Matthew Leavitt
Chief Science Officer, Co-founder at DatologyAI
San Francisco Bay Area United States
Summary
🤩
Rockstar🎓
Top SchoolMatthew Leavitt is a neuroscience-trained AI researcher and entrepreneur with nine years of experience building and scaling empirical deep learning systems. As Co-founder and Chief Science Officer at DatologyAI, he translates research-grade understanding of neural mechanisms into practical ML products, after leading data research at MosaicML and contributing to high-profile open-source projects like Facebook Research's VISSL. His work focuses on making ML more efficient—improving vision transformer implementations, self-supervised learning pipelines, and training tooling—while maintaining a strong emphasis on reproducibility and documentation. A former FAIR AI Resident and postdoc, he bridges cognitive neuroscience and machine learning to bring experimental rigor to model development and analysis. He also has a long history mentoring and organizing technical communities, from student conferences to industry research teams.
8 years of coding experience
7 years of employment as a software developer
The Athenian SchoolDoctor of Philosophy (PhD), Neuroscience, Doctor of Philosophy (PhD), Neuroscience at McGill University
Github Skills (11)
transformers10
computer-vision10
transformer10
machine-learning10
pytorch10
deep-learning10
self-supervised-learning10
resnet10
python10
vision-transformer10
documentation9
Programming languages (4)
Github contributions (5)
mosaicml/composer
Oct 2021 - Dec 2022
Supercharge Your Model Training
Role in this project:
ML Engineer
ML EngineerContributions:150 reviews, 35 commits, 78 PRs in 1 year 2 months
Contributions summary:Matthew primarily contributes to the documentation of the library, updating method cards and fixing formatting issues within the documentation. They also added a new ResNet20 model to the project, including the necessary files and parameters. Additionally, the user improved data transformation capabilities by making the `add_dataset_transform()` function more flexible, and updated demo notebook examples.
pytorchml-systemsdeep-learningneural-networksmachine-learning
facebookresearch/vissl
Feb 2021 - Jul 2021
VISSL is FAIR's library of extensible, modular and scalable components for SOTA Self-Supervised Learning with images.
Role in this project:ML Engineer
ML EngineerContributions:4 reviews, 6 commits, 6 PRs in 5 months
Contributions summary:Matthew primarily contributed to the development and enhancement of vision transformer models within the VISSL framework. Their work included implementing and integrating the ConViT architecture, fixing compatibility issues with MoCo and SimCLR within the CutMixUp collator, and making minor documentation updates related to vision transformers. These contributions demonstrate a focus on advancing self-supervised learning techniques, particularly with vision transformer models.
pytorchscalablesupervised-learningvision-transformersupervised
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.