FastOPD: On-Policy Distillation for Lightweight VLA Deployment
arXiv preprint, 2026.
About
I am a Ph.D. student at the KAIST Graduate School of AI, working in the Bio Imaging, Signal Processing & Learning Lab (BISPL) under the supervision of Prof. Jong Chul Ye.
My research focuses on generative models (3D/4D, videos) and robotics: how the dynamic world looks, how it changes over time, and how agents act in it.
Previously I was a Research Intern at NAVER AI Lab (Generation Research Team), hosted by Jin-Hwa Kim.
News
- 2026FlowLong selected as an Oral at the AVGenL Workshop @ ECCV 2026.
- 2026Zero4D accepted to CVPR Findings 2026.
- 2025Joined NAVER AI Lab, Generation Research Team, as a Research Intern (Mar–Sep).
- 2024ED-NeRF presented at ICLR 2024.
Experience
-
2025
Research Intern, NAVER AI Lab
Publications
DriftOPD: Sequence-Level Reverse-KL Distillation for One-Step VLA Policies
arXiv preprint, 2026.
Adjoint Guidance Flow: Amortized Critic Guidance for VLA Policies
arXiv preprint, 2026.
Don't Throw Away the Tail: Action Upcycling for Policy Acceleration
arXiv preprint, 2026.
Dynamic Manipulation with World-Action Models via Counterfactual Planning
arXiv preprint, 2026.
FlowLong: Inference-time Long Video Generation via Manifold-constrained Tweedie Matching
arXiv preprint, 2026.
Zero4D: Training-Free 4D Video Generation From a Single Video Using Off-the-Shelf Video Diffusion
IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Findings, 2026.
ED-NeRF: Efficient Text-guided Editing of 3D Scene Using Latent Space NeRF
International Conference on Learning Representations (ICLR), 2024.
Academic Service
Reviewer NeurIPS 2025 · ICLR 2025, 2026 · ICCV 2025 · CVPR 2026 · ECCV 2026