DAG Structure Learning with Node-Varying Heavy-Tailed Disturbances
The research proposes a node-varying Student-t DAGMA to learn directed acyclic graphs from linear models where disturbances have different scales and tail behaviors across nodes. The method jointly estimates the graph structure, node-specific disturbance scales, and degrees of freedom. Experiments show improved structural recovery under heterogeneous heavy-tailed noise on synthetic and real financial data.