Assistant Professor, Yonsei University
3 papers at NeurIPS 2025
A two‐stage active learning pipeline that uses diffusion‐based feature sampling and entropy‐augmented disagreement to pick the most informative pixels under extreme labeling constraints.
Instance-level adaptive KL penalty control method for Direct Preference Optimization
Our Orthogonal Residual Update improves deep networks by adding only novel, stream-orthogonal module outputs, boosting generalization, stability, and efficiency.
