Minhak Song

Undergraduate Student

KAIST

Hi, I am an undergraduate student at KAIST, double majoring in Industrial & Systems Engineering and Mathematical Sciences. My research interest lies in building the theoretical foundations of modern machine learning through the lens of optimization theory and statistics. Recently, I am focusing on understanding the training dynamics of gradient-based optimization methods in deep learning, closely working with Prof. Chulhee Yun.

I am on leave of absence from my studies at KAIST to fulfill my mandatory military service in South Korea, from February 2023 to November 2024.

If you are interested in collaboration or wish to engage in discussions with me, feel free to contact me at: minhaksong {at} kaist {dot} ac {dot} kr.

Interests

Machine Learning Theory
Optimization
Statistics

Education

BSc in Industrial & Systems Engineering and Mathematical Sciences, 2020 - Present
KAIST

Publications

Kwangjun Ahn, Xiang Cheng, Minhak Song, Chulhee Yun, Ali Jadbabaie, Suvrit Sra (2023). Linear attention is (maybe) all you need (to understand Transformer optimization). ICLR 2024, short version at NeurIPS 2023 Workshop on Mathematics of Modern Machine Learning (Oral).

arXiv

Minhak Song, Chulhee Yun (2023). Trajectory Alignment: Understanding the Edge of Stability Phenomenon via Bifurcation Theory. NeurIPS 2023.

arXiv