Sangin Lee

My research focuses on building effective and efficient multimodal AI systems for real-world environments. My goal is to develop AI/ML that can reliably assist humans in perceiving, understanding, and interacting with the real world, with long-term applications in autonomous driving and intelligent infotainment systems. I am deeply interested in exploring three key areas for a comprehensive understanding of multimodal:

1. Effective Perception: Developing reliable and effective perception (e.g., object detection) systems through sensor fusion and domain adaptation/generalization approaches.

2. Efficient VLMs: Reducing the computational cost of large vision-language models through token pruning and efficient representation learning while preserving visual grounding capabilities.

3. Scale-efficient Training: Exploring scalable and data-efficient training strategies for AI/ML systems to enable practical deployment in real-world applications with limited computational resources.