Education
Dartmouth College
PhD Student in Computer Science
(Sep. 2024 – Present, Hanover, NH, US)
Supervised by Prof. Yu-Wing TAI, Associate Professor at Dartmouth College
The Hong Kong University of Science and Technology
B.Sc. in Data Science and Technology, First Class Honor (top 10% )
(Sep. 2020 – Jun. 2024, Hong Kong SAR)
École polytechnique fédérale de Lausanne (EPFL)
Regular Term Exchange
(Sep. 2022 – Feb. 2023, Lausanne, Switzerland)
Publications
(ArXiv Preprint, Co-First Author)
VP-LLM: Text-Driven 3D Volume Completion with Large Language Models through Patchification(ArXiv Preprint, Co-First Author)
Prompt2NeRF-PIL: NeRF Generation via Pretrained Implicit Latent(MICCAI 2023 (DART), Co-First Author)
Self-prompting Large Vision Models for Few-Shot Medical Image Segmentation
Working Experience
Research Assistant @ HKUST
(Jun. 2023 – May. 2024)
Co-supervised by Prof. Chi Keung Tang (HKUST)
and Prof. Yu-Wing TAI (Dartmouth College)
- VP-LLM: Text-Driven 3D Volume Completion with Large Language Models through Patchification
- Proposed a novel method of encoding 3D models into “patches” that can be parsed by LLMs, the first work to leverage LLMs for 3D objects.
- Finetuned LLMs to achieve complex text-guided 3D object completion and denoising, with ~20% improvement over existing methods in recovery quality.
- Prompt2NeRF-PIL: NeRF Generation via Pretrained Implicit Latent
- Proposed and implemented “Prompt2NeRF-PIL,” a fast method for generating 3D objects in NeRF from text/image prompts in one forward pass, increasing speed by 3–5 times over current state-of-the-art.
Teaching Assistant (Database) @ HKUST
(Sep. 2023 – Dec. 2023)
COMP3311 Database Management Systems — Department of Computer Science and Engineering
- Guided students through lab assignments and answered questions during tutorial sessions
- Prepared lab and tutorial materials
- Investigated potential usage of ChatGPT for DBMS design
Projects
CT Image Segmentation
(Feb. 2023 – Jul. 2023, HKUST)
Supervised by Prof. Xiaomeng Li
- Proposed a self-prompting method for few-shot medical image segmentation
- Compared performance against multiple state-of-the-art methods
- Published a paper as Co-First Author:
Self-prompting Large Vision Models for Few-Shot Medical Image Segmentation
Crack Segmentation on FRP CT Images
(Oct. 2022 – Dec. 2022, EPFL CS433 Course Project)
GitHub Project Link
- Collaborated with CCLab (EPFL) to collect and annotate FRP CT images
- Proposed Multi-scale Dilated Convolution (MDC), significantly improving state-of-the-art performance; method was put into service
Trends in the American Beer Market
(Oct. 2022 – Dec. 2022, EPFL CS401 Course Project)
GitHub Project Link
- Employed sentiment analysis and NLP to analyze consumer comments
- Utilized statistical and time series modeling to explore the US beer market trends