Hi, I am Yoonsoo Nam, a Postdoctoral Researcher at KAIST and a member of the Center for Large Language Model Innovation (InnoCORE-LLM). My research focuses on uncovering the mechanisms of neural network dynamics and inductive bias in the simplest, most intuitive ways possible (yes, I was a physicist). After all, shouldn’t we understand how these models learn before we trust them blindly?

Download my CV

Research Interests

Dynamics of neural networks, linear networks, grokking, neural collapse, emergence, feature learning, inductive bias of neural networks

Education

University of Oxford (2020-2025)

• DPhil in Theoretical Physics (Advisor: Ard A. Louis)

University of Cambridge (2013-2014)

• MASt in Theoretical Physics

University of Oxford (2008-2013)

• BA in Physics

Work Experience

KAIST (2025-Present)

• postdoc researcher on ML dynamics and inductive bias (Advisor: Chulhee Yun)

NAVER Corporation (2017-2020)

• ML Researcher/Engineer (Restaurant Recommendation / NAS)

TMAX Data (2015-2017)

• Backend Software Engineer (RDBMS Engine)