Junu Kim

I'm Junu Kim, a Ph.D. student at the KAIST Graduate School of AI, advised by Professor Edward Choi and Han-Gil Jeong. My research focuses on long context modeling and healthcare AI. I study (1) efficient Transformer-based methods for modeling long context data, including an ongoing project using test-time training for text and REMed for electronic health records; and (2) specialized large language models for clinical data, including C-Reason, a reasoning model trained on a sepsis registry, and Asclepius, one of the earliest clinical QA LLMs. Keywords: Long Context Modeling, ML4H, Clinical LLMs

Spotlights

Enhancing LLMs' Clinical Reasoning with Real-World Data from a Nationwide Sepsis Registry

Junu Kim, Chaeeun Shim, Sungjin Park, Su Yeon Lee, Gee Young Suh, Chae-Man Lim, Seong Jin Choi, Song Mi Moon, Kyoung-Ho Song, Eu Suk Kim, Hong Bin Kim, Sejoong Kim, Chami Im, Dong-Wan Kang, Yong Soo Kim, Hee-Joon Bae, Sung Yoon Lim, Han-Gil Jeong, Edward Choi [Arxiv] [Github]

💡
We propose enhancing the clinical reasoning capabilities of LLMs by leveraging real-world clinical data.

LayerNorm Induces Recency Bias in Transformer Decoders

Junu Kim, Xiao Liu, Zhenghao Lin, Lei Ji, Yeyun Gong, Edward Choi [Arxiv (ACL 2026 Findings)] [Github]

💡
We discover that LayerNorm induces recency bias in Transformer decoders.

General-Purpose Retrieval-Enhanced Medical Prediction Model Using Near-Infinite History

Junu Kim, Chaeeun Shim, Bosco Seong Kyu Yang, Chami Im, Sung Yoon Lim, Han-Gil Jeong, Edward Choi [Arxiv (MLHC 2024)] [Github]

💡
We propose REMed, the first retrieval-based model for medical prediction tasks, which reduces the need for domain experts in the development process.

Asclepius: Publicly Shareable Clinical Large Language Model Built on Synthetic Clinical Notes

Sunjun Kweon*, Junu Kim*, Jiyoun Kim, Sujeong Im, Eunbyeol Cho, Seongsu Bae, Jungwoo Oh, Gyubok Lee, Jong Hak Moon, Seng Chan You, Seungjin Baek, Chang Hoon Han, Yoon Bin Jung, Yohan Jo, Edward Choi [ACL 2024 Findings] [Github] [Huggingface]

💡
We introduce the first publicly sharable clinical large language model, Asclepius.

Publications

  • Towards Predicting Temporal Changes in a Patient's Chest X-ray Images based on Electronic Health Records [Arxiv (CHIL 2025)]
    Daeun Kyung, Junu Kim, Tackeun Kim, Edward Choi
  • Federated Learning for Heterogeneous Electronic Health Record Systems with Cost Effective Participant Selection [Scientific Reports]
    Jiyoun Kim, Junu Kim, Kyunghoon Hur, Edward Choi
  • GenHPF: Universal Healthcare Predictive Framework with Zero Domain Knowledge [JBHI] [Arxiv]
    Kyunghoon Hur, Jungwoo Oh, Junu Kim, Jiyoun Kim, Min Jae Lee, Eunbyeol Cho, Seong-Eun Moon, Young-Hak Kim, Louis Atallah, Edward Choi

Experiences

  • Institute of Foundation Models, Abu Dhabi, UAE (Remote)
    Advised by Haonan Li, Apr 2026 - Current
  • Microsoft Research Asia, Beijing, China (On-site)
    Artificial Intelligence & Reasoning Team, Advised by Xiao Liu, Jul 2025 - Dec 2025

Projects & Abstracts

  • Toward general‑purpose ICU predictive model: without task‑specific feature selection and infinite observation windows [ESICM LIVES 2023 Abstract]
    Junu Kim, Chaeeun Shim, Edward Choi, Han-Gil Jeong
  • CAMEL: Clinically Adapted Model Enhanced from LLaMA [Blog] [Github] [Demo]
    Sunjun Kweon*,Junu Kim*, Seongsu Bae**, Eunbyeol Cho**, Sujeong Im**, Jiyoun Kim**, Gyubok Lee**, JongHak Moon**, JeongWoo Oh**, Edward Choi
  • EHRFL: Federated Learning Framework for Heterogeneous EHRs and Precision-guided Selection of Participating Clients [Arxiv (ML4H 2024 Findings)] [Github]
    Jiyoun Kim, Junu Kim, Kyunghoon Hur, Edward Choi

Education

KAIST Graduate School for AI

MS-Ph.D. Integrated Course, 2022-Current

KAIST Computer Science

BSc Computer Science, 2018-2022

Misc.

Reviewer

ICLR, ARR, CoLM, CHIL, ML4H, MLHC

Sessions

  • Korean Digital Convergent Neurosurgery Research Society (KDCNRS) Symposium
  • Seoul AI Hub AI+Healthcare Tutorial Session [Link]
  • KOSMI 2024 Winter School - Make your own LLM [Github]
  • KoSAIM 2023 Summer School Hands-on Session: Transformer [Github]
  • KoSAIM RGEMM (생성모델연구회) Seminar [Link]

Honors

  • Presidential Science Scholarship (2018-2022)
    Prestigious scholarship awarded by the Korean government to students in science and engineering fields

Programming Skills

Python, C