Naoto Usuyama
Principal Researcher at Microsoft
Seattle, Washington, United States
Summary
🤩
Rockstar🎓
Top SchoolNaoto Usuyama is a Principal Researcher at Microsoft Research in Seattle with 14 years of experience building AI systems for precision health, pharmacy, and applied computer vision. He leads work on multimodal foundation models for radiology and pathology (including projects like GigaPath, BiomedParse, LLaVA-Med and BioViL) and develops biomedical LLMs such as Medprompt GPT-4 and domain-specific BERT variants. His research has produced practical, high-impact outputs—e.g., fast medication identification published in Nature Digital Medicine and the ePillID low-shot pill identification benchmark presented at CVPR. Prior roles span applied research and software engineering at Microsoft and Microsoft Japan, where he contributed to widely used mobile imaging apps like Office Lens. He is also an active ML engineer in open source, implementing PyTorch U-Net and ResNet-based segmentation models that reflect hands-on expertise in medical image analysis. Combining deep research, product-minded engineering, and open-source practice, he focuses on translating multimodal AI into clinically useful tools.
14 years of coding experience
11 years of employment as a software developer
University of Tokyo
Japanese, English
Stackoverflow
855reputation
53kreached
6answers
9questions
Github Skills (14)
semantic-segmentation10
net10
pytorch10
image-segmentation10
segmentation10
keras6
deep-learning6
cntk6
tilde6
latex6
tensorflow6
python6
regex6
java6
Programming languages (14)
Github contributions (5)
usuyama/pytorch-unet
Jun 2018 - Aug 2020
Simple PyTorch implementations of U-Net/FullyConvNet (FCN) for image segmentation
Role in this project:
ML Engineer
ML EngineerContributions:18 commits, 1 PR, 11 pushes in 2 years 2 months
Contributions summary:Naoto implemented a U-Net model using PyTorch for image segmentation. Their work involved defining the FCN (Fully Convolutional Network) and training procedures within a Jupyter Notebook. The user also worked on creating a ResNet50-based U-Net architecture, and corrected some layer dimensions for the ResNet18 version.
image-segmentationpytorchunetfully-convolutional-networkssemantic-segmentation
usuyama/emacs-like-key-bindings-windows
Sep 2019 - Sep 2021
System-wide Emacs-like/Mac-like key-bindings on Windows 10 using AutoHotKey
Contributions:3 releases, 15 commits, 13 pushes in 1 year 11 months
autohotkeyemacskeybindingswindows
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.