Efficient training & inference.
Multimodal & generative AI.
I am a Ph.D. student in Electrical and Computer Engineering at Seoul National University, working with the Intelligent Computing Lab. I develop methods for faster, memory-efficient training and inference in multimodal and generative AI. My goal is to reduce compute costs and memory requirements while preserving model quality, making capable models practical under real-world resource constraints.
Research in focus
Strong Helps Weak: Directional Cross-Modal Alignment Transfer in Multi-modal LLMs
Transferring alignment from strong vision models to audio and video MLLMs without further fine-tuning.
Explore paper
Training-free Mixed-Resolution Latent Upsampling for Spatially Accelerated Diffusion Transformers
Region-adaptive latent upsampling accelerates diffusion transformers with minimal quality degradation.
Explore paper
Erasing Thousands of Concepts: Towards Scalable and Practical Concept Erasure for Text-to-Image Diffusion Model
Scalable concept erasure that removes thousands of concepts while preserving generation quality.
Explore paper
On Epistemic Uncertainty of Visual Tokens for Object Hallucinations in Large Vision-Language Models
Identifying and masking uncertain visual tokens to reduce object hallucinations without training.
Explore paperWhat I work on
Multimodal Language-Vision Models
Understanding and reducing object hallucinations with uncertainty-aware token analysis.
Diffusion Model Efficiency
Improving speed and memory efficiency with adaptive sampling and lightweight optimization.
Text-to-Image Alignment
Studying token geometry and cross-attention behavior for stronger semantic binding.
Practical Robustness
Designing methods that remain effective under limited compute and real-world constraints.
Recent news
Oct 15, 2026
Our paper "Recovering Diversity in Distilled Diffusion Models via Low-Rank Policy-Optimized Merging" was accepted to WACV 2027!
Oct 09, 2026
Selected as a Top Reviewer at NeurIPS 2026.
Sep 30, 2026
Our paper "Strong Helps Weak: Directional Cross-Modal Alignment Transfer in Multi-modal LLMs" (DCAT) was accepted to NeurIPS 2026 as a Spotlight (Top 1.3%)!
May 01, 2026
Our paper "Triadic Dynamics Aware Diffusion Posterior Sampling for Inverse Problems: Optimizing Guidance and Stochasticity Schedules" accepted to ICML 2026! See you in Seoul, Korea (my hometown)!
Mar 12, 2026
3 Papers accepted to CVPR 2026 (2 first-authored, 1 co-authored)! RALU was selected as a Highlight (Top 3.4%). See you in Denver, Colorado!