Xing Liu
Senior Staff Research Scientist at Meta
Menlo Park, California, United States
Summary
🤩
Rockstar🎓
Top SchoolXing Liu is a Senior Staff Research Scientist in Menlo Park with 11 years of experience building high-performance ML infrastructure and embedding systems. At Meta he drives research-to-production work on embedding lookup optimizations and scalable sparse/dense operations, building on prior research roles at Intel Labs and IBM. His contributions to prominent open-source projects like PyTorch's FBGEMM and TorchRec show a focus on kernel-level performance, unpooled table batching, and quantized communication for embeddings—work that materially improves recommendation system throughput. He holds a Ph.D. in Computational Science and Engineering from Georgia Tech and a bachelor's in Electronic & Information Engineering from HUST. Known for bridging deep research with pragmatic engineering, he often surfaces subtle algorithmic improvements (e.g., permuting sparse data and table-wise sequence embeddings) that yield outsized system gains.
11 years of coding experience
6 years of employment as a software developer
Ph.D., Computational Science and Engineering, Ph.D., Computational Science and Engineering at Georgia Institute of TechnologyBachelor, Electronic & Information Engineering, Bachelor, Electronic & Information Engineering at Huazhong University of Science and Technology
Github Skills (14)
cuda10
quantization10
gpu-programming10
pytorch10
c-language10
recommendation-system10
fba10
f10
cprogramming-language10
emm10
deeplearning-ai9
machine-learning9
distributed-systems9
deep-learning9
Programming languages (4)
Github contributions (5)
pytorch/torchrec
Dec 2021 - Sep 2022
Pytorch domain library for recommendation systems
Role in this project:
Back-end Developer / ML Engineer
Back-end Developer / ML EngineerContributions:62 commits, 42 PRs, 15 comments in 9 months
Contributions summary:Xing contributed to the PyTorch domain library for recommendation systems, specifically focusing on optimizing the embedding lookup operations. The user implemented and refined unpooled table batched FBGEMM ops for embedding lookups, leading to performance improvements. Furthermore, the user was involved in enabling quantized communication for unpooled embeddings. The user also made changes to incorporate FSDP.
cudapytorchrecommendation-systemsdeep-learninggpu
pytorch/FBGEMM
Feb 2021 - Jul 2022
FB (Facebook) + GEMM (General Matrix-Matrix Multiplication) - https://code.fb.com/ml-applications/fbgemm/
Role in this project:ML Engineer
ML EngineerContributions:17 commits, 12 PRs in 1 year 4 months
Contributions summary:Xing primarily contributed to the fbgemm library, focusing on optimizing and enhancing sparse and dense embedding operations. Their work included fixing and improving kernel implementations for various embedding table configurations, such as those shared by multiple features. They also introduced new functionalities, including functions for permuting sparse data and added a feature related to table-wise sequence embeddings.
matrix-multiplicationfacebookmultiplicationmatrixml-applications
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.