Sehyun Kwon

I am a Staff Engineer at Samsung Research, where I develop on-device AI models.

I did my Ph.D. in Artificial Intelligence at Seoul National University, where I was advised by Ernest K. Ryu and M.S. in the Mathematical Sciences at Seoul National University and B.S. in Mathematics Education at Dankook University.

Also, I worked as a research scientist intern at NAVER AI Lab and KRAFTON AI.

Here is my CV.

Email  /  Google Scholar  /  X  /  Github  /  LinkedIn

I am interested in Language Models, Multi-modal Learning, Information Retrieval, and Clustering. My research aims for a fundamental understanding of these models to tackle inherent ambiguity. Some papers are highlighted.

Providing theoretical and empirical evidence of the inherent limitations of dense retrieval for handling compositional intents.

Revealing that training on multiple diverse In-context Learning tasks simultaneously shortens the loss plateaus, making each task easier to learn.

Prosposing the first framework to resolve the fundamental ambiguity of clustering by explicitly conditioning the process on user-provided text criteria.

We develop invariant representation learning methods using implicit neural representations for geometric transformations.

We prove that Wasserstein GANs with infinitely wide generators have no spurious stationary points in the optimization landscape.

We study the memorization and generalization properties of diffusion probabilistic models.

Youlchon AI Star Fellowship, 2024

Outstanding TA Award for the Mathematical Foundations of Deep Neural Networks course, Seoul National University, 2022

Website template from Jon Barron