Taco Cohen
AI Research Scientist at Meta
Amsterdam, North Holland, Netherlands
Summary
🤩
Rockstar🎓
Top SchoolTaco Cohen is an AI research scientist based in Amsterdam with 13 years of experience bridging deep learning research and applied ML engineering. He holds a PhD (cum laude) from the University of Amsterdam on equivariant convolutional networks and co-founded a deep active learning startup that was acquired by Qualcomm, where he later progressed to principal research roles. Currently at Meta, he focuses on reinforcement learning and large language models for code generation, drawing on prior internships at OpenAI and DeepMind. Taco is also an early contributor to the Chainer deep-learning framework, adding crucial tensor ops like swapaxes and transpose and improving negative-dimension support. His background spans academic rigor, product-driven research, and low-level array/kernel work, giving him a rare blend of theory and practical systems fluency. Colleagues describe him as a researcher who consistently turns mathematical insights about symmetry into robust, production-ready models.
13 years of coding experience
17 years of employment as a software developer
Bachelors Computer science / Informatica (cum laude), Bachelors Computer science / Informatica (cum laude) at Utrecht UniversityDoctor of Philosophy - PhD Machine Learning, Doctor of Philosophy - PhD Machine Learning at University of Amsterdam
English
Stackoverflow
195reputation
18kreached
1answer
3questions
Github Skills (16)
cuda10
machine-learning10
deeplearning-ai10
deep-learning10
python10
cupy10
chainer10
numpy10
testing9
gpu9
fftw6
iterator6
cufft6
cython6
theano6
Programming languages (3)
Github contributions (5)
chainer/chainer
Oct 2015 - Oct 2015
A flexible framework of neural networks for deep learning
Role in this project:
Back-end Developer
Back-end DeveloperContributions:16 commits, 7 PRs, 27 comments in 26 days
Contributions summary:Taco primarily contributed to the Chainer deep learning framework by implementing and refining core array manipulation functions. They added `swapaxes` and `transpose` functions, crucial for tensor operations. Their work included writing unit tests and addressing code style issues. The user also incorporated improvements for negative dimension support within the `transpose` function.
cudapythonmxnetcaffe2flexible-framework
tscohen/GrouPy
Jun 2016 - Oct 2017
Group Equivariant Convolutional Neural Networks
Contributions:14 commits, 1 PR, 10 pushes in 1 year 4 months
deep-learningconvolutionalmachine-learningneural-networksneural-network
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.