Scott Fujimoto
Research Scientist at Meta
Montreal, Quebec, Canada
Summary
🤩
Rockstar🎓
Top SchoolScott Fujimoto is a research scientist based in Montreal with a decade of experience specializing in deep reinforcement learning, now working at Meta after completing a PhD at McGill and research positions at Mila. His work spans both academia and industry, including internships and research roles at Google and Facebook, and he has contributed production-grade RL implementations such as SAC and TD3 to prominent open-source projects like Facebook Research’s ReAgent. Scott combines rigorous theoretical research with practical engineering—fixing GPU and value-network bugs, adding automatic entropy tuning, and iterating on PyTorch implementations for gym environments. He also has a strong teaching background at McGill, reflecting an aptitude for communicating complex ideas. Notably, his open-source contributions demonstrate a focus on robustness and hyperparameter efficiency that makes cutting-edge RL algorithms more practical in real-world systems.
10 years of coding experience
7 years of employment as a software developer
Diploma of Collegial Studies (DEC), Science, Diploma of Collegial Studies (DEC), Science at CEGEP - John Abbott CollegeHigh School, Science, High School, Science at Beaconsfield High SchoolDoctor of Philosophy - PhD, Computer Science, Doctor of Philosophy - PhD, Computer Science at McGill University
English, French
Github Skills (12)
algorithms10
pytorch10
machine-learning10
deep-learning10
tdd10
python10
reinforcement-learning10
gymnasium9
openai-gym9
data-structures8
algorithm8
data-structure8
Programming languages (1)
Github contributions (5)
sfujim/TD3
Feb 2018 - Jun 2021
Author's PyTorch implementation of TD3 for OpenAI gym tasks
Role in this project:
ML Engineer
ML EngineerContributions:28 commits, 20 pushes, 42 comments in 3 years 3 months
Contributions summary:Scott primarily focused on implementing and refining a Twin Delayed Deep Deterministic Policy Gradients (TD3) algorithm, a reinforcement learning technique. They implemented the core TD3 algorithm in PyTorch, including actor and critic networks. The commits reveal bug fixes, such as input adjustments for the critic, and adaptations for different environments, demonstrating an iterative development process.
pytorchreinforcement-learningopenaipytorch-implementationtd3
facebookresearch/ReAgent
May 2019 - Oct 2019
A platform for Reasoning systems (Reinforcement Learning, Contextual Bandits, etc.)
Role in this project:ML Engineer
ML EngineerContributions:11 commits, 1 PR, 1 comment in 4 months
Contributions summary:Scott significantly contributed to the reinforcement learning platform by implementing and refining the Soft Actor-Critic (SAC) algorithm. They addressed bugs related to GPU usage and value network implementations, improving the algorithm's performance. Furthermore, the user introduced automatic entropy tuning for SAC, enhancing its efficiency and reducing the need for hyperparameter optimization. The user also implemented Twin Delayed DDPG (TD3) for gym tasks.
reinforcement-learningcontextualbanditscontextual-banditsreinforcement
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.