Mu Li
Co-Founder at Boson AI
Palo Alto, California, United States
Summary
🤩
Rockstar🎓
Top School
Top expert inHigh-Performance Machine Learning Computing
Mu Li is a machine learning engineer and founder with 13 years of experience building distributed training systems and developer-facing ML infrastructure. As Co-Founder of Boson AI and former senior scientist at AWS, he has led teams to make large-scale ML cheaper and easier to use while also teaching practical deep learning at Stanford and Berkeley. His deep systems expertise is reflected in substantial open-source work on MXNet, PS-Lite, and mshadow—contributions that shaped parameter-server design and high-performance tensor kernels. He combines low-level C++/CUDA backend development with DevOps and documentation work, ensuring both performant engines and usable tooling (notably contributing to the widely adopted Dive into Deep Learning materials). Known for pragmatic refactors and performance-minded logging and serialization fixes, he excels at turning research-grade ideas into production-ready distributed systems. Based in Palo Alto, he pairs a CMU PhD in machine learning and SJTU engineering roots with a founder’s appetite for productizing ML at scale.
13 years of coding experience
10 years of employment as a software developer
Doctor of Philosophy (Ph.D.) Machine Learning, Doctor of Philosophy (Ph.D.) Machine Learning at Carnegie Mellon UniversityMaster Computer Science, Master Computer Science at Shanghai Jiao Tong University
Github Skills (69)
transformers10
benchmark10
restructuredtext10
operation10
python10
bash10
dm10
deep-learning10
ci-cd-pipeline10
build-automation10
distributed-training10
rs10
performance-optimization10
pipeline10
params10
Programming languages (10)
Github contributions (5)
d2l-ai/berkeley-stat-157
Dec 2018 - May 2019
Homepage for STAT 157 at UC Berkeley
Role in this project:
DevOps Engineer
DevOps EngineerContributions:118 commits, 124 pushes, 1 branch in 4 months
Contributions summary:Mu primarily focused on configuring and automating the build and deployment process for the project. Their contributions included modifying build scripts (build.sh) to include environment updates and setting up the deployment script (upload.sh) to synchronize the built documentation to an S3 bucket. Furthermore, the user adjusted configurations to integrate with the Jenkins CI/CD system, indicating a focus on infrastructure management. These changes likely streamline the build and release cycle.
berkeleyjekylluc-berkeleystat
dmlc/ps-lite
Apr 2015 - Jun 2018
A lightweight parameter server interface
Role in this project:Back-end Developer
Back-end DeveloperContributions:223 commits, 39 PRs, 152 pushes in 3 years 2 months
Contributions summary:Mu primarily contributed to the core functionality of the parameter server interface. Their initial commit established the project's foundation, and subsequent commits involved implementing core functionalities. These included adding a distributed monitor, updating configuration, and refactoring existing code. They also worked on refactoring the KV-Store component including adding dynamic-sized values and integrating with DMLC's serialization.
parameter-serverparameter
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.