at École des Ponts and École Polytechnique) with applied engineering. He has contributed to high-profile open-source ML tooling—most notably enhancements to PyTorch’s TorchRL (reward rescaling, TensorDictModule tutorials and multi-logger support)—and interned on model-based RL at Meta. His background spans vision and generative adversarial research, anomaly detection systems in finance, and scalable data engineering with Spark and Hadoop, reflecting comfort across theory, experiments and production code. Notably, he blends rigorous probabilistic and neural approaches (work on semi‑Markov models and neural tangent perspectives) with practical implementations, making him effective at turning cutting‑edge research into reusable software.
8 years of coding experience
Msc MVA (Mathematics / Vision / Machine Learning), Msc MVA (Mathematics / Vision / Machine Learning) at ENS Paris-SaclayMPSI-MP, MPSI-MP at CPGE JoffreBaccalauréat Scientifique, Baccalauréat Scientifique at Lycée Français Paul Valéry CaliEngineering Diploma (Equivalent to a Master of Science), Engineering Diploma (Equivalent to a Master of Science) at Télécom SudParisDoctor of Philosophy - PhD Mathématiques et informatique, Doctor of Philosophy - PhD Mathématiques et informatique at École nationale des ponts et chausséesDoctorat de philosophie Mathématiques et informatique, Doctorat de philosophie Mathématiques et informatique at École Polytechnique
Spanish, French, English, Portuguese
Stackoverflow
203reputation
633reached
1answer
6questions
Github Skills (15)
pytorch10
machine-learning10
python10
reinforcement-learning10
pytest9
unit-test9
unit-testing9
documentation8
wandb7
content-management-system6
tensorboard6
build-process6
api6
monitoring6
ant6
Programming languages (5)
Github contributions (5)
pytorch/rl
Jul 2022 - Oct 2022
A modular, primitive-first, python-first PyTorch library for Reinforcement Learning.
Role in this project:
ML Engineer
ML EngineerContributions:85 reviews, 32 commits, 31 PRs in 3 months
Contributions summary:Nicolas contributed to the PyTorch RL library by implementing and testing a reward rescaling transform. This involved modifying the test suite to include tests for the new transform and modifying the transforms to allow the user to easily scale and shift the rewards. Furthermore, they added a tutorial on the `TensorDictModule` which is a crucial part of the library, which involved the use of the core functionalities of the library. The user also added new loggers, Wandb, and refactored existing logging to better support multiple loggers and logging.
pythonpytorchreinforcement-learningaidecision-making
nicolas-dufour/torchrl
Jul 2022 - Oct 2022
A modular, primitive-first, python-first PyTorch library for Reinforcement Learning.
Contributions:6 PRs, 592 pushes, 58 branches in 3 months
pytorchpythonreinforcement-learningprimitivedeep-reinforcement-learning
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.