Boxiang WANG

I am currently a student at Harvard University pursuing my Master degree in Computational Science and Engineering. I just graduated from Nanyang Technological University with a Bachelor degree of Electrical and Electronic Engineering and a minor in Mathematics. I have received two Dean's List (top 5%) during my study at NTU.

My research interests include Machine Learning, Deep Learning, especially systems for Machine Learning. Please feel free to contact me.

Email / LinkedIn / Github / CV / Bilibili

Tesseract: Treat Deep Neural Networks with Scalable Efficient Model Parallelism
Boxiang Wang, Qifan Xu, Zhengda Bian, Yang You
Proceedings of the 51th International Conference on Parallel Processing
[pdf]

We propose a efficient scalable tensor parallelism structure, it reached a speedup of 1.375x compared to SOTA 1d parallelism Megatron-LM.

Cubework: An Efficient Model Parallelism Framework for Training Huge Neural Networks
Zhengda Bian, Qifan Xu, Boxiang Wang, Yang You
arXiv:2105.14450, 2021
[pdf]

Our work is the first to introduce a 3-dimensional model parallelism for expediting huge language models.

Colossal-AI
Developed an integrated large-scale model training system with efficient parallelization techniques which has attracted about 4000 stars so far.
Implemented Tesseract and Cubework
[Documentation] [GitHub]

Detection of Non-human Faces
with Home Team Science & Technology Agency (HTX), Singapore
Deployed EfficientNet and ResNet with TensorFlow to reach an 98% accuracy on the classification task. Achieved top accuracy among all participants.