---

Kayvan Sadeghi -  University  College of London - 12 ottobre, ore 17:00, aula B del Palazzo delle Scienze

Abstract: Using a characterization of faithfulness, we describe the foundational graph orientation rule in constraint-based causal structure learning and the assumptions under which it recovers the "correct" graph. We present the theory for the class of directed acyclic graphs (DAGs) and then generalize it to the broader class of mixed graphs, which can encode independence structures arising from feedback, latent variables, and selection mechanisms. We also introduce a general and computationally efficient structure learning algorithm for mixed graphs.