Youngsun Lim
Hello :) I am Youngsun Lim, a Master's student at KAIST AI. My research interests lie in Responsible, Reliable & Safe AI, within the domains of Multimodal and Generative AI. I am a researcher who is constantly mindful of the societal impacts of my work, respecting minorities, and striving to make a positive contribution to the world. You can check my CV here.
Contact: youngsun_ai@kaist.ac.kr, firefpswl@naver.com
Research Interests
- Responsible, Reliable & Safe AI
- Generative AI & AI Alignment
- Computer Vision
- Human-Computer Interaction
news
| Jan, 2025 |
The paper "Label-Augmented Dataset Distillation " has been selected for an **Oral Presentation** at WACV 2025!  |
|---|---|
| Jan, 2025 |
The paper "Evaluating Image Hallucination in Text-to-Image Generation with Question-Answering " has been selected for an **Oral Presentation** at AAAI 2025! |
| Dec, 2024 |
The paper "Evaluating Image Hallucination in Text-to-Image Generation with Question-Answering " has been accepted to AAAI 2025!  |
| Sep, 2024 |
The paper "Label-Augmented Dataset Distillation " has been accepted to WACV 2025! |
| Sep, 2024 |
I was selected as an alternate candidate for the 2025 Fulbright U.S.-Korea Presidential STEM Initiative Award!  |
| June, 2024 |
The paper "Addressing Image Hallucination in Text-to-Image Generation through Factual Image Retrieval " has been accepted for an oral presentation at the IJCAI 2024 Workshop! |
| June, 2024 |
I was selected as a recipient of the M.S. Fellowship from the National Research Foundation (NRF) of Korea!  |
Education
M.S. in Artificial Intelligence, KAIST AI
|
Aug 2023 - present GPA: 96.5/100 |
B.S. in Physics and Artificial Intelligence (Double Major), Sogang University
|
Feb 2017 - Aug 2023 GPA: 93.9/100 |
publications
-
Evaluating Image Hallucination in Text-to-Image Generation with Question-AnsweringIn 39th Annual AAAI Conference on Artificial Intelligence (AAAI '25) 2024 (Oral presentation)
-
FUS (Find Undetectable Sound): An application to indicate sound directionIn Journal of Digital Contents Society 2022 (Korean Citation Index)
-
FUS (Find Undetectable Sound): Machine Learning Application to indicate the sound directionIn Proceedings of the Korean HCI Conference 2022
Research Experiences
Responsible, Reliable & Safe AI / Generative AI
-
Master’s Research Assistant at CVML Lab, KAIST AI (Aug 2023 - present)
- First proposed the concept of “image hallucination” to address the failure of text-to-image models in generating factual information from input prompts, and conducted research on benchmarks, evaluation metrics, and solutions.
- Investigated dataset distillation to address privacy issues in datasets and reduce training costs.
- Currently researching uncertainty in multimodal generation as a cause of hallucination, aiming to reduce hallucinations and generate reliable datasets.
-
Undergraduate Researcher at ISDS Lab, Sogang University (Aug 2022 - Jul 2023)
- Led the first project to build a benchmark for the Korean legal domain and eliminate hallucinations from large language models using chain-of-thought prompting to replicate legal experts’ reasoning.
Human-Computer Interaction & Machine Learning
-
Undergraduate Researcher at Lifestyle Media Lab, Sogang University (Apr 2021 - Jul 2022)
- Conducted research on a machine learning application to detect undetectable sounds and prevent hearing loss, achieving 73% accuracy in sound localization. Presented my work at HCI Korea 2022 and published two first-author papers.
-
“Wearable Assistive Devices for Deaf-Blind People” Project (Mar 2019 - Jun 2019)
- Led the project in the Physics Capstone Design I class as a team leader.
- Developed wearable devices using Arduino and ultrasonic sensors to assist deaf-blind individuals.
Professional and Industrial Experiences
-
Industry-Academic Cooperation with Daimler Truck AG, Japan (Aug 2022 - Dec 2022)
- Optimized machine learning models to predict trucks’ estimated time of arrival, improving accuracy by 15%.
- Project Reported in the News (English): Link to the article
-
AI Research & Development Intern at INEEJI Corporation, Korea (Feb 2022 - Aug 2022)
- Developed the Korean national project to estimate traffic flow states using a GNN-based model (STGCN, etc.).
- Investigated an adaptive adjacency matrix based on real road data, improving the mean absolute error by 5%.
- Project Reported in the News (Korean): Link to the article
Awards and Honors
- 2025 Fulbright U.S.-Korea Presidential STEM Initiative Award (Alternate Candidate, approximately 30 people in South Korea)
- M.S. Fellowship from the National Research Foundation (NRF) of Korea (Research scholarship, awarded to projects with a 4.4:1 acceptance rate in South Korea)
- Best Presentation Paper Award, 2023 Korea Computer Congress (KCC) (Awarded for outstanding research and presentation)
- South Korea National Scholarship (Financial support for outstanding students)
- Sogang University Alumni Association Academic Scholarship
- Sogang University Dasan Scholarship
- Industry-academic Scholarship by LG Electronics (Awarded with a selection rate of 10:1 in South Korea)
- 1st Prize in Data Analytics Competition (Held by “Insight” Club, Sogang University)
- Lotte Scholarship Recipient, 4th Cohort (Financial aid awarded to 1,000 outstanding undergraduates)
Professional Service
- Reviewer for NeurIPS 2024, ICLR 2025, ICML 2025