Kexin Yu
Software Engineer at Google
San Francisco Bay Area United States
Summary
🤩
Rockstar🎓
Top SchoolKexin Yu is a software engineer with 10 years of experience building scalable machine learning and large-scale data systems across industry leaders in the Bay Area, currently at Google after a deep-learning frameworks role at NVIDIA. She holds an MS from Stanford ICME and a BS in Mathematics of Computation from UCLA, blending strong theoretical foundations with production engineering. Her open-source contributions include significant optimizer work on NVIDIA/apex—improving mixed-precision and distributed training primitives like FusedSGD and FusedLAMB—demonstrating both low-level CUDA-aware optimization and practical ML tooling. Past roles span research and applied engineering at Baidu USA and Verizon, where she delivered GPU-accelerated algorithms, adaptive sampling strategies, and automated model-selection systems. Colleagues describe her as someone who bridges research and production: she moves models from prototype to robust, deployable components. Outside core development she’s reviewed career-track submissions for the Grace Hopper Conference, reflecting a commitment to community and mentorship.
10 years of coding experience
3 years of employment as a software developer
Qingdao No.2 Middle SchoolBachelor's Degree, Mathematics of Computation, Bachelor's Degree, Mathematics of Computation at University of California, Los AngelesMaster of Science - MS, Computational and Mathematical Engineering, Master of Science - MS, Computational and Mathematical Engineering at Stanford University
Chinese, English
Github Skills (6)
multiprecision10
pytorch10
distributed-training10
apex10
optimization10
cuda9
Programming languages (2)
Github contributions (5)
NVIDIA/apex
Dec 2019 - Sep 2021
A PyTorch Extension: Tools for easy mixed precision and distributed training in Pytorch
Role in this project:
ML Engineer
ML EngineerContributions:3 reviews, 98 commits, 22 PRs in 1 year 8 months
Contributions summary:Kexin primarily contributed to optimizing and extending the `apex.contrib.optimizers` module, a PyTorch extension for mixed precision and distributed training. Their work involved significant modifications to the `FusedSGD` and `FusedLAMB` optimizers, incorporating features like global gradient clipping, and addressing state-dict mismatches. Further contributions included the integration of multi-tensor L2 norm calculation and debugging/testing. They also made relevant edits to example files and build scripts.
distributed-trainingpytorch
yws12/CS341_cryptocurrency_prediction
Apr 2018 - Jun 2018
Contributions:73 commits, 67 pushes in 1 month
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.