Bayesian structure learning in closed skew-normal graphical models

Rezaeitabar, Vahid

doi:10.1080/00949655.2026.2673048

Bayesian structure learning in closed skew-normal graphical models

2026-05-01

0

PRE

AI

R

Rezaeitabar, Vahid *

DOI：10.1080/00949655.2026.2673048

Original

Original request for help

Save

Abstract

En 中文

Classical methods for learning Directed Acyclic Graphs (DAGs) typically rely on Gaussian assumptions, under which conditional independencies correspond to sparsity in the precision matrix. However, many real-world datasets exhibit skewness, violating normality and potentially degrading structural learning. We propose a Bayesian framework for DAG structure learning under multivariate closed skew-normal (CSN) distributions. The approach integrates hierarchical mixture priors for structure selection with a Gibbs sampling scheme for posterior inference, enabling simultaneous estimation of model parameters and edge inclusion probabilities. Simulation studies demonstrate that the proposed method achieves improved structural recovery compared to Gaussian-based approaches, particularly in the presence of skewness.

Keywords:

Skew-normal DAG

hierarchical mixture prior

Bayesian graph selection

Bayesian structure learning in closed skew-normal graphical models

Abstract

Journal

Organization

Cited Papers