Jacob Kelly
Research Engineer at Google DeepMind
Stony Stratford, England, United Kingdom
Summary
🤩
Rockstar🎓
Top SchoolJacob Kelly is a Research Engineer at Google DeepMind with a decade of experience building and optimizing machine learning systems across industry and academia. He has contributed to large-scale generative projects like Lyria and Imagen 2 while also improving core ML infrastructure—most notably adding differentiation rules and optimizers to JAX and contributing to TensorFlow Probability. His background spans probabilistic modelling, energy-based models, and model compression for genomics, translating research ideas into practical, faster models (e.g., 3.7x smaller, 3.3x faster models at Deep Genomics). Comfortable on clusters and in production, he combines strong math/stat foundations from U of T with hands-on engineering in ML primitives, ODE solvers, and biologically informed pipelines. Based in the UK, he brings a blend of low-level autodiff expertise and applied generative research that’s rare among research engineers.
10 years of coding experience
1 year of employment as a software developer
Bachelor of Science - BS Computer Science Mathematics Statistics, Bachelor of Science - BS Computer Science Mathematics Statistics at University of Toronto
Github Skills (15)
automatic-differentiation10
machine-learning10
probabilistic-programming10
tensorflow10
jax10
python10
numpy10
deep-learning9
statistics9
bayesian-methods8
algorithm8
optimization8
linear-algebra8
optimisation8
optimizers8
Programming languages (4)
Github contributions (5)
jax-ml/jax
Sep 2019 - Dec 2022
Composable transformations of Python+NumPy programs: differentiate, vectorize, JIT to GPU/TPU, and more
Role in this project:
ML Engineer
ML EngineerContributions:20 PRs, 48 comments, 15 issues in 3 years 3 months
Contributions summary:Jacob primarily contributed to the JAX library, focusing on implementing and improving jet rules for various mathematical functions, crucial for automatic differentiation in machine learning contexts. Their contributions involved adding jet rules for functions like erf, erfc, erf_inv, and other math operations. They also introduced Adamax, a type of optimizer used for gradient descent. Furthermore, the user added and tested various primitives to implement the core functionality in jax.
gpujittpujax
tensorflow/probability
Feb 2020 - Jun 2020
Probabilistic reasoning and statistical analysis in TensorFlow
Role in this project:ML Engineer
ML EngineerContributions:26 commits in 3 months
Contributions summary:Jacob contributed to the `tensorflow/probability` repository, a project focused on probabilistic reasoning and statistical analysis. Their commits primarily involved implementing and refining features related to the `jax` framework, which is used for high-performance numerical computing. This work included adding and modifying jet rules, which are used for automatic differentiation and higher-order derivatives, particularly for functions like erf, erfc, and erf_inv, as well as test cases involving those functions.
probabilistic-reasoningtensorflowbayesian-methodsdeep-learningmachine-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.