Hangyeol (Han) Jung

I’m a first-year M.S. student at the MLILAB (Machine Learning and Intelligence Lab), KAIST (advisor: Prof. Eunho Yang).

My recent research focuses on understanding the mechanisms underlying generalization and misalignment in neural networks. More specifically, I am interested in the following research questions:

• Can we understand how post-training shifts a model’s high-dimensional intrinsic geometry?
• How can we leverage mechanistic interpretability to build trustworthy neural networks that remain reliable in real-world settings?
• (A deeper question on my mind) How can we build AI models that exhibit characteristics of complex biological organisms, such as evolution, self-replication, adaptation, and socialization?