Johann Brehmer
Member Of Technical Staff at CuspAI
Amsterdam, North Holland, Netherlands
Summary
👤
Senior🎓
Top SchoolJohann Brehmer is a machine learning researcher and engineer with a PhD in physics and a decade of experience applying ML and statistical methods to scientific problems, now focused on ML-driven material discovery for carbon capture. He has published 29 papers with over 2,100 citations and maintains open-source Python tooling, contributing robustness and cross-platform fixes to notable projects like PySR and nflows. At Qualcomm AI Research he led small teams across geometric deep learning, diffusion models, causality and offline RL, supervising interns and direct reports while bridging research and engineering. Comfortable across PyTorch, probabilistic modeling and scientific codebases, he excels at interdisciplinary problem solving that translates complex models into reliable, tested software. An uncommon asset: he pairs particle-physics rigor with hands-on backend improvements that reduce friction in open-source ML tooling.
10 years of coding experience
7 years of employment as a software developer
Abitur, Abitur at Ökumenisches Gymnasium BremenPhysics, Physics at Imperial College LondonDoctor of Philosophy (Ph.D.) Physics, Doctor of Philosophy (Ph.D.) Physics at Heidelberg University
English, German, Dutch
Github Skills (16)
algorithm10
file-handling10
pytorch10
machine-learning10
pathlib10
python10
data-science10
julia10
testing10
subprocess9
develop8
scikit7
scikit-learn7
automl7
unit-testing7
Programming languages (7)
Github contributions (5)
bayesiains/nflows
May 2020 - Oct 2020
Normalizing flows in PyTorch
Role in this project:
QA Engineer / Test Automation Engineer
QA Engineer / Test Automation EngineerContributions:5 commits, 2 PRs, 2 comments in 5 months
Contributions summary:Johann focused on improving the testing infrastructure within the repository. Their commits added and expanded tests for the ActNorm transform, specifically checking for consistency after reloading state dictionaries and verifying the behavior of the transform after modifications. The user also addressed issues with spline transforms, ensuring proper handling of edge cases and adding tests to confirm their functionality.
pytorchflowsdeep-learninggenerative-modelnormalizing
MilesCranmer/PySR
Oct 2020 - Oct 2020
High-Performance Symbolic Regression in Python and Julia
Role in this project:Back-end Developer & Data Scientist
Back-end Developer & Data ScientistContributions:7 commits, 1 PR, 12 comments in 1 day
Contributions summary:Johann primarily focused on modifying the `pysr` codebase to improve its functionality and system compatibility. Key contributions included implementing system-independent temporary file handling, making the deletion of temporary files optional, and ensuring the use of absolute paths for critical files. Further changes involved escaping backslashes to ensure cross-platform compatibility, along with modifying default behaviors. These changes suggest a focus on refining the software's underlying structure and improving its operational robustness.
regressionpythonnumpysymbolic-regressionmachine-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.