Han Cai
Senior Research Scientist at NVIDIA
Cambridge, Massachusetts, United States
Summary
🤩
Rockstar🎓
Top SchoolHan Cai is a Senior Research Scientist at NVIDIA with a decade of experience bridging academic research and production ML systems, holding a PhD from MIT and earlier degrees from Shanghai Jiao Tong University. He has a strong track record in hardware-aware neural architecture search and model deployment, contributing to influential open-source projects such as Once-For-All and ProxylessNAS where he improved cross-framework implementations and ensured pretrained model accessibility. His work spans PyTorch and TensorFlow, low-level layer implementations, and practical engineering like NCHW/NHWC conversions and checkpoint tooling, reflecting a focus on making cutting-edge research usable in real-world settings. Based in Cambridge, MA, he combines deep research experience from MIT with iterative industry impact through internships and roles at OmniML, IBM, and NVIDIA. Notably, he repeatedly moves results from prototype to production-ready code, adding example notebooks and configuration fixes that lower the barrier for practitioners to adopt efficient models.
10 years of coding experience
8 years of employment as a software developer
Doctor of Philosophy - PhD, Electrical and Electronics Engineering, Doctor of Philosophy - PhD, Electrical and Electronics Engineering at Massachusetts Institute of TechnologyMaster's degree, Computer Science, Master's degree, Computer Science at Shanghai Jiao Tong University
Github Skills (20)
continuous-deployment10
pytorch10
convolutional-neural-networks10
python10
architecture10
acceleration10
machine-learning10
ml-deployment10
mask-rcnn10
deep-learning10
trainings10
tensorflow10
automl10
neural-network10
faster-rcnn10
Programming languages (2)
Github contributions (5)
mit-han-lab/once-for-all
Jan 2020 - Jul 2022
[ICLR 2020] Once for All: Train One Network and Specialize it for Efficient Deployment
Role in this project:
ML Engineer
ML EngineerContributions:1 release, 60 commits, 5 PRs in 2 years 6 months
Contributions summary:Han primarily updated the codebase to utilize alternative links for pre-trained models, specifically focusing on the OFA (Once for All) framework. These updates involved modifying files related to model loading and configuration within the `model_zoo.py` and `train_ofa_net.py` files, indicating an effort to ensure the accessibility and usability of pre-trained models for efficient deployment. Additionally, the user made minor updates to URL paths for model checkpoints, which is crucial for maintaining the functionality of the OFA framework. Furthermore, the user added an example notebook to show the usage of OFA-ResNet50.
pytorchiclrnastinymldeep-learning
MIT-HAN-LAB/ProxylessNAS
Dec 2018 - Dec 2018
Role in this project:ML Engineer
ML EngineerContributions:1 push, 1 branch in 1 day
Contributions summary:Han contributed significantly to the ProxylessNAS project, focusing on the implementation of different network architectures and supporting TensorFlow versions. They added and modified core layers, including convolutional and depthwise convolutional layers, pooling layers and MBInvertedConvLayer, for both PyTorch and TensorFlow frameworks. The user also updated links and model configurations and implemented a search module, indicating involvement in architecture search and model development.
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.