Research Interests

Our research broadly lies in statistics and data science. We aim to develop principled statistical methods for analyzing high-dimensional and complex data, along with foundational theories to understand and justify these methods. Our interests span both methodological development and theoretical analysis, often guided by practical challenges arising in real-world applications.

Some of our current research directions include:

Statistical perspectives on deep learning
Distributionally robust inference
Bayesian online learning and domain adaptation
Frequentist theory for Bayesian procedures

We strive to maintain a balance between theoretical and applied research. Much of our applied work is driven by collaborations with researchers in both academia and industry. Below are a few notable recent collaborations:

Collaboration with Seungho Song: We are collaborating with Professor Seungho Song, a colorectal surgeon at Kyungpook National University School of Medicine, to model the risk of postoperative complications following colorectal cancer surgery. A key aspect of this project is leveraging datasets from different domains, such as surgical data from Korea and Japan, to enable effective modeling even in regions with limited data. The goal is to build robust and generalizable models that can support clinical decision-making in diverse healthcare settings.

Collaboration with Yogiyo and Dong Gu Choi: South Korea’s food delivery industry has experienced rapid growth over the past decade, presenting a range of interesting data science challenges. We collaborated with Yogiyo, one of the country’s largest food delivery companies, to address some of these problems. Our main focus was on predicting demand for food delivery, with the goal of informing the dynamic pricing of rider commissions. The collaboration itself was also featured in media coverage.

Collaboration with Samsung C&T and Minseok Song: We collaborated with Samsung C&T to develop a fashion recommendation system for SSF Shop. The project involved utilizing various data sources, including customer click history, product images, and a curated set of compatible fashion items collected by domain experts. The collaboration was also featured in media coverage.