Hyunseok Seung
Postdoctoral Research Associate · University of Wisconsin–Madison
I am a postdoctoral researcher in the Department of Statistics at the University of Wisconsin–Madison, where I work with Professor Matthias Katzfuss. I received my Ph.D. in Statistics from the University of Georgia, advised by Professor Jaewoo Lee and Professor Yuan Ke.
My research is driven by a fundamental scalability challenge in modern machine learning. Gradients and curvature often contain the local information needed to make learning more statistically and computationally efficient, but exploiting this information directly can be prohibitively expensive in high-dimensional settings. I develop methods that retain the useful structure in such information while reducing the memory, computation, and approximation costs that usually limit its use.
This principle connects my work on Gaussian processes, Bayesian optimization, and deep learning optimization. In Gaussian processes, I develop methods that use gradient observations for scalable prediction and data-efficient decision making without forming prohibitively large gradient covariance systems. In deep learning optimization, I develop curvature-aware methods for pretraining and fine-tuning that capture useful second-order structure under the memory and compute constraints of large-scale neural networks.
My current research directions include:
- scalable Gaussian processes with gradient observations
- Bayesian optimization and sample-efficient decision making
- memory-efficient fine-tuning of large language models
- curvature-aware optimization for deep neural networks
news
| Oct 2026 | Preprint “Derivative Gaussian processes on a two-direction budget” is now available on arXiv. |
|---|---|
| Sep 2026 | Conference “Scalable derivative Gaussian processes via exact gradient reduction” was accepted to the Conference on Neural Information Processing Systems (NeurIPS 2026). |
| Jun 2026 | Preprint “Scalable derivative Gaussian processes via exact gradient reduction” is now available on arXiv. |
| May 2026 | Journal “Enhancing COVID-19 mortality prediction with online autocovariance change points detection” was accepted for publication in Stat. |
| Apr 2026 | Journal “Mean activation curvature for scalable second-order optimization in deep networks” was accepted for publication in Knowledge and Information Systems. |
| Nov 2025 | Conference “Low-rank curvature for zeroth-order optimization in LLM fine-tuning” was accepted to the AAAI Conference on Artificial Intelligence (AAAI 2026). |
| Aug 2025 | Conference “MAC: An efficient gradient preconditioning using mean activation approximated curvature” was accepted to the IEEE International Conference on Data Mining (ICDM 2025). |
| Aug 2025 | Position Joined the Katzfuss group as a postdoctoral researcher. |