Research Scientist at 
“Standing on the shoulders of giants.” — Isaac Newton
Research Interest
Building efficient, high-performing Vision-Language Models (VLMs), with focus on:
- Distillation
- Pruning
- RL
- Agent
- Harness
University Collaboration or Intern & Full-time
Feel free to reach out for collaboration by email if:
- Your research interests align with mine.
- You already have a draft idea to discuss and develop together.
For university collaboration, alignment is the primary criterion.
We are actively looking for talented internship and full-time candidates. Preferred qualifications:
- Strong first author publications at top-tier main-track conferences (not workshops); co-first author papers are discouraged.
- Deep expertise in one focused area: in my view, five-to-ten first author papers at top-tier venues.
- Top-tier venues: CVPR, ICCV, ECCV, NeurIPS, ICLR, ICML, ACL, EMNLP. I am not familiar with other conferences and journals.
If you meet these criteria, please send me an email with your resume/CV attached.
Work Experience
NVIDIA
Research Scientist
Oct. 2025 — Current
-
LEAD
Mid-Harness: Scaling Actions Between Model and Harness
[Completed]
—
Action-level test-time scaling for terminal agents that samples and verifies candidate commands at the model–harness boundary before execution, lifting trajectory success without changing the generator or harness.
Under Review U.S. Provisional Patent Filed Internal Tech Transfer
-
LEAD
ZPPO: Zone of Proximal Policy Optimization
[Completed]
—
RL post-training that keeps the teacher inside the prompt rather than the policy gradient, using BCQ/NCQ question reformulations and a prompt replay buffer to lift small students on hard questions without policy drift.
Under Review U.S. Provisional Patent Filed
-
LEAD
AXPO: Agent eXplorative Policy Optimization
[Completed]
—
Agentic RL training that recovers tool usage through tool-call resampling, improving multimodal reasoning performance against larger baselines.
NeurIPS 2026 U.S. Provisional Patent Filed Internal Tech Transfer
-
LEAD
Masking Teacher and Reinforcing Student
[Completed]
—
Mask-progressive RL distillation that gradually unmasks teacher weights and uses offline RL with accuracy & distillation rewards.
CVPR 2026 U.S. Non-Provisional Patent Filed
-
LEAD
Unified RL & Imitation Learning for VLMs
[Completed]
—
Combines RL with adversarial imitation, using an LLM-based discriminator and multi-teacher guidance to build lightweight yet powerful VLMs.
NeurIPS 2025 U.S. Patent Upgraded to Non-Provisional
-
LEAD
GenRecal
[Completed]
—
Cross-architecture VLM distillation via a Recalibrator that aligns heterogeneous token representations regardless of vocabulary, token splits, or index ordering.
ECCV 2026 U.S. Patent Upgraded to Non-Provisional Internal Tech Transfer
-
LEAD
VLsI: Verbalized Layers-to-Interactions
[Completed]
—
Hardened the verbalizer-based layer-wise distillation into a production-ready recipe and led its internal tech transfer into NVIDIA's VLM development pipeline.
CVPR 2025 U.S. Patent Upgraded to Non-Provisional Internal Tech Transfer
NVIDIA
Research Intern
Oct. 2024 — Oct. 2025
-
LEAD
Unified RL & Imitation Learning for VLMs
[Initiated]
—
First formulation of the RL + adversarial imitation training pipeline; initial experiments and team setup.
U.S. Provisional Patent Filed
-
LEAD
GenRecal
[Initiated]
—
Initial design of the Recalibrator framework for cross-tokenizer VLM distillation; first proof-of-concept and internal demo.
U.S. Provisional Patent Filed
-
LEAD
VLsI: Verbalized Layers-to-Interactions
[Initiated]
—
Layer-wise distillation using intermediate verbalizers, enabling small VLMs (2B/7B) to align with large VLMs' reasoning progression and outperform GPT-4V.
CVPR 2025 U.S. Provisional Patent Filed
Education
KAIST
Mar. 2020 — Aug. 2025
Ph.D., School of Electrical Engineering GPA 3.77 / 4.3
Dissertation: Building High-performing, Efficient-size Vision Language Models: Merge, Modify, and Distill
[Link]
[Degree Certificate]
KAIST
Mar. 2018 — Feb. 2020
M.S., The Cho Chun Sik Graduate School of Green Transportation GPA 3.72 / 4.3
Thesis: Training Encoder-Attention through Fully-Connected CRFs for Efficient End-to-End Lane Detection Model
[Link]
[Degree Certificate]
Hanyang University
Mar. 2014 — Feb. 2018
B.S., Mathematics and Electronic Engineering GPA 3.86 / 4.5
Thesis: Learning Rank of Evaluated Songs for Classifying Unknown Music by Convolutional Neural Network
[Link]
[Degree Certificate]
Publications
| Overall | 21 Accepts · 3 Pending · 5 Tech Reports |
| CV | 5 CVPR · 2 ICCV · 3 ECCV · 1 ICIP |
| ML | 6 NeurIPS · 1 ICLR |
| NLP | 1 ACL · 1 EMNLP |
| Journal | 1 Pattern Recognition |
| In Progress | 16 On-going Projects |
Publication Profile
69%
31%
- First & Lead Author 20 / 29 = 69%
- Co-Author 9 / 29 = 31%
-
-
-
-
-
-
-
-
-
Masking Teacher and Reinforcing Student for Distilling Vision-Language ModelsIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2026 [Paper]U.S. Patent Application Filed (Non-Provisional), NVIDIA Research -
-
-
-
-
-
-
-
-
SPARK: Multi-Vision Sensor Perception and Reasoning Benchmark for Large-scale Vision-Language Models -
-
-
-
-
-
-
-
-
-
-
Reviewer Experience
Journal
- IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI)
- IEEE Transactions on Neural Networks and Learning Systems (TNNLS)
- IEEE Transactions on Circuits and Systems for Video Technology (TCSVT)
Conference
- IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
- IEEE/CVF International Conference on Computer Vision (ICCV)
- European Conference on Computer Vision (ECCV)
- International Conference on Learning Representations (ICLR)
- Neural Information Processing Systems (NeurIPS)
- International Conference on Machine Learning (ICML)
- AAAI Conference on Artificial Intelligence (AAAI)
- Association for Computational Linguistic (ACL)
- Empirical Methods in Natural Language Processing (EMNLP)
Invited Talks & Awards
- Invited Talk at Chung-Ang University2026
- Invited Talk at Kongju National University2026
- Invited Talk at Kyonggi University2026
- Invited Talk at Kookmin University2026
- Outstanding Reviewer Award, European Conference on Computer Vision (ECCV)2026
- Silver Reviewer Award, International Conference on Machine Learning (ICML)2026
- Best Runner-Up Award (Oral, Top 1%), Multi-Turn Interactions in LLMs Workshop @ NeurIPS2025
- Invited Talk at Kongju National University2025
- Invited Talk at Hanyang University ERICA2025
- Invited Talk at Kookmin University2025
- Invited Talk at Dongseo University2025
- 31st Samsung HumanTech Paper Awards in Computer Science & Engineering2025
- KCC XAI Workshop, Best Paper Awards2024
- Invited Talk at Korea Institute of Science and Technology Information (KISTI)2024
- Invited Talk at NAVER HyperCLOVA X2024
- Invited Talk at Kakao Brain2024
- 1st Award for Research Presentation, Center for Applied Research in AI (CARAI)2023
- GIRE Research Mentor, Seoul International School — "MUsE", Journal of Student Research [Publication]2023
- KAIST SW IT Academy, AI Project Director of 1st Award Topic2023
- Invitation to Presenter in Korean Conference on Computer Vision (KCCV)2023
- KAIST SW IT Academy, Lecturer of Data Structure & Algorithm, Deep Learning, Computer Vision2023
- Invitation to Presenter in Korean Conference on Computer Vision (KCCV)2022
- 2nd Award for Research Presentation, Center for Applied Research in AI (CARAI)2022
- KAIST FellowshipMar. 2020 — Feb. 2024
- National Government FellowshipMar. 2018 — Feb. 2020
- 1st President's Award for International Autonomous Vehicle Competition, Team Leader [YouTube][Talk]2018
NVIDIA Stock
Current Price & Today Change
TradingView