Haotian Zhang

Haotian Zhang

Ph.D. Candidate, Research Intern at MSR AI

ECE, University of Washington

Biography

“Be Boundless.”

Haotian (Carl) Zhang is now a 4th-year Ph.D candidate of Department of Electrical and Computer Engineering at the University of Washington. His research interests basically include Computer Vision and Deep Learning. He’s now a member of Infomation Processing Lab IPL, supervised by Prof. Jenq-Neng Hwang. Before that, he received his BS.c degree in Electrical Engineering at Shanghai Jiao Tong University in 2017, supervised by Prof. Jun-Fa Mao.

Haotian believes that living an interesting life is done by doing interesting things, and that’s what he hopes to do.

Interests
  • 3D Vision
  • Object Detection and Tracking
  • Vision & Language
Education
  • Ph.D. in Electrical & Computer Engineering, 2022

    University of Washington

  • Master in Electrical & Computer Engineering, 2018

    University of Washington

  • BS.c. in Electrical Engineering, 2017

    Shanghai Jiao Tong University

Experience

 
 
 
 
 
Microsoft Research
Research Intern, Deep Learning
Jun 2021 – Sep 2021 Redmond, USA
supervised by Pengchuan Zhang, Jianwei Yang, Chunyuan Li, Xiyang Dai, Xena Zhu, Cheng Wu, Yuan Lu, and Jianfeng Gao.
 
 
 
 
 
Microsoft AI & Cloud
Research Intern, Computer Vision
Microsoft AI & Cloud
Jun 2020 – Sep 2020 Redmond, USA

My research project is to investigate pre-training models with additional visual modalities by involving image embeddings in the pre-training steps, supervised by Guoxin Wang, Yijuan Lu and Dinei Florencio.

  • Adopted a simple yet powerful Transformer model as the backbone and extends it to take both visual and text embedded features.
  • Designed the Masked Language Model (MLM) and Image-Text Matching (ITM) to jointly model interactions between language, layout and rich visual information.
  • Visual-LayoutLM model has shown its potential to outperform the original LayoutLM and other SOTA models in several document understanding tasks.

Recent Publications

Quickly discover relevant content by filtering publications.

Popular Topics