Eugene Lee

Currently a Postdoctoral Researcher at the College of Medicine, my work bridges fundamental and applied AI. My fundamental research explores the mechanisms of in-context learning (ICL) and test-time adaptation (TTA) for LLMs to enhance model flexibility without retraining. On the applied side, I develop Vision-Language Models (VLMs) for biological applications, building imaging pipelines for organelle dynamics and utilizing LLMs to accelerate complex biological workflows through segmentation and temporal modeling.

At National Yang Ming Chiao Tung University (NYCU), I directed architectural studies optimizing Vision-Language Pipelines to improve token efficiency, reducing visual token count by 60% through adaptive token reduction. During this period, I also served as CEO and co-led BioFPGA commercialization at Advanced Bio Chips, driving cross-functional integration across IC design, embedded software, and bioassay teams.

I founded Paidge, an object-intelligence platform designed to index physical assets via scan-based conversational interaction. I led the development of a multi-frame aggregation model and a Perception Encoder with FAISS for efficient retrieval, integrated with a GraphRAG system for high-throughput indexing. The technology was successfully validated through multiple pilots with museums in the Greater Cincinnati area.

Note: paidge.com is no longer the main product page for this startup effort and now serves as an active hub for my personal research notes.

I earned my Ph.D. in Electronics Engineering from National Yang Ming Chiao Tung University (2023) under Prof. Chen-Yi Lee. My doctoral research focused on efficient deep learning, meta-learning, and resource-constrained neural networks, resulting in publications at top-tier venues including CVPR, ICCV, ECCV, and WACV.