Peizhao Zhang
Senior Staff Research Scientist at Facebook
Menlo Park, California, United States
Summary
🤩
Rockstar🎓
Top SchoolPeizhao Zhang is a founding member and experienced research scientist with 11 years driving mobile vision and generative AI at Meta/Reality Labs, now building a stealth startup in Menlo Park. He holds a PhD in Computer Science and has led production-focused research on image and video foundation models, bridging cutting-edge research with deployable mobile and core AI systems. His open-source contributions include nontrivial optimizations to Caffe2 and mobile-vision—implementing graph-ordering and dynamic programming techniques to reduce memory use and ensure model fidelity with torchvision—highlighting both theoretical depth and pragmatic engineering. Known for spotting and fixing subtle edge cases (e.g., uninitialized blob fetching and IRFBlock shuffle issues), he combines systems-level optimization with model correctness.
11 years of coding experience
4 years of employment as a software developer
Doctor of Philosophy (PhD), Computer Science, Doctor of Philosophy (PhD), Computer Science at Texas A&M UniversityBachelor of Engineering (B.E.), Computer Software Engineering, Bachelor of Engineering (B.E.), Computer Software Engineering at Sun Yat-Sen University
Github Skills (17)
pytorch10
caffe10
python10
machine-learning10
deep-learning10
resnet10
ai10
computer-vision10
optimization10
data-structure9
algorithm9
algorithms9
graph-theory9
datastructures-algorithms9
data-structures9
Programming languages (3)
Github contributions (5)
facebookresearch/mobile-vision
May 2019 - Sep 2022
Mobile vision models and code
Role in this project:
ML Engineer
ML EngineerContributions:5 reviews, 42 commits, 45 PRs in 3 years 4 months
Contributions summary:Peizhao primarily focused on implementing and optimizing ResNet models within the mobile-vision framework. Their work involved ensuring exact matches with torchvision models, which included adjustments to model definitions and block structures. The user also addressed the potential skipping of shuffle operations within the IRFBlock, and explored the application of static quantization aware training (QAT) for LSTM models. The user also refactored recursive iteration utilities.
computer-visionvision
facebookarchive/caffe2
Mar 2017 - Jan 2018
Caffe2 is a lightweight, modular, and scalable deep learning framework.
Role in this project:ML Engineer
ML EngineerContributions:16 commits in 10 months
Contributions summary:Peizhao contributed to the Caffe2 deep learning framework by fixing a bug related to the `ModelTrainerLog` class, which likely involved debugging and modifying Python code within the framework's utilities. They implemented an ordering function, `topological_sort_traversal_longest_path()`, to reduce memory usage by optimizing the execution order of operations, showing an understanding of graph theory and optimization techniques for deep learning. Furthermore, the user optimized the framework by implementing a dynamic programming algorithm to find the optimal blob assignments and also addressed potential errors related to fetching uninitialized blobs, demonstrating an ability to find and fix edge cases.
pytorchscalablecaffe2deep-learningml
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.