Search papers, labs, and topics across Lattice.
This paper explores separable graphical models, which represent independence structures through vertex separation in mixed graphs that include directed, undirected, and bidirected edges. The authors introduce the concepts of separable and essentially separable graphs, providing multiple characterizations and demonstrating that these models encompass various existing graph families. A key result is the development of an algorithm that identifies the equivalence class of any essentially separable graph, enhancing our understanding of their structural properties and relationships.
Separable graphs reveal a hidden structure in graphical models that could unify diverse independence frameworks and streamline model identification.
We study a broad class of graphical models whose independencies correspond to vertex separation in mixed graphs with directed, undirected, and bidirected edges, that are capable of encoding independence structures arising from feedback, latent and selection mechanisms. In particular, we introduce separable graphs, in which each missing edge implies the existence of a separating set for its endpoints, and essentially separable graphs, those graphs separation equivalent to a separable graph. We show that these models include many existing graph families used to define graphical models an provide several characterizations of separable graphs and essentially separable graphs. We also provide multiple characterizations of separation equivalence for separable graphs. One is a graphical characterization in terms of ordinary graph properties, extending earlier results for specific subfamilies Another is a separational characterization depending only on graph separation properties. Finally, we provide a canonical representation for the equivalence classes of essentially separable graphs and develop an algorithm that, under suitable assumptions, identifies the equivalence class of any essentially separable graph.
UW