Tianbing Xu
Research Scientist Software Engineer at Facebook
San Francisco Bay Area United States
Summary
🤩
Rockstar
Top expert inHigh-Performance Machine Learning Computing
Tianbing Xu is a research scientist and software engineer based in the San Francisco Bay Area with about 12 years of experience building and optimizing machine learning systems in industry, currently at Facebook. He blends deep ML research with production-grade engineering, contributing to core libraries and compilers such as XGBoost, MXNet, TVM and AITemplate to improve performance on CPU and GPU targets. His work spans low-level C++/CUDA kernel development, compiler fixes, and scalable model training and evaluation—demonstrated by multi-threading optimizations, cross-validation integrations, and hardware-specific tuning. Tianbing’s contributions to widely used open-source projects show a focus on inference performance and build reliability, including resolving LLVM issues and enhancing build/amalgamation processes. He often operates at the intersection of research and tooling, turning algorithmic insight into practical, deployable improvements. An understated strength is his knack for debugging complex compilation and dependency flows that unblock production deployments.
12 years of coding experience
Chinese, English, wu chinese
Github Skills (31)
xgboost10
performance-monitor10
c-language10
operation10
python10
data-science10
llvm10
tensorrt10
gpu-programming10
machine-learning10
blas10
dm10
numpy10
gradient-boosting10
deeplearning-ai10
Programming languages (9)
Github contributions (5)
apache/mxnet
Jun 2015 - Jan 2017
Lightweight, Portable, Flexible Distributed/Mobile Deep Learning with Dynamic, Mutation-aware Dataflow Dep Scheduler; for Python, R, Julia, Scala, Go, Javascript and more
Role in this project:
Back-end Developer & ML Engineer
Back-end Developer & ML EngineerContributions:535 commits, 368 PRs, 318 pushes in 1 year 7 months
Contributions summary:Tianbing's commits focused on significant code modifications within the MXNet deep learning library, primarily concerning the core components. The changes include restructuring the NArray class, implementing a "clip" operation, adding support for an LSTM network, and developing functions for setting and retrieving elements within NDArrays. They also involved the creation of functions, related to matrix multiplication.
pythonschedulerdataflowmutationdata-science
dmlc/mshadow
Dec 2013 - Apr 2017
Matrix Shadow:Lightweight CPU/GPU Matrix and Tensor Template Library in C++/CUDA for (Deep) Machine Learning
Role in this project:Back-end Developer
Back-end DeveloperContributions:128 commits, 18 PRs, 46 pushes in 3 years 4 months
Contributions summary:Tianbing contributed to the development of a lightweight matrix and tensor template library in C++ and CUDA for deep machine learning. Their work involved implementing and modifying tensor-related functionalities, including 1D, 2D, and 3D tensor structures, along with related operations like map functions and binary operations. The user also integrated new operations like take, concat and other operator overloads into the library. Additionally, the user added support for various BLAS functions using Intel's MKL library and some cuDNN functionalities.
cudapytorchcpucppdeep-learning
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.