Chen Cheng

Last updated in . Get the latest version at www.chencheng.me/cv/.

https://www.chencheng.me
ischen.cheng@gmail.com
@chencheng

Research Interest

To facilitate a deeper understanding and more efficient communication of information across various interfaces, my research focuses on the synergy between different information modules, particularly in the realms of AI and human-AI interactions. I am dedicated to exploring the transformation and exchange of information, aiming to align these elements to enhance mutual understanding between diverse modules. My approach integrates techniques from human-computer interaction, visualization, and artificial intelligence to develop tools and systems that not only improve communication between AI agents but also between humans and AI agents. This work, at the intersection of different disciplines, aims to advance our capacity to interpret, manage, and utilize complex data streams in a variety of contexts, from everyday interactions to sophisticated data analysis.

Educational Background

Sep.2020–Jun.2024 Bachelor of Engineering in Computer Science and Technology, ShanghaiTech University Shanghai, China

GPA: 3.88/4.0 Rank 3/248

Aug.2023–Feb.2024 Exchange Student in the Department of Electrical Engineering and Computer Sciences, University of California, Berkeley Berkeley, CA

GPA: 4.0/4.0

Research Experience

Jun. 2023–Present Bridging the Comprehension Gap: A Deep Dive into LLM-Generated Code and the Design of CodeCognoscenti Remote Independent Research

Human-Centered Software Systems Lab  |  Prof. Tianyi Zhang  |  Purdue University

  • Iteratively improved the mock-up and designed CodeCognoscenti, a VSCode extension that assists users in building an understanding of function-to-class level code generated by LLM.
  • Conducted a formative study including literature review and semi-structured interviews with 15 developers.
  • Designed a mock-up of a VSCode extension based on the GitHub Copilot Chat interface with features to enhance code understanding.
  • Constructed a user flow based on observations with 3 programmers using LLM for code generation, comprehension, and debugging.
  • Designed an adaptive copilot for programming utilizing interactive machine teaching and LLM self-reflection based on the pAIr programming model.
  • Proposed a humans and AI pair programming (pAIr programming) interaction model.
  • Proposed a conceptual prototype – A Sensemaking-Based Code Block Validation Tool integrating chatbots, API documentation, and live programming.
Nov. 2022–Mar. 2023 Searching for Optimal Heterogeneous Graph Neural Networks: A Comparative and Explainable Approach with VAC-HGNN Shanghai, China

ViSeer LAB  |  Prof. Quan Li  |  ShanghaiTech University

  • Designed and implemented VAC-HGNN, a visual analytics system for HGNN comparison and analysis.
  • Developed a pipeline for NAS dataset analysis, enabling understanding and comparative analysis of HGNNs.
  • Proposed a nested unsupervised decision tree algorithm for HGNN design space partition.
  • Utilized OpenHGNN for real-time HGNN training, comparison, and hypothesis validation.
  • Conducted interviews to find user requirements for using heterogeneous neural networks.
Oct. 2022–Dec. 2022 FMLens: Towards Better Scaffolding the Process of Fund Manager Selection in Actively Managed Equity Fund Investments Shanghai, China

ViSeer LAB  |  Prof. Quan Li  |  ShanghaiTech University

  • Implemented FMLens, a visual analytics system for the fund manager selection process.
  • Constructed regression equations for fund position simulation and compared three regression methods.
Jan. 2022–Sep. 2022 ALens: An Adaptive Training System for Academic Abstract Writing Shanghai, China

ViSeer LAB  |  Prof. Quan Li  |  ShanghaiTech University

  • Developed chapters of the paper, organized ideas, and presented the work.
  • Built ALens, a web-based application for academic abstract writing.
  • Designed an abstract writing training process.
  • Conducted a formative study on the challenges faced by L2 junior researchers in academic abstract writing.

Publications [Interactive Version]

Conference

C4
Bridging the Comprehension Gap: A Deep Dive into LLM-Generated Code and the Design of CodeCognoscenti
Chen Cheng, Tianyi Zhang. Upcoming, 2024.
C2
Searching for Optimal Heterogeneous Graph Neural Networks: A Comparative and Explainable Approach with VAC-HGNN
Chen Cheng, Junlei Zhu, Yufei Zhang, Quan Li. Under Revison, 2023.
Video

Journal

Honors & Awards

Dec.2023 Undergraduate Special Scholarship ShanghaiTech University
Aug.2023 Best Paper Honorable Mention Harbin, China

For "ALens: An Adaptive Domain-Oriented Abstract Writing Training Tool for Novice Researchers"

Dec.2022 Undergraduate Special Scholarship ShanghaiTech University
Jul.2022 Data Visualization Competition 2nd Prize Xining, China
Dec.2021 Undergraduate Special Scholarship ShanghaiTech University

Coursework

Black Asset Network Visual Analytic System  |  Data Visualization
The system employs visual analysis solutions to investigate the network assets and correlations of black and gray industry gangs, such as domain names and IP addresses, to identify suspicious activities, understand their operational mechanisms, and develop strategies to combat their detrimental impact on network ecology and social security.
Code Video
Linear Programming Solver  |  Numerical Optimization
Implemented a linear programming solver using Python.Employed a two-phase approach to simplex algorithms.
Chrome Dinosaur Game in RISC-V  |  Computer Architecture I
Use RISC-V to implement the Chrome Dinosaur Game on Sipeed Longan Nano development board.
Video
Meta-Path Discovery Based on Temporal Equivariant Graph  |  Artificial Intelligence
Added temporal information to static graph representation by GRU and used DQN to discover meta-paths.
Code PDF
Hand Gesture Recognition using DD-Net & Knowledge Distillation  |  Computer Vision
Collected a hand gesture recognition dataset, built DD-Net from research and compress the model with knowledge distillation.
Code PDF
Linking Tweets with NYT Articles using ChatGPT & BERT  |  Data Mining
Mitigated data imbalance in tweet-news linkage by utilizing ChatGPT for text augmentation and use Sentence-BERT-based model to link tweet and news.
Code PDF
Robustness of In-Context Learning with Noisy Labels  | 
Explored the resilience of Transformers in In-Context Learning (ICL) against noisy labels in training corpora and prompt demonstrations.
Code PDF

Service

References

Quan Li
Assistant Professor at School of Information Science and Technology
ShanghaiTech University
faculty.sist.shanghaitech.edu.cn/liquan/

Tianyi Zhang
Assistant Professor in Computer Science
Purdue University
www.cs.purdue.edu/people/faculty/tianyi.html