Towards Safe and Reliable Foundation Models

Hello, I’m Jeonghyeon Kim, a Ph.D. student in Data Science at Seoul National University of Science and Technology (SeoulTech), where I am privileged to be advised by Prof. Sangheum Hwang of DAINTLAB.

My research is driven by the ambition to advance the frontier of trustworthy AI, with a focus on making machine learning systems more robust, reliable, and interpretable in real-world scenarios. I specialize in:

• Out-of-distribution detection: Developing novel algorithms to identify when AI models encounter unfamiliar data, a critical capability for safe deployment in open-world applications.
• Cross-modal representation learning: Innovating techniques to align and leverage multimodal data (e.g., vision and language) for improved generalization and semantic understanding.
• Machine unlearning: Exploring methods to enable AI systems to selectively forget sensitive or outdated information, ensuring privacy and adaptability.

My ultimate goal is to pioneer AI solutions that are not only state-of-the-art in performance but also transparent, secure, and trustworthy—empowering their adoption in mission-critical domains such as healthcare, autonomous systems, and beyond.