Joon Hyeok Kim
Always enthusiastically, considerately, and jocundly!
· ICLR 2026 DeLTa Workshop [Open Review]
· Grokking of Diffusion Models: Case Study on Modular Addition (2026): Co-led research on Rectified Flow models (DiT backbone) by implementing training framework and conducting mechanistic interpretability (MI) analysis during inference to investigate grokking and algorithmic phase transitions.
· Literature Review and Theoretical Framework Design: Conducting extensive literature review and theoretical analysis on Discrete Diffusion (CTMC) and Flow Matching (e.g., DFM, Editflow) to propose efficient multi-token prediction frameworks within AutoRegressive structures.
· Managed system integration and ERP development for the LG Electronics Global ERP (GERP) system. · Developed PL/SQL procedures and Java applications; designed RDBMS structures and optimized complex queries for enterprise-level performance.
· Proudly discharged as a sergeant
· Specialty in information tactics, signal corps, and mid-sized military vehicle driving
· Squad Leader of the Information and Tactics Squad, 53rd Logistics Support Group
· CGPA: 4.0/4.0
· CGPA: 4.04/4.3
· Ranked 2nd out of 102
· Summa Cum Laude
· Financial Economics Track
· ML Packages: PyTorch, Slurm, Wandb, GPyTorch, BoTorch, PyMoo
· Programming Languages: Python, PL/SQL, C/C++, Java, JavaScript, R
· Database Management: Oracle DBMS, MariaDB, MongoDB
· Tools: Linux, AWS Cloud, Microsoft Azure, Docker, Django, Stata, Oracle ERP
I’m an outdoor enthusiast at heart. Whether it’s bikepacking through forest trails or going on long-distance treks with my gravel bike, I love the freedom of the open road. Lately, I’ve also been heavily focused on boxing. I’m training hard to step into the ring for my first official match as soon as I’m ready!
Global Korea Scholarship, Ministry of Education, Republic of Korea, Jul. 2023