Koki Saitoh
エンジニア at Preferred Networks, Inc.
Kawasaki, Japan
Summary
🤩
Rockstar🎓
Top SchoolKoki Saitoh is an engineer specializing in deep learning and computer vision with 11+ years of experience across industry research and product teams, currently developing AI at Preferred Networks. A Tokyo Institute of Technology and University of Tokyo graduate, he previously conducted research at Toshiba and built engineering projects at teamLab. He is the author of the bestselling Japanese series "Deep Learning from Scratch," whose codebase he actively contributed to—implementing core layers, optimizers, RNN/LSTM blocks, an autodiff engine and reinforcement-learning examples—demonstrating deep low-level understanding of neural network mechanics. His work spans both theoretical foundations and practical implementations, from custom backpropagation engines to training loops for real tasks. Recognition includes multiple IT engineering book awards in Japan, and his books have collectively exceeded hundreds of thousands of readers. Colleagues describe him as someone who prefers building systems "from zero" to truly grasp and teach how complex models work.
11 years of coding experience
学士, 機械知能システム工学科, 学士, 機械知能システム工学科 at 東京工業大学修士, 情報学環, 修士, 情報学環 at 東京大学
Chinese, Japanese
Github Skills (27)
pytorch10
convolutional-neural-networks10
bilstm10
python10
machine-learning10
rnn-model10
n10
reinforcement-learning10
numpy10
automatic-differentiation10
lstm10
deep-learning10
neural-network10
relu10
softmax10
Programming languages (8)
Github contributions (5)
oreilly-japan/deep-learning-from-scratch-4
Oct 2021 - Jul 2022
『ゼロから作る Deep Learning ❹』(O'Reilly Japan, 2022)
Role in this project:
Data Scientist
Data ScientistContributions:72 commits, 9 PRs, 61 pushes in 9 months
Contributions summary:Koki contributed code related to implementing and comparing machine learning algorithms within the context of reinforcement learning. They added code for a non-stationary bandit problem, including the implementation of sample average and alpha constant update agents. Further contributions involved the implementation of policy evaluation and value iteration algorithms within a grid world environment, demonstrating experience with dynamic programming techniques. The user's work included the development of various reinforcement learning concepts.
oreilly-japan/deep-learning-from-scratch-2
Feb 2018 - Oct 2021
『ゼロから作る Deep Learning ❷』(O'Reilly Japan, 2018)
Role in this project:ML Engineer
ML EngineerContributions:65 commits, 5 PRs, 56 pushes in 3 years 8 months
Contributions summary:Koki primarily contributed to the implementation and refinement of deep learning models within the repository. Their work involved defining and implementing various time-series layers, including RNN, LSTM, and BiLSTM, as well as associated functionalities like embedding, affine transformations, and softmax with loss. These contributions suggest a focus on building the core components of a deep learning framework for sequence modeling, with the user actively involved in code additions and modifications related to model structure and functionality. The commit history strongly implies an understanding of backpropagation and gradient calculations, as shown in the inclusion of custom layer implementations.
deep-learningmachine-learningtensorflowo-reilly
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.