Gunwoong Park will be giving a talk on "Identifiability of Directed Acyclic Graphical Models" on Wednesday, October 1 in HSSB 1173.
Event Date Wednesday, October 1, 2025 - 3:30pm to 4:30pm Venue HSSB 1173 Speaker Gunwoong Park - Associate Professor in the Department of Statistics at Seoul National University and the Director of the Institute for Data Innovation in Science, Data Discovery Center.
Title:
Identifiability of Directed Acyclic Graphical Models
Abstract:
A fundamental task in various fields of science is to uncover the underlying structural relations among variables. Directed acyclic graphical (DAG) models provide a natural representation for such relations. While experimental interventions can reveal the structure, they are often infeasible in practice. Therefore, it is important to study conditions and methods under which the structure of DAG models can be recovered from purely observational data. In this talk, I introduce DAG models and review statistical approaches for structure learning, together with conditions that guarantee identifiability without requiring interventions. I introduce the Gaussian and Poisson DAG models for the causal discovery. In addition, I provide how these models can be identifiable without any experiments or interventions, and how the model can be learned in finite sample settings.
Short Bio:
Gunwoong Park is an Associate Professor in the Department of Statistics at Seoul National University and the Director of the Institute for Data Innovation in Science, Data Discovery Center. He received his Ph.D. in Statistics from the University of Wisconsin–Madison under the supervision of Garvesh Raskutti and was a postdoctoral fellow at the University of Michigan with Liza Levina and Ji Zhu. His research interests lie in statistical machine learning and artificial intelligence, with a particular focus on graphical models, high-dimensional inference, and robust learning methods. Recently, his work has centered on structure learning in graphical models under challenging settings, including the presence of outliers and measurement error. He also studies methods for enhancing the interpretability and reliability of AI models.