On the identification of discrete graphical models with hidden nodes
Barbara Vantaggi, Elena Stanghellini · arXiv (Cornell University) · 2010
Conditions are presented for local identifiability of discrete undirected graphical models with a binary hidden node. These models can be obtained by extending the latent class model to allow for conditional associations between the observed variables. We establish a necessary and sufficient condition for the model to be locally identified almost everywhere in the parameter space and we provide expressions of the subspace where identifiability breaks down. The condition is based on the topology of the undirected graph and relies on the faithfulness assumption.