Lukasz Kaiser
Member Of Technical Staff at CNRS
San Francisco, California, United States
Summary
🤩
Rockstar🎓
Top SchoolLukasz Kaiser is a research-driven machine learning scientist and engineer with over a decade of experience building foundational ML systems and state-of-the-art NLP models. After a rigorous academic career in logic, automata and program synthesis—including proofs resolving long-standing theoretical problems—he helped design and ship core ML infrastructure at Google Brain (TensorFlow) and later advanced research and products at OpenAI. His work spans from demonstrating that neural nets can learn complex discrete algorithms to practical contributions on Transformer inference and sampling (notably on the well-known trax library). Based in San Francisco, he blends deep theoretical insight with hands-on engineering to bridge research and production-scale ML. An often-overlooked strength is his background in symbolic methods and satisfiability solvers, which informs robust approaches to verification and program synthesis in ML contexts.
10 years of coding experience
10 years of employment as a software developer
Master of Science, Mathematics, Master of Science, Mathematics at University of WroclawPh.D, Computer Science, Ph.D, Computer Science at RWTH Aachen
English, French, German, Polish
Stackoverflow
2,196reputation
164kreached
26answers
0questions
neural-network
top-5%
tensorflow
top-1%
deep-learning
top-5%
python
top-5%
Github Skills (19)
machine-learning10
deeplearning-ai10
deep-learning10
tensorflow10
transformer10
jax10
python9
neural-network9
numpy8
nlp6
sequence6
lstm6
scopes6
recurrent-neural-networks6
language-model6
Programming languages (3)
Github contributions (5)
google/trax
Aug 2021 - Sep 2021
Trax — Deep Learning with Clear Code and Speed
Role in this project:
ML Engineer
ML EngineerContributions:4 reviews, 7 commits, 10 PRs in 1 month
Contributions summary:Lukasz primarily contributed to the `trax` repository by implementing and modifying features related to autoregressive sampling in the context of Transformer models. These changes included adding and refining the `eval_mode` functionality for decoding, incorporating depthwise convolutions, and making T5 imports dynamic. The work also involved adjustments to the sampling process, including handling minimum lengths and padding for evaluation, demonstrating a focus on model inference and performance.
pytorchspeeddeep-learningnumpyreinforcement-learning
lukaszkaiser/NeuralGPU
Nov 2015 - Mar 2016
Contributions:3 commits, 2 pushes, 1 branch in 3 months
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.