Jiwan Kim

I am Jiwan Kim, a Ph.D. student in the Department of Industrial & Systems Engineering at KAIST, where I am advised by Prof. Chanyoung Park at the Data Science and Artificial Intelligence Lab (DSAIL). I received my M.S. degree from KAIST under the same advisor, and my B.S. degree in Statistics (major) and Computer Science (double major) from Seoul National University.

🔬 Core Research Focus

Efficient Multimodal Large Language Models (MLLMs)

My primary research focuses on building high-performance yet efficient multimodal LLMs (image and video) at both training and inference time, and leveraging them to solve real-world deployment challenges.

Keywords: Multimodal LLM, Knowledge Distillation, Visual Token Pruning, On-Device AI

Key Focus:

• On-Device MLLM: efficiency for both training and inference
  • Knowledge Distillation: effective training to strengthen small MLLMs
  • Visual Token Pruning & Compression: taming large visual-token cost at inference
• MLLM Applications
  • Agentic & Tool-Integrated MLLM: agent / tool-use ability for small MLLMs
  • Autonomous Driving, Vision-Language-Action (VLA)

Previous Research: Recommendation Systems

Multimodal and LLM-based recommender systems — generating missing modalities, building token-efficient item representations, and understanding user behavior with LLMs.