Yang Sui
Member Of Technical Staff at Microsoft AI
San Francisco Bay Area United States
Summary
🤩
Rockstar🎓
Top SchoolYang Sui is a Member of Technical Staff on Microsoft’s SuperIntelligence team, specializing in efficient multimodal training and inference systems. With nine years of experience spanning academia and industry, he focuses on model compression, token compression, and efficient reasoning to make large models faster and cheaper to run. A former Rice University postdoc and NeurIPS/CVPR/TMLR author, he combines rigorous research (e.g., lossless LLM compression and token pruning) with production-grade ML systems engineering. He was an initial contributor to the widely used Paddle-Lite inference engine, adding core operators and performance-focused tests, and has driven quantization work on diffusion models and vision transformers during internships at Snap and Tencent. Based in the Bay Area, he bridges cutting-edge research and deployable AI infra, often finding efficiency gains at the operator and token level rather than just model architecture.
9 years of coding experience
2 years of employment as a software developer
Exchange Student, Exchange Student at Princeton UniversityMaster's degree, Master's degree at Jilin UniversityDoctor of Philosophy - PhD, Doctor of Philosophy - PhD at Rutgers University
Stackoverflow
1reputation
0reached
0answers
0questions
Github Skills (18)
operation10
tensorrt10
machine-learning10
deep-learning10
tensorflow10
tensor10
model-driven9
unit-testing9
performance-optimization9
model-driven-development9
modeling9
model-building9
cprogramming-language8
convolutional-neural-networks8
c-language8
Programming languages (2)
Github contributions (5)
PaddlePaddle/Paddle-Lite
May 2018 - Nov 2018
PaddlePaddle High Performance Deep Learning Inference Engine for Mobile and Edge (飞桨高性能深度学习端侧推理引擎）
Role in this project:
ML Engineer
ML EngineerContributions:298 commits, 84 PRs, 40 pushes in 6 months
Contributions summary:Yang contributed to the implementation of the `elementwise_add_op` and `mul_op`, showcasing a focus on core operator functionality. They also added tests for these operators, demonstrating a commitment to testing and ensuring the correctness of the implemented functionalities. The addition of `fusion_fc_op`, and supporting files such as `test_fushion_fc_op.cpp` and `test_pe.cpp` indicate efforts to improve efficiency and performance. Further, they added new models for testing such as `yolo`, `mobilenet` and the addition of the `relu`, `bn`, `lrn`, and other operations indicate a strong understanding of the model's architecture and the underlying deep learning framework.
inference-enginemobilebaidutensorflowfpga
Eclipsess/paddle-mobile
May 2018 - Nov 2018
This research aims at simply deploying deeplearning on mobile devices, with low complexity and high speed.
Contributions:134 pushes in 6 months
deployingcomplexityspeedcaffe2deep-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.