Hi there - My name is Shujun Xia, and I also go by Erin. I’m currently a junior majoring in CS at Columbia University. Previously, I was fortunate to work with Prof. Antoni Chan on explainable AI for computer vision. I have also worked on Continual Learning and Knowledge Editing in LLMs with two first-authored papers under review. My current research interests lie in multimodal understanding and learning, and interpretability of deep learning models.
I am actively seeking research collaborations. If you are interested in potential collaborations or have related interests, feel free to email me.
Email: shujun dot xia at columbia dot edu
📣 News
- [2025/11] MedREK has been accepted to ResponsibleFM@NeurIPS 2025! Huge thanks to all my coauthors.
- [2025/09] Awarded the ICCV Broadening Participation Travel Grant. See you in Hawaii!
- [2025/07] My first paper accepted to eXCV@ICCV 2025!
✨ Experiences
Hong Kong Generative AI Research and Development Center
Machine Learning Engineer Intern | Mar. 2025 – Jun. 2025
Expand details
- Trained a Transformer-based punctuation prediction model using PyTorch, improving punctuation restoration for speech transcripts.
- Constructed an automated data generation pipeline using Pandas for speech transcript data.
Department of Computer Science, City University of Hong Kong
Software Developer Intern | Jun. 2024 – Mar. 2025
Expand details
- Built a chatbot web application for students with a Python Streamlit front-end and Django backend, integrating the OpenAI API for GPT-powered chatbots.
- Developed a course management system for teachers using Spring Boot and React, enabling student progress monitoring.
- Designed and optimized SQL and MongoDB schemas for user and chat data, improving query efficiency.
- Deployed both systems with Docker on AWS Cloud and implemented automated testing using Pytest, JUnit 5, and Jest.
Video, Image, and Sound Analysis Lab, City University of Hong Kong
Student Researcher | Sep. 2023 – Jun. 2024
Expand details
- Developed explainable AI methods for interpreting predictions from Transformer-based (DETR, PVT) and one- and two-stage object detectors (FCOS, Faster R-CNN).
- Generated explanation heatmaps using gradients w.r.t. feature maps, and proposed a measure of context influence to quantify object relationships.
📝 Publications
Explaining Object Detection Through Difference Map
ICCV exCV Workshop (Proceedings Track), 2025
Shujun Xia, Chenyang Zhao, Antoni Chan
MedREK: Retrieval-Based Editing for Medical LLMs with Key-Aware Prompts
NeurIPS ResponsibleFM Workshop, 2025
Shujun Xia*, Haokun Lin*, Yichen Wu, Yinan Zhou, Zixuan Li, Zhongwei Wan, Xingrun Xing, Yefeng Zheng, Xiang Li, Caifeng Shan, Zhenan Sun, Quanzheng Li
📖 Educations
- 2025.08 - 2027.06 (expected): Columbia University (Dual Degree Program).
- 2022.08 - 2027.06 (expected): City University of Hong Kong (Dual Degree Program).
🏆 Honors and Awards
- ICCV 2025 Broadening Participation Travel Grant (2025)
- Columbia University Distinguished New Student Scholoarship (2025)
- National First Prize, Wutong Cup Big Data Innovation Competition (2024)
- HKSAR Talent Development Scholarship (2023 & 2025)