Tonghe Zhang
PHD Student at SEDI Education Group
Leuven, Flemish Brabant, Belgium
Summary
🤩
Rockstar🎓
Top SchoolTonghe Zhang is a PhD student at KU Leuven with nine years of technical and educational experience, blending machine learning engineering and backend development with academic research in movement sciences. He has hands-on MLOps and distributed training experience—contributing to the notable VideoMAE NeurIPS spotlight project by improving fine-tuning/pretraining scripts, fixing training bugs, and adapting pipelines across ViT model scales and datasets. Based in Leuven, he also teaches English, demonstrating strong communication skills and an ability to translate complex ideas for diverse audiences. Tonghe’s profile reflects a rare hybrid of applied deep learning engineering and movement-science research, equipped to bridge experimental ML workflows and reproducible research. Colleagues can expect pragmatic code contributions focused on scalable training and dataset handling coupled with scientific rigor from his PhD work.
9 years of coding experience
Doctor of Philosophy - PhD, Faculty of Movement Sciences, Doctor of Philosophy - PhD, Faculty of Movement Sciences at KU LeuvenMaster of Education - MEd, Educational leadership and policy, Master of Education - MEd, Educational leadership and policy at Monash University
Github Skills (13)
computer-vision10
pytorch10
transformer10
machine-learning10
self-supervised-learning10
action-recognition10
video-understanding10
python9
mlops9
distributed-training9
dockers7
adam7
docker7
Programming languages (5)
Github contributions (5)
MCG-NJU/VideoMAE
Mar 2022 - Jan 2023
[NeurIPS 2022 Spotlight] VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training
Role in this project:
MLOps Engineer & Backend Developer
MLOps Engineer & Backend DeveloperContributions:38 commits, 1 PR, 51 pushes in 9 months
Contributions summary:Tonghe primarily contributed to the project by adding and modifying scripts related to fine-tuning and pre-training for the VideoMAE model, particularly focusing on different datasets (Kinetics-400, SSV2) and model configurations (ViT-Base, ViT-Large, ViT-Small, ViT-Huge). They updated scripts for distributed training using torch.distributed.launch. They also fixed bugs related to mixup and multi-clip testing, suggesting an involvement in model training and evaluation pipelines. Additionally, they modified a core Python file related to the dataset, suggesting backend logic work.
autoencoderslearnersvisual-recognitiontrainingself-supervised-learning
yztongzhan/SlowFast
Apr 2020 - Nov 2022
PySlowFast: video understanding codebase from FAIR for reproducing state-of-the-art video models.
Contributions:20 PRs, 12 pushes, 2 branches in 2 years 7 months
pytorchunderstandingpythonartstate-of-the-art
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.