About

I am interested in building foundation models that can interact efficiently with the physical world. My work combines robot foundation models, embodied AI, and efficient AI systems.

I am an integrated M.S./Ph.D. student in Software at Sungkyunkwan University, advised by Honguk Woo in CSI Agent Group. My current research examines efficient multimodal inference, including VLA models operating under continuously changing observations.

My earlier research focused on memory management and ML systems. That background led me to study computation and state reuse in long-running AI systems, and how those mechanisms can support responsive embodied agents.

Background

I joined CSI Lab as a graduate researcher in 2024 and began the integrated M.S./Ph.D. program in 2025. Previously, I completed my B.S. in Software at Sungkyunkwan University and worked with VLDB Lab on memory management for HTAP workloads.

My research includes HeMeR (NeurIPS 2026 · Oral) and two ICML 2026 · Poster papers. I also work on efficient system support for foundation-model and agent workloads through the Samsung Electronics HiPER collaboration.

My academic CV contains the full publication record, projects, teaching experience, and awards.

Contact

For research discussions or collaboration: thousae@skku.edu.