Joon Hyeok Kim

I am a graduate research assistant at GMLR Lab, led by Prof. Jiatao Gu at the University of Pennsylvania.

My research is driven by a fundamental question: How do complex neural systems learn, generalize, and organize their computation? I am particularly interested in understanding deep generative models, with my current work focusing on the mechanistic interpretability of diffusion models and how they learn algorithmic structure, including phenomena such as grokking.

Prior to joining Penn, I spent three years as a software engineer at LG CNS, where I worked on LG Electronics’ global ERP system. This experience sharpened my interest in the contrast between explicitly engineered systems and learned models whose internal computation is far less transparent.

In the long term, I hope to contribute to generative models that are not only increasingly capable, but also sufficiently understandable and reliable to support complex, high-stakes decision making. I am particularly interested in how deeper scientific understanding of learned systems can eventually enable their responsible use in domains where transparency and accountability matter.