Justin Fu
Research Scientist at Google DeepMind
Berkeley, California, United States
Summary
🤩
Rockstar🎓
Top SchoolJustin Fu is a Research Scientist based in Berkeley with 11 years of experience at the intersection of machine learning research and systems engineering. He has contributed to high-impact projects at Waymo and now Google DeepMind’s JAX Core team, specializing in TPU-focused optimizations and distributed ML workflows (notably through Pallas and shard_map enhancements). His background includes inverse reinforcement learning work at DeepMind and Google, and open-source contributions to D4RL and JAX that span offline RL environments and advanced TPU tutorials. Justin pairs deep academic training—a PhD in AI from UC Berkeley and coursework at Stanford—with hands-on engineering that moves cutting-edge algorithms toward production. Colleagues describe him as comfortable shifting between low-level kernel optimizations and algorithmic research, often surfacing practical implementation details that improve large-scale training performance.
11 years of coding experience
4 years of employment as a software developer
Doctor of Philosophy (Ph.D.) Artificial Intelligence, Doctor of Philosophy (Ph.D.) Artificial Intelligence at University of California, BerkeleyComputer Science, Computer Science at Stanford University
English, Chinese
Github Skills (12)
gymnasium10
openai-gym10
environmental10
dev-environment10
jax10
environ10
python10
tpu10
reinforcement-learning10
enviroment10
distributed-computing10
machine-learning9
Programming languages (3)
Github contributions (5)
Farama-Foundation/D4RL
Apr 2020 - May 2022
A collection of reference environments for offline reinforcement learning
Role in this project:
Back-end Developer
Back-end DeveloperContributions:145 commits, 77 PRs, 86 pushes in 2 years 1 month
Contributions summary:Justin contributed to the development of reference environments for offline reinforcement learning by implementing and modifying code related to Minigrid and PointMaze environments. Specifically, the user added the implementation of several classes, methods, and attributes to extend and utilize the environment. The user also made adjustments to existing environment specifications by modifying the existing file to reflect the changes in the environment and making a couple of changes to the environments.
offline-reinforcement-learningroboticspybulletreinforcement-learninggymnasium
jax-ml/jax
Jun 2019 - Mar 2025
Composable transformations of Python+NumPy programs: differentiate, vectorize, JIT to GPU/TPU, and more
Role in this project:ML Engineer
ML EngineerContributions:82 reviews, 18 PRs, 4 pushes in 5 years 9 months
Contributions summary:Justin primarily contributes to the Pallas framework within the JAX project, focusing on TPU-specific optimizations and features. Their work includes implementing distributed computing tutorials and examples utilizing the RDMA model, enhancing the capabilities of collective operations with shard_map, and developing block-sparse kernel tutorials. These contributions showcase the user's expertise in leveraging Pallas for advanced TPU programming, particularly for machine learning workloads.
pytorchpythonjitautomatic-differentiationgpu
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.