Seunghoon Paik

Stat PhD @ UC Berkeley

Research Interests

My research interests lie in statistical analysis and methodology on deep learning. I aim (i) to understand statistical properties of modern deep learning models and phenomena, and (ii) to draw statistically rigorous conclusions, covering generalization, forecasting, and uncertainty quantification.

Recently, I have grown interested in the AI safety, evaluations, and AI for sciences, and am exploring the field through a statistical perspective. I am happy to chat about the questions and agenda in this space. For those interested, the Berkeley AI Risk Speaker Series is running on campus this semester (organized by Will Fithian and Wes Hoilday).

Publications and Preprints

Here is my Google Scholar.

  • Seunghoon Paik, Kangjie Zhou, Matus Telgarsky, and Ryan Tibshirani. Basic Inequalities for First-Order Optimization with Applications to Statistical Risk Analysis. 2025.  arXiv

  • Seunghoon Paik, Michael Celentano, Alden Green, and Ryan Tibshirani. Integral Probability Metrics Meets Neural Networks: The Radon-Kolmogorov-Smirnov Test. Journal of Machine Learning Research (JMLR). 2025.  Journal  arXiv

  • Young-Geun Choi, Gi-Soo Kim, Seunghoon Paik, and Myunghee Cho Paik. Semi-Parametric Contextual Bandits with Graph-Laplacian Regularization. Information Sciences, 645. 2023.  Journal  arXiv