📑 About Me

🧠 Interests

Artificial General Intelligence, Multimodal Large Language Models, Vision-Language-Speech Generation, Object-Centric Representation Learning, Text-to-Image Generation, Multimodal Alignment

🎓 Education

MS in Artificial Intelligence, Korea University
2025.03 – Present
GPA: 4.44 / 4.5

BS in Computer Science and Engineering, Korea University
2019.03 – 2025.02
GPA: 4.11 / 4.5
Leave of absence for mandatory military service, Republic of Korea Air Force (2021.01 – 2023.01)

📄 Publications

Slot-MLLM: Object-Centric Visual Tokenization for Multimodal LLM
Donghwan Chi*, Hyomin Kim*, Yoonjin Oh, Yongjin Kim, Donghoon Lee, Daejin Jo, Jongmin Kim, Junyeob Baek, Sungjin Ahn, Sungwoong Kim
Submitted to IEEE Transactions on Multimedia
Introduced Slot Q-Former to enhance multimodal LLMs in object-centric visual understanding, generation, and editing tasks.
Object-centric Self-improving Preference Optimization for Text-to-Image Generation
Yoonjin Oh, Yongjin Kim, Hyomin Kim, Donghwan Chi, Sungwoong Kim
Accepted to CVPR 2026
Proposed the OSPO framework to mitigate object-level hallucinations in text-to-image generation models.

🔬 Research Experience

WAVLab, Carnegie Mellon University
Visiting Collaborator (Advisor: Prof. Shinji Watanabe)
2025.10 – Present

Conducting research on vision-speech-language omni-modal models for unified multimodal understanding and generation
Developing alignment-preserving training methods to extend large vision-language models with discrete speech tokenizers for ITS2ITS omni-modal understanding and generation across image, text, and speech

AGI Lab, Korea University
Graduate Researcher / Undergraduate Intern (Advisor: Prof. Sungwoong Kim and Prof. Sungbin Lim)
2023.09 – Present

Led research on object-centric visual tokenization for multimodal large language models, resulting in Slot-MLLM
Designed a Slot Q-Former-based visual tokenizer to improve object-level grounding, generation, and editing
Conducted large-scale multimodal training on image-text and interleaved datasets using multi-GPU distributed training
Evaluated object-centric reasoning and generation on benchmarks including spatial relation understanding, text-to-image generation, and image editing
Managed lab GPU servers and supported large-scale training infrastructure

👨‍🏫 Activities

Artificial Intelligence Korea University (AIKU)
First Cohort Senior Member, Technical & Academic Management Team (2023.01 – 2024.01)

Managed the club’s servers, ensuring reliable and efficient operation
Served as a committee member for multiple hackathon question designs
Conducted and participated in a project related to diffusion models
Organized and taught introductory courses for new members

🏅 Honors & Awards

Chief of Naval Operations Award, Republic of Korea Navy, Open Source Academy Military (OSAM) Hackathon, 2021
Excellence Award, Capstone Design Competition, Korea University, Fall 2024
Academic Excellence Scholarship, Korea University, Spring 2024
Special Scholarship by Student Affairs Office, Korea University, Fall 2023 & Spring 2024
Semester High Honors, Korea University, 2019, 2020, 2023, 2024

💻 Skills

Languages: Python, Bash, LaTeX
ML Frameworks: PyTorch, PyTorch Lightning, Hugging Face Transformers
ML Systems: Distributed Training, Multi-GPU Training, Mixed Precision Training, DeepSpeed, Weights & Biases
Research Areas: Multimodal LLMs, Vision-Language Models, Diffusion Models, Speech Tokenization
Language Proficiency: TOEIC 905