Biography
Xuhui Zhan, a second-year Master’s student in Data Science at Vanderbilt University and a Lead AI Data Science Researcher there, embarked on his data science journey during undergraduate studies. This early foundation set the stage for his professional and academic growth.
Formerly an Algorithm Development Engineer, Xuhui specialized in automating AGV robots and developing machine learning systems for detecting cargo status. This role deepened his practical insights into data science applications.
In his current academic pursuit, Xuhui focuses on deep learning areas like Graph Neural Network, NLP and recommandation system. His passion lies in creating machines that can understand the world and perform complex, multi-modal tasks. His ultimate goal is ambitious: to enable machines to dream, emulating one of humanity’s most profound abilities.
Driven by the principle that “Any sufficiently advanced technology is indistinguishable from magic,” Xuhui aims to transcend the boundaries of technology, merging the real with the fantastical. He seeks to deepen his data science expertise, contributing to innovations that turn today’s science fiction into tomorrow’s reality.
- Graph Neural Network
- Computer Vision
- Natural Language Processing
- Recommendation System
- Foundation Models
- Neural-Symbolic AI
- AI4Science
- Quantum Computing
MSc in Data Science, 2023-Present
Vanderbilt University
BSc in Data Science, 2018-2022
BNU-HKBU United International College
BSc in Data Science, 2018-2022
Hong Kong Baptist University
Experience
Advisor: Ray Friedman (AINegotiation Lab)
- Served as Lead AI Data Science Researcher in the AINegotiation Lab. (Please check out lab details at https://www.ainegotiationlab-vanderbilt.com).
- Full-time during summer and part-time during semesters.
- Develop programs leveraging LLMs for negotiation applications.
Advisor: Tyler Derr (NDS Lab)
Develop a fusion network that integrates Graph Neural Networks (TGAT) and LLMs (Llama 3), not from a token based fusion but a highly customized knowledge injection way with adoption of ideas from PEFT (LoRA and DoRA) and differential transformer to enable the LLMs understanding the topological information and messages for personalized generation in social network settings.
Implement from scratch, need to rewrite the transformers library for customization need, first on Venmo Dataset (Spend about 3 months to collect, 3 million users) then switch to Amazon Review Dataset (Adopted from UCSD collections, raw dataset more than 78 GB after compression)
Showing potential to expand to unified modalities fusion.
Idea is cheap, experiments are expensive.
Graduate Teaching Fellowship
Responsible for organizing labs, creating quizzes, grading materials and hosting office hours.
Graduate level:
- DS 5620: Probability and Statistical Inference
- DS 5690: Gen AI in Theory and Practice
Undergrad level:
- DS 3100: Fundamentals of Data Science
Advisor: Markus Eberl
- Apply vision transformers on ancient mortars classification.
- Develop algorithms to enable sample similarities comparison based on the pure vision input. (Estimate moving traces between samples collected from different places)
Work as an algorithm engineer and take in charge of all machine learning applications and efficient data analysis.
Responsibilities include:
- Optimize the end localizer program and deploy on new products (ROS)
- Develop the vision algorithm for detecting whether two shelves are aligned (opencv, pytorch)
- Develop the end-to-end storage status detection system, implement the whole pipeline including data collection, data labeling, model training, model deployment, database construction, communication policies and api for updating status to central server (WCS) and display results on the front-end website with scaling abilities to aggregate and update results in real-time on constraint computation resources and deployment process (roboflow, django, redis, celery, pytorch, ONNX)