Hi, I am Hee-Sung Kim, a Ph.D. student in the Department of Computer Science at Hanyang University, advised by Prof. Sungyoon Lee. I received my B.S. in Computer Science from Hanyang University in February 2025.

My research aims to develop improved machine learning methodologies by deepening the understanding of deep learning itself, which is my current focus. I am particularly interested in the implicit bias of optimization, generalization, and training dynamics of neural networks.

News

• Jul 2026: Presented two papers at ICML 2026: Inconsistency-Aware Minimization and Gradient Descent with Large Step Size Restores Symmetry in Deep Linear Networks with Multi-Pathway.
• May 2026: Two papers accepted to ICML 2026 as first author! 🎉
• Mar 2025: Started my M.S./Ph.D. integrated program at Hanyang University, advised by Prof. Sungyoon Lee.
• Feb 2025: Received my B.S. in Computer Science from Hanyang University.

Selected Publications

• Gradient Descent with Large Step Size Restores Symmetry in Deep Linear Networks with Multi-Pathway
  Hee-Sung Kim, Sungyoon Lee.
  International Conference on Machine Learning (ICML), 2026. [paper] [arXiv] [poster] [slides]

• Inconsistency-Aware Minimization: Improving Generalization with Unlabeled Data
  Hee-Sung Kim, Hyeonseong Kim, Sungyoon Lee.
  International Conference on Machine Learning (ICML), 2026. [paper] [arXiv] [poster] [slides]

Research Interests

Currently working on understanding how large learning rates and sharpness-aware methods shape what neural networks learn.

Education

Hanyang University (Mar 2025 - Present)

• M.S./Ph.D. integrated program in Computer Science (Advisor: Sungyoon Lee)

Hanyang University (Mar 2021 - Feb 2025)

• B.S. in Computer Science