Template Infilling
Template Infilling is a conditioning method for diffusion language models that places structural anchors throughout the target response before filling masked segments.
Ph.D. Candidate · Seoul National University
Junhoo Lee
Machine learning researcher at MIPAL, advised by Prof. Nojun Kwak. Ph.D. expected August 2026.
I work on diffusion language models, meta-learning, and methods for analyzing and attributing pretrained models. I also work on vision-language-action models.
Selected Research
CSF
CSF attributes fine-tuned text-to-image models to their base-model lineages using only query access and compositional, underspecified prompts.
Deep Support Vectors
DeepKKT adapts the Karush–Kuhn–Tucker condition to identify or generate Deep Support Vectors from trained deep classifiers.
Selected Publications
ACL2026
Unlocking the Potential of Diffusion Language Models through Template Infilling
, Seungyeon Kim, Nojun Kwak · Long Paper, Oral Presentation
CVPR2026
CSF: Black-box Fingerprinting via Compositional Semantics for Text-to-Image Models
, Mijin Koo, Nojun Kwak · Query-Only Attribution
NeurIPS2025
Deep Edge Filter †
Dongkwan Lee†, , Nojun Kwak · Representation Analysis
ICCV2025
What's Making That Sound Right Now? Video-centric Audio-Visual Localization
Hahyeon Choi, , Nojun Kwak · Audio-Visual Localization
NeurIPS2024
Deep Support Vectors
, Hyunho Lee, Kyomin Hwang, Nojun Kwak · Isometry
AAAI2024
Any-Way Meta Learning
, Yearim Kim, Hyunho Lee, Nojun Kwak · Few-Shot Learning
NeurIPS2023
SHOT: Suppressing the Hessian along the Optimization Trajectory
, Jayeon Yoo, Nojun Kwak · Optimization
Talks & Academic Service
- Compositional Semantic Fingerprinting for Black-box Attribution
SWCS 2026 - Deep Support Vectors: Interpreting Deep Models via Support Vector Extraction
KCC 2024 - ICML Gold Reviewer
- Conference Reviewer
CVPR, ICCV, ECCV, ICLR, ICML, NeurIPS
Recent News
- I will join ALIN-LAB at KAIST as a postdoctoral researcher in September 2026.
- Template Infilling will be presented at ACL 2026 as a long-paper oral presentation.
- CSF was presented at CVPR 2026.
- Selected as an ICML 2026 Gold Reviewer.
- Template Infilling was accepted to ACL 2026 as a long paper and oral presentation.
Education & Selected Honors
Education
Sep 2021 – Aug 2026
Ph.D. Candidate in Intelligence and Information
Seoul National University · Advisor: Prof. Nojun Kwak
Mar 2017 – Sep 2021
B.Sc. in Electrical and Computer Engineering
Seoul National University
Selected honors
2023
BK21 Future Innovation Talent Bronze Prize
2023
BK21 Outstanding Research Talent Fellowship
2022
Yulchon AI Star Scholarship