Directional Evidence Guided Search-Space Reduction for Exact DAG Learning
DECO develops a hybrid method to learn Directed Acyclic Graphs (DAGs) by reducing the search space for exact learning. It extracts directional evidence from observational data using metrics like Mutual Information and Maximized Conditional Divergence to prune impossible connections before running an optimization framework. This process significantly cuts down the number of possible graph structures to check, leading to faster and more accurate DAG recovery.