Heesu Kim

I am an undergraduate at Sungkyunkwan University (B.Eng. in Applied Artificial Intelligence, expected August 2027), currently on a year-long exchange at Hitotsubashi University in Tokyo, where I work on NLP research in Prof. Mamoru Komachi’s lab.

I am interested in how linguistic, cultural, and multimodal representations emerge in modern AI systems. My current research spans LLM evaluation, semantic change, multilingual NLP, and vision-language models. My long-term goal is to better understand what neural representations capture and how these representations can be evaluated and improved.

News

• May 2026 — First-authored paper on multilingual cultural evaluation of LLMs submitted to ACL Rolling Review (May 2026 cycle).
• Oct 2025 — Hi-DARTS, a hierarchical reinforcement-learning trading system I co-authored, published at ICTC 2025.
• Sep 2025 — Began exchange year at Hitotsubashi University; joined Prof. Mamoru Komachi’s lab as an undergraduate research intern.
• Jul 2025 — Research intern at MARS Lab, Samsung Medical Center, working on medical object detection and segmentation.
• Jun 2025 — Won 3rd place at Bias-a-Thon 2025 (Track 1), hosted by SKKU’s Center for Multimedia Intelligence, for identifying and mitigating demographic biases in widely used LLMs.

Current Research

Cultural Evaluation of LLMs (first author; under review at ACL Rolling Review, May 2026 cycle)
We extend cultural-bias benchmarks for LLMs — previously limited to Japanese — to a multilingual setting. I led the construction of the Korean subset and evaluated model behavior under zero-shot prompting, chain-of-thought prompting, and supervised fine-tuning.

Semantic Shift and Representation Change
Working on semantic shift detection projects in Komachi Lab: measuring how lexical meaning evolves across domains and languages through the geometry of contextual embeddings.

LLM Pre-training and Machine Translation
Investigating how pre-training objectives and data composition influence multilingual transfer and translation performance.

Linguistic Structure in Vision-Language Models
Analyzing image embeddings from vision-language models to understand what linguistic information is encoded in multimodal representations.

PhD Applications

I am applying in the Fall 2026 admissions cycle for PhD programs beginning in Fall 2027 (B.Eng. expected August 2027). I am particularly interested in NLP evaluation and benchmarking, semantic change and language dynamics, multilingual NLP, and multimodal foundation models.

If your lab works on these problems, I would love to connect! Please reach out at [email protected].

Background

I grew up in Seoul, South Korea, with time spent in Maryland during elementary school. I am pursuing my undergraduate degree at Sungkyunkwan University (SKKU), took a leave for military service (2023–2024) during which I served as a translator/interpreter supporting US–ROK joint operations, and am currently spending a year as an exchange student at Hitotsubashi University in Japan.

On the research side, I interned at MARS Lab at Samsung Medical Center, where I fine-tuned models for detection and segmentation of post-surgery skin flaps. I then joined Prof. Mamoru Komachi’s lab at Hitotsubashi University, where I work on cultural evaluation of LLMs and semantic shift detection. I also co-authored Hi-DARTS, a hierarchical reinforcement-learning trading system published at ICTC 2025.

Hobbies

Outside of research I practice kendo and Asian calligraphy, play FPS games, tinker with keyboards and headphones (chasing the HD800s), and am working toward fluency in Japanese.