Discounted Beta-Bernoulli Reward Estimation for Sample-Efficient Reinforcement Learning with Verifiable Rewards
International Conference on Machine Learning (ICML), 2026
About Me
I am a Ph.D. student at KAIST AI, advised by Prof. Eunho Yang at the Machine Learning and Intelligence Lab (MLILAB).
My research focuses on understanding and improving Vision-Language Models (VLMs), particularly their reliability and reasoning capabilities. I study their failure modes and limitations, from vision-language alignment and hallucination to spatial reasoning, and develop methods to address them through data-centric approaches and post-training.
Publications
Unveiling the Response of Large Vision-Language Models to Visually Absent Tokens
Conference on Empirical Methods in Natural Language Processing (EMNLP), 2025
* denotes equal contribution.
Work Experience
Research Intern · NCSoft
Language AI Lab, Language Model Team
- Reproduced and analyzed representative pre-trained language models (e.g., Transformer, BERT), and designed a time-aware dialogue generation model.
Education
Integrated M.S./Ph.D. in Artificial Intelligence · KAIST
Kim Jaechul Graduate School of AI · Advisor: Prof. Eunho Yang
B.S. in Computer Science · KAIST
Double major in Mathematical Sciences
Honors and Awards
Presidential Science Scholarship
대통령과학장학금, Ministry of Science and ICT / KOSAF
KAIST Presidential Fellowship
KAIST 총장장학생
KAIST Dean’s List
Spring 2020
Projects
Meta-Learning Applicable to Real-World Problems
IITP-funded national R&D program · KAIST
- Developing meta-learning models and algorithms that hold up under real-world conditions, across large-scale, unsupervised, few-task, multimodal, and cross-domain settings.
Unified Multi-Performance Prediction Model Based on Tire Pattern Feature Extraction
Industry collaboration with Hankook Tire · KAIST
- Replaced categorical pattern inputs, which could not represent unseen tread designs, with learned embeddings of tread images and pattern codes, and consolidated the per-metric models into a single multi-task predictor.
Academic Service
ICML, ECCV, EMNLP