Presentation 2-4, 4:30PM~4:50PM KST.
Speaker
Youngmin Ahn (Seoul National University)
Title
A Robust Approach to Recover Gaussian DAG Models in Quantile-Based Data Discretization.
Abstract
Directed acyclic graphical (DAG) models are used to represent the relationships between variables in datasets across various fields. However, due to data security concerns, these datasets are often provided with transformed variables, which can make it difficult to accurately understand the relationships between the variables. To address this issue, this study introduces quantile-based discretization, a type of data transformation, within a Gaussian setting. Moreover, a method called Disc-PC is proposed for determining the Markov equivalent class (MEC) of the original data using the transformed datasets. Through numerical experiments, the proposed algorithm is demonstrated to significantly outperform other algorithms in this context.