Roman Novak
Research Scientist at OpenAI
San Francisco, California, United States
Summary
🤩
Rockstar🎓
Top SchoolRoman Novak is a research scientist in deep learning with 7 years of experience bridging theory and production-grade ML systems, currently working on the science of deep learning at OpenAI after research roles at Google Brain and DeepMind. He has a strong mathematical background from École Polytechnique and ENS/Paris-Saclay and a track record of improving core ML tooling—contributing stability, numerical fixes and new features to JAX, Flax, TensorFlow Probability and the influential Neural Tangents library. Roman’s work spans probabilistic modelling, infinite-width neural network theory and practical engineering, evidenced by tasks added to BIG-bench and low-level fixes that reduce NaNs and rank-promotion bugs in XLA/JAX. Colleagues describe him as equally comfortable proving theorems in functional analysis and shipping robust, well-tested examples and layers used by the wider ML community.
7 years of coding experience
3 years of employment as a software developer
Igor Sikorsky Kyiv Polytechnic InstituteEngineer’s Degree, Computer Science, Engineer’s Degree, Computer Science at Ecole polytechniqueMaster’s Degree, Mathematics, Vision, Learning, Master’s Degree, Mathematics, Vision, Learning at École Normale Supérieure de CachanMaster’s Degree, Mathematics, Vision, Learning, Master’s Degree, Mathematics, Vision, Learning at Université Paris-Saclay
English, French, Ukrainian, Russian
Github Skills (26)
convolutional-neural-networks10
probabilistic-programming10
python10
testing10
machine-learning10
flax10
numpy10
tensorflow10
neural-network10
workbench10
nlp10
testbench10
jax10
operation9
tensorrt9
Programming languages (4)
Github contributions (5)
google/neural-tangents
May 2019 - Jan 2023
Fast and Easy Infinite Neural Networks in Python
Role in this project:
ML Engineer
ML EngineerContributions:16 releases, 33 reviews, 446 commits in 3 years 8 months
Contributions summary:Roman updated tests and examples within the `google/neural-tangents` repository, focusing on compatibility with the new JAX optimizers API and related API updates. The changes included refactoring and cleaning up code, removing unused imports, and aligning the testing framework with changes in the underlying JAX library. The user also modified existing examples and added a new one, updating them to reflect the latest JAX optimizers API, and ensuring functionality and accuracy.
kernelpythonbayesian-inferencebayesian-networksgaussian-processes
google/flax
Sep 2021 - Feb 2022
Flax is a neural network library for JAX that is designed for flexibility.
Role in this project:ML Engineer
ML EngineerContributions:8 reviews, 9 commits, 2 PRs in 5 months
Contributions summary:Roman primarily contributed to the `google/flax` repository, a neural network library for JAX, focusing on implementing and testing various convolutional layers and functionalities. Their work includes adding the `ConvLocal` layer, incorporating circular padding, and performing testing of convolution operations. They also made internal changes and refactored code to switch to subclassing in `flax.linen.Conv[Local]`.
deep-learningneural-networksneural-networkflaxjax
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.