SeongHyeon Kim

Master’s student at KAIST AI · Intern at Kakao

About

I am a master’s student at KAIST AI, advised by Prof. Juho Lee, and currently an intern at Kakao. My research focuses on model consolidation for large language models.

I explore model merging and knowledge distillation to combine knowledge and complementary capabilities learned by different models. This includes approaches such as multi-teacher on-policy distillation (MOPD), with the broader goal of building unified models that perform well across diverse tasks and objectives.

shkim0824@kaist.ac.krGitHub

Publications

(*) denotes equal contribution

No Pain, More Gain: Iterative Merging for Effective Multi-Teacher On-Policy Distillation

SeongHyeon Kim*, Chaeyun Jang*, Noah Lee, Boseop Kim, Juho Lee

arXiv preprint

paper

Mixture-Trained Merging for Unified Multi-Objective Models

SeongHyeon Kim*, Chaeyun Jang*, Seungyoo Lee, Jiyeon Ham, Yunju Bak, Boseop Kim, Juho Lee

NeurIPS 2026

paper

Learning to Explore for Stochastic Gradient MCMC

SeungHyun Kim*, Seohyeon Jung*, SeongHyeon Kim, Juho Lee

ICML 2024

paper

Education

KAIST AI

Master’s program

In progress