Liyuan Liu
Member Of Technical Staff at Thinking Machines Lab
Redmond, Washington, United States
Summary
🤩
Rockstar🎓
Top SchoolLiyuan Liu is a Member of Technical Staff and machine learning engineer with a PhD in Computer Science from UIUC and a decade of experience building and debugging production-grade ML systems. Based in Redmond, he progressed through senior research roles at Microsoft Research before joining Thinking Machines Lab, blending research rigor with practical engineering. His open-source work includes contributions to sequence labeling models (LM-LSTM-CRF) and optimizer research (RAdam), where he implemented bug fixes, performance improvements, multi-document prediction support, and a novel RAdam_4step variant. Comfortable across model internals, training scripts, and deployment-ready utilities, he focuses on making research artifacts robust and usable. Colleagues value his attention to detail in evaluator and embedding handling—areas that often make or break real-world ML pipelines. He maintains a technical home page showcasing projects and reproducible work that bridge academic insight and production needs.
10 years of coding experience
4 years of employment as a software developer
University of Science and Technology of ChinaUniversity of Illinois Urbana-Champaign
Github Skills (12)
sequence-labeling10
algorithm10
pytorch10
machine-learning10
optimisation10
python10
crf10
optimizers10
adam10
optimization10
deep-learning9
language-model9
Programming languages (5)
Github contributions (5)
LiyuanLucasLiu/LM-LSTM-CRF
Sep 2017 - Jun 2022
Empower Sequence Labeling with Task-Aware Language Model
Role in this project:
ML Engineer
ML EngineerContributions:2 releases, 88 commits, 31 PRs in 4 years 10 months
Contributions summary:Liyuan primarily contributed to improving and debugging the sequence labeling model. They fixed bugs in the evaluator, addressed issues related to embedding loading and shrinking the embedding dictionary, and debugged the model. They also added a predict function and support for multi-document input, enhancing the model's usability.
pytorchnlplanguage-modelnersequence
LiyuanLucasLiu/RAdam
Jul 2019 - Jul 2021
On the Variance of the Adaptive Learning Rate and Beyond
Role in this project:ML Engineer
ML EngineerContributions:57 commits, 20 PRs, 46 pushes in 2 years
Contributions summary:Liyuan contributed to the development of a machine learning library, RAdam, and its related functionalities. Their work involved modifying the core optimization algorithm, including bug fixes, performance improvements, and the addition of new features such as the RAdam_4step variant. The user also made changes to training scripts, potentially for testing or integrating RAdam into existing projects, demonstrating practical application. Furthermore, they made adjustments to the documentation and setup files.
information-theoryadversarial-learninglearning-ratevariancelanguage-modeling
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.