Yasuhiro Fujita
Engineer at Preferred Networks, Inc.
Japan
Summary
🤩
Rockstar🎓
Top SchoolYasuhiro Fujita is an engineer with 13 years of experience building and refining machine learning and reinforcement learning systems, currently developing ML/RL algorithms for self-driving cars and industrial robots at Preferred Networks in Japan. He holds advanced degrees in artificial intelligence from The University of Tokyo and brings deep practical expertise in core deep learning frameworks, contributing to high-profile open-source projects like Chainer and CuPy. His contributions include implementing optimizers (RMSpropGraves), activation functions (ELU, CReLU), and fixing loss/backprop routines, alongside improving categorical DQN and A3C implementations—work that underscores a focus on numerical robustness and algorithmic correctness. Comfortable across backend engineering and test automation, he blends rigorous unit testing with low-level numerical work to make ML libraries production-ready. Fluent in both research-grade algorithm development and engineering hygiene, he pairs academic grounding with hands-on fixes that accelerate real-world RL deployments.
13 years of coding experience
Master of Information Science and Technology, Artificial Intelligence, Master of Information Science and Technology, Artificial Intelligence at Graduate School of Information Science and Technology, The University of TokyoBachelor of Engineering (BE), Artificial Intelligence, Bachelor of Engineering (BE), Artificial Intelligence at 東京大学 / The University of Tokyo
Github Skills (20)
algorithm10
pytorch10
python10
optimizers10
chainer10
testing10
machine-learning10
optimizer10
rms10
reinforcement-learning10
numpy10
dqn10
deeplearning-ai10
deep-learning10
optimisation10
Programming languages (4)
Github contributions (5)
pfnet/pfrl
Jun 2020 - Sep 2022
PFRL: a PyTorch-based deep reinforcement learning library
Role in this project:
Back-end Developer / ML Engineer
Back-end Developer / ML EngineerContributions:5 releases, 48 reviews, 110 commits in 2 years 3 months
Contributions summary:Yasuhiro contributed to the deep reinforcement learning library, focusing on core functionality. Their work included setting data types for returns in the A3C algorithm, adding test cases for non-scalar event shapes, and correctly handling event shapes in a sample with log probability function. They also implemented changes related to a SlimeVolley training script, enhancing the library's usability.
pytorchactor-criticdeep-learningreinforcement-learningdeep-reinforcement-learning
chainer/chainerrl
Feb 2016 - Apr 2021
ChainerRL is a deep reinforcement learning library built on top of Chainer.
Role in this project:ML Engineer
ML EngineerContributions:8 releases, 5 reviews, 2215 commits in 5 years 2 months
Contributions summary:Yasuhiro made several contributions to the categorical deep Q-network (DQN) implementation within the ChainerRL library. Their work focused on refining the core components of the algorithm, including the application of categorical projection and implementing methods for action evaluation and the correction of a typo. Further, they were involved in passing key arguments. This suggests a focus on refining and improving the performance of the DQN architecture.
actor-criticdeep-learningreinforcement-learningdeep-reinforcement-learningchainer
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.