CV

Yucheng Zhang

• Email: yz90@illinois.edu
• Mob: +1 2177214643

EDUCATION

Zhejiang University - University of Illinois Urbana-Champaign Institute, Zhejiang, China

• Bachelor of Engineering in Computer Engineering
  • Duration: Sept. 2021 - Present
  • Overall GPA: 3.78/4.0
  • Core Courses: Data Structures, Computer Systems & Programming, Computer Systems Engineering, Database Systems, Intro to Algs & Models of Comp (Taken this semester), Machine Learning (Taken this semester)
  • Overall GPA (UIUC): 3.97 /4.0
  • Awards and Honors:
    • Innovation and Entrepreneurship Pacesetter, 2021-2022
    • Dean’s List, 2021-2022

PUBLICATIONS

• Authored as the second author in a paper titled “MovieChat: From Dense Token to Sparse Memory for Long Video Understanding” in CVPR 2024.
• The first author in a paper titled “InterAct: Advancing Large-Scale Versatile 3D Human-Object Interaction Generation” under review of NeurIPS 2024

RESEARCH EXPERIENCES

Research on Human-Object Interaction Benchmark build and representation | Individual Research

• Duration: Feb. 2024 - now
• Advisor: Yuxiong Wang (Professor of University of Illinois Urbana Champaign)
• There’re few datasets that cover the human-object iteraction with detailed hand pose.
• We build a dataset with 22.94 hours’ of human-object interaction data as well as human-annotated detailed annotation for each split short video.
• We also present a new marker based method to repreesent the human pose.
• The paper is under review of NeurIPS 2024.

Research on weak supervised motion detection | Team Leader (Group of 3)

• Duration: Mar. 2023 - Nov. 2024
• Advisor: Gaoang Wang (Professor of ZJU-UIUC Institute)
• Current methods to do motion detection or image classification under weak supervision has a drawback of low accuracy compared with those of strong supervision which is unavoidable.
• But we still want to improve the accuracy of the weak supervision model so we try to modify the models and use the newest advances from other models to help improve our model.
• Using neural network method to detect and identify motions with the learning data that’s not well labeled.
• The improvement of our low supervision model is still in progress and we believe it will have a good improvement on current models.
• We also cooperate with other PhD students and contribute to a paper under submission of CVPR as the second authors while the paper titled “MovieChat: From Dense Token to Sparse Memory for Long Video Understanding.”

Research on encrypting DNN | Individual Research

• Duration: Jul. 2021 - Aug. 2021
• Advisor: Jian Liu (Professor of Zhejiang University)
• There’s a need for provider of both model and data so prevent their model or data from being leaked so comes the algorithm to encrypt the process of DNN.
• The latest advance at that time is Cheetah model which is already a big improvement in time compared with previous models.
• But we find that with the newest way of OT—the VOLE style OT the speed of the model can be even faster.

SKILLS AND OTHERS

• Computer: MATLAB, C/C++, Python, Sagemath
• Language: English, Chinese
• Interests: Travel, Read books
• Other Awards:
  • Gold Medal in the 46th ICPC Asia Regional Contest Jinan Site, National 2021
  • Silver Medal in 2021 China Collegiate Programming Contest Weihai Site, National 2021