Mingyuan Li

霆霓快雨

Download Resume

About Me

霆霓快雨 describes the scene of a summer thunderstorm followed by immediate clear skies, which also can be referred to as Nimble as Lightning . This term accurately summarizes my undergraduate research:

1) reproduced a fracture detection model within two weeks; 2) identified research gaps in few-shot classification, building, experimenting, writing, and publishing a paper within three weeks; 3) processed idea development, and prototyping a generative AI project for programming educationin one month, which won in a campus competition ; 4) learned React from scratch in one month and developing eight chatbot-response algorithm visualization modules in one month, integrating them into a 30,000+ line project for large-scale experiments, leading to an international competition win; 5) developed four graph algorithm visualization modules in three weeks for a large-scale experiment involving over 300 participants. 6) currently working on two manuscripts for submission to IEEE TSE and BJET.

Throughout this period, I successfully managed eight courses and up to three group projects and assisted with various other research tasks assigned by my supervisor.

Education

Xi'an Jiaotong-Liverpool University

September/2021 - July/2025

BSc Information and Computing Science

WES GPA: 3.77/4.0    Check the WES unofficial transcript

Major GPA: 3.82/4.0

Ranked in the top 1.97% in Java Programming.

Ranked approximately in the top 9% in Operating Systems Concepts.

Ranked approximately in the top 5% in Advanced OO Programming.


Check the full official transcript and marking criteria

Publications

VisualCodeMOOC: A Course Platform for Algorithms and Data Structures Integrating a Conversational Agent for Enhanced Learning Through Dynamic Visualizations

Mingyuan Li; Duan Wang; Erick Purwanto; Thomas Selig; Qing Zhang; Hai-Ning Liang

SoftwareX (IF: 2.8) [Under Review: Submitted Nov. 2024]

Chatbot-Powered Learning for Sustainable Education in Programming

Erick Purwanto; Na Li; Qing Zhang; Thomas Selig; Yihong Wang; Teng Ma; Filbert Juwono; Pengfei Fan, (student member) Mingyuan Li and Duan Wang

Global Impact Grants 2024 - Education for Sustainable Development and Building Future Leaders

Access the full paper

Patch-Based Multi-Level Attention Mechanism for Few-Shot Multi-Label Medical Image Classification

Mingyuan Li; Yichuan Wang; Junfeng Huang; Erick Purwanto; Ka Lok Man

The 15th International Conference on Cyber-Enabled Distributed Computing and Knowledge Discovery, 2023

Access the full paper

Cervical Spine Fracture Detection Through Two-Stage Approach of Mask Segmentation and Windowing Based on Convolutional Neural Network

Doyeon Kim; Xujia Ning; Kaicheng Liang; Yi Ni; Duan Wang; Mingyuan Li ; Yichuan Wang; Erick Purwanto; Ka Lok Man

International Conference on Platform Technology and Service (PlatCon), 2023

Access the full paper

Accomplishments

Achieved Honourable Mention for MCM/ICM in 2023

Top 21%

Recognized as a Summer Undergraduate Research Fellow (2023) following participation in the program and poster competition

Check the Poster of the SURF (2023)

Achieved Third Prize in the IEEE CyberC 2023 Data Analysis Competition

Top 3

Achieved Second Prize at the 2023 XJTLU Student Research-Oriented Learning Summit

1000 RMB

Achieved the Global Impact Grants 2023-24 from AdvanceHE

Top 15 / £1000

Access the the Global Impact Grants 2023-24

Access the full study (Volume 1, Page 5) on advanceHE

Honored with an Honorable Mention in the CPT208 Human-Centric Computing

Top 10%

Presented project outcomes as a Guest Speaker at EdVenture: Exploring Innovative Practices in Higher Education

Achieved the Excellent Poster in SURF 2024 and presented research results at 2024 SURF Poster Fair as a School Winner.

Top 5%

Check the Poster of the SURF (2024)

Research Experience

Summer Undergraduate Research Fellowship 2023 (SURF)

Research Fellow

Supervisor:

Erick Purwanto

    Replication and Improvement of the Cervical Spine Fracture Detection Research (Publish a paper at PlatCon)
    • Implemented advanced pre-processing techniques, e.g., windowing, image cropping with Yolov5, and voxel clipping, to enhance image quality and prepare data for analysis.
    • Developed a two-stage approach for cervical spine CT scan analysis, achieving a combined accuracy of 94.9%, and a BCE logits coefficient of 0.20 ± 0.01.
    • Utilized UNet-EfficientNet in Stage 1 for precise CT image segmentation, attaining an accuracy of 99.91%. Applied CrackNet-LSTM in Stage 2 for accurate cervical spine fracture detection, achieving an accuracy of 94.9%.
    Few-Shot Multi-Label Medical Image Classification Research (Publish a paper at CyberC)
    • Training a few-shot classification model using the VPT model with a backbone of Swin-transformer. Evaluated the model's robustness using three datasets: ChestDR (chest X-ray), Endo (genuine colonoscopy), and Colon (colon cells).
    • Developed multi-level attention patch-based preprocessing technique for the model, enhancing the model's ability to detect minute details, e.g., overlapping information.
    • Improved mAP (1.2%-1.7%) and AUC (4.1%-5.2%) on two datasets (ChestDR, and Endo), compared to the baseline.

GenAI-Powered ChatBot in Programming and Algorithm Education

Research Assistant (Team of Two)

Supervisor:

Erick Purwanto

Yihong Wang

Thomas Selig

    Phase One: Created a programming tutor bot based on open-source projects tailored for programming and algorithm; winning the second prize at the XJTLU Student Research-Oriented Learning Summit
    • Utilized Prompt Engineering to call OpenAI APIs, creating an educational chatbot for programming with 8 different topics. Implemented Chain of Thought and Moderation to prevent non-educational output, making it suitable for beginners.
    • The bot was tested with an Indonesian student across four topics with two languages, achieving a 91% effectiveness rate, which demonstrates its pedagogical value and multilingual capabilities.
    Phase Two: Developed algorithm visualization module based on bot responses to aid teaching; received the Global Impact Grants 2023-24 from Advance HE with present findings
    • Created seven algorithm visualizations using d3.node, including for loops and array sorting lessons. Refactored the backend logic of a React-based TypeScript project to synchronize bot responses with dynamic visualizations, leading to the development of VisualCodeChat, a teaching-focused chatbot platform.
    • Developed a pilot MOOC with VisualCodeChat and integrated our chatbot into an existing mature platform to create VisualCodeMOOC.
    • Conducted 2 controlled experiments consecutively with 16 preA high school students and 88 non-programming background students from the CPT206 course. Reliable questionnaire results (Cronbach's alpha: 0.891) and a high average score of 4.18, supported by qualitative coding, demonstrate the bot met all evaluation criteria.
    • Authered a manuscript based on the current findings, currently under review by SoftwareX (IF: 2.8).
    Phase Three: Researched the effectiveness and HCI criterias of our design compared to standard GPT in teaching graph algorithms.
    • Iterated three graph algorithm visualizations using d3.force, including DFS-based cycle and connectivity checks, and BFS-based connected components check. Added data recording and user-friendly guidance modules.
    • Conducted 8 controlled experiments to verify its effectiveness: experiments involved 330 students with varying programming backgrounds from the CPT204 course at XJTLU. One group used our bot for the set learning tasks, while the other group used AI based on the original GPT. Ours outperforms ChatGPT across all dimensions, with T-test p-values below 0.05, confirming statistically significant improvements in usability, effectiveness, and engagement. The project was showcased at EdVenture and internal sharing seminars.
    • (In progress) Authoring a manuscript for British Journal of Educational Technology (IF: 6.7), with an anticipated abstract submission by the 1st Jan. 2025.
    • (In progress) Iterating VisualCodeChat to version 2.0, incorporating multimodal capabilities for experimental validation in multimodal-powered education.
    • (In progress) Evaluating code quality and algorithm visualization quality of GPT-4o-powered VisualCodeChat 2.0.
    • (In progress) Conducting 10 experiments, demonstrating its effectiveness in producing satisfactory code quality, effective algorithm visualization, and valid multimodal-powered test case education.
    • (In progress) Authoring a manuscript for IEEE Transactions on Software Engineering (IF: 7.4), with an anticipated submission by the end of 2024.

Course Projects

Access the details of all course projects on GitHub

Java Programming | Java

Individual Project

  • Use of object-oriented techniques, file processing, and data structures to accomplish pathological sequence detection and identification of DNA base sequences of interest.
  • Introduction to Databases | SQL

    Group Member of the Group Project

  • Completion of a third-paradigm-compliant drink information dataset and extension of the development of a database on drink packaging material information.
  • Artificial Intelligence | Python

    Individual Project

  • After performing dimensionality reduction through PCA, utilizing classification methods such as CNN, SVM, and Naive Bayes, along with unsupervised learning via K-means clustering.
  • Introduction to Computer Networking | Python

    Group Member of the Group Project

  • Implemented a client application for user authorization, file upload, and download using Python Socket programming based on a given protocol.

  • Created a simple SDN network topology with Mininet, simulating traffic control using SDN flow tables, enabling the SDN controller to forward/redirect traffic without client awareness.
  • Computer Graphics | C++, OpenGL

    Individual Project

  • Created a 2D graduation invitation card using OpenGL functions, including static and dynamic objects with interactive inputs.

  • Developed a 3D school scene using OpenGL, covering geometry creation, hierarchical modeling, transformations, viewing and projection, lighting and materials, texture mapping, animation, and interaction.
  • Software Engineering Group Project | Spring Boot, MySQL, HTML, CSS, JavaScript

    Group Leader of the Group Project

  • As the group leader, I led an 8-member team to develop an integrated sports center booking system using Spring Boot, based on the Model View Controller architecture, which includes both user and administrator systems consisting of 10 subsystems. I completed two full admin modules with full-stack development.

  • Human-Centric Computing | MoDao

    Group Member of the Group Project

  • Collaborated in a 5-member team to complete requirement analysis, design, prototyping, evaluation, and iteration using various HCI techniques, culminating in the production of a project presentation, report, and poster, and earning an Honorable Mention (1 of 8).
  • Advanced OO Programming | Java

    Group Member of the Group Project

  • Collaborated with a teammate to develop a multi-agent pathfinding game using BFS, employing Agile methodology, maintaining consistent code conventions, and implementing a user-friendly graphical interface. Demonstrated algorithm superiority through interactive validation and applied comprehensive OOP principles.
  • Big Data Analytics | Python

    Individual Project

  • Scraped 400 movie datasets from TMDB using BeautifulSoup. After literature reviews and idea exploration, performed exploratory data analysis with visualizations of post-Hollywood film attributes. Data cleaning and preprocessing led to eight visualizations addressing the research questions, followed by hypothesis validation using PLS-SEM.

  • Developed a Random Forest model with appropriate preprocessing for multi-class classification to predict student academic trajectory. The model outperformed others, achieving superior evaluation metrics and a competitive score in a Kaggle competition.
  • Machine Learning | Python

    Individual Project

  • Improved the GECCO architecture for handwritten digit recognition on the MNIST dataset by implementing three key modifications, enhancing model accuracy while reducing its size. The modified GECCO was compared with baseline models, including CNN, SpinalNet, ResNet, and Efficient-CapsNet. Results showed that the improved GECCO outperformed traditional CNN models, achieving higher accuracy, lower error rates, and a more compact parameter size.
  • Internship

    Guangzhou Hehui Technology Co.

    Java Software Engineer Intern; Mar-Jun/2023

  • Contributed to the development of a pharmaceutical ordering system.
  • Optimized the code structure with a low coupling design principle to ensure maintainability and scalability.
  • Effectively managed MySQL databases, and designed database structures that adhere to the Third Normal Form
  • Skills